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_irqfds_allowed;
111 bool kvm_msi_via_irqfd_allowed;
112 bool kvm_gsi_routing_allowed;
114 bool kvm_readonly_mem_allowed;
116 static const KVMCapabilityInfo kvm_required_capabilites[] = {
117 KVM_CAP_INFO(USER_MEMORY),
118 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS),
122 static KVMSlot *kvm_alloc_slot(KVMState *s)
126 for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
127 if (s->slots[i].memory_size == 0) {
132 fprintf(stderr, "%s: no free slot available\n", __func__);
136 static KVMSlot *kvm_lookup_matching_slot(KVMState *s,
142 for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
143 KVMSlot *mem = &s->slots[i];
145 if (start_addr == mem->start_addr &&
146 end_addr == mem->start_addr + mem->memory_size) {
155 * Find overlapping slot with lowest start address
157 static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s,
161 KVMSlot *found = NULL;
164 for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
165 KVMSlot *mem = &s->slots[i];
167 if (mem->memory_size == 0 ||
168 (found && found->start_addr < mem->start_addr)) {
172 if (end_addr > mem->start_addr &&
173 start_addr < mem->start_addr + mem->memory_size) {
181 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram,
186 for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
187 KVMSlot *mem = &s->slots[i];
189 if (ram >= mem->ram && ram < mem->ram + mem->memory_size) {
190 *phys_addr = mem->start_addr + (ram - mem->ram);
198 static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot)
200 struct kvm_userspace_memory_region mem;
202 mem.slot = slot->slot;
203 mem.guest_phys_addr = slot->start_addr;
204 mem.userspace_addr = (unsigned long)slot->ram;
205 mem.flags = slot->flags;
206 if (s->migration_log) {
207 mem.flags |= KVM_MEM_LOG_DIRTY_PAGES;
210 if (slot->memory_size && mem.flags & KVM_MEM_READONLY) {
211 /* Set the slot size to 0 before setting the slot to the desired
212 * value. This is needed based on KVM commit 75d61fbc. */
214 kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
216 mem.memory_size = slot->memory_size;
217 return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
220 static void kvm_reset_vcpu(void *opaque)
222 CPUState *cpu = opaque;
224 kvm_arch_reset_vcpu(cpu);
227 int kvm_init_vcpu(CPUState *cpu)
229 KVMState *s = kvm_state;
233 DPRINTF("kvm_init_vcpu\n");
235 ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)kvm_arch_vcpu_id(cpu));
237 DPRINTF("kvm_create_vcpu failed\n");
243 cpu->kvm_vcpu_dirty = true;
245 mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
248 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
252 cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
254 if (cpu->kvm_run == MAP_FAILED) {
256 DPRINTF("mmap'ing vcpu state failed\n");
260 if (s->coalesced_mmio && !s->coalesced_mmio_ring) {
261 s->coalesced_mmio_ring =
262 (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE;
265 ret = kvm_arch_init_vcpu(cpu);
267 qemu_register_reset(kvm_reset_vcpu, cpu);
268 kvm_arch_reset_vcpu(cpu);
275 * dirty pages logging control
278 static int kvm_mem_flags(KVMState *s, bool log_dirty, bool readonly)
281 flags = log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0;
282 if (readonly && kvm_readonly_mem_allowed) {
283 flags |= KVM_MEM_READONLY;
288 static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty)
290 KVMState *s = kvm_state;
291 int flags, mask = KVM_MEM_LOG_DIRTY_PAGES;
294 old_flags = mem->flags;
296 flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty, false);
299 /* If nothing changed effectively, no need to issue ioctl */
300 if (s->migration_log) {
301 flags |= KVM_MEM_LOG_DIRTY_PAGES;
304 if (flags == old_flags) {
308 return kvm_set_user_memory_region(s, mem);
311 static int kvm_dirty_pages_log_change(hwaddr phys_addr,
312 ram_addr_t size, bool log_dirty)
314 KVMState *s = kvm_state;
315 KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size);
318 fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-"
319 TARGET_FMT_plx "\n", __func__, phys_addr,
320 (hwaddr)(phys_addr + size - 1));
323 return kvm_slot_dirty_pages_log_change(mem, log_dirty);
326 static void kvm_log_start(MemoryListener *listener,
327 MemoryRegionSection *section)
331 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
332 int128_get64(section->size), true);
338 static void kvm_log_stop(MemoryListener *listener,
339 MemoryRegionSection *section)
343 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
344 int128_get64(section->size), false);
350 static int kvm_set_migration_log(int enable)
352 KVMState *s = kvm_state;
356 s->migration_log = enable;
358 for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
361 if (!mem->memory_size) {
364 if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) {
367 err = kvm_set_user_memory_region(s, mem);
375 /* get kvm's dirty pages bitmap and update qemu's */
376 static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section,
377 unsigned long *bitmap)
380 unsigned long page_number, c;
382 unsigned int pages = int128_get64(section->size) / getpagesize();
383 unsigned int len = (pages + HOST_LONG_BITS - 1) / HOST_LONG_BITS;
384 unsigned long hpratio = getpagesize() / TARGET_PAGE_SIZE;
387 * bitmap-traveling is faster than memory-traveling (for addr...)
