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;
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;
114 bool kvm_gsi_direct_mapping;
116 bool kvm_readonly_mem_allowed;
118 static const KVMCapabilityInfo kvm_required_capabilites[] = {
119 KVM_CAP_INFO(USER_MEMORY),
120 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS),
124 static KVMSlot *kvm_alloc_slot(KVMState *s)
128 for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
129 if (s->slots[i].memory_size == 0) {
134 fprintf(stderr, "%s: no free slot available\n", __func__);
138 static KVMSlot *kvm_lookup_matching_slot(KVMState *s,
144 for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
145 KVMSlot *mem = &s->slots[i];
147 if (start_addr == mem->start_addr &&
148 end_addr == mem->start_addr + mem->memory_size) {
157 * Find overlapping slot with lowest start address
159 static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s,
163 KVMSlot *found = NULL;
166 for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
167 KVMSlot *mem = &s->slots[i];
169 if (mem->memory_size == 0 ||
170 (found && found->start_addr < mem->start_addr)) {
174 if (end_addr > mem->start_addr &&
175 start_addr < mem->start_addr + mem->memory_size) {
183 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram,
188 for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
189 KVMSlot *mem = &s->slots[i];
191 if (ram >= mem->ram && ram < mem->ram + mem->memory_size) {
192 *phys_addr = mem->start_addr + (ram - mem->ram);
200 static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot)
202 struct kvm_userspace_memory_region mem;
204 mem.slot = slot->slot;
205 mem.guest_phys_addr = slot->start_addr;
206 mem.userspace_addr = (unsigned long)slot->ram;
207 mem.flags = slot->flags;
208 if (s->migration_log) {
209 mem.flags |= KVM_MEM_LOG_DIRTY_PAGES;
212 if (slot->memory_size && mem.flags & KVM_MEM_READONLY) {
213 /* Set the slot size to 0 before setting the slot to the desired
214 * value. This is needed based on KVM commit 75d61fbc. */
216 kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
218 mem.memory_size = slot->memory_size;
219 return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
222 static void kvm_reset_vcpu(void *opaque)
224 CPUState *cpu = opaque;
226 kvm_arch_reset_vcpu(cpu);
229 int kvm_init_vcpu(CPUState *cpu)
231 KVMState *s = kvm_state;
235 DPRINTF("kvm_init_vcpu\n");
237 ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)kvm_arch_vcpu_id(cpu));
239 DPRINTF("kvm_create_vcpu failed\n");
245 cpu->kvm_vcpu_dirty = true;
247 mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
250 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
254 cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
256 if (cpu->kvm_run == MAP_FAILED) {
258 DPRINTF("mmap'ing vcpu state failed\n");
262 if (s->coalesced_mmio && !s->coalesced_mmio_ring) {
263 s->coalesced_mmio_ring =
264 (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE;
267 ret = kvm_arch_init_vcpu(cpu);
269 qemu_register_reset(kvm_reset_vcpu, cpu);
270 kvm_arch_reset_vcpu(cpu);
277 * dirty pages logging control
280 static int kvm_mem_flags(KVMState *s, bool log_dirty, bool readonly)
283 flags = log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0;
284 if (readonly && kvm_readonly_mem_allowed) {
285 flags |= KVM_MEM_READONLY;
290 static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty)
292 KVMState *s = kvm_state;
293 int flags, mask = KVM_MEM_LOG_DIRTY_PAGES;
296 old_flags = mem->flags;
298 flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty, false);
301 /* If nothing changed effectively, no need to issue ioctl */
302 if (s->migration_log) {
303 flags |= KVM_MEM_LOG_DIRTY_PAGES;
306 if (flags == old_flags) {
310 return kvm_set_user_memory_region(s, mem);
313 static int kvm_dirty_pages_log_change(hwaddr phys_addr,
314 ram_addr_t size, bool log_dirty)
316 KVMState *s = kvm_state;
317 KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size);
320 fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-"
321 TARGET_FMT_plx "\n", __func__, phys_addr,
322 (hwaddr)(phys_addr + size - 1));
325 return kvm_slot_dirty_pages_log_change(mem, log_dirty);
328 static void kvm_log_start(MemoryListener *listener,
329 MemoryRegionSection *section)
333 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
334 int128_get64(section->size), true);
340 static void kvm_log_stop(MemoryListener *listener,
341 MemoryRegionSection *section)
345 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
346 int128_get64(section->size), false);
352 static int kvm_set_migration_log(int enable)
354 KVMState *s = kvm_state;
358 s->migration_log = enable;
360 for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
363 if (!mem->memory_size) {
366 if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) {
369 err = kvm_set_user_memory_region(s, mem);
377 /* get kvm's dirty pages bitmap and update qemu's */
378 static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section,
379 unsigned long *bitmap)
382 unsigned long page_number, c;
384 unsigned int pages = int128_get64(section->size) / getpagesize();
385 unsigned int len = (pages + HOST_LONG_BITS - 1) / HOST_LONG_BITS;
386 unsigned long hpratio = getpagesize() / TARGET_PAGE_SIZE;
389 * bitmap-traveling is faster than memory-traveling (for addr...)
