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/ram_addr.h"
35 #include "exec/address-spaces.h"
36 #include "qemu/event_notifier.h"
39 /* This check must be after config-host.h is included */
41 #include <sys/eventfd.h>
44 #ifdef CONFIG_VALGRIND_H
45 #include <valgrind/memcheck.h>
48 /* KVM uses PAGE_SIZE in its definition of COALESCED_MMIO_MAX */
49 #define PAGE_SIZE TARGET_PAGE_SIZE
54 #define DPRINTF(fmt, ...) \
55 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
57 #define DPRINTF(fmt, ...) \
61 #define KVM_MSI_HASHTAB_SIZE 256
63 typedef struct KVMSlot
66 ram_addr_t memory_size;
72 typedef struct kvm_dirty_log KVMDirtyLog;
81 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
82 bool coalesced_flush_in_progress;
83 int broken_set_mem_region;
86 int robust_singlestep;
88 #ifdef KVM_CAP_SET_GUEST_DEBUG
89 struct kvm_sw_breakpoint_head kvm_sw_breakpoints;
95 /* The man page (and posix) say ioctl numbers are signed int, but
96 * they're not. Linux, glibc and *BSD all treat ioctl numbers as
97 * unsigned, and treating them as signed here can break things */
98 unsigned irq_set_ioctl;
99 #ifdef KVM_CAP_IRQ_ROUTING
100 struct kvm_irq_routing *irq_routes;
101 int nr_allocated_irq_routes;
102 uint32_t *used_gsi_bitmap;
103 unsigned int gsi_count;
104 QTAILQ_HEAD(msi_hashtab, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE];
110 bool kvm_kernel_irqchip;
111 bool kvm_async_interrupts_allowed;
112 bool kvm_halt_in_kernel_allowed;
113 bool kvm_irqfds_allowed;
114 bool kvm_msi_via_irqfd_allowed;
115 bool kvm_gsi_routing_allowed;
116 bool kvm_gsi_direct_mapping;
118 bool kvm_readonly_mem_allowed;
120 static const KVMCapabilityInfo kvm_required_capabilites[] = {
121 KVM_CAP_INFO(USER_MEMORY),
122 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS),
126 static KVMSlot *kvm_alloc_slot(KVMState *s)
130 for (i = 0; i < s->nr_slots; i++) {
131 if (s->slots[i].memory_size == 0) {
136 fprintf(stderr, "%s: no free slot available\n", __func__);
140 static KVMSlot *kvm_lookup_matching_slot(KVMState *s,
146 for (i = 0; i < s->nr_slots; i++) {
147 KVMSlot *mem = &s->slots[i];
149 if (start_addr == mem->start_addr &&
150 end_addr == mem->start_addr + mem->memory_size) {
159 * Find overlapping slot with lowest start address
161 static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s,
165 KVMSlot *found = NULL;
168 for (i = 0; i < s->nr_slots; i++) {
169 KVMSlot *mem = &s->slots[i];
171 if (mem->memory_size == 0 ||
172 (found && found->start_addr < mem->start_addr)) {
176 if (end_addr > mem->start_addr &&
177 start_addr < mem->start_addr + mem->memory_size) {
185 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram,
190 for (i = 0; i < s->nr_slots; i++) {
191 KVMSlot *mem = &s->slots[i];
193 if (ram >= mem->ram && ram < mem->ram + mem->memory_size) {
194 *phys_addr = mem->start_addr + (ram - mem->ram);
202 static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot)
204 struct kvm_userspace_memory_region mem;
206 mem.slot = slot->slot;
207 mem.guest_phys_addr = slot->start_addr;
208 mem.userspace_addr = (unsigned long)slot->ram;
209 mem.flags = slot->flags;
210 if (s->migration_log) {
211 mem.flags |= KVM_MEM_LOG_DIRTY_PAGES;
214 if (slot->memory_size && mem.flags & KVM_MEM_READONLY) {
215 /* Set the slot size to 0 before setting the slot to the desired
216 * value. This is needed based on KVM commit 75d61fbc. */
218 kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
220 mem.memory_size = slot->memory_size;
221 return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
224 static void kvm_reset_vcpu(void *opaque)
226 CPUState *cpu = opaque;
228 kvm_arch_reset_vcpu(cpu);
231 int kvm_init_vcpu(CPUState *cpu)
233 KVMState *s = kvm_state;
237 DPRINTF("kvm_init_vcpu\n");
239 ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)kvm_arch_vcpu_id(cpu));
241 DPRINTF("kvm_create_vcpu failed\n");
247 cpu->kvm_vcpu_dirty = true;
249 mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
252 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
256 cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
258 if (cpu->kvm_run == MAP_FAILED) {
260 DPRINTF("mmap'ing vcpu state failed\n");
264 if (s->coalesced_mmio && !s->coalesced_mmio_ring) {
265 s->coalesced_mmio_ring =
266 (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE;
269 ret = kvm_arch_init_vcpu(cpu);
271 qemu_register_reset(kvm_reset_vcpu, cpu);
272 kvm_arch_reset_vcpu(cpu);
279 * dirty pages logging control
282 static int kvm_mem_flags(KVMState *s, bool log_dirty, bool readonly)
285 flags = log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0;
286 if (readonly && kvm_readonly_mem_allowed) {
287 flags |= KVM_MEM_READONLY;
292 static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty)
294 KVMState *s = kvm_state;
295 int flags, mask = KVM_MEM_LOG_DIRTY_PAGES;
298 old_flags = mem->flags;
300 flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty, false);
303 /* If nothing changed effectively, no need to issue ioctl */
304 if (s->migration_log) {
305 flags |= KVM_MEM_LOG_DIRTY_PAGES;
308 if (flags == old_flags) {
312 return kvm_set_user_memory_region(s, mem);
