4 * Copyright IBM, Corp. 2008
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 <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;
79 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
80 bool coalesced_flush_in_progress;
81 int broken_set_mem_region;
84 int robust_singlestep;
86 #ifdef KVM_CAP_SET_GUEST_DEBUG
87 struct kvm_sw_breakpoint_head kvm_sw_breakpoints;
93 /* The man page (and posix) say ioctl numbers are signed int, but
94 * they're not. Linux, glibc and *BSD all treat ioctl numbers as
95 * unsigned, and treating them as signed here can break things */
96 unsigned irq_set_ioctl;
97 #ifdef KVM_CAP_IRQ_ROUTING
98 struct kvm_irq_routing *irq_routes;
99 int nr_allocated_irq_routes;
100 uint32_t *used_gsi_bitmap;
101 unsigned int gsi_count;
102 QTAILQ_HEAD(msi_hashtab, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE];
108 bool kvm_kernel_irqchip;
109 bool kvm_async_interrupts_allowed;
110 bool kvm_halt_in_kernel_allowed;
111 bool kvm_irqfds_allowed;
112 bool kvm_msi_via_irqfd_allowed;
113 bool kvm_gsi_routing_allowed;
115 bool kvm_readonly_mem_allowed;
117 static const KVMCapabilityInfo kvm_required_capabilites[] = {
118 KVM_CAP_INFO(USER_MEMORY),
119 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS),
123 static KVMSlot *kvm_alloc_slot(KVMState *s)
127 for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
128 if (s->slots[i].memory_size == 0) {
133 fprintf(stderr, "%s: no free slot available\n", __func__);
137 static KVMSlot *kvm_lookup_matching_slot(KVMState *s,
143 for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
144 KVMSlot *mem = &s->slots[i];
146 if (start_addr == mem->start_addr &&
147 end_addr == mem->start_addr + mem->memory_size) {
156 * Find overlapping slot with lowest start address
158 static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s,
162 KVMSlot *found = NULL;
165 for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
166 KVMSlot *mem = &s->slots[i];
168 if (mem->memory_size == 0 ||
169 (found && found->start_addr < mem->start_addr)) {
173 if (end_addr > mem->start_addr &&
174 start_addr < mem->start_addr + mem->memory_size) {
182 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram,
187 for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
188 KVMSlot *mem = &s->slots[i];
190 if (ram >= mem->ram && ram < mem->ram + mem->memory_size) {
191 *phys_addr = mem->start_addr + (ram - mem->ram);
199 static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot)
201 struct kvm_userspace_memory_region mem;
203 mem.slot = slot->slot;
204 mem.guest_phys_addr = slot->start_addr;
205 mem.userspace_addr = (unsigned long)slot->ram;
206 mem.flags = slot->flags;
207 if (s->migration_log) {
208 mem.flags |= KVM_MEM_LOG_DIRTY_PAGES;
211 if (slot->memory_size && mem.flags & KVM_MEM_READONLY) {
212 /* Set the slot size to 0 before setting the slot to the desired
213 * value. This is needed based on KVM commit 75d61fbc. */
215 kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
217 mem.memory_size = slot->memory_size;
218 return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
221 static void kvm_reset_vcpu(void *opaque)
223 CPUState *cpu = opaque;
225 kvm_arch_reset_vcpu(cpu);
228 int kvm_init_vcpu(CPUState *cpu)
230 KVMState *s = kvm_state;
234 DPRINTF("kvm_init_vcpu\n");
236 ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)kvm_arch_vcpu_id(cpu));
238 DPRINTF("kvm_create_vcpu failed\n");
244 cpu->kvm_vcpu_dirty = true;
246 mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
249 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
253 cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
255 if (cpu->kvm_run == MAP_FAILED) {
257 DPRINTF("mmap'ing vcpu state failed\n");
261 if (s->coalesced_mmio && !s->coalesced_mmio_ring) {
262 s->coalesced_mmio_ring =
263 (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE;
266 ret = kvm_arch_init_vcpu(cpu);
268 qemu_register_reset(kvm_reset_vcpu, cpu);
269 kvm_arch_reset_vcpu(cpu);
276 * dirty pages logging control
279 static int kvm_mem_flags(KVMState *s, bool log_dirty, bool readonly)
282 flags = log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0;
283 if (readonly && kvm_readonly_mem_allowed) {
284 flags |= KVM_MEM_READONLY;
289 static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty)
291 KVMState *s = kvm_state;
292 int flags, mask = KVM_MEM_LOG_DIRTY_PAGES;
295 old_flags = mem->flags;
297 flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty, false);
300 /* If nothing changed effectively, no need to issue ioctl */
301 if (s->migration_log) {
302 flags |= KVM_MEM_LOG_DIRTY_PAGES;
305 if (flags == old_flags) {
309 return kvm_set_user_memory_region(s, mem);
312 static int kvm_dirty_pages_log_change(hwaddr phys_addr,
313 ram_addr_t size, bool log_dirty)
315 KVMState *s = kvm_state;
316 KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size);
319 fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-"
320 TARGET_FMT_plx "\n", __func__, phys_addr,
321 (hwaddr)(phys_addr + size - 1));
324 return kvm_slot_dirty_pages_log_change(mem, log_dirty);
327 static void kvm_log_start(MemoryListener *listener,
328 MemoryRegionSection *section)
332 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
333 int128_get64(section->size), true);
339 static void kvm_log_stop(MemoryListener *listener,
340 MemoryRegionSection *section)
344 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
345 int128_get64(section->size), false);
351 static int kvm_set_migration_log(int enable)
353 KVMState *s = kvm_state;
357 s->migration_log = enable;
359 for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
362 if (!mem->memory_size) {
365 if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) {
368 err = kvm_set_user_memory_region(s, mem);
376 /* get kvm's dirty pages bitmap and update qemu's */
377 static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section,
378 unsigned long *bitmap)
381 unsigned long page_number, c;
383 unsigned int pages = int128_get64(section->size) / getpagesize();
384 unsigned int len = (pages + HOST_LONG_BITS - 1) / HOST_LONG_BITS;
385 unsigned long hpratio = getpagesize() / TARGET_PAGE_SIZE;
388 * bitmap-traveling is faster than memory-traveling (for addr...)
