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 "hw/s390x/adapter.h"
31 #include "exec/gdbstub.h"
32 #include "sysemu/kvm.h"
33 #include "qemu/bswap.h"
34 #include "exec/memory.h"
35 #include "exec/ram_addr.h"
36 #include "exec/address-spaces.h"
37 #include "qemu/event_notifier.h"
40 #include "hw/boards.h"
42 /* This check must be after config-host.h is included */
44 #include <sys/eventfd.h>
47 #ifdef CONFIG_VALGRIND_H
48 #include <valgrind/memcheck.h>
51 /* KVM uses PAGE_SIZE in its definition of COALESCED_MMIO_MAX */
52 #define PAGE_SIZE TARGET_PAGE_SIZE
57 #define DPRINTF(fmt, ...) \
58 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
60 #define DPRINTF(fmt, ...) \
64 #define KVM_MSI_HASHTAB_SIZE 256
66 typedef struct KVMSlot
69 ram_addr_t memory_size;
75 typedef struct kvm_dirty_log KVMDirtyLog;
84 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
85 bool coalesced_flush_in_progress;
86 int broken_set_mem_region;
89 int robust_singlestep;
91 #ifdef KVM_CAP_SET_GUEST_DEBUG
92 struct kvm_sw_breakpoint_head kvm_sw_breakpoints;
98 /* The man page (and posix) say ioctl numbers are signed int, but
99 * they're not. Linux, glibc and *BSD all treat ioctl numbers as
100 * unsigned, and treating them as signed here can break things */
101 unsigned irq_set_ioctl;
102 #ifdef KVM_CAP_IRQ_ROUTING
103 struct kvm_irq_routing *irq_routes;
104 int nr_allocated_irq_routes;
105 uint32_t *used_gsi_bitmap;
106 unsigned int gsi_count;
107 QTAILQ_HEAD(msi_hashtab, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE];
113 bool kvm_kernel_irqchip;
114 bool kvm_async_interrupts_allowed;
115 bool kvm_halt_in_kernel_allowed;
116 bool kvm_eventfds_allowed;
117 bool kvm_irqfds_allowed;
118 bool kvm_msi_via_irqfd_allowed;
119 bool kvm_gsi_routing_allowed;
120 bool kvm_gsi_direct_mapping;
122 bool kvm_readonly_mem_allowed;
124 static const KVMCapabilityInfo kvm_required_capabilites[] = {
125 KVM_CAP_INFO(USER_MEMORY),
126 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS),
130 static KVMSlot *kvm_alloc_slot(KVMState *s)
134 for (i = 0; i < s->nr_slots; i++) {
135 if (s->slots[i].memory_size == 0) {
140 fprintf(stderr, "%s: no free slot available\n", __func__);
144 static KVMSlot *kvm_lookup_matching_slot(KVMState *s,
150 for (i = 0; i < s->nr_slots; i++) {
151 KVMSlot *mem = &s->slots[i];
153 if (start_addr == mem->start_addr &&
154 end_addr == mem->start_addr + mem->memory_size) {
163 * Find overlapping slot with lowest start address
165 static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s,
169 KVMSlot *found = NULL;
172 for (i = 0; i < s->nr_slots; i++) {
173 KVMSlot *mem = &s->slots[i];
175 if (mem->memory_size == 0 ||
176 (found && found->start_addr < mem->start_addr)) {
180 if (end_addr > mem->start_addr &&
181 start_addr < mem->start_addr + mem->memory_size) {
189 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram,
194 for (i = 0; i < s->nr_slots; i++) {
195 KVMSlot *mem = &s->slots[i];
197 if (ram >= mem->ram && ram < mem->ram + mem->memory_size) {
198 *phys_addr = mem->start_addr + (ram - mem->ram);
206 static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot)
208 struct kvm_userspace_memory_region mem;
210 mem.slot = slot->slot;
211 mem.guest_phys_addr = slot->start_addr;
212 mem.userspace_addr = (unsigned long)slot->ram;
213 mem.flags = slot->flags;
214 if (s->migration_log) {
215 mem.flags |= KVM_MEM_LOG_DIRTY_PAGES;
218 if (slot->memory_size && mem.flags & KVM_MEM_READONLY) {
219 /* Set the slot size to 0 before setting the slot to the desired
220 * value. This is needed based on KVM commit 75d61fbc. */
222 kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
224 mem.memory_size = slot->memory_size;
225 return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
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);
272 * dirty pages logging control
275 static int kvm_mem_flags(KVMState *s, bool log_dirty, bool readonly)
278 flags = log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0;
279 if (readonly && kvm_readonly_mem_allowed) {
280 flags |= KVM_MEM_READONLY;
285 static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty)
287 KVMState *s = kvm_state;
288 int flags, mask = KVM_MEM_LOG_DIRTY_PAGES;
291 old_flags = mem->flags;