315 static int kvm_dirty_pages_log_change(hwaddr phys_addr,
316 ram_addr_t size, bool log_dirty)
318 KVMState *s = kvm_state;
319 KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size);
322 fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-"
323 TARGET_FMT_plx "\n", __func__, phys_addr,
324 (hwaddr)(phys_addr + size - 1));
327 return kvm_slot_dirty_pages_log_change(mem, log_dirty);
330 static void kvm_log_start(MemoryListener *listener,
331 MemoryRegionSection *section)
335 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
336 int128_get64(section->size), true);
342 static void kvm_log_stop(MemoryListener *listener,
343 MemoryRegionSection *section)
347 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
348 int128_get64(section->size), false);
354 static int kvm_set_migration_log(int enable)
356 KVMState *s = kvm_state;
360 s->migration_log = enable;
362 for (i = 0; i < s->nr_slots; i++) {
365 if (!mem->memory_size) {
368 if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) {
371 err = kvm_set_user_memory_region(s, mem);
379 /* get kvm's dirty pages bitmap and update qemu's */
380 static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section,
381 unsigned long *bitmap)
383 ram_addr_t start = section->offset_within_region + section->mr->ram_addr;
384 ram_addr_t pages = int128_get64(section->size) / getpagesize();
386 cpu_physical_memory_set_dirty_lebitmap(bitmap, start, pages);
390 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
393 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
394 * This function updates qemu's dirty bitmap using
395 * memory_region_set_dirty(). This means all bits are set
398 * @start_add: start of logged region.
399 * @end_addr: end of logged region.
401 static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection *section)
403 KVMState *s = kvm_state;
404 unsigned long size, allocated_size = 0;
408 hwaddr start_addr = section->offset_within_address_space;
409 hwaddr end_addr = start_addr + int128_get64(section->size);
411 d.dirty_bitmap = NULL;
412 while (start_addr < end_addr) {
413 mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr);
418 /* XXX bad kernel interface alert
419 * For dirty bitmap, kernel allocates array of size aligned to
420 * bits-per-long. But for case when the kernel is 64bits and
421 * the userspace is 32bits, userspace can't align to the same
422 * bits-per-long, since sizeof(long) is different between kernel
423 * and user space. This way, userspace will provide buffer which
424 * may be 4 bytes less than the kernel will use, resulting in
425 * userspace memory corruption (which is not detectable by valgrind
426 * too, in most cases).
427 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
428 * a hope that sizeof(long) wont become >8 any time soon.
430 size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS),
431 /*HOST_LONG_BITS*/ 64) / 8;
432 if (!d.dirty_bitmap) {
433 d.dirty_bitmap = g_malloc(size);
434 } else if (size > allocated_size) {
435 d.dirty_bitmap = g_realloc(d.dirty_bitmap, size);
437 allocated_size = size;
438 memset(d.dirty_bitmap, 0, allocated_size);
442 if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
443 DPRINTF("ioctl failed %d\n", errno);
448 kvm_get_dirty_pages_log_range(section, d.dirty_bitmap);
449 start_addr = mem->start_addr + mem->memory_size;
451 g_free(d.dirty_bitmap);
456 static void kvm_coalesce_mmio_region(MemoryListener *listener,
457 MemoryRegionSection *secion,
458 hwaddr start, hwaddr size)
460 KVMState *s = kvm_state;
462 if (s->coalesced_mmio) {
463 struct kvm_coalesced_mmio_zone zone;
469 (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
473 static void kvm_uncoalesce_mmio_region(MemoryListener *listener,
474 MemoryRegionSection *secion,
475 hwaddr start, hwaddr size)
477 KVMState *s = kvm_state;
479 if (s->coalesced_mmio) {
480 struct kvm_coalesced_mmio_zone zone;
486 (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
490 int kvm_check_extension(KVMState *s, unsigned int extension)
494 ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
502 static int kvm_set_ioeventfd_mmio(int fd, hwaddr addr, uint32_t val,
503 bool assign, uint32_t size, bool datamatch)
506 struct kvm_ioeventfd iofd;
508 iofd.datamatch = datamatch ? val : 0;
514 if (!kvm_enabled()) {
519 iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
522 iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
525 ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd);
534 static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val,
535 bool assign, uint32_t size, bool datamatch)
537 struct kvm_ioeventfd kick = {
538 .datamatch = datamatch ? val : 0,
540 .flags = KVM_IOEVENTFD_FLAG_PIO,
545 if (!kvm_enabled()) {
549 kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
552 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
554 r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
562 static int kvm_check_many_ioeventfds(void)
564 /* Userspace can use ioeventfd for io notification. This requires a host
565 * that supports eventfd(2) and an I/O thread; since eventfd does not
566 * support SIGIO it cannot interrupt the vcpu.