390 * especially when most of the memory is not dirty.
392 for (i = 0; i < len; i++) {
393 if (bitmap[i] != 0) {
394 c = leul_to_cpu(bitmap[i]);
398 page_number = (i * HOST_LONG_BITS + j) * hpratio;
399 addr1 = page_number * TARGET_PAGE_SIZE;
400 addr = section->offset_within_region + addr1;
401 memory_region_set_dirty(section->mr, addr,
402 TARGET_PAGE_SIZE * hpratio);
409 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
412 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
413 * This function updates qemu's dirty bitmap using
414 * memory_region_set_dirty(). This means all bits are set
417 * @start_add: start of logged region.
418 * @end_addr: end of logged region.
420 static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection *section)
422 KVMState *s = kvm_state;
423 unsigned long size, allocated_size = 0;
427 hwaddr start_addr = section->offset_within_address_space;
428 hwaddr end_addr = start_addr + int128_get64(section->size);
430 d.dirty_bitmap = NULL;
431 while (start_addr < end_addr) {
432 mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr);
437 /* XXX bad kernel interface alert
438 * For dirty bitmap, kernel allocates array of size aligned to
439 * bits-per-long. But for case when the kernel is 64bits and
440 * the userspace is 32bits, userspace can't align to the same
441 * bits-per-long, since sizeof(long) is different between kernel
442 * and user space. This way, userspace will provide buffer which
443 * may be 4 bytes less than the kernel will use, resulting in
444 * userspace memory corruption (which is not detectable by valgrind
445 * too, in most cases).
446 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
447 * a hope that sizeof(long) wont become >8 any time soon.
449 size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS),
450 /*HOST_LONG_BITS*/ 64) / 8;
451 if (!d.dirty_bitmap) {
452 d.dirty_bitmap = g_malloc(size);
453 } else if (size > allocated_size) {
454 d.dirty_bitmap = g_realloc(d.dirty_bitmap, size);
456 allocated_size = size;
457 memset(d.dirty_bitmap, 0, allocated_size);
461 if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
462 DPRINTF("ioctl failed %d\n", errno);
467 kvm_get_dirty_pages_log_range(section, d.dirty_bitmap);
468 start_addr = mem->start_addr + mem->memory_size;
470 g_free(d.dirty_bitmap);
475 static void kvm_coalesce_mmio_region(MemoryListener *listener,
476 MemoryRegionSection *secion,
477 hwaddr start, hwaddr size)
479 KVMState *s = kvm_state;
481 if (s->coalesced_mmio) {
482 struct kvm_coalesced_mmio_zone zone;
488 (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
492 static void kvm_uncoalesce_mmio_region(MemoryListener *listener,
493 MemoryRegionSection *secion,
494 hwaddr start, hwaddr size)
496 KVMState *s = kvm_state;
498 if (s->coalesced_mmio) {
499 struct kvm_coalesced_mmio_zone zone;
505 (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
509 int kvm_check_extension(KVMState *s, unsigned int extension)
513 ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
521 static int kvm_set_ioeventfd_mmio(int fd, uint32_t addr, uint32_t val,
522 bool assign, uint32_t size, bool datamatch)
525 struct kvm_ioeventfd iofd;
527 iofd.datamatch = datamatch ? val : 0;
533 if (!kvm_enabled()) {
538 iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
541 iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
544 ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd);
553 static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val,
554 bool assign, uint32_t size, bool datamatch)
556 struct kvm_ioeventfd kick = {
557 .datamatch = datamatch ? val : 0,
559 .flags = KVM_IOEVENTFD_FLAG_PIO,
564 if (!kvm_enabled()) {
568 kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
571 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
573 r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
581 static int kvm_check_many_ioeventfds(void)
583 /* Userspace can use ioeventfd for io notification. This requires a host
584 * that supports eventfd(2) and an I/O thread; since eventfd does not
585 * support SIGIO it cannot interrupt the vcpu.
587 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
588 * can avoid creating too many ioeventfds.