388 * especially when most of the memory is not dirty.
390 for (i = 0; i < len; i++) {
391 if (bitmap[i] != 0) {
392 c = leul_to_cpu(bitmap[i]);
396 page_number = (i * HOST_LONG_BITS + j) * hpratio;
397 addr1 = page_number * TARGET_PAGE_SIZE;
398 addr = section->offset_within_region + addr1;
399 memory_region_set_dirty(section->mr, addr,
400 TARGET_PAGE_SIZE * hpratio);
407 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
410 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
411 * This function updates qemu's dirty bitmap using
412 * memory_region_set_dirty(). This means all bits are set
415 * @start_add: start of logged region.
416 * @end_addr: end of logged region.
418 static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection *section)
420 KVMState *s = kvm_state;
421 unsigned long size, allocated_size = 0;
425 hwaddr start_addr = section->offset_within_address_space;
426 hwaddr end_addr = start_addr + int128_get64(section->size);
428 d.dirty_bitmap = NULL;
429 while (start_addr < end_addr) {
430 mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr);
435 /* XXX bad kernel interface alert
436 * For dirty bitmap, kernel allocates array of size aligned to
437 * bits-per-long. But for case when the kernel is 64bits and
438 * the userspace is 32bits, userspace can't align to the same
439 * bits-per-long, since sizeof(long) is different between kernel
440 * and user space. This way, userspace will provide buffer which
441 * may be 4 bytes less than the kernel will use, resulting in
442 * userspace memory corruption (which is not detectable by valgrind
443 * too, in most cases).
444 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
445 * a hope that sizeof(long) wont become >8 any time soon.
447 size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS),
448 /*HOST_LONG_BITS*/ 64) / 8;
449 if (!d.dirty_bitmap) {
450 d.dirty_bitmap = g_malloc(size);
451 } else if (size > allocated_size) {
452 d.dirty_bitmap = g_realloc(d.dirty_bitmap, size);
454 allocated_size = size;
455 memset(d.dirty_bitmap, 0, allocated_size);
459 if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
460 DPRINTF("ioctl failed %d\n", errno);
465 kvm_get_dirty_pages_log_range(section, d.dirty_bitmap);
466 start_addr = mem->start_addr + mem->memory_size;
468 g_free(d.dirty_bitmap);
473 static void kvm_coalesce_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_REGISTER_COALESCED_MMIO, &zone);
490 static void kvm_uncoalesce_mmio_region(MemoryListener *listener,
491 MemoryRegionSection *secion,
492 hwaddr start, hwaddr size)
494 KVMState *s = kvm_state;
496 if (s->coalesced_mmio) {
497 struct kvm_coalesced_mmio_zone zone;
503 (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
507 int kvm_check_extension(KVMState *s, unsigned int extension)
511 ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
519 static int kvm_set_ioeventfd_mmio(int fd, uint32_t addr, uint32_t val,
520 bool assign, uint32_t size, bool datamatch)
523 struct kvm_ioeventfd iofd;
525 iofd.datamatch = datamatch ? val : 0;
531 if (!kvm_enabled()) {
536 iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
539 iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
542 ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd);
551 static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val,
552 bool assign, uint32_t size, bool datamatch)
554 struct kvm_ioeventfd kick = {
555 .datamatch = datamatch ? val : 0,
557 .flags = KVM_IOEVENTFD_FLAG_PIO,
562 if (!kvm_enabled()) {
566 kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
569 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
571 r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
579 static int kvm_check_many_ioeventfds(void)
581 /* Userspace can use ioeventfd for io notification. This requires a host
582 * that supports eventfd(2) and an I/O thread; since eventfd does not
583 * support SIGIO it cannot interrupt the vcpu.
585 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
586 * can avoid creating too many ioeventfds.