568 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
569 * can avoid creating too many ioeventfds.
571 #if defined(CONFIG_EVENTFD)
574 for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) {
575 ioeventfds[i] = eventfd(0, EFD_CLOEXEC);
576 if (ioeventfds[i] < 0) {
579 ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true);
581 close(ioeventfds[i]);
586 /* Decide whether many devices are supported or not */
587 ret = i == ARRAY_SIZE(ioeventfds);
590 kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true);
591 close(ioeventfds[i]);
599 static const KVMCapabilityInfo *
600 kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list)
603 if (!kvm_check_extension(s, list->value)) {
611 static void kvm_set_phys_mem(MemoryRegionSection *section, bool add)
613 KVMState *s = kvm_state;
616 MemoryRegion *mr = section->mr;
617 bool log_dirty = memory_region_is_logging(mr);
618 bool writeable = !mr->readonly && !mr->rom_device;
619 bool readonly_flag = mr->readonly || memory_region_is_romd(mr);
620 hwaddr start_addr = section->offset_within_address_space;
621 ram_addr_t size = int128_get64(section->size);
625 /* kvm works in page size chunks, but the function may be called
626 with sub-page size and unaligned start address. */
627 delta = TARGET_PAGE_ALIGN(size) - size;
633 size &= TARGET_PAGE_MASK;
634 if (!size || (start_addr & ~TARGET_PAGE_MASK)) {
638 if (!memory_region_is_ram(mr)) {
639 if (writeable || !kvm_readonly_mem_allowed) {
641 } else if (!mr->romd_mode) {
642 /* If the memory device is not in romd_mode, then we actually want
643 * to remove the kvm memory slot so all accesses will trap. */
648 ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta;
651 mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
656 if (add && start_addr >= mem->start_addr &&
657 (start_addr + size <= mem->start_addr + mem->memory_size) &&
658 (ram - start_addr == mem->ram - mem->start_addr)) {
659 /* The new slot fits into the existing one and comes with
660 * identical parameters - update flags and done. */
661 kvm_slot_dirty_pages_log_change(mem, log_dirty);
667 if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
668 kvm_physical_sync_dirty_bitmap(section);
671 /* unregister the overlapping slot */
672 mem->memory_size = 0;
673 err = kvm_set_user_memory_region(s, mem);
675 fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
676 __func__, strerror(-err));
680 /* Workaround for older KVM versions: we can't join slots, even not by
681 * unregistering the previous ones and then registering the larger
682 * slot. We have to maintain the existing fragmentation. Sigh.
684 * This workaround assumes that the new slot starts at the same
685 * address as the first existing one. If not or if some overlapping
686 * slot comes around later, we will fail (not seen in practice so far)
687 * - and actually require a recent KVM version. */
688 if (s->broken_set_mem_region &&
689 old.start_addr == start_addr && old.memory_size < size && add) {
690 mem = kvm_alloc_slot(s);
691 mem->memory_size = old.memory_size;
692 mem->start_addr = old.start_addr;
694 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
696 err = kvm_set_user_memory_region(s, mem);
698 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
703 start_addr += old.memory_size;
704 ram += old.memory_size;
705 size -= old.memory_size;
709 /* register prefix slot */
710 if (old.start_addr < start_addr) {
711 mem = kvm_alloc_slot(s);
712 mem->memory_size = start_addr - old.start_addr;
713 mem->start_addr = old.start_addr;
715 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
717 err = kvm_set_user_memory_region(s, mem);
719 fprintf(stderr, "%s: error registering prefix slot: %s\n",
720 __func__, strerror(-err));
722 fprintf(stderr, "%s: This is probably because your kernel's " \
723 "PAGE_SIZE is too big. Please try to use 4k " \
724 "PAGE_SIZE!\n", __func__);
730 /* register suffix slot */
731 if (old.start_addr + old.memory_size > start_addr + size) {
732 ram_addr_t size_delta;
734 mem = kvm_alloc_slot(s);
735 mem->start_addr = start_addr + size;
736 size_delta = mem->start_addr - old.start_addr;
737 mem->memory_size = old.memory_size - size_delta;
738 mem->ram = old.ram + size_delta;
739 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
741 err = kvm_set_user_memory_region(s, mem);
743 fprintf(stderr, "%s: error registering suffix slot: %s\n",
744 __func__, strerror(-err));
750 /* in case the KVM bug workaround already "consumed" the new slot */
757 mem = kvm_alloc_slot(s);
758 mem->memory_size = size;
759 mem->start_addr = start_addr;
761 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
763 err = kvm_set_user_memory_region(s, mem);
765 fprintf(stderr, "%s: error registering slot: %s\n", __func__,
771 static void kvm_region_add(MemoryListener *listener,
772 MemoryRegionSection *section)
774 memory_region_ref(section->mr);
775 kvm_set_phys_mem(section, true);
778 static void kvm_region_del(MemoryListener *listener,
779 MemoryRegionSection *section)
781 kvm_set_phys_mem(section, false);
782 memory_region_unref(section->mr);
785 static void kvm_log_sync(MemoryListener *listener,
786 MemoryRegionSection *section)