590 #if defined(CONFIG_EVENTFD)
593 for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) {
594 ioeventfds[i] = eventfd(0, EFD_CLOEXEC);
595 if (ioeventfds[i] < 0) {
598 ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true);
600 close(ioeventfds[i]);
605 /* Decide whether many devices are supported or not */
606 ret = i == ARRAY_SIZE(ioeventfds);
609 kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true);
610 close(ioeventfds[i]);
618 static const KVMCapabilityInfo *
619 kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list)
622 if (!kvm_check_extension(s, list->value)) {
630 static void kvm_set_phys_mem(MemoryRegionSection *section, bool add)
632 KVMState *s = kvm_state;
635 MemoryRegion *mr = section->mr;
636 bool log_dirty = memory_region_is_logging(mr);
637 bool writeable = !mr->readonly && !mr->rom_device;
638 bool readonly_flag = mr->readonly || memory_region_is_romd(mr);
639 hwaddr start_addr = section->offset_within_address_space;
640 ram_addr_t size = int128_get64(section->size);
644 /* kvm works in page size chunks, but the function may be called
645 with sub-page size and unaligned start address. */
646 delta = TARGET_PAGE_ALIGN(size) - size;
652 size &= TARGET_PAGE_MASK;
653 if (!size || (start_addr & ~TARGET_PAGE_MASK)) {
657 if (!memory_region_is_ram(mr)) {
658 if (writeable || !kvm_readonly_mem_allowed) {
660 } else if (!mr->romd_mode) {
661 /* If the memory device is not in romd_mode, then we actually want
662 * to remove the kvm memory slot so all accesses will trap. */
667 ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta;
670 mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
675 if (add && start_addr >= mem->start_addr &&
676 (start_addr + size <= mem->start_addr + mem->memory_size) &&
677 (ram - start_addr == mem->ram - mem->start_addr)) {
678 /* The new slot fits into the existing one and comes with
679 * identical parameters - update flags and done. */
680 kvm_slot_dirty_pages_log_change(mem, log_dirty);
686 if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
687 kvm_physical_sync_dirty_bitmap(section);
690 /* unregister the overlapping slot */
691 mem->memory_size = 0;
692 err = kvm_set_user_memory_region(s, mem);
694 fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
695 __func__, strerror(-err));
699 /* Workaround for older KVM versions: we can't join slots, even not by
700 * unregistering the previous ones and then registering the larger
701 * slot. We have to maintain the existing fragmentation. Sigh.
703 * This workaround assumes that the new slot starts at the same
704 * address as the first existing one. If not or if some overlapping
705 * slot comes around later, we will fail (not seen in practice so far)
706 * - and actually require a recent KVM version. */
707 if (s->broken_set_mem_region &&
708 old.start_addr == start_addr && old.memory_size < size && add) {
709 mem = kvm_alloc_slot(s);
710 mem->memory_size = old.memory_size;
711 mem->start_addr = old.start_addr;
713 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
715 err = kvm_set_user_memory_region(s, mem);
717 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
722 start_addr += old.memory_size;
723 ram += old.memory_size;
724 size -= old.memory_size;
728 /* register prefix slot */
729 if (old.start_addr < start_addr) {
730 mem = kvm_alloc_slot(s);
731 mem->memory_size = start_addr - old.start_addr;
732 mem->start_addr = old.start_addr;
734 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
736 err = kvm_set_user_memory_region(s, mem);
738 fprintf(stderr, "%s: error registering prefix slot: %s\n",
739 __func__, strerror(-err));
741 fprintf(stderr, "%s: This is probably because your kernel's " \
742 "PAGE_SIZE is too big. Please try to use 4k " \
743 "PAGE_SIZE!\n", __func__);
749 /* register suffix slot */
750 if (old.start_addr + old.memory_size > start_addr + size) {
751 ram_addr_t size_delta;
753 mem = kvm_alloc_slot(s);
754 mem->start_addr = start_addr + size;
755 size_delta = mem->start_addr - old.start_addr;
756 mem->memory_size = old.memory_size - size_delta;
757 mem->ram = old.ram + size_delta;
758 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
760 err = kvm_set_user_memory_region(s, mem);
762 fprintf(stderr, "%s: error registering suffix slot: %s\n",
763 __func__, strerror(-err));
769 /* in case the KVM bug workaround already "consumed" the new slot */
776 mem = kvm_alloc_slot(s);
777 mem->memory_size = size;
778 mem->start_addr = start_addr;
780 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
782 err = kvm_set_user_memory_region(s, mem);
784 fprintf(stderr, "%s: error registering slot: %s\n", __func__,
790 static void kvm_region_add(MemoryListener *listener,
791 MemoryRegionSection *section)
793 memory_region_ref(section->mr);
794 kvm_set_phys_mem(section, true);
797 static void kvm_region_del(MemoryListener *listener,
798 MemoryRegionSection *section)
800 kvm_set_phys_mem(section, false);
801 memory_region_unref(section->mr);
804 static void kvm_log_sync(MemoryListener *listener,
805 MemoryRegionSection *section)
809 r = kvm_physical_sync_dirty_bitmap(section);