389 * especially when most of the memory is not dirty.
391 for (i = 0; i < len; i++) {
392 if (bitmap[i] != 0) {
393 c = leul_to_cpu(bitmap[i]);
397 page_number = (i * HOST_LONG_BITS + j) * hpratio;
398 addr1 = page_number * TARGET_PAGE_SIZE;
399 addr = section->offset_within_region + addr1;
400 memory_region_set_dirty(section->mr, addr,
401 TARGET_PAGE_SIZE * hpratio);
408 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
411 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
412 * This function updates qemu's dirty bitmap using
413 * memory_region_set_dirty(). This means all bits are set
416 * @start_add: start of logged region.
417 * @end_addr: end of logged region.
419 static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection *section)
421 KVMState *s = kvm_state;
422 unsigned long size, allocated_size = 0;
426 hwaddr start_addr = section->offset_within_address_space;
427 hwaddr end_addr = start_addr + int128_get64(section->size);
429 d.dirty_bitmap = NULL;
430 while (start_addr < end_addr) {
431 mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr);
436 /* XXX bad kernel interface alert
437 * For dirty bitmap, kernel allocates array of size aligned to
438 * bits-per-long. But for case when the kernel is 64bits and
439 * the userspace is 32bits, userspace can't align to the same
440 * bits-per-long, since sizeof(long) is different between kernel
441 * and user space. This way, userspace will provide buffer which
442 * may be 4 bytes less than the kernel will use, resulting in
443 * userspace memory corruption (which is not detectable by valgrind
444 * too, in most cases).
445 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
446 * a hope that sizeof(long) wont become >8 any time soon.
448 size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS),
449 /*HOST_LONG_BITS*/ 64) / 8;
450 if (!d.dirty_bitmap) {
451 d.dirty_bitmap = g_malloc(size);
452 } else if (size > allocated_size) {
453 d.dirty_bitmap = g_realloc(d.dirty_bitmap, size);
455 allocated_size = size;
456 memset(d.dirty_bitmap, 0, allocated_size);
460 if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
461 DPRINTF("ioctl failed %d\n", errno);
466 kvm_get_dirty_pages_log_range(section, d.dirty_bitmap);
467 start_addr = mem->start_addr + mem->memory_size;
469 g_free(d.dirty_bitmap);
474 static void kvm_coalesce_mmio_region(MemoryListener *listener,
475 MemoryRegionSection *secion,
476 hwaddr start, hwaddr size)
478 KVMState *s = kvm_state;
480 if (s->coalesced_mmio) {
481 struct kvm_coalesced_mmio_zone zone;
487 (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
491 static void kvm_uncoalesce_mmio_region(MemoryListener *listener,
492 MemoryRegionSection *secion,
493 hwaddr start, hwaddr size)
495 KVMState *s = kvm_state;
497 if (s->coalesced_mmio) {
498 struct kvm_coalesced_mmio_zone zone;
504 (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
508 int kvm_check_extension(KVMState *s, unsigned int extension)
512 ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
520 static int kvm_set_ioeventfd_mmio(int fd, uint32_t addr, uint32_t val,
521 bool assign, uint32_t size, bool datamatch)
524 struct kvm_ioeventfd iofd;
526 iofd.datamatch = datamatch ? val : 0;
532 if (!kvm_enabled()) {
537 iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
540 iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
543 ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd);
552 static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val,
553 bool assign, uint32_t size, bool datamatch)
555 struct kvm_ioeventfd kick = {
556 .datamatch = datamatch ? val : 0,
558 .flags = KVM_IOEVENTFD_FLAG_PIO,
563 if (!kvm_enabled()) {
567 kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
570 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
572 r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
580 static int kvm_check_many_ioeventfds(void)
582 /* Userspace can use ioeventfd for io notification. This requires a host
583 * that supports eventfd(2) and an I/O thread; since eventfd does not
584 * support SIGIO it cannot interrupt the vcpu.