790 r = kvm_physical_sync_dirty_bitmap(section);
796 static void kvm_log_global_start(struct MemoryListener *listener)
800 r = kvm_set_migration_log(1);
804 static void kvm_log_global_stop(struct MemoryListener *listener)
808 r = kvm_set_migration_log(0);
812 static void kvm_mem_ioeventfd_add(MemoryListener *listener,
813 MemoryRegionSection *section,
814 bool match_data, uint64_t data,
817 int fd = event_notifier_get_fd(e);
820 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
821 data, true, int128_get64(section->size),
824 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
825 __func__, strerror(-r));
830 static void kvm_mem_ioeventfd_del(MemoryListener *listener,
831 MemoryRegionSection *section,
832 bool match_data, uint64_t data,
835 int fd = event_notifier_get_fd(e);
838 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
839 data, false, int128_get64(section->size),
846 static void kvm_io_ioeventfd_add(MemoryListener *listener,
847 MemoryRegionSection *section,
848 bool match_data, uint64_t data,
851 int fd = event_notifier_get_fd(e);
854 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
855 data, true, int128_get64(section->size),
858 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
859 __func__, strerror(-r));
864 static void kvm_io_ioeventfd_del(MemoryListener *listener,
865 MemoryRegionSection *section,
866 bool match_data, uint64_t data,
870 int fd = event_notifier_get_fd(e);
873 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
874 data, false, int128_get64(section->size),
881 static MemoryListener kvm_memory_listener = {
882 .region_add = kvm_region_add,
883 .region_del = kvm_region_del,
884 .log_start = kvm_log_start,
885 .log_stop = kvm_log_stop,
886 .log_sync = kvm_log_sync,
887 .log_global_start = kvm_log_global_start,
888 .log_global_stop = kvm_log_global_stop,
889 .eventfd_add = kvm_mem_ioeventfd_add,
890 .eventfd_del = kvm_mem_ioeventfd_del,
891 .coalesced_mmio_add = kvm_coalesce_mmio_region,
892 .coalesced_mmio_del = kvm_uncoalesce_mmio_region,
896 static MemoryListener kvm_io_listener = {
897 .eventfd_add = kvm_io_ioeventfd_add,
898 .eventfd_del = kvm_io_ioeventfd_del,
902 static void kvm_handle_interrupt(CPUState *cpu, int mask)
904 cpu->interrupt_request |= mask;
906 if (!qemu_cpu_is_self(cpu)) {
911 int kvm_set_irq(KVMState *s, int irq, int level)
913 struct kvm_irq_level event;
916 assert(kvm_async_interrupts_enabled());
920 ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event);
922 perror("kvm_set_irq");
926 return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status;
929 #ifdef KVM_CAP_IRQ_ROUTING
930 typedef struct KVMMSIRoute {
931 struct kvm_irq_routing_entry kroute;
932 QTAILQ_ENTRY(KVMMSIRoute) entry;
935 static void set_gsi(KVMState *s, unsigned int gsi)
937 s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32);
940 static void clear_gsi(KVMState *s, unsigned int gsi)
942 s->used_gsi_bitmap[gsi / 32] &= ~(1U << (gsi % 32));
945 void kvm_init_irq_routing(KVMState *s)
949 gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING);
951 unsigned int gsi_bits, i;
953 /* Round up so we can search ints using ffs */
954 gsi_bits = ALIGN(gsi_count, 32);
955 s->used_gsi_bitmap = g_malloc0(gsi_bits / 8);
956 s->gsi_count = gsi_count;
958 /* Mark any over-allocated bits as already in use */
959 for (i = gsi_count; i < gsi_bits; i++) {
964 s->irq_routes = g_malloc0(sizeof(*s->irq_routes));
965 s->nr_allocated_irq_routes = 0;
967 if (!s->direct_msi) {
968 for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) {
969 QTAILQ_INIT(&s->msi_hashtab[i]);
973 kvm_arch_init_irq_routing(s);
976 void kvm_irqchip_commit_routes(KVMState *s)
980 s->irq_routes->flags = 0;
981 ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes);
985 static void kvm_add_routing_entry(KVMState *s,
986 struct kvm_irq_routing_entry *entry)
988 struct kvm_irq_routing_entry *new;
991 if (s->irq_routes->nr == s->nr_allocated_irq_routes) {
992 n = s->nr_allocated_irq_routes * 2;
996 size = sizeof(struct kvm_irq_routing);
997 size += n * sizeof(*new);
998 s->irq_routes = g_realloc(s->irq_routes, size);
999 s->nr_allocated_irq_routes = n;
1001 n = s->irq_routes->nr++;
1002 new = &s->irq_routes->entries[n];
1006 set_gsi(s, entry->gsi);
1009 static int kvm_update_routing_entry(KVMState *s,
1010 struct kvm_irq_routing_entry *new_entry)
1012 struct kvm_irq_routing_entry *entry;
1015 for (n = 0; n < s->irq_routes->nr; n++) {
1016 entry = &s->irq_routes->entries[n];
1017 if (entry->gsi != new_entry->gsi) {
1021 if(!memcmp(entry, new_entry, sizeof *entry)) {
1025 *entry = *new_entry;
1027 kvm_irqchip_commit_routes(s);
1035 void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin)
1037 struct kvm_irq_routing_entry e = {};
1039 assert(pin < s->gsi_count);
1042 e.type = KVM_IRQ_ROUTING_IRQCHIP;
1044 e.u.irqchip.irqchip = irqchip;
1045 e.u.irqchip.pin = pin;
1046 kvm_add_routing_entry(s, &e);
1049 void kvm_irqchip_release_virq(KVMState *s, int virq)
1051 struct kvm_irq_routing_entry *e;
1054 if (kvm_gsi_direct_mapping()) {
1058 for (i = 0; i < s->irq_routes->nr; i++) {
1059 e = &s->irq_routes->entries[i];
1060 if (e->gsi == virq) {