815 static void kvm_log_global_start(struct MemoryListener *listener)
819 r = kvm_set_migration_log(1);
823 static void kvm_log_global_stop(struct MemoryListener *listener)
827 r = kvm_set_migration_log(0);
831 static void kvm_mem_ioeventfd_add(MemoryListener *listener,
832 MemoryRegionSection *section,
833 bool match_data, uint64_t data,
836 int fd = event_notifier_get_fd(e);
839 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
840 data, true, int128_get64(section->size),
843 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
844 __func__, strerror(-r));
849 static void kvm_mem_ioeventfd_del(MemoryListener *listener,
850 MemoryRegionSection *section,
851 bool match_data, uint64_t data,
854 int fd = event_notifier_get_fd(e);
857 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
858 data, false, int128_get64(section->size),
865 static void kvm_io_ioeventfd_add(MemoryListener *listener,
866 MemoryRegionSection *section,
867 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, true, int128_get64(section->size),
877 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
878 __func__, strerror(-r));
883 static void kvm_io_ioeventfd_del(MemoryListener *listener,
884 MemoryRegionSection *section,
885 bool match_data, uint64_t data,
889 int fd = event_notifier_get_fd(e);
892 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
893 data, false, int128_get64(section->size),
900 static MemoryListener kvm_memory_listener = {
901 .region_add = kvm_region_add,
902 .region_del = kvm_region_del,
903 .log_start = kvm_log_start,
904 .log_stop = kvm_log_stop,
905 .log_sync = kvm_log_sync,
906 .log_global_start = kvm_log_global_start,
907 .log_global_stop = kvm_log_global_stop,
908 .eventfd_add = kvm_mem_ioeventfd_add,
909 .eventfd_del = kvm_mem_ioeventfd_del,
910 .coalesced_mmio_add = kvm_coalesce_mmio_region,
911 .coalesced_mmio_del = kvm_uncoalesce_mmio_region,
915 static MemoryListener kvm_io_listener = {
916 .eventfd_add = kvm_io_ioeventfd_add,
917 .eventfd_del = kvm_io_ioeventfd_del,
921 static void kvm_handle_interrupt(CPUState *cpu, int mask)
923 cpu->interrupt_request |= mask;
925 if (!qemu_cpu_is_self(cpu)) {
930 int kvm_set_irq(KVMState *s, int irq, int level)
932 struct kvm_irq_level event;
935 assert(kvm_async_interrupts_enabled());
939 ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event);
941 perror("kvm_set_irq");
945 return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status;
948 #ifdef KVM_CAP_IRQ_ROUTING
949 typedef struct KVMMSIRoute {
950 struct kvm_irq_routing_entry kroute;
951 QTAILQ_ENTRY(KVMMSIRoute) entry;
954 static void set_gsi(KVMState *s, unsigned int gsi)
956 s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32);
959 static void clear_gsi(KVMState *s, unsigned int gsi)
961 s->used_gsi_bitmap[gsi / 32] &= ~(1U << (gsi % 32));
964 void kvm_init_irq_routing(KVMState *s)
968 gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING);
970 unsigned int gsi_bits, i;
972 /* Round up so we can search ints using ffs */
973 gsi_bits = ALIGN(gsi_count, 32);
974 s->used_gsi_bitmap = g_malloc0(gsi_bits / 8);
975 s->gsi_count = gsi_count;
977 /* Mark any over-allocated bits as already in use */
978 for (i = gsi_count; i < gsi_bits; i++) {
983 s->irq_routes = g_malloc0(sizeof(*s->irq_routes));
984 s->nr_allocated_irq_routes = 0;
986 if (!s->direct_msi) {
987 for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) {
988 QTAILQ_INIT(&s->msi_hashtab[i]);
992 kvm_arch_init_irq_routing(s);
995 void kvm_irqchip_commit_routes(KVMState *s)
999 s->irq_routes->flags = 0;
1000 ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes);
1004 static void kvm_add_routing_entry(KVMState *s,
1005 struct kvm_irq_routing_entry *entry)
1007 struct kvm_irq_routing_entry *new;
1010 if (s->irq_routes->nr == s->nr_allocated_irq_routes) {
1011 n = s->nr_allocated_irq_routes * 2;
1015 size = sizeof(struct kvm_irq_routing);
1016 size += n * sizeof(*new);
1017 s->irq_routes = g_realloc(s->irq_routes, size);
1018 s->nr_allocated_irq_routes = n;
1020 n = s->irq_routes->nr++;
1021 new = &s->irq_routes->entries[n];
1025 set_gsi(s, entry->gsi);
1028 static int kvm_update_routing_entry(KVMState *s,
1029 struct kvm_irq_routing_entry *new_entry)
1031 struct kvm_irq_routing_entry *entry;
1034 for (n = 0; n < s->irq_routes->nr; n++) {
1035 entry = &s->irq_routes->entries[n];
1036 if (entry->gsi != new_entry->gsi) {
1040 if(!memcmp(entry, new_entry, sizeof *entry)) {
1044 *entry = *new_entry;
1046 kvm_irqchip_commit_routes(s);
1054 void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin)
1056 struct kvm_irq_routing_entry e = {};
1058 assert(pin < s->gsi_count);
1061 e.type = KVM_IRQ_ROUTING_IRQCHIP;
1063 e.u.irqchip.irqchip = irqchip;
1064 e.u.irqchip.pin = pin;
1065 kvm_add_routing_entry(s, &e);
1068 void kvm_irqchip_release_virq(KVMState *s, int virq)
1070 struct kvm_irq_routing_entry *e;
1073 if (kvm_gsi_direct_mapping()) {
1077 for (i = 0; i < s->irq_routes->nr; i++) {
1078 e = &s->irq_routes->entries[i];
1079 if (e->gsi == virq) {
1080 s->irq_routes->nr--;
1081 *e = s->irq_routes->entries[s->irq_routes->nr];