586 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
587 * can avoid creating too many ioeventfds.
589 #if defined(CONFIG_EVENTFD)
592 for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) {
593 ioeventfds[i] = eventfd(0, EFD_CLOEXEC);
594 if (ioeventfds[i] < 0) {
597 ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true);
599 close(ioeventfds[i]);
604 /* Decide whether many devices are supported or not */
605 ret = i == ARRAY_SIZE(ioeventfds);
608 kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true);
609 close(ioeventfds[i]);
617 static const KVMCapabilityInfo *
618 kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list)
621 if (!kvm_check_extension(s, list->value)) {
629 static void kvm_set_phys_mem(MemoryRegionSection *section, bool add)
631 KVMState *s = kvm_state;
634 MemoryRegion *mr = section->mr;
635 bool log_dirty = memory_region_is_logging(mr);
636 bool writeable = !mr->readonly && !mr->rom_device;
637 bool readonly_flag = mr->readonly || memory_region_is_romd(mr);
638 hwaddr start_addr = section->offset_within_address_space;
639 ram_addr_t size = int128_get64(section->size);
643 /* kvm works in page size chunks, but the function may be called
644 with sub-page size and unaligned start address. */
645 delta = TARGET_PAGE_ALIGN(size) - size;
651 size &= TARGET_PAGE_MASK;
652 if (!size || (start_addr & ~TARGET_PAGE_MASK)) {
656 if (!memory_region_is_ram(mr)) {
657 if (writeable || !kvm_readonly_mem_allowed) {
659 } else if (!mr->romd_mode) {
660 /* If the memory device is not in romd_mode, then we actually want
661 * to remove the kvm memory slot so all accesses will trap. */
666 ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta;
669 mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
674 if (add && start_addr >= mem->start_addr &&
675 (start_addr + size <= mem->start_addr + mem->memory_size) &&
676 (ram - start_addr == mem->ram - mem->start_addr)) {
677 /* The new slot fits into the existing one and comes with
678 * identical parameters - update flags and done. */
679 kvm_slot_dirty_pages_log_change(mem, log_dirty);
685 if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
686 kvm_physical_sync_dirty_bitmap(section);
689 /* unregister the overlapping slot */
690 mem->memory_size = 0;
691 err = kvm_set_user_memory_region(s, mem);
693 fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
694 __func__, strerror(-err));
698 /* Workaround for older KVM versions: we can't join slots, even not by
699 * unregistering the previous ones and then registering the larger
700 * slot. We have to maintain the existing fragmentation. Sigh.
702 * This workaround assumes that the new slot starts at the same
703 * address as the first existing one. If not or if some overlapping
704 * slot comes around later, we will fail (not seen in practice so far)
705 * - and actually require a recent KVM version. */
706 if (s->broken_set_mem_region &&
707 old.start_addr == start_addr && old.memory_size < size && add) {
708 mem = kvm_alloc_slot(s);
709 mem->memory_size = old.memory_size;
710 mem->start_addr = old.start_addr;
712 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
714 err = kvm_set_user_memory_region(s, mem);
716 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
721 start_addr += old.memory_size;
722 ram += old.memory_size;
723 size -= old.memory_size;
727 /* register prefix slot */
728 if (old.start_addr < start_addr) {
729 mem = kvm_alloc_slot(s);
730 mem->memory_size = start_addr - old.start_addr;
731 mem->start_addr = old.start_addr;
733 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
735 err = kvm_set_user_memory_region(s, mem);
737 fprintf(stderr, "%s: error registering prefix slot: %s\n",
738 __func__, strerror(-err));
740 fprintf(stderr, "%s: This is probably because your kernel's " \
741 "PAGE_SIZE is too big. Please try to use 4k " \
742 "PAGE_SIZE!\n", __func__);
748 /* register suffix slot */
749 if (old.start_addr + old.memory_size > start_addr + size) {
750 ram_addr_t size_delta;
752 mem = kvm_alloc_slot(s);
753 mem->start_addr = start_addr + size;
754 size_delta = mem->start_addr - old.start_addr;
755 mem->memory_size = old.memory_size - size_delta;
756 mem->ram = old.ram + size_delta;
757 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
759 err = kvm_set_user_memory_region(s, mem);
761 fprintf(stderr, "%s: error registering suffix slot: %s\n",
762 __func__, strerror(-err));
768 /* in case the KVM bug workaround already "consumed" the new slot */
775 mem = kvm_alloc_slot(s);
776 mem->memory_size = size;
777 mem->start_addr = start_addr;
779 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
781 err = kvm_set_user_memory_region(s, mem);
783 fprintf(stderr, "%s: error registering slot: %s\n", __func__,
789 static void kvm_region_add(MemoryListener *listener,
790 MemoryRegionSection *section)
792 memory_region_ref(section->mr);
793 kvm_set_phys_mem(section, true);
796 static void kvm_region_del(MemoryListener *listener,
797 MemoryRegionSection *section)
799 kvm_set_phys_mem(section, false);
800 memory_region_unref(section->mr);