1061 s->irq_routes->nr--;
1062 *e = s->irq_routes->entries[s->irq_routes->nr];
1068 static unsigned int kvm_hash_msi(uint32_t data)
1070 /* This is optimized for IA32 MSI layout. However, no other arch shall
1071 * repeat the mistake of not providing a direct MSI injection API. */
1075 static void kvm_flush_dynamic_msi_routes(KVMState *s)
1077 KVMMSIRoute *route, *next;
1080 for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) {
1081 QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) {
1082 kvm_irqchip_release_virq(s, route->kroute.gsi);
1083 QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry);
1089 static int kvm_irqchip_get_virq(KVMState *s)
1091 uint32_t *word = s->used_gsi_bitmap;
1092 int max_words = ALIGN(s->gsi_count, 32) / 32;
1097 /* Return the lowest unused GSI in the bitmap */
1098 for (i = 0; i < max_words; i++) {
1099 bit = ffs(~word[i]);
1104 return bit - 1 + i * 32;
1106 if (!s->direct_msi && retry) {
1108 kvm_flush_dynamic_msi_routes(s);
1115 static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg)
1117 unsigned int hash = kvm_hash_msi(msg.data);
1120 QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) {
1121 if (route->kroute.u.msi.address_lo == (uint32_t)msg.address &&
1122 route->kroute.u.msi.address_hi == (msg.address >> 32) &&
1123 route->kroute.u.msi.data == le32_to_cpu(msg.data)) {
1130 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1135 if (s->direct_msi) {
1136 msi.address_lo = (uint32_t)msg.address;
1137 msi.address_hi = msg.address >> 32;
1138 msi.data = le32_to_cpu(msg.data);
1140 memset(msi.pad, 0, sizeof(msi.pad));
1142 return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi);
1145 route = kvm_lookup_msi_route(s, msg);
1149 virq = kvm_irqchip_get_virq(s);
1154 route = g_malloc0(sizeof(KVMMSIRoute));
1155 route->kroute.gsi = virq;
1156 route->kroute.type = KVM_IRQ_ROUTING_MSI;
1157 route->kroute.flags = 0;
1158 route->kroute.u.msi.address_lo = (uint32_t)msg.address;
1159 route->kroute.u.msi.address_hi = msg.address >> 32;
1160 route->kroute.u.msi.data = le32_to_cpu(msg.data);
1162 kvm_add_routing_entry(s, &route->kroute);
1163 kvm_irqchip_commit_routes(s);
1165 QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route,
1169 assert(route->kroute.type == KVM_IRQ_ROUTING_MSI);
1171 return kvm_set_irq(s, route->kroute.gsi, 1);
1174 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1176 struct kvm_irq_routing_entry kroute = {};
1179 if (kvm_gsi_direct_mapping()) {
1180 return msg.data & 0xffff;
1183 if (!kvm_gsi_routing_enabled()) {
1187 virq = kvm_irqchip_get_virq(s);
1193 kroute.type = KVM_IRQ_ROUTING_MSI;
1195 kroute.u.msi.address_lo = (uint32_t)msg.address;
1196 kroute.u.msi.address_hi = msg.address >> 32;
1197 kroute.u.msi.data = le32_to_cpu(msg.data);
1199 kvm_add_routing_entry(s, &kroute);
1200 kvm_irqchip_commit_routes(s);
1205 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1207 struct kvm_irq_routing_entry kroute = {};
1209 if (kvm_gsi_direct_mapping()) {
1213 if (!kvm_irqchip_in_kernel()) {
1218 kroute.type = KVM_IRQ_ROUTING_MSI;
1220 kroute.u.msi.address_lo = (uint32_t)msg.address;
1221 kroute.u.msi.address_hi = msg.address >> 32;
1222 kroute.u.msi.data = le32_to_cpu(msg.data);
1224 return kvm_update_routing_entry(s, &kroute);
1227 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int rfd, int virq,
1230 struct kvm_irqfd irqfd = {
1233 .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN,
1237 irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE;
1238 irqfd.resamplefd = rfd;
1241 if (!kvm_irqfds_enabled()) {
1245 return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd);
1248 #else /* !KVM_CAP_IRQ_ROUTING */
1250 void kvm_init_irq_routing(KVMState *s)
1254 void kvm_irqchip_release_virq(KVMState *s, int virq)
1258 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1263 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1268 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign)
1273 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1277 #endif /* !KVM_CAP_IRQ_ROUTING */
1279 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n,
1280 EventNotifier *rn, int virq)
1282 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n),
1283 rn ? event_notifier_get_fd(rn) : -1, virq, true);
1286 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, int virq)
1288 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), -1, virq,
1292 static int kvm_irqchip_create(KVMState *s)
1296 if (!qemu_opt_get_bool(qemu_get_machine_opts(), "kernel_irqchip", true) ||
1297 !kvm_check_extension(s, KVM_CAP_IRQCHIP)) {
1301 ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP);
1303 fprintf(stderr, "Create kernel irqchip failed\n");
1307 kvm_kernel_irqchip = true;
1308 /* If we have an in-kernel IRQ chip then we must have asynchronous
1309 * interrupt delivery (though the reverse is not necessarily true)
1311 kvm_async_interrupts_allowed = true;
1312 kvm_halt_in_kernel_allowed = true;
1314 kvm_init_irq_routing(s);
1319 /* Find number of supported CPUs using the recommended
1320 * procedure from the kernel API documentation to cope with
1321 * older kernels that may be missing capabilities.