588 #if defined(CONFIG_EVENTFD)
591 for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) {
592 ioeventfds[i] = eventfd(0, EFD_CLOEXEC);
593 if (ioeventfds[i] < 0) {
596 ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true);
598 close(ioeventfds[i]);
603 /* Decide whether many devices are supported or not */
604 ret = i == ARRAY_SIZE(ioeventfds);
607 kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true);
608 close(ioeventfds[i]);
616 static const KVMCapabilityInfo *
617 kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list)
620 if (!kvm_check_extension(s, list->value)) {
628 static void kvm_set_phys_mem(MemoryRegionSection *section, bool add)
630 KVMState *s = kvm_state;
633 MemoryRegion *mr = section->mr;
634 bool log_dirty = memory_region_is_logging(mr);
635 bool writeable = !mr->readonly && !mr->rom_device;
636 bool readonly_flag = mr->readonly || memory_region_is_romd(mr);
637 hwaddr start_addr = section->offset_within_address_space;
638 ram_addr_t size = int128_get64(section->size);
642 /* kvm works in page size chunks, but the function may be called
643 with sub-page size and unaligned start address. */
644 delta = TARGET_PAGE_ALIGN(size) - size;
650 size &= TARGET_PAGE_MASK;
651 if (!size || (start_addr & ~TARGET_PAGE_MASK)) {
655 if (!memory_region_is_ram(mr)) {
656 if (writeable || !kvm_readonly_mem_allowed) {
658 } else if (!mr->romd_mode) {
659 /* If the memory device is not in romd_mode, then we actually want
660 * to remove the kvm memory slot so all accesses will trap. */
665 ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta;
668 mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
673 if (add && start_addr >= mem->start_addr &&
674 (start_addr + size <= mem->start_addr + mem->memory_size) &&
675 (ram - start_addr == mem->ram - mem->start_addr)) {
676 /* The new slot fits into the existing one and comes with
677 * identical parameters - update flags and done. */
678 kvm_slot_dirty_pages_log_change(mem, log_dirty);
684 if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
685 kvm_physical_sync_dirty_bitmap(section);
688 /* unregister the overlapping slot */
689 mem->memory_size = 0;
690 err = kvm_set_user_memory_region(s, mem);
692 fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
693 __func__, strerror(-err));
697 /* Workaround for older KVM versions: we can't join slots, even not by
698 * unregistering the previous ones and then registering the larger
699 * slot. We have to maintain the existing fragmentation. Sigh.
701 * This workaround assumes that the new slot starts at the same
702 * address as the first existing one. If not or if some overlapping
703 * slot comes around later, we will fail (not seen in practice so far)
704 * - and actually require a recent KVM version. */
705 if (s->broken_set_mem_region &&
706 old.start_addr == start_addr && old.memory_size < size && add) {
707 mem = kvm_alloc_slot(s);
708 mem->memory_size = old.memory_size;
709 mem->start_addr = old.start_addr;
711 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
713 err = kvm_set_user_memory_region(s, mem);
715 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
720 start_addr += old.memory_size;
721 ram += old.memory_size;
722 size -= old.memory_size;
726 /* register prefix slot */
727 if (old.start_addr < start_addr) {
728 mem = kvm_alloc_slot(s);
729 mem->memory_size = start_addr - old.start_addr;
730 mem->start_addr = old.start_addr;
732 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
734 err = kvm_set_user_memory_region(s, mem);
736 fprintf(stderr, "%s: error registering prefix slot: %s\n",
737 __func__, strerror(-err));
739 fprintf(stderr, "%s: This is probably because your kernel's " \
740 "PAGE_SIZE is too big. Please try to use 4k " \
741 "PAGE_SIZE!\n", __func__);
747 /* register suffix slot */
748 if (old.start_addr + old.memory_size > start_addr + size) {
749 ram_addr_t size_delta;
751 mem = kvm_alloc_slot(s);
752 mem->start_addr = start_addr + size;
753 size_delta = mem->start_addr - old.start_addr;
754 mem->memory_size = old.memory_size - size_delta;
755 mem->ram = old.ram + size_delta;
756 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
758 err = kvm_set_user_memory_region(s, mem);
760 fprintf(stderr, "%s: error registering suffix slot: %s\n",
761 __func__, strerror(-err));
767 /* in case the KVM bug workaround already "consumed" the new slot */
774 mem = kvm_alloc_slot(s);
775 mem->memory_size = size;
776 mem->start_addr = start_addr;
778 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
780 err = kvm_set_user_memory_region(s, mem);
782 fprintf(stderr, "%s: error registering slot: %s\n", __func__,
788 static void kvm_region_add(MemoryListener *listener,
789 MemoryRegionSection *section)
791 kvm_set_phys_mem(section, true);
794 static void kvm_region_del(MemoryListener *listener,
795 MemoryRegionSection *section)
797 kvm_set_phys_mem(section, false);
800 static void kvm_log_sync(MemoryListener *listener,