1087 static unsigned int kvm_hash_msi(uint32_t data)
1089 /* This is optimized for IA32 MSI layout. However, no other arch shall
1090 * repeat the mistake of not providing a direct MSI injection API. */
1094 static void kvm_flush_dynamic_msi_routes(KVMState *s)
1096 KVMMSIRoute *route, *next;
1099 for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) {
1100 QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) {
1101 kvm_irqchip_release_virq(s, route->kroute.gsi);
1102 QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry);
1108 static int kvm_irqchip_get_virq(KVMState *s)
1110 uint32_t *word = s->used_gsi_bitmap;
1111 int max_words = ALIGN(s->gsi_count, 32) / 32;
1116 /* Return the lowest unused GSI in the bitmap */
1117 for (i = 0; i < max_words; i++) {
1118 bit = ffs(~word[i]);
1123 return bit - 1 + i * 32;
1125 if (!s->direct_msi && retry) {
1127 kvm_flush_dynamic_msi_routes(s);
1134 static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg)
1136 unsigned int hash = kvm_hash_msi(msg.data);
1139 QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) {
1140 if (route->kroute.u.msi.address_lo == (uint32_t)msg.address &&
1141 route->kroute.u.msi.address_hi == (msg.address >> 32) &&
1142 route->kroute.u.msi.data == le32_to_cpu(msg.data)) {
1149 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1154 if (s->direct_msi) {
1155 msi.address_lo = (uint32_t)msg.address;
1156 msi.address_hi = msg.address >> 32;
1157 msi.data = le32_to_cpu(msg.data);
1159 memset(msi.pad, 0, sizeof(msi.pad));
1161 return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi);
1164 route = kvm_lookup_msi_route(s, msg);
1168 virq = kvm_irqchip_get_virq(s);
1173 route = g_malloc0(sizeof(KVMMSIRoute));
1174 route->kroute.gsi = virq;
1175 route->kroute.type = KVM_IRQ_ROUTING_MSI;
1176 route->kroute.flags = 0;
1177 route->kroute.u.msi.address_lo = (uint32_t)msg.address;
1178 route->kroute.u.msi.address_hi = msg.address >> 32;
1179 route->kroute.u.msi.data = le32_to_cpu(msg.data);
1181 kvm_add_routing_entry(s, &route->kroute);
1182 kvm_irqchip_commit_routes(s);
1184 QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route,
1188 assert(route->kroute.type == KVM_IRQ_ROUTING_MSI);
1190 return kvm_set_irq(s, route->kroute.gsi, 1);
1193 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1195 struct kvm_irq_routing_entry kroute = {};
1198 if (kvm_gsi_direct_mapping()) {
1199 return msg.data & 0xffff;
1202 if (!kvm_gsi_routing_enabled()) {
1206 virq = kvm_irqchip_get_virq(s);
1212 kroute.type = KVM_IRQ_ROUTING_MSI;
1214 kroute.u.msi.address_lo = (uint32_t)msg.address;
1215 kroute.u.msi.address_hi = msg.address >> 32;
1216 kroute.u.msi.data = le32_to_cpu(msg.data);
1218 kvm_add_routing_entry(s, &kroute);
1219 kvm_irqchip_commit_routes(s);
1224 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1226 struct kvm_irq_routing_entry kroute = {};
1228 if (kvm_gsi_direct_mapping()) {
1232 if (!kvm_irqchip_in_kernel()) {
1237 kroute.type = KVM_IRQ_ROUTING_MSI;
1239 kroute.u.msi.address_lo = (uint32_t)msg.address;
1240 kroute.u.msi.address_hi = msg.address >> 32;
1241 kroute.u.msi.data = le32_to_cpu(msg.data);
1243 return kvm_update_routing_entry(s, &kroute);
1246 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int rfd, int virq,
1249 struct kvm_irqfd irqfd = {
1252 .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN,
1256 irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE;
1257 irqfd.resamplefd = rfd;
1260 if (!kvm_irqfds_enabled()) {
1264 return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd);
1267 #else /* !KVM_CAP_IRQ_ROUTING */
1269 void kvm_init_irq_routing(KVMState *s)
1273 void kvm_irqchip_release_virq(KVMState *s, int virq)
1277 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1282 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1287 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign)
1292 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1296 #endif /* !KVM_CAP_IRQ_ROUTING */
1298 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n,
1299 EventNotifier *rn, int virq)
1301 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n),
1302 rn ? event_notifier_get_fd(rn) : -1, virq, true);
1305 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, int virq)
1307 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), -1, virq,
1311 static int kvm_irqchip_create(KVMState *s)
1315 if (!qemu_opt_get_bool(qemu_get_machine_opts(), "kernel_irqchip", true) ||
1316 !kvm_check_extension(s, KVM_CAP_IRQCHIP)) {
1320 ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP);
1322 fprintf(stderr, "Create kernel irqchip failed\n");
1326 kvm_kernel_irqchip = true;
1327 /* If we have an in-kernel IRQ chip then we must have asynchronous
1328 * interrupt delivery (though the reverse is not necessarily true)
1330 kvm_async_interrupts_allowed = true;
1331 kvm_halt_in_kernel_allowed = true;
1333 kvm_init_irq_routing(s);
1338 /* Find number of supported CPUs using the recommended
1339 * procedure from the kernel API documentation to cope with
1340 * older kernels that may be missing capabilities.