803 static void kvm_log_sync(MemoryListener *listener,
804 MemoryRegionSection *section)
808 r = kvm_physical_sync_dirty_bitmap(section);
814 static void kvm_log_global_start(struct MemoryListener *listener)
818 r = kvm_set_migration_log(1);
822 static void kvm_log_global_stop(struct MemoryListener *listener)
826 r = kvm_set_migration_log(0);
830 static void kvm_mem_ioeventfd_add(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, true, int128_get64(section->size),
842 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
843 __func__, strerror(-r));
848 static void kvm_mem_ioeventfd_del(MemoryListener *listener,
849 MemoryRegionSection *section,
850 bool match_data, uint64_t data,
853 int fd = event_notifier_get_fd(e);
856 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
857 data, false, int128_get64(section->size),
864 static void kvm_io_ioeventfd_add(MemoryListener *listener,
865 MemoryRegionSection *section,
866 bool match_data, uint64_t data,
869 int fd = event_notifier_get_fd(e);
872 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
873 data, true, int128_get64(section->size),
876 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
877 __func__, strerror(-r));
882 static void kvm_io_ioeventfd_del(MemoryListener *listener,
883 MemoryRegionSection *section,
884 bool match_data, uint64_t data,
888 int fd = event_notifier_get_fd(e);
891 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
892 data, false, int128_get64(section->size),
899 static MemoryListener kvm_memory_listener = {
900 .region_add = kvm_region_add,
901 .region_del = kvm_region_del,
902 .log_start = kvm_log_start,
903 .log_stop = kvm_log_stop,
904 .log_sync = kvm_log_sync,
905 .log_global_start = kvm_log_global_start,
906 .log_global_stop = kvm_log_global_stop,
907 .eventfd_add = kvm_mem_ioeventfd_add,
908 .eventfd_del = kvm_mem_ioeventfd_del,
909 .coalesced_mmio_add = kvm_coalesce_mmio_region,
910 .coalesced_mmio_del = kvm_uncoalesce_mmio_region,
914 static MemoryListener kvm_io_listener = {
915 .eventfd_add = kvm_io_ioeventfd_add,
916 .eventfd_del = kvm_io_ioeventfd_del,
920 static void kvm_handle_interrupt(CPUState *cpu, int mask)
922 cpu->interrupt_request |= mask;
924 if (!qemu_cpu_is_self(cpu)) {
929 int kvm_set_irq(KVMState *s, int irq, int level)
931 struct kvm_irq_level event;
934 assert(kvm_async_interrupts_enabled());
938 ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event);
940 perror("kvm_set_irq");
944 return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status;
947 #ifdef KVM_CAP_IRQ_ROUTING
948 typedef struct KVMMSIRoute {
949 struct kvm_irq_routing_entry kroute;
950 QTAILQ_ENTRY(KVMMSIRoute) entry;
953 static void set_gsi(KVMState *s, unsigned int gsi)
955 s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32);
958 static void clear_gsi(KVMState *s, unsigned int gsi)
960 s->used_gsi_bitmap[gsi / 32] &= ~(1U << (gsi % 32));
963 void kvm_init_irq_routing(KVMState *s)
967 gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING);
969 unsigned int gsi_bits, i;
971 /* Round up so we can search ints using ffs */
972 gsi_bits = ALIGN(gsi_count, 32);
973 s->used_gsi_bitmap = g_malloc0(gsi_bits / 8);
974 s->gsi_count = gsi_count;
976 /* Mark any over-allocated bits as already in use */
977 for (i = gsi_count; i < gsi_bits; i++) {
982 s->irq_routes = g_malloc0(sizeof(*s->irq_routes));
983 s->nr_allocated_irq_routes = 0;
985 if (!s->direct_msi) {
986 for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) {
987 QTAILQ_INIT(&s->msi_hashtab[i]);
991 kvm_arch_init_irq_routing(s);
994 void kvm_irqchip_commit_routes(KVMState *s)
998 s->irq_routes->flags = 0;
999 ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes);
1003 static void kvm_add_routing_entry(KVMState *s,
1004 struct kvm_irq_routing_entry *entry)
1006 struct kvm_irq_routing_entry *new;
1009 if (s->irq_routes->nr == s->nr_allocated_irq_routes) {
1010 n = s->nr_allocated_irq_routes * 2;
1014 size = sizeof(struct kvm_irq_routing);
1015 size += n * sizeof(*new);
1016 s->irq_routes = g_realloc(s->irq_routes, size);
1017 s->nr_allocated_irq_routes = n;
1019 n = s->irq_routes->nr++;
1020 new = &s->irq_routes->entries[n];
1024 set_gsi(s, entry->gsi);
1027 static int kvm_update_routing_entry(KVMState *s,
1028 struct kvm_irq_routing_entry *new_entry)
1030 struct kvm_irq_routing_entry *entry;
1033 for (n = 0; n < s->irq_routes->nr; n++) {
1034 entry = &s->irq_routes->entries[n];
1035 if (entry->gsi != new_entry->gsi) {
1039 if(!memcmp(entry, new_entry, sizeof *entry)) {
1043 *entry = *new_entry;
1045 kvm_irqchip_commit_routes(s);
1053 void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin)
1055 struct kvm_irq_routing_entry e = {};
1057 assert(pin < s->gsi_count);
1060 e.type = KVM_IRQ_ROUTING_IRQCHIP;
1062 e.u.irqchip.irqchip = irqchip;
1063 e.u.irqchip.pin = pin;
1064 kvm_add_routing_entry(s, &e);
1067 void kvm_irqchip_release_virq(KVMState *s, int virq)
1069 struct kvm_irq_routing_entry *e;
1072 for (i = 0; i < s->irq_routes->nr; i++) {
1073 e = &s->irq_routes->entries[i];
1074 if (e->gsi == virq) {
1075 s->irq_routes->nr--;