801 MemoryRegionSection *section)
805 r = kvm_physical_sync_dirty_bitmap(section);
811 static void kvm_log_global_start(struct MemoryListener *listener)
815 r = kvm_set_migration_log(1);
819 static void kvm_log_global_stop(struct MemoryListener *listener)
823 r = kvm_set_migration_log(0);
827 static void kvm_mem_ioeventfd_add(MemoryListener *listener,
828 MemoryRegionSection *section,
829 bool match_data, uint64_t data,
832 int fd = event_notifier_get_fd(e);
835 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
836 data, true, int128_get64(section->size),
843 static void kvm_mem_ioeventfd_del(MemoryListener *listener,
844 MemoryRegionSection *section,
845 bool match_data, uint64_t data,
848 int fd = event_notifier_get_fd(e);
851 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
852 data, false, int128_get64(section->size),
859 static void kvm_io_ioeventfd_add(MemoryListener *listener,
860 MemoryRegionSection *section,
861 bool match_data, uint64_t data,
864 int fd = event_notifier_get_fd(e);
867 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
868 data, true, int128_get64(section->size),
875 static void kvm_io_ioeventfd_del(MemoryListener *listener,
876 MemoryRegionSection *section,
877 bool match_data, uint64_t data,
881 int fd = event_notifier_get_fd(e);
884 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
885 data, false, int128_get64(section->size),
892 static MemoryListener kvm_memory_listener = {
893 .region_add = kvm_region_add,
894 .region_del = kvm_region_del,
895 .log_start = kvm_log_start,
896 .log_stop = kvm_log_stop,
897 .log_sync = kvm_log_sync,
898 .log_global_start = kvm_log_global_start,
899 .log_global_stop = kvm_log_global_stop,
900 .eventfd_add = kvm_mem_ioeventfd_add,
901 .eventfd_del = kvm_mem_ioeventfd_del,
902 .coalesced_mmio_add = kvm_coalesce_mmio_region,
903 .coalesced_mmio_del = kvm_uncoalesce_mmio_region,
907 static MemoryListener kvm_io_listener = {
908 .eventfd_add = kvm_io_ioeventfd_add,
909 .eventfd_del = kvm_io_ioeventfd_del,
913 static void kvm_handle_interrupt(CPUState *cpu, int mask)
915 cpu->interrupt_request |= mask;
917 if (!qemu_cpu_is_self(cpu)) {
922 int kvm_set_irq(KVMState *s, int irq, int level)
924 struct kvm_irq_level event;
927 assert(kvm_async_interrupts_enabled());
931 ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event);
933 perror("kvm_set_irq");
937 return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status;
940 #ifdef KVM_CAP_IRQ_ROUTING
941 typedef struct KVMMSIRoute {
942 struct kvm_irq_routing_entry kroute;
943 QTAILQ_ENTRY(KVMMSIRoute) entry;
946 static void set_gsi(KVMState *s, unsigned int gsi)
948 s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32);
951 static void clear_gsi(KVMState *s, unsigned int gsi)
953 s->used_gsi_bitmap[gsi / 32] &= ~(1U << (gsi % 32));
956 static void kvm_init_irq_routing(KVMState *s)
960 gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING);
962 unsigned int gsi_bits, i;
964 /* Round up so we can search ints using ffs */
965 gsi_bits = ALIGN(gsi_count, 32);
966 s->used_gsi_bitmap = g_malloc0(gsi_bits / 8);
967 s->gsi_count = gsi_count;
969 /* Mark any over-allocated bits as already in use */
970 for (i = gsi_count; i < gsi_bits; i++) {
975 s->irq_routes = g_malloc0(sizeof(*s->irq_routes));
976 s->nr_allocated_irq_routes = 0;
978 if (!s->direct_msi) {
979 for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) {
980 QTAILQ_INIT(&s->msi_hashtab[i]);
984 kvm_arch_init_irq_routing(s);
987 static void kvm_irqchip_commit_routes(KVMState *s)
991 s->irq_routes->flags = 0;
992 ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes);
996 static void kvm_add_routing_entry(KVMState *s,
997 struct kvm_irq_routing_entry *entry)
999 struct kvm_irq_routing_entry *new;
1002 if (s->irq_routes->nr == s->nr_allocated_irq_routes) {
1003 n = s->nr_allocated_irq_routes * 2;
1007 size = sizeof(struct kvm_irq_routing);
1008 size += n * sizeof(*new);
1009 s->irq_routes = g_realloc(s->irq_routes, size);
1010 s->nr_allocated_irq_routes = n;
1012 n = s->irq_routes->nr++;
1013 new = &s->irq_routes->entries[n];
1014 memset(new, 0, sizeof(*new));
1015 new->gsi = entry->gsi;
1016 new->type = entry->type;
1017 new->flags = entry->flags;
1020 set_gsi(s, entry->gsi);
1022 kvm_irqchip_commit_routes(s);
1025 static int kvm_update_routing_entry(KVMState *s,
1026 struct kvm_irq_routing_entry *new_entry)
1028 struct kvm_irq_routing_entry *entry;
1031 for (n = 0; n < s->irq_routes->nr; n++) {
1032 entry = &s->irq_routes->entries[n];
1033 if (entry->gsi != new_entry->gsi) {
1037 entry->type = new_entry->type;
1038 entry->flags = new_entry->flags;
1039 entry->u = new_entry->u;
1041 kvm_irqchip_commit_routes(s);
1049 void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin)
1051 struct kvm_irq_routing_entry e;
1053 assert(pin < s->gsi_count);
1056 e.type = KVM_IRQ_ROUTING_IRQCHIP;
1058 e.u.irqchip.irqchip = irqchip;
1059 e.u.irqchip.pin = pin;