1342 static int kvm_recommended_vcpus(KVMState *s)
1344 int ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS);
1345 return (ret) ? ret : 4;
1348 static int kvm_max_vcpus(KVMState *s)
1350 int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS);
1351 return (ret) ? ret : kvm_recommended_vcpus(s);
1356 static const char upgrade_note[] =
1357 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1358 "(see http://sourceforge.net/projects/kvm).\n";
1363 { "SMP", smp_cpus },
1364 { "hotpluggable", max_cpus },
1367 int soft_vcpus_limit, hard_vcpus_limit;
1369 const KVMCapabilityInfo *missing_cap;
1373 s = g_malloc0(sizeof(KVMState));
1376 * On systems where the kernel can support different base page
1377 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1378 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1379 * page size for the system though.
1381 assert(TARGET_PAGE_SIZE <= getpagesize());
1383 #ifdef KVM_CAP_SET_GUEST_DEBUG
1384 QTAILQ_INIT(&s->kvm_sw_breakpoints);
1386 for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
1387 s->slots[i].slot = i;
1390 s->fd = qemu_open("/dev/kvm", O_RDWR);
1392 fprintf(stderr, "Could not access KVM kernel module: %m\n");
1397 ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
1398 if (ret < KVM_API_VERSION) {
1402 fprintf(stderr, "kvm version too old\n");
1406 if (ret > KVM_API_VERSION) {
1408 fprintf(stderr, "kvm version not supported\n");
1412 /* check the vcpu limits */
1413 soft_vcpus_limit = kvm_recommended_vcpus(s);
1414 hard_vcpus_limit = kvm_max_vcpus(s);
1417 if (nc->num > soft_vcpus_limit) {
1419 "Warning: Number of %s cpus requested (%d) exceeds "
1420 "the recommended cpus supported by KVM (%d)\n",
1421 nc->name, nc->num, soft_vcpus_limit);
1423 if (nc->num > hard_vcpus_limit) {
1425 fprintf(stderr, "Number of %s cpus requested (%d) exceeds "
1426 "the maximum cpus supported by KVM (%d)\n",
1427 nc->name, nc->num, hard_vcpus_limit);
1434 s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0);
1437 fprintf(stderr, "Please add the 'switch_amode' kernel parameter to "
1438 "your host kernel command line\n");
1444 missing_cap = kvm_check_extension_list(s, kvm_required_capabilites);
1447 kvm_check_extension_list(s, kvm_arch_required_capabilities);
1451 fprintf(stderr, "kvm does not support %s\n%s",
1452 missing_cap->name, upgrade_note);
1456 s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
1458 s->broken_set_mem_region = 1;
1459 ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
1461 s->broken_set_mem_region = 0;
1464 #ifdef KVM_CAP_VCPU_EVENTS
1465 s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS);
1468 s->robust_singlestep =
1469 kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP);
1471 #ifdef KVM_CAP_DEBUGREGS
1472 s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS);
1475 #ifdef KVM_CAP_XSAVE
1476 s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE);
1480 s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS);
1483 #ifdef KVM_CAP_PIT_STATE2
1484 s->pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2);
1487 #ifdef KVM_CAP_IRQ_ROUTING
1488 s->direct_msi = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0);
1491 s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3);
1493 s->irq_set_ioctl = KVM_IRQ_LINE;
1494 if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) {
1495 s->irq_set_ioctl = KVM_IRQ_LINE_STATUS;
1498 #ifdef KVM_CAP_READONLY_MEM
1499 kvm_readonly_mem_allowed =
1500 (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0);
1503 ret = kvm_arch_init(s);
1508 ret = kvm_irqchip_create(s);
1514 memory_listener_register(&kvm_memory_listener, &address_space_memory);
1515 memory_listener_register(&kvm_io_listener, &address_space_io);
1517 s->many_ioeventfds = kvm_check_many_ioeventfds();
1519 cpu_interrupt_handler = kvm_handle_interrupt;
1535 static void kvm_handle_io(uint16_t port, void *data, int direction, int size,
1539 uint8_t *ptr = data;
1541 for (i = 0; i < count; i++) {
1542 address_space_rw(&address_space_io, port, ptr, size,
1543 direction == KVM_EXIT_IO_OUT);
1548 static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run)
1550 fprintf(stderr, "KVM internal error.");
1551 if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) {
1554 fprintf(stderr, " Suberror: %d\n", run->internal.suberror);
1555 for (i = 0; i < run->internal.ndata; ++i) {
1556 fprintf(stderr, "extra data[%d]: %"PRIx64"\n",
1557 i, (uint64_t)run->internal.data[i]);
1560 fprintf(stderr, "\n");
1562 if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) {
1563 fprintf(stderr, "emulation failure\n");
1564 if (!kvm_arch_stop_on_emulation_error(cpu)) {
1565 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1566 return EXCP_INTERRUPT;
1569 /* FIXME: Should trigger a qmp message to let management know
1570 * something went wrong.