1323 static int kvm_recommended_vcpus(KVMState *s)
1325 int ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS);
1326 return (ret) ? ret : 4;
1329 static int kvm_max_vcpus(KVMState *s)
1331 int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS);
1332 return (ret) ? ret : kvm_recommended_vcpus(s);
1337 static const char upgrade_note[] =
1338 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1339 "(see http://sourceforge.net/projects/kvm).\n";
1344 { "SMP", smp_cpus },
1345 { "hotpluggable", max_cpus },
1348 int soft_vcpus_limit, hard_vcpus_limit;
1350 const KVMCapabilityInfo *missing_cap;
1354 s = g_malloc0(sizeof(KVMState));
1357 * On systems where the kernel can support different base page
1358 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1359 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1360 * page size for the system though.
1362 assert(TARGET_PAGE_SIZE <= getpagesize());
1365 #ifdef KVM_CAP_SET_GUEST_DEBUG
1366 QTAILQ_INIT(&s->kvm_sw_breakpoints);
1369 s->fd = qemu_open("/dev/kvm", O_RDWR);
1371 fprintf(stderr, "Could not access KVM kernel module: %m\n");
1376 ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
1377 if (ret < KVM_API_VERSION) {
1381 fprintf(stderr, "kvm version too old\n");
1385 if (ret > KVM_API_VERSION) {
1387 fprintf(stderr, "kvm version not supported\n");
1391 s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
1393 /* If unspecified, use the default value */
1398 s->slots = g_malloc0(s->nr_slots * sizeof(KVMSlot));
1400 for (i = 0; i < s->nr_slots; i++) {
1401 s->slots[i].slot = i;
1404 /* check the vcpu limits */
1405 soft_vcpus_limit = kvm_recommended_vcpus(s);
1406 hard_vcpus_limit = kvm_max_vcpus(s);
1409 if (nc->num > soft_vcpus_limit) {
1411 "Warning: Number of %s cpus requested (%d) exceeds "
1412 "the recommended cpus supported by KVM (%d)\n",
1413 nc->name, nc->num, soft_vcpus_limit);
1415 if (nc->num > hard_vcpus_limit) {
1417 fprintf(stderr, "Number of %s cpus requested (%d) exceeds "
1418 "the maximum cpus supported by KVM (%d)\n",
1419 nc->name, nc->num, hard_vcpus_limit);
1427 ret = kvm_ioctl(s, KVM_CREATE_VM, 0);
1428 } while (ret == -EINTR);
1431 fprintf(stderr, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -s->vmfd,
1435 fprintf(stderr, "Please add the 'switch_amode' kernel parameter to "
1436 "your host kernel command line\n");
1442 missing_cap = kvm_check_extension_list(s, kvm_required_capabilites);
1445 kvm_check_extension_list(s, kvm_arch_required_capabilities);
1449 fprintf(stderr, "kvm does not support %s\n%s",
1450 missing_cap->name, upgrade_note);
1454 s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
1456 s->broken_set_mem_region = 1;
1457 ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
1459 s->broken_set_mem_region = 0;
1462 #ifdef KVM_CAP_VCPU_EVENTS
1463 s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS);
1466 s->robust_singlestep =
1467 kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP);
1469 #ifdef KVM_CAP_DEBUGREGS
1470 s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS);
1473 #ifdef KVM_CAP_XSAVE
1474 s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE);
1478 s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS);
1481 #ifdef KVM_CAP_PIT_STATE2
1482 s->pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2);
1485 #ifdef KVM_CAP_IRQ_ROUTING
1486 s->direct_msi = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0);
1489 s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3);
1491 s->irq_set_ioctl = KVM_IRQ_LINE;
1492 if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) {
1493 s->irq_set_ioctl = KVM_IRQ_LINE_STATUS;
1496 #ifdef KVM_CAP_READONLY_MEM
1497 kvm_readonly_mem_allowed =
1498 (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0);
1501 ret = kvm_arch_init(s);
1506 ret = kvm_irqchip_create(s);
1512 memory_listener_register(&kvm_memory_listener, &address_space_memory);
1513 memory_listener_register(&kvm_io_listener, &address_space_io);
1515 s->many_ioeventfds = kvm_check_many_ioeventfds();
1517 cpu_interrupt_handler = kvm_handle_interrupt;
1534 static void kvm_handle_io(uint16_t port, void *data, int direction, int size,
1538 uint8_t *ptr = data;
1540 for (i = 0; i < count; i++) {
1541 address_space_rw(&address_space_io, port, ptr, size,
1542 direction == KVM_EXIT_IO_OUT);
1547 static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run)
1549 fprintf(stderr, "KVM internal error.");
1550 if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) {
1553 fprintf(stderr, " Suberror: %d\n", run->internal.suberror);
1554 for (i = 0; i < run->internal.ndata; ++i) {
1555 fprintf(stderr, "extra data[%d]: %"PRIx64"\n",
1556 i, (uint64_t)run->internal.data[i]);
1559 fprintf(stderr, "\n");
1561 if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) {
1562 fprintf(stderr, "emulation failure\n");
1563 if (!kvm_arch_stop_on_emulation_error(cpu)) {
1564 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1565 return EXCP_INTERRUPT;
1568 /* FIXME: Should trigger a qmp message to let management know
1569 * something went wrong.