1060 kvm_add_routing_entry(s, &e);
1063 void kvm_irqchip_release_virq(KVMState *s, int virq)
1065 struct kvm_irq_routing_entry *e;
1068 for (i = 0; i < s->irq_routes->nr; i++) {
1069 e = &s->irq_routes->entries[i];
1070 if (e->gsi == virq) {
1071 s->irq_routes->nr--;
1072 *e = s->irq_routes->entries[s->irq_routes->nr];
1078 static unsigned int kvm_hash_msi(uint32_t data)
1080 /* This is optimized for IA32 MSI layout. However, no other arch shall
1081 * repeat the mistake of not providing a direct MSI injection API. */
1085 static void kvm_flush_dynamic_msi_routes(KVMState *s)
1087 KVMMSIRoute *route, *next;
1090 for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) {
1091 QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) {
1092 kvm_irqchip_release_virq(s, route->kroute.gsi);
1093 QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry);
1099 static int kvm_irqchip_get_virq(KVMState *s)
1101 uint32_t *word = s->used_gsi_bitmap;
1102 int max_words = ALIGN(s->gsi_count, 32) / 32;
1107 /* Return the lowest unused GSI in the bitmap */
1108 for (i = 0; i < max_words; i++) {
1109 bit = ffs(~word[i]);
1114 return bit - 1 + i * 32;
1116 if (!s->direct_msi && retry) {
1118 kvm_flush_dynamic_msi_routes(s);
1125 static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg)
1127 unsigned int hash = kvm_hash_msi(msg.data);
1130 QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) {
1131 if (route->kroute.u.msi.address_lo == (uint32_t)msg.address &&
1132 route->kroute.u.msi.address_hi == (msg.address >> 32) &&
1133 route->kroute.u.msi.data == msg.data) {
1140 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1145 if (s->direct_msi) {
1146 msi.address_lo = (uint32_t)msg.address;
1147 msi.address_hi = msg.address >> 32;
1148 msi.data = msg.data;
1150 memset(msi.pad, 0, sizeof(msi.pad));
1152 return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi);
1155 route = kvm_lookup_msi_route(s, msg);
1159 virq = kvm_irqchip_get_virq(s);
1164 route = g_malloc(sizeof(KVMMSIRoute));
1165 route->kroute.gsi = virq;
1166 route->kroute.type = KVM_IRQ_ROUTING_MSI;
1167 route->kroute.flags = 0;
1168 route->kroute.u.msi.address_lo = (uint32_t)msg.address;
1169 route->kroute.u.msi.address_hi = msg.address >> 32;
1170 route->kroute.u.msi.data = msg.data;
1172 kvm_add_routing_entry(s, &route->kroute);
1174 QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route,
1178 assert(route->kroute.type == KVM_IRQ_ROUTING_MSI);
1180 return kvm_set_irq(s, route->kroute.gsi, 1);
1183 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1185 struct kvm_irq_routing_entry kroute;
1188 if (!kvm_gsi_routing_enabled()) {
1192 virq = kvm_irqchip_get_virq(s);
1198 kroute.type = KVM_IRQ_ROUTING_MSI;
1200 kroute.u.msi.address_lo = (uint32_t)msg.address;
1201 kroute.u.msi.address_hi = msg.address >> 32;
1202 kroute.u.msi.data = msg.data;
1204 kvm_add_routing_entry(s, &kroute);
1209 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1211 struct kvm_irq_routing_entry kroute;
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 = msg.data;
1224 return kvm_update_routing_entry(s, &kroute);
1227 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign)
1229 struct kvm_irqfd irqfd = {
1232 .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN,
1235 if (!kvm_irqfds_enabled()) {
1239 return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd);
1242 #else /* !KVM_CAP_IRQ_ROUTING */
1244 static void kvm_init_irq_routing(KVMState *s)
1248 void kvm_irqchip_release_virq(KVMState *s, int virq)
1252 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1257 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1262 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign)
1267 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1271 #endif /* !KVM_CAP_IRQ_ROUTING */
1273 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n, int virq)
1275 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), virq, true);
1278 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, int virq)
1280 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), virq, false);
1283 static int kvm_irqchip_create(KVMState *s)
1285 QemuOptsList *list = qemu_find_opts("machine");
1288 if (QTAILQ_EMPTY(&list->head) ||
1289 !qemu_opt_get_bool(QTAILQ_FIRST(&list->head),
1290 "kernel_irqchip", true) ||
1291 !kvm_check_extension(s, KVM_CAP_IRQCHIP)) {
1295 ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP);
1297 fprintf(stderr, "Create kernel irqchip failed\n");
1301 kvm_kernel_irqchip = true;
1302 /* If we have an in-kernel IRQ chip then we must have asynchronous
1303 * interrupt delivery (though the reverse is not necessarily true)
1305 kvm_async_interrupts_allowed = true;
1307 kvm_init_irq_routing(s);
1312 static int kvm_max_vcpus(KVMState *s)
1316 /* Find number of supported CPUs using the recommended
1317 * procedure from the kernel API documentation to cope with
1318 * older kernels that may be missing capabilities.