1575 void kvm_flush_coalesced_mmio_buffer(void)
1577 KVMState *s = kvm_state;
1579 if (s->coalesced_flush_in_progress) {
1583 s->coalesced_flush_in_progress = true;
1585 if (s->coalesced_mmio_ring) {
1586 struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring;
1587 while (ring->first != ring->last) {
1588 struct kvm_coalesced_mmio *ent;
1590 ent = &ring->coalesced_mmio[ring->first];
1592 cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
1594 ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
1598 s->coalesced_flush_in_progress = false;
1601 static void do_kvm_cpu_synchronize_state(void *arg)
1603 CPUState *cpu = arg;
1605 if (!cpu->kvm_vcpu_dirty) {
1606 kvm_arch_get_registers(cpu);
1607 cpu->kvm_vcpu_dirty = true;
1611 void kvm_cpu_synchronize_state(CPUState *cpu)
1613 if (!cpu->kvm_vcpu_dirty) {
1614 run_on_cpu(cpu, do_kvm_cpu_synchronize_state, cpu);
1618 void kvm_cpu_synchronize_post_reset(CPUState *cpu)
1620 kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE);
1621 cpu->kvm_vcpu_dirty = false;
1624 void kvm_cpu_synchronize_post_init(CPUState *cpu)
1626 kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE);
1627 cpu->kvm_vcpu_dirty = false;
1630 int kvm_cpu_exec(CPUState *cpu)
1632 struct kvm_run *run = cpu->kvm_run;
1635 DPRINTF("kvm_cpu_exec()\n");
1637 if (kvm_arch_process_async_events(cpu)) {
1638 cpu->exit_request = 0;
1643 if (cpu->kvm_vcpu_dirty) {
1644 kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE);
1645 cpu->kvm_vcpu_dirty = false;
1648 kvm_arch_pre_run(cpu, run);
1649 if (cpu->exit_request) {
1650 DPRINTF("interrupt exit requested\n");
1652 * KVM requires us to reenter the kernel after IO exits to complete
1653 * instruction emulation. This self-signal will ensure that we
1656 qemu_cpu_kick_self();
1658 qemu_mutex_unlock_iothread();
1660 run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0);
1662 qemu_mutex_lock_iothread();
1663 kvm_arch_post_run(cpu, run);
1666 if (run_ret == -EINTR || run_ret == -EAGAIN) {
1667 DPRINTF("io window exit\n");
1668 ret = EXCP_INTERRUPT;
1671 fprintf(stderr, "error: kvm run failed %s\n",
1672 strerror(-run_ret));
1676 trace_kvm_run_exit(cpu->cpu_index, run->exit_reason);
1677 switch (run->exit_reason) {
1679 DPRINTF("handle_io\n");
1680 kvm_handle_io(run->io.port,
1681 (uint8_t *)run + run->io.data_offset,
1688 DPRINTF("handle_mmio\n");
1689 cpu_physical_memory_rw(run->mmio.phys_addr,
1692 run->mmio.is_write);
1695 case KVM_EXIT_IRQ_WINDOW_OPEN:
1696 DPRINTF("irq_window_open\n");
1697 ret = EXCP_INTERRUPT;
1699 case KVM_EXIT_SHUTDOWN:
1700 DPRINTF("shutdown\n");
1701 qemu_system_reset_request();
1702 ret = EXCP_INTERRUPT;
1704 case KVM_EXIT_UNKNOWN:
1705 fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n",
1706 (uint64_t)run->hw.hardware_exit_reason);
1709 case KVM_EXIT_INTERNAL_ERROR:
1710 ret = kvm_handle_internal_error(cpu, run);
1713 DPRINTF("kvm_arch_handle_exit\n");
1714 ret = kvm_arch_handle_exit(cpu, run);
1720 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1721 vm_stop(RUN_STATE_INTERNAL_ERROR);
1724 cpu->exit_request = 0;
1728 int kvm_ioctl(KVMState *s, int type, ...)
1735 arg = va_arg(ap, void *);
1738 trace_kvm_ioctl(type, arg);
1739 ret = ioctl(s->fd, type, arg);
1746 int kvm_vm_ioctl(KVMState *s, int type, ...)