1076 *e = s->irq_routes->entries[s->irq_routes->nr];
1082 static unsigned int kvm_hash_msi(uint32_t data)
1084 /* This is optimized for IA32 MSI layout. However, no other arch shall
1085 * repeat the mistake of not providing a direct MSI injection API. */
1089 static void kvm_flush_dynamic_msi_routes(KVMState *s)
1091 KVMMSIRoute *route, *next;
1094 for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) {
1095 QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) {
1096 kvm_irqchip_release_virq(s, route->kroute.gsi);
1097 QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry);
1103 static int kvm_irqchip_get_virq(KVMState *s)
1105 uint32_t *word = s->used_gsi_bitmap;
1106 int max_words = ALIGN(s->gsi_count, 32) / 32;
1111 /* Return the lowest unused GSI in the bitmap */
1112 for (i = 0; i < max_words; i++) {
1113 bit = ffs(~word[i]);
1118 return bit - 1 + i * 32;
1120 if (!s->direct_msi && retry) {
1122 kvm_flush_dynamic_msi_routes(s);
1129 static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg)
1131 unsigned int hash = kvm_hash_msi(msg.data);
1134 QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) {
1135 if (route->kroute.u.msi.address_lo == (uint32_t)msg.address &&
1136 route->kroute.u.msi.address_hi == (msg.address >> 32) &&
1137 route->kroute.u.msi.data == le32_to_cpu(msg.data)) {
1144 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1149 if (s->direct_msi) {
1150 msi.address_lo = (uint32_t)msg.address;
1151 msi.address_hi = msg.address >> 32;
1152 msi.data = le32_to_cpu(msg.data);
1154 memset(msi.pad, 0, sizeof(msi.pad));
1156 return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi);
1159 route = kvm_lookup_msi_route(s, msg);
1163 virq = kvm_irqchip_get_virq(s);
1168 route = g_malloc0(sizeof(KVMMSIRoute));
1169 route->kroute.gsi = virq;
1170 route->kroute.type = KVM_IRQ_ROUTING_MSI;
1171 route->kroute.flags = 0;
1172 route->kroute.u.msi.address_lo = (uint32_t)msg.address;
1173 route->kroute.u.msi.address_hi = msg.address >> 32;
1174 route->kroute.u.msi.data = le32_to_cpu(msg.data);
1176 kvm_add_routing_entry(s, &route->kroute);
1177 kvm_irqchip_commit_routes(s);
1179 QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route,
1183 assert(route->kroute.type == KVM_IRQ_ROUTING_MSI);
1185 return kvm_set_irq(s, route->kroute.gsi, 1);
1188 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1190 struct kvm_irq_routing_entry kroute = {};
1193 if (!kvm_gsi_routing_enabled()) {
1197 virq = kvm_irqchip_get_virq(s);
1203 kroute.type = KVM_IRQ_ROUTING_MSI;
1205 kroute.u.msi.address_lo = (uint32_t)msg.address;
1206 kroute.u.msi.address_hi = msg.address >> 32;
1207 kroute.u.msi.data = le32_to_cpu(msg.data);
1209 kvm_add_routing_entry(s, &kroute);
1210 kvm_irqchip_commit_routes(s);
1215 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1217 struct kvm_irq_routing_entry kroute = {};
1219 if (!kvm_irqchip_in_kernel()) {
1224 kroute.type = KVM_IRQ_ROUTING_MSI;
1226 kroute.u.msi.address_lo = (uint32_t)msg.address;
1227 kroute.u.msi.address_hi = msg.address >> 32;
1228 kroute.u.msi.data = le32_to_cpu(msg.data);
1230 return kvm_update_routing_entry(s, &kroute);
1233 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int rfd, int virq,
1236 struct kvm_irqfd irqfd = {
1239 .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN,
1243 irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE;
1244 irqfd.resamplefd = rfd;
1247 if (!kvm_irqfds_enabled()) {
1251 return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd);
1254 #else /* !KVM_CAP_IRQ_ROUTING */
1256 void kvm_init_irq_routing(KVMState *s)
1260 void kvm_irqchip_release_virq(KVMState *s, int virq)
1264 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1269 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1274 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign)
1279 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1283 #endif /* !KVM_CAP_IRQ_ROUTING */
1285 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n,
1286 EventNotifier *rn, int virq)
1288 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n),
1289 rn ? event_notifier_get_fd(rn) : -1, virq, true);
1292 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, int virq)
1294 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), -1, virq,
1298 static int kvm_irqchip_create(KVMState *s)
1302 if (!qemu_opt_get_bool(qemu_get_machine_opts(), "kernel_irqchip", true) ||
1303 !kvm_check_extension(s, KVM_CAP_IRQCHIP)) {
1307 ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP);
1309 fprintf(stderr, "Create kernel irqchip failed\n");
1313 kvm_kernel_irqchip = true;
1314 /* If we have an in-kernel IRQ chip then we must have asynchronous
1315 * interrupt delivery (though the reverse is not necessarily true)
1317 kvm_async_interrupts_allowed = true;
1318 kvm_halt_in_kernel_allowed = true;
1320 kvm_init_irq_routing(s);
1325 static int kvm_max_vcpus(KVMState *s)
1329 /* Find number of supported CPUs using the recommended
1330 * procedure from the kernel API documentation to cope with
1331 * older kernels that may be missing capabilities.