1574 void kvm_flush_coalesced_mmio_buffer(void)
1576 KVMState *s = kvm_state;
1578 if (s->coalesced_flush_in_progress) {
1582 s->coalesced_flush_in_progress = true;
1584 if (s->coalesced_mmio_ring) {
1585 struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring;
1586 while (ring->first != ring->last) {
1587 struct kvm_coalesced_mmio *ent;
1589 ent = &ring->coalesced_mmio[ring->first];
1591 cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
1593 ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
1597 s->coalesced_flush_in_progress = false;
1600 static void do_kvm_cpu_synchronize_state(void *arg)
1602 CPUState *cpu = arg;
1604 if (!cpu->kvm_vcpu_dirty) {
1605 kvm_arch_get_registers(cpu);
1606 cpu->kvm_vcpu_dirty = true;
1610 void kvm_cpu_synchronize_state(CPUState *cpu)
1612 if (!cpu->kvm_vcpu_dirty) {
1613 run_on_cpu(cpu, do_kvm_cpu_synchronize_state, cpu);
1617 void kvm_cpu_synchronize_post_reset(CPUState *cpu)
1619 kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE);
1620 cpu->kvm_vcpu_dirty = false;
1623 void kvm_cpu_synchronize_post_init(CPUState *cpu)
1625 kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE);
1626 cpu->kvm_vcpu_dirty = false;
1629 int kvm_cpu_exec(CPUState *cpu)
1631 struct kvm_run *run = cpu->kvm_run;
1634 DPRINTF("kvm_cpu_exec()\n");
1636 if (kvm_arch_process_async_events(cpu)) {
1637 cpu->exit_request = 0;
1642 if (cpu->kvm_vcpu_dirty) {
1643 kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE);
1644 cpu->kvm_vcpu_dirty = false;
1647 kvm_arch_pre_run(cpu, run);
1648 if (cpu->exit_request) {
1649 DPRINTF("interrupt exit requested\n");
1651 * KVM requires us to reenter the kernel after IO exits to complete
1652 * instruction emulation. This self-signal will ensure that we
1655 qemu_cpu_kick_self();
1657 qemu_mutex_unlock_iothread();
1659 run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0);
1661 qemu_mutex_lock_iothread();
1662 kvm_arch_post_run(cpu, run);
1665 if (run_ret == -EINTR || run_ret == -EAGAIN) {
1666 DPRINTF("io window exit\n");
1667 ret = EXCP_INTERRUPT;
1670 fprintf(stderr, "error: kvm run failed %s\n",
1671 strerror(-run_ret));
1675 trace_kvm_run_exit(cpu->cpu_index, run->exit_reason);
1676 switch (run->exit_reason) {
1678 DPRINTF("handle_io\n");
1679 kvm_handle_io(run->io.port,
1680 (uint8_t *)run + run->io.data_offset,
1687 DPRINTF("handle_mmio\n");
1688 cpu_physical_memory_rw(run->mmio.phys_addr,
1691 run->mmio.is_write);
1694 case KVM_EXIT_IRQ_WINDOW_OPEN:
1695 DPRINTF("irq_window_open\n");
1696 ret = EXCP_INTERRUPT;
1698 case KVM_EXIT_SHUTDOWN:
1699 DPRINTF("shutdown\n");
1700 qemu_system_reset_request();
1701 ret = EXCP_INTERRUPT;
1703 case KVM_EXIT_UNKNOWN:
1704 fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n",
1705 (uint64_t)run->hw.hardware_exit_reason);
1708 case KVM_EXIT_INTERNAL_ERROR:
1709 ret = kvm_handle_internal_error(cpu, run);
1712 DPRINTF("kvm_arch_handle_exit\n");
1713 ret = kvm_arch_handle_exit(cpu, run);
1719 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1720 vm_stop(RUN_STATE_INTERNAL_ERROR);
1723 cpu->exit_request = 0;
1727 int kvm_ioctl(KVMState *s, int type, ...)