1753 arg = va_arg(ap, void *);
1756 trace_kvm_vm_ioctl(type, arg);
1757 ret = ioctl(s->vmfd, type, arg);
1764 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...)
1771 arg = va_arg(ap, void *);
1774 trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg);
1775 ret = ioctl(cpu->kvm_fd, type, arg);
1782 int kvm_has_sync_mmu(void)
1784 return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU);
1787 int kvm_has_vcpu_events(void)
1789 return kvm_state->vcpu_events;
1792 int kvm_has_robust_singlestep(void)
1794 return kvm_state->robust_singlestep;
1797 int kvm_has_debugregs(void)
1799 return kvm_state->debugregs;
1802 int kvm_has_xsave(void)
1804 return kvm_state->xsave;
1807 int kvm_has_xcrs(void)
1809 return kvm_state->xcrs;
1812 int kvm_has_pit_state2(void)
1814 return kvm_state->pit_state2;
1817 int kvm_has_many_ioeventfds(void)
1819 if (!kvm_enabled()) {
1822 return kvm_state->many_ioeventfds;
1825 int kvm_has_gsi_routing(void)
1827 #ifdef KVM_CAP_IRQ_ROUTING
1828 return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING);
1834 int kvm_has_intx_set_mask(void)
1836 return kvm_state->intx_set_mask;
1839 void *kvm_ram_alloc(ram_addr_t size)
1844 mem = kvm_arch_ram_alloc(size);
1849 return qemu_anon_ram_alloc(size);
1852 void kvm_setup_guest_memory(void *start, size_t size)
1854 #ifdef CONFIG_VALGRIND_H
1855 VALGRIND_MAKE_MEM_DEFINED(start, size);
1857 if (!kvm_has_sync_mmu()) {
1858 int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK);
1861 perror("qemu_madvise");
1863 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1869 #ifdef KVM_CAP_SET_GUEST_DEBUG
1870 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu,
1873 struct kvm_sw_breakpoint *bp;
1875 QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) {
1883 int kvm_sw_breakpoints_active(CPUState *cpu)
1885 return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints);
1888 struct kvm_set_guest_debug_data {
1889 struct kvm_guest_debug dbg;
1894 static void kvm_invoke_set_guest_debug(void *data)
1896 struct kvm_set_guest_debug_data *dbg_data = data;
1898 dbg_data->err = kvm_vcpu_ioctl(dbg_data->cpu, KVM_SET_GUEST_DEBUG,
1902 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
1904 struct kvm_set_guest_debug_data data;
1906 data.dbg.control = reinject_trap;
1908 if (cpu->singlestep_enabled) {
1909 data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
1911 kvm_arch_update_guest_debug(cpu, &data.dbg);
1914 run_on_cpu(cpu, kvm_invoke_set_guest_debug, &data);
1918 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
1919 target_ulong len, int type)
1921 struct kvm_sw_breakpoint *bp;
1924 if (type == GDB_BREAKPOINT_SW) {
1925 bp = kvm_find_sw_breakpoint(cpu, addr);
1931 bp = g_malloc(sizeof(struct kvm_sw_breakpoint));
1938 err = kvm_arch_insert_sw_breakpoint(cpu, bp);
1944 QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
1946 err = kvm_arch_insert_hw_breakpoint(addr, len, type);
1953 err = kvm_update_guest_debug(cpu, 0);
1961 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
1962 target_ulong len, int type)
1964 struct kvm_sw_breakpoint *bp;
1967 if (type == GDB_BREAKPOINT_SW) {
1968 bp = kvm_find_sw_breakpoint(cpu, addr);
1973 if (bp->use_count > 1) {
1978 err = kvm_arch_remove_sw_breakpoint(cpu, bp);
1983 QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
1986 err = kvm_arch_remove_hw_breakpoint(addr, len, type);
1993 err = kvm_update_guest_debug(cpu, 0);
2001 void kvm_remove_all_breakpoints(CPUState *cpu)
2003 struct kvm_sw_breakpoint *bp, *next;
2004 KVMState *s = cpu->kvm_state;
2006 QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
2007 if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) {
2008 /* Try harder to find a CPU that currently sees the breakpoint. */
2010 if (kvm_arch_remove_sw_breakpoint(cpu, bp) == 0) {
2015 QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry);
2018 kvm_arch_remove_all_hw_breakpoints();
2021 kvm_update_guest_debug(cpu, 0);
2025 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2027 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2032 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2033 target_ulong len, int type)
2038 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2039 target_ulong len, int type)
2044 void kvm_remove_all_breakpoints(CPUState *cpu)
2047 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2049 int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset)
2051 struct kvm_signal_mask *sigmask;
2055 return kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, NULL);
2058 sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset));
2061 memcpy(sigmask->sigset, sigset, sizeof(*sigset));
2062 r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask);
2067 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2069 return kvm_arch_on_sigbus_vcpu(cpu, code, addr);
2072 int kvm_on_sigbus(int code, void *addr)
2074 return kvm_arch_on_sigbus(code, addr);