1333 ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS);
1337 ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS);
1347 static const char upgrade_note[] =
1348 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1349 "(see http://sourceforge.net/projects/kvm).\n";
1351 const KVMCapabilityInfo *missing_cap;
1356 s = g_malloc0(sizeof(KVMState));
1359 * On systems where the kernel can support different base page
1360 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1361 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1362 * page size for the system though.
1364 assert(TARGET_PAGE_SIZE <= getpagesize());
1366 #ifdef KVM_CAP_SET_GUEST_DEBUG
1367 QTAILQ_INIT(&s->kvm_sw_breakpoints);
1369 for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
1370 s->slots[i].slot = i;
1373 s->fd = qemu_open("/dev/kvm", O_RDWR);
1375 fprintf(stderr, "Could not access KVM kernel module: %m\n");
1380 ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
1381 if (ret < KVM_API_VERSION) {
1385 fprintf(stderr, "kvm version too old\n");
1389 if (ret > KVM_API_VERSION) {
1391 fprintf(stderr, "kvm version not supported\n");
1395 max_vcpus = kvm_max_vcpus(s);
1396 if (smp_cpus > max_vcpus) {
1398 fprintf(stderr, "Number of SMP cpus requested (%d) exceeds max cpus "
1399 "supported by KVM (%d)\n", smp_cpus, max_vcpus);
1403 if (max_cpus > max_vcpus) {
1405 fprintf(stderr, "Number of hotpluggable cpus requested (%d) exceeds max cpus "
1406 "supported by KVM (%d)\n", max_cpus, max_vcpus);
1410 s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0);
1413 fprintf(stderr, "Please add the 'switch_amode' kernel parameter to "
1414 "your host kernel command line\n");
1420 missing_cap = kvm_check_extension_list(s, kvm_required_capabilites);
1423 kvm_check_extension_list(s, kvm_arch_required_capabilities);
1427 fprintf(stderr, "kvm does not support %s\n%s",
1428 missing_cap->name, upgrade_note);
1432 s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
1434 s->broken_set_mem_region = 1;
1435 ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
1437 s->broken_set_mem_region = 0;
1440 #ifdef KVM_CAP_VCPU_EVENTS
1441 s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS);
1444 s->robust_singlestep =
1445 kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP);
1447 #ifdef KVM_CAP_DEBUGREGS
1448 s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS);
1451 #ifdef KVM_CAP_XSAVE
1452 s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE);
1456 s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS);
1459 #ifdef KVM_CAP_PIT_STATE2
1460 s->pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2);
1463 #ifdef KVM_CAP_IRQ_ROUTING
1464 s->direct_msi = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0);
1467 s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3);
1469 s->irq_set_ioctl = KVM_IRQ_LINE;
1470 if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) {
1471 s->irq_set_ioctl = KVM_IRQ_LINE_STATUS;
1474 #ifdef KVM_CAP_READONLY_MEM
1475 kvm_readonly_mem_allowed =
1476 (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0);
1479 ret = kvm_arch_init(s);
1484 ret = kvm_irqchip_create(s);
1490 memory_listener_register(&kvm_memory_listener, &address_space_memory);
1491 memory_listener_register(&kvm_io_listener, &address_space_io);
1493 s->many_ioeventfds = kvm_check_many_ioeventfds();
1495 cpu_interrupt_handler = kvm_handle_interrupt;
1511 static void kvm_handle_io(uint16_t port, void *data, int direction, int size,
1515 uint8_t *ptr = data;
1517 for (i = 0; i < count; i++) {
1518 address_space_rw(&address_space_io, port, ptr, size,
1519 direction == KVM_EXIT_IO_OUT);
1524 static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run)
1526 fprintf(stderr, "KVM internal error.");
1527 if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) {
1530 fprintf(stderr, " Suberror: %d\n", run->internal.suberror);
1531 for (i = 0; i < run->internal.ndata; ++i) {
1532 fprintf(stderr, "extra data[%d]: %"PRIx64"\n",
1533 i, (uint64_t)run->internal.data[i]);
1536 fprintf(stderr, "\n");
1538 if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) {
1539 fprintf(stderr, "emulation failure\n");
1540 if (!kvm_arch_stop_on_emulation_error(cpu)) {
1541 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1542 return EXCP_INTERRUPT;
1545 /* FIXME: Should trigger a qmp message to let management know
1546 * something went wrong.