1320 ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS);
1324 ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS);
1334 static const char upgrade_note[] =
1335 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1336 "(see http://sourceforge.net/projects/kvm).\n";
1338 const KVMCapabilityInfo *missing_cap;
1343 s = g_malloc0(sizeof(KVMState));
1346 * On systems where the kernel can support different base page
1347 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1348 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1349 * page size for the system though.
1351 assert(TARGET_PAGE_SIZE <= getpagesize());
1353 #ifdef KVM_CAP_SET_GUEST_DEBUG
1354 QTAILQ_INIT(&s->kvm_sw_breakpoints);
1356 for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
1357 s->slots[i].slot = i;
1360 s->fd = qemu_open("/dev/kvm", O_RDWR);
1362 fprintf(stderr, "Could not access KVM kernel module: %m\n");
1367 ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
1368 if (ret < KVM_API_VERSION) {
1372 fprintf(stderr, "kvm version too old\n");
1376 if (ret > KVM_API_VERSION) {
1378 fprintf(stderr, "kvm version not supported\n");
1382 max_vcpus = kvm_max_vcpus(s);
1383 if (smp_cpus > max_vcpus) {
1385 fprintf(stderr, "Number of SMP cpus requested (%d) exceeds max cpus "
1386 "supported by KVM (%d)\n", smp_cpus, max_vcpus);
1390 s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0);
1393 fprintf(stderr, "Please add the 'switch_amode' kernel parameter to "
1394 "your host kernel command line\n");
1400 missing_cap = kvm_check_extension_list(s, kvm_required_capabilites);
1403 kvm_check_extension_list(s, kvm_arch_required_capabilities);
1407 fprintf(stderr, "kvm does not support %s\n%s",
1408 missing_cap->name, upgrade_note);
1412 s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
1414 s->broken_set_mem_region = 1;
1415 ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
1417 s->broken_set_mem_region = 0;
1420 #ifdef KVM_CAP_VCPU_EVENTS
1421 s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS);
1424 s->robust_singlestep =
1425 kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP);
1427 #ifdef KVM_CAP_DEBUGREGS
1428 s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS);
1431 #ifdef KVM_CAP_XSAVE
1432 s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE);
1436 s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS);
1439 #ifdef KVM_CAP_PIT_STATE2
1440 s->pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2);
1443 #ifdef KVM_CAP_IRQ_ROUTING
1444 s->direct_msi = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0);
1447 s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3);
1449 s->irq_set_ioctl = KVM_IRQ_LINE;
1450 if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) {
1451 s->irq_set_ioctl = KVM_IRQ_LINE_STATUS;
1454 #ifdef KVM_CAP_READONLY_MEM
1455 kvm_readonly_mem_allowed =
1456 (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0);
1459 ret = kvm_arch_init(s);
1464 ret = kvm_irqchip_create(s);
1470 memory_listener_register(&kvm_memory_listener, &address_space_memory);
1471 memory_listener_register(&kvm_io_listener, &address_space_io);
1473 s->many_ioeventfds = kvm_check_many_ioeventfds();
1475 cpu_interrupt_handler = kvm_handle_interrupt;
1491 static void kvm_handle_io(uint16_t port, void *data, int direction, int size,
1495 uint8_t *ptr = data;
1497 for (i = 0; i < count; i++) {
1498 if (direction == KVM_EXIT_IO_IN) {
1501 stb_p(ptr, cpu_inb(port));
1504 stw_p(ptr, cpu_inw(port));
1507 stl_p(ptr, cpu_inl(port));
1513 cpu_outb(port, ldub_p(ptr));
1516 cpu_outw(port, lduw_p(ptr));
1519 cpu_outl(port, ldl_p(ptr));
1528 static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run)
1530 fprintf(stderr, "KVM internal error.");
1531 if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) {
1534 fprintf(stderr, " Suberror: %d\n", run->internal.suberror);
1535 for (i = 0; i < run->internal.ndata; ++i) {
1536 fprintf(stderr, "extra data[%d]: %"PRIx64"\n",
1537 i, (uint64_t)run->internal.data[i]);
1540 fprintf(stderr, "\n");
1542 if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) {
1543 fprintf(stderr, "emulation failure\n");
1544 if (!kvm_arch_stop_on_emulation_error(cpu)) {
1545 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1546 return EXCP_INTERRUPT;
1549 /* FIXME: Should trigger a qmp message to let management know
1550 * something went wrong.