1734 arg = va_arg(ap, void *);
1737 trace_kvm_ioctl(type, arg);
1738 ret = ioctl(s->fd, type, arg);
1745 int kvm_vm_ioctl(KVMState *s, int type, ...)
1752 arg = va_arg(ap, void *);
1755 trace_kvm_vm_ioctl(type, arg);
1756 ret = ioctl(s->vmfd, type, arg);
1763 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...)
1770 arg = va_arg(ap, void *);
1773 trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg);
1774 ret = ioctl(cpu->kvm_fd, type, arg);
1781 int kvm_has_sync_mmu(void)
1783 return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU);
1786 int kvm_has_vcpu_events(void)
1788 return kvm_state->vcpu_events;
1791 int kvm_has_robust_singlestep(void)
1793 return kvm_state->robust_singlestep;
1796 int kvm_has_debugregs(void)
1798 return kvm_state->debugregs;
1801 int kvm_has_xsave(void)
1803 return kvm_state->xsave;
1806 int kvm_has_xcrs(void)
1808 return kvm_state->xcrs;
1811 int kvm_has_pit_state2(void)
1813 return kvm_state->pit_state2;
1816 int kvm_has_many_ioeventfds(void)
1818 if (!kvm_enabled()) {
1821 return kvm_state->many_ioeventfds;
1824 int kvm_has_gsi_routing(void)
1826 #ifdef KVM_CAP_IRQ_ROUTING
1827 return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING);
1833 int kvm_has_intx_set_mask(void)
1835 return kvm_state->intx_set_mask;
1838 void kvm_setup_guest_memory(void *start, size_t size)
1840 #ifdef CONFIG_VALGRIND_H
1841 VALGRIND_MAKE_MEM_DEFINED(start, size);
1843 if (!kvm_has_sync_mmu()) {
1844 int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK);
1847 perror("qemu_madvise");
1849 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1855 #ifdef KVM_CAP_SET_GUEST_DEBUG
1856 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu,
1859 struct kvm_sw_breakpoint *bp;
1861 QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) {
1869 int kvm_sw_breakpoints_active(CPUState *cpu)
1871 return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints);
1874 struct kvm_set_guest_debug_data {
1875 struct kvm_guest_debug dbg;
1880 static void kvm_invoke_set_guest_debug(void *data)
1882 struct kvm_set_guest_debug_data *dbg_data = data;
1884 dbg_data->err = kvm_vcpu_ioctl(dbg_data->cpu, KVM_SET_GUEST_DEBUG,
1888 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
1890 struct kvm_set_guest_debug_data data;
1892 data.dbg.control = reinject_trap;
1894 if (cpu->singlestep_enabled) {
1895 data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
1897 kvm_arch_update_guest_debug(cpu, &data.dbg);
1900 run_on_cpu(cpu, kvm_invoke_set_guest_debug, &data);
1904 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
1905 target_ulong len, int type)
1907 struct kvm_sw_breakpoint *bp;
1910 if (type == GDB_BREAKPOINT_SW) {
1911 bp = kvm_find_sw_breakpoint(cpu, addr);
1917 bp = g_malloc(sizeof(struct kvm_sw_breakpoint));
1924 err = kvm_arch_insert_sw_breakpoint(cpu, bp);
1930 QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
1932 err = kvm_arch_insert_hw_breakpoint(addr, len, type);
1939 err = kvm_update_guest_debug(cpu, 0);
1947 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
1948 target_ulong len, int type)
1950 struct kvm_sw_breakpoint *bp;
1953 if (type == GDB_BREAKPOINT_SW) {
1954 bp = kvm_find_sw_breakpoint(cpu, addr);
1959 if (bp->use_count > 1) {
1964 err = kvm_arch_remove_sw_breakpoint(cpu, bp);
1969 QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
1972 err = kvm_arch_remove_hw_breakpoint(addr, len, type);
1979 err = kvm_update_guest_debug(cpu, 0);
1987 void kvm_remove_all_breakpoints(CPUState *cpu)
1989 struct kvm_sw_breakpoint *bp, *next;
1990 KVMState *s = cpu->kvm_state;
1992 QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
1993 if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) {
1994 /* Try harder to find a CPU that currently sees the breakpoint. */
1996 if (kvm_arch_remove_sw_breakpoint(cpu, bp) == 0) {
2001 QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry);
2004 kvm_arch_remove_all_hw_breakpoints();
2007 kvm_update_guest_debug(cpu, 0);
2011 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2013 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2018 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2019 target_ulong len, int type)
2024 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2025 target_ulong len, int type)
2030 void kvm_remove_all_breakpoints(CPUState *cpu)
2033 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2035 int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset)
2037 struct kvm_signal_mask *sigmask;
2041 return kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, NULL);
2044 sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset));
2047 memcpy(sigmask->sigset, sigset, sizeof(*sigset));
2048 r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask);
2053 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2055 return kvm_arch_on_sigbus_vcpu(cpu, code, addr);
2058 int kvm_on_sigbus(int code, void *addr)
2060 return kvm_arch_on_sigbus(code, addr);