1551 void kvm_flush_coalesced_mmio_buffer(void)
1553 KVMState *s = kvm_state;
1555 if (s->coalesced_flush_in_progress) {
1559 s->coalesced_flush_in_progress = true;
1561 if (s->coalesced_mmio_ring) {
1562 struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring;
1563 while (ring->first != ring->last) {
1564 struct kvm_coalesced_mmio *ent;
1566 ent = &ring->coalesced_mmio[ring->first];
1568 cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
1570 ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
1574 s->coalesced_flush_in_progress = false;
1577 static void do_kvm_cpu_synchronize_state(void *arg)
1579 CPUState *cpu = arg;
1581 if (!cpu->kvm_vcpu_dirty) {
1582 kvm_arch_get_registers(cpu);
1583 cpu->kvm_vcpu_dirty = true;
1587 void kvm_cpu_synchronize_state(CPUState *cpu)
1589 if (!cpu->kvm_vcpu_dirty) {
1590 run_on_cpu(cpu, do_kvm_cpu_synchronize_state, cpu);
1594 void kvm_cpu_synchronize_post_reset(CPUState *cpu)
1596 kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE);
1597 cpu->kvm_vcpu_dirty = false;
1600 void kvm_cpu_synchronize_post_init(CPUState *cpu)
1602 kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE);
1603 cpu->kvm_vcpu_dirty = false;
1606 int kvm_cpu_exec(CPUState *cpu)
1608 struct kvm_run *run = cpu->kvm_run;
1611 DPRINTF("kvm_cpu_exec()\n");
1613 if (kvm_arch_process_async_events(cpu)) {
1614 cpu->exit_request = 0;
1619 if (cpu->kvm_vcpu_dirty) {
1620 kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE);
1621 cpu->kvm_vcpu_dirty = false;
1624 kvm_arch_pre_run(cpu, run);
1625 if (cpu->exit_request) {
1626 DPRINTF("interrupt exit requested\n");
1628 * KVM requires us to reenter the kernel after IO exits to complete
1629 * instruction emulation. This self-signal will ensure that we
1632 qemu_cpu_kick_self();
1634 qemu_mutex_unlock_iothread();
1636 run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0);
1638 qemu_mutex_lock_iothread();
1639 kvm_arch_post_run(cpu, run);
1642 if (run_ret == -EINTR || run_ret == -EAGAIN) {
1643 DPRINTF("io window exit\n");
1644 ret = EXCP_INTERRUPT;
1647 fprintf(stderr, "error: kvm run failed %s\n",
1648 strerror(-run_ret));
1652 trace_kvm_run_exit(cpu->cpu_index, run->exit_reason);
1653 switch (run->exit_reason) {
1655 DPRINTF("handle_io\n");
1656 kvm_handle_io(run->io.port,
1657 (uint8_t *)run + run->io.data_offset,
1664 DPRINTF("handle_mmio\n");
1665 cpu_physical_memory_rw(run->mmio.phys_addr,
1668 run->mmio.is_write);
1671 case KVM_EXIT_IRQ_WINDOW_OPEN:
1672 DPRINTF("irq_window_open\n");
1673 ret = EXCP_INTERRUPT;
1675 case KVM_EXIT_SHUTDOWN:
1676 DPRINTF("shutdown\n");
1677 qemu_system_reset_request();
1678 ret = EXCP_INTERRUPT;
1680 case KVM_EXIT_UNKNOWN:
1681 fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n",
1682 (uint64_t)run->hw.hardware_exit_reason);
1685 case KVM_EXIT_INTERNAL_ERROR:
1686 ret = kvm_handle_internal_error(cpu, run);
1689 DPRINTF("kvm_arch_handle_exit\n");
1690 ret = kvm_arch_handle_exit(cpu, run);
1696 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1697 vm_stop(RUN_STATE_INTERNAL_ERROR);
1700 cpu->exit_request = 0;
1704 int kvm_ioctl(KVMState *s, int type, ...)
1711 arg = va_arg(ap, void *);
1714 trace_kvm_ioctl(type, arg);
1715 ret = ioctl(s->fd, type, arg);
1722 int kvm_vm_ioctl(KVMState *s, int type, ...)