1555 void kvm_flush_coalesced_mmio_buffer(void)
1557 KVMState *s = kvm_state;
1559 if (s->coalesced_flush_in_progress) {
1563 s->coalesced_flush_in_progress = true;
1565 if (s->coalesced_mmio_ring) {
1566 struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring;
1567 while (ring->first != ring->last) {
1568 struct kvm_coalesced_mmio *ent;
1570 ent = &ring->coalesced_mmio[ring->first];
1572 cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
1574 ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
1578 s->coalesced_flush_in_progress = false;
1581 static void do_kvm_cpu_synchronize_state(void *arg)
1583 CPUState *cpu = arg;
1585 if (!cpu->kvm_vcpu_dirty) {
1586 kvm_arch_get_registers(cpu);
1587 cpu->kvm_vcpu_dirty = true;
1591 void kvm_cpu_synchronize_state(CPUState *cpu)
1593 if (!cpu->kvm_vcpu_dirty) {
1594 run_on_cpu(cpu, do_kvm_cpu_synchronize_state, cpu);
1598 void kvm_cpu_synchronize_post_reset(CPUState *cpu)
1600 kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE);
1601 cpu->kvm_vcpu_dirty = false;
1604 void kvm_cpu_synchronize_post_init(CPUState *cpu)
1606 kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE);
1607 cpu->kvm_vcpu_dirty = false;
1610 int kvm_cpu_exec(CPUState *cpu)
1612 struct kvm_run *run = cpu->kvm_run;
1615 DPRINTF("kvm_cpu_exec()\n");
1617 if (kvm_arch_process_async_events(cpu)) {
1618 cpu->exit_request = 0;
1623 if (cpu->kvm_vcpu_dirty) {
1624 kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE);
1625 cpu->kvm_vcpu_dirty = false;
1628 kvm_arch_pre_run(cpu, run);
1629 if (cpu->exit_request) {
1630 DPRINTF("interrupt exit requested\n");
1632 * KVM requires us to reenter the kernel after IO exits to complete
1633 * instruction emulation. This self-signal will ensure that we
1636 qemu_cpu_kick_self();
1638 qemu_mutex_unlock_iothread();
1640 run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0);
1642 qemu_mutex_lock_iothread();
1643 kvm_arch_post_run(cpu, run);
1646 if (run_ret == -EINTR || run_ret == -EAGAIN) {
1647 DPRINTF("io window exit\n");
1648 ret = EXCP_INTERRUPT;
1651 fprintf(stderr, "error: kvm run failed %s\n",
1652 strerror(-run_ret));
1656 trace_kvm_run_exit(cpu->cpu_index, run->exit_reason);
1657 switch (run->exit_reason) {
1659 DPRINTF("handle_io\n");
1660 kvm_handle_io(run->io.port,
1661 (uint8_t *)run + run->io.data_offset,
1668 DPRINTF("handle_mmio\n");
1669 cpu_physical_memory_rw(run->mmio.phys_addr,
1672 run->mmio.is_write);
1675 case KVM_EXIT_IRQ_WINDOW_OPEN:
1676 DPRINTF("irq_window_open\n");
1677 ret = EXCP_INTERRUPT;
1679 case KVM_EXIT_SHUTDOWN:
1680 DPRINTF("shutdown\n");
1681 qemu_system_reset_request();
1682 ret = EXCP_INTERRUPT;
1684 case KVM_EXIT_UNKNOWN:
1685 fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n",
1686 (uint64_t)run->hw.hardware_exit_reason);
1689 case KVM_EXIT_INTERNAL_ERROR:
1690 ret = kvm_handle_internal_error(cpu, run);
1693 DPRINTF("kvm_arch_handle_exit\n");
1694 ret = kvm_arch_handle_exit(cpu, run);
1700 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1701 vm_stop(RUN_STATE_INTERNAL_ERROR);
1704 cpu->exit_request = 0;
1708 int kvm_ioctl(KVMState *s, int type, ...)
1715 arg = va_arg(ap, void *);
1718 trace_kvm_ioctl(type, arg);
1719 ret = ioctl(s->fd, type, arg);
1726 int kvm_vm_ioctl(KVMState *s, int type, ...)
1733 arg = va_arg(ap, void *);
1736 trace_kvm_vm_ioctl(type, arg);
1737 ret = ioctl(s->vmfd, type, arg);
1744 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...)