1729 arg = va_arg(ap, void *);
1732 trace_kvm_vm_ioctl(type, arg);
1733 ret = ioctl(s->vmfd, type, arg);
1740 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...)
1747 arg = va_arg(ap, void *);
1750 trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg);
1751 ret = ioctl(cpu->kvm_fd, type, arg);
1758 int kvm_has_sync_mmu(void)
1760 return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU);
1763 int kvm_has_vcpu_events(void)
1765 return kvm_state->vcpu_events;
1768 int kvm_has_robust_singlestep(void)
1770 return kvm_state->robust_singlestep;
1773 int kvm_has_debugregs(void)
1775 return kvm_state->debugregs;
1778 int kvm_has_xsave(void)
1780 return kvm_state->xsave;
1783 int kvm_has_xcrs(void)
1785 return kvm_state->xcrs;
1788 int kvm_has_pit_state2(void)
1790 return kvm_state->pit_state2;
1793 int kvm_has_many_ioeventfds(void)
1795 if (!kvm_enabled()) {
1798 return kvm_state->many_ioeventfds;
1801 int kvm_has_gsi_routing(void)
1803 #ifdef KVM_CAP_IRQ_ROUTING
1804 return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING);
1810 int kvm_has_intx_set_mask(void)
1812 return kvm_state->intx_set_mask;
1815 void *kvm_ram_alloc(ram_addr_t size)
1820 mem = kvm_arch_ram_alloc(size);
1825 return qemu_anon_ram_alloc(size);
1828 void kvm_setup_guest_memory(void *start, size_t size)
1830 #ifdef CONFIG_VALGRIND_H
1831 VALGRIND_MAKE_MEM_DEFINED(start, size);
1833 if (!kvm_has_sync_mmu()) {
1834 int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK);
1837 perror("qemu_madvise");
1839 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1845 #ifdef KVM_CAP_SET_GUEST_DEBUG
1846 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu,
1849 struct kvm_sw_breakpoint *bp;
1851 QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) {
1859 int kvm_sw_breakpoints_active(CPUState *cpu)
1861 return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints);
1864 struct kvm_set_guest_debug_data {
1865 struct kvm_guest_debug dbg;
1870 static void kvm_invoke_set_guest_debug(void *data)
1872 struct kvm_set_guest_debug_data *dbg_data = data;
1874 dbg_data->err = kvm_vcpu_ioctl(dbg_data->cpu, KVM_SET_GUEST_DEBUG,
1878 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
1880 struct kvm_set_guest_debug_data data;
1882 data.dbg.control = reinject_trap;
1884 if (cpu->singlestep_enabled) {
1885 data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
1887 kvm_arch_update_guest_debug(cpu, &data.dbg);
1890 run_on_cpu(cpu, kvm_invoke_set_guest_debug, &data);
1894 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
1895 target_ulong len, int type)
1897 struct kvm_sw_breakpoint *bp;
1900 if (type == GDB_BREAKPOINT_SW) {
1901 bp = kvm_find_sw_breakpoint(cpu, addr);
1907 bp = g_malloc(sizeof(struct kvm_sw_breakpoint));
1914 err = kvm_arch_insert_sw_breakpoint(cpu, bp);
1920 QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
1922 err = kvm_arch_insert_hw_breakpoint(addr, len, type);
1929 err = kvm_update_guest_debug(cpu, 0);
1937 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
1938 target_ulong len, int type)
1940 struct kvm_sw_breakpoint *bp;
1943 if (type == GDB_BREAKPOINT_SW) {
1944 bp = kvm_find_sw_breakpoint(cpu, addr);
1949 if (bp->use_count > 1) {
1954 err = kvm_arch_remove_sw_breakpoint(cpu, bp);
1959 QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
1962 err = kvm_arch_remove_hw_breakpoint(addr, len, type);
1969 err = kvm_update_guest_debug(cpu, 0);
1977 void kvm_remove_all_breakpoints(CPUState *cpu)
1979 struct kvm_sw_breakpoint *bp, *next;
1980 KVMState *s = cpu->kvm_state;
1982 QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
1983 if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) {
1984 /* Try harder to find a CPU that currently sees the breakpoint. */
1986 if (kvm_arch_remove_sw_breakpoint(cpu, bp) == 0) {
1991 QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry);
1994 kvm_arch_remove_all_hw_breakpoints();
1997 kvm_update_guest_debug(cpu, 0);
2001 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2003 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2008 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2009 target_ulong len, int type)
2014 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2015 target_ulong len, int type)
2020 void kvm_remove_all_breakpoints(CPUState *cpu)
2023 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2025 int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset)
2027 struct kvm_signal_mask *sigmask;
2031 return kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, NULL);
2034 sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset));
2037 memcpy(sigmask->sigset, sigset, sizeof(*sigset));
2038 r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask);
2043 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2045 return kvm_arch_on_sigbus_vcpu(cpu, code, addr);
2048 int kvm_on_sigbus(int code, void *addr)
2050 return kvm_arch_on_sigbus(code, addr);