1751 arg = va_arg(ap, void *);
1754 trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg);
1755 ret = ioctl(cpu->kvm_fd, type, arg);
1762 int kvm_has_sync_mmu(void)
1764 return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU);
1767 int kvm_has_vcpu_events(void)
1769 return kvm_state->vcpu_events;
1772 int kvm_has_robust_singlestep(void)
1774 return kvm_state->robust_singlestep;
1777 int kvm_has_debugregs(void)
1779 return kvm_state->debugregs;
1782 int kvm_has_xsave(void)
1784 return kvm_state->xsave;
1787 int kvm_has_xcrs(void)
1789 return kvm_state->xcrs;
1792 int kvm_has_pit_state2(void)
1794 return kvm_state->pit_state2;
1797 int kvm_has_many_ioeventfds(void)
1799 if (!kvm_enabled()) {
1802 return kvm_state->many_ioeventfds;
1805 int kvm_has_gsi_routing(void)
1807 #ifdef KVM_CAP_IRQ_ROUTING
1808 return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING);
1814 int kvm_has_intx_set_mask(void)
1816 return kvm_state->intx_set_mask;
1819 void *kvm_ram_alloc(ram_addr_t size)
1824 mem = kvm_arch_ram_alloc(size);
1829 return qemu_anon_ram_alloc(size);
1832 void kvm_setup_guest_memory(void *start, size_t size)
1834 #ifdef CONFIG_VALGRIND_H
1835 VALGRIND_MAKE_MEM_DEFINED(start, size);
1837 if (!kvm_has_sync_mmu()) {
1838 int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK);
1841 perror("qemu_madvise");
1843 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1849 #ifdef KVM_CAP_SET_GUEST_DEBUG
1850 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu,
1853 struct kvm_sw_breakpoint *bp;
1855 QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) {
1863 int kvm_sw_breakpoints_active(CPUState *cpu)
1865 return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints);
1868 struct kvm_set_guest_debug_data {
1869 struct kvm_guest_debug dbg;
1874 static void kvm_invoke_set_guest_debug(void *data)
1876 struct kvm_set_guest_debug_data *dbg_data = data;
1878 dbg_data->err = kvm_vcpu_ioctl(dbg_data->cpu, KVM_SET_GUEST_DEBUG,
1882 int kvm_update_guest_debug(CPUArchState *env, unsigned long reinject_trap)
1884 CPUState *cpu = ENV_GET_CPU(env);
1885 struct kvm_set_guest_debug_data data;
1887 data.dbg.control = reinject_trap;
1889 if (env->singlestep_enabled) {
1890 data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
1892 kvm_arch_update_guest_debug(cpu, &data.dbg);
1895 run_on_cpu(cpu, kvm_invoke_set_guest_debug, &data);
1899 int kvm_insert_breakpoint(CPUArchState *current_env, target_ulong addr,
1900 target_ulong len, int type)
1902 CPUState *current_cpu = ENV_GET_CPU(current_env);
1903 struct kvm_sw_breakpoint *bp;
1907 if (type == GDB_BREAKPOINT_SW) {
1908 bp = kvm_find_sw_breakpoint(current_cpu, addr);
1914 bp = g_malloc(sizeof(struct kvm_sw_breakpoint));
1921 err = kvm_arch_insert_sw_breakpoint(current_cpu, bp);
1927 QTAILQ_INSERT_HEAD(¤t_cpu->kvm_state->kvm_sw_breakpoints,
1930 err = kvm_arch_insert_hw_breakpoint(addr, len, type);
1936 for (env = first_cpu; env != NULL; env = env->next_cpu) {
1937 err = kvm_update_guest_debug(env, 0);
1945 int kvm_remove_breakpoint(CPUArchState *current_env, target_ulong addr,
1946 target_ulong len, int type)
1948 CPUState *current_cpu = ENV_GET_CPU(current_env);
1949 struct kvm_sw_breakpoint *bp;
1953 if (type == GDB_BREAKPOINT_SW) {
1954 bp = kvm_find_sw_breakpoint(current_cpu, addr);
1959 if (bp->use_count > 1) {
1964 err = kvm_arch_remove_sw_breakpoint(current_cpu, bp);
1969 QTAILQ_REMOVE(¤t_cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
1972 err = kvm_arch_remove_hw_breakpoint(addr, len, type);
1978 for (env = first_cpu; env != NULL; env = env->next_cpu) {
1979 err = kvm_update_guest_debug(env, 0);
1987 void kvm_remove_all_breakpoints(CPUArchState *current_env)
1989 CPUState *current_cpu = ENV_GET_CPU(current_env);
1990 struct kvm_sw_breakpoint *bp, *next;
1991 KVMState *s = current_cpu->kvm_state;
1995 QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
1996 if (kvm_arch_remove_sw_breakpoint(current_cpu, bp) != 0) {
1997 /* Try harder to find a CPU that currently sees the breakpoint. */
1998 for (env = first_cpu; env != NULL; env = env->next_cpu) {
1999 cpu = ENV_GET_CPU(env);
2000 if (kvm_arch_remove_sw_breakpoint(cpu, bp) == 0) {
2005 QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry);
2008 kvm_arch_remove_all_hw_breakpoints();
2010 for (env = first_cpu; env != NULL; env = env->next_cpu) {
2011 kvm_update_guest_debug(env, 0);
2015 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2017 int kvm_update_guest_debug(CPUArchState *env, unsigned long reinject_trap)
2022 int kvm_insert_breakpoint(CPUArchState *current_env, target_ulong addr,
2023 target_ulong len, int type)
2028 int kvm_remove_breakpoint(CPUArchState *current_env, target_ulong addr,
2029 target_ulong len, int type)
2034 void kvm_remove_all_breakpoints(CPUArchState *current_env)
2037 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2039 int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset)
2041 struct kvm_signal_mask *sigmask;
2045 return kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, NULL);
2048 sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset));
2051 memcpy(sigmask->sigset, sigset, sizeof(*sigset));
2052 r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask);
2057 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2059 return kvm_arch_on_sigbus_vcpu(cpu, code, addr);
2062 int kvm_on_sigbus(int code, void *addr)
2064 return kvm_arch_on_sigbus(code, addr);