293 flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty, false);
296 /* If nothing changed effectively, no need to issue ioctl */
297 if (s->migration_log) {
298 flags |= KVM_MEM_LOG_DIRTY_PAGES;
301 if (flags == old_flags) {
305 return kvm_set_user_memory_region(s, mem);
308 static int kvm_dirty_pages_log_change(hwaddr phys_addr,
309 ram_addr_t size, bool log_dirty)
311 KVMState *s = kvm_state;
312 KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size);
315 fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-"
316 TARGET_FMT_plx "\n", __func__, phys_addr,
317 (hwaddr)(phys_addr + size - 1));
320 return kvm_slot_dirty_pages_log_change(mem, log_dirty);
323 static void kvm_log_start(MemoryListener *listener,
324 MemoryRegionSection *section)
328 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
329 int128_get64(section->size), true);
335 static void kvm_log_stop(MemoryListener *listener,
336 MemoryRegionSection *section)
340 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
341 int128_get64(section->size), false);
347 static int kvm_set_migration_log(int enable)
349 KVMState *s = kvm_state;
353 s->migration_log = enable;
355 for (i = 0; i < s->nr_slots; i++) {
358 if (!mem->memory_size) {
361 if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) {
364 err = kvm_set_user_memory_region(s, mem);
372 /* get kvm's dirty pages bitmap and update qemu's */
373 static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section,
374 unsigned long *bitmap)
376 ram_addr_t start = section->offset_within_region + section->mr->ram_addr;
377 ram_addr_t pages = int128_get64(section->size) / getpagesize();
379 cpu_physical_memory_set_dirty_lebitmap(bitmap, start, pages);
383 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
386 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
387 * This function updates qemu's dirty bitmap using
388 * memory_region_set_dirty(). This means all bits are set
391 * @start_add: start of logged region.
392 * @end_addr: end of logged region.
394 static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection *section)
396 KVMState *s = kvm_state;
397 unsigned long size, allocated_size = 0;
401 hwaddr start_addr = section->offset_within_address_space;
402 hwaddr end_addr = start_addr + int128_get64(section->size);
404 d.dirty_bitmap = NULL;
405 while (start_addr < end_addr) {
406 mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr);
411 /* XXX bad kernel interface alert
412 * For dirty bitmap, kernel allocates array of size aligned to
413 * bits-per-long. But for case when the kernel is 64bits and
414 * the userspace is 32bits, userspace can't align to the same
415 * bits-per-long, since sizeof(long) is different between kernel
416 * and user space. This way, userspace will provide buffer which
417 * may be 4 bytes less than the kernel will use, resulting in
418 * userspace memory corruption (which is not detectable by valgrind
419 * too, in most cases).
420 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
421 * a hope that sizeof(long) wont become >8 any time soon.
423 size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS),
424 /*HOST_LONG_BITS*/ 64) / 8;
425 if (!d.dirty_bitmap) {
426 d.dirty_bitmap = g_malloc(size);
427 } else if (size > allocated_size) {
428 d.dirty_bitmap = g_realloc(d.dirty_bitmap, size);
430 allocated_size = size;
431 memset(d.dirty_bitmap, 0, allocated_size);
435 if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
436 DPRINTF("ioctl failed %d\n", errno);
441 kvm_get_dirty_pages_log_range(section, d.dirty_bitmap);
442 start_addr = mem->start_addr + mem->memory_size;
444 g_free(d.dirty_bitmap);
449 static void kvm_coalesce_mmio_region(MemoryListener *listener,
450 MemoryRegionSection *secion,
451 hwaddr start, hwaddr size)
453 KVMState *s = kvm_state;
455 if (s->coalesced_mmio) {
456 struct kvm_coalesced_mmio_zone zone;
462 (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
466 static void kvm_uncoalesce_mmio_region(MemoryListener *listener,
467 MemoryRegionSection *secion,
468 hwaddr start, hwaddr size)
470 KVMState *s = kvm_state;
472 if (s->coalesced_mmio) {
473 struct kvm_coalesced_mmio_zone zone;
479 (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
483 int kvm_check_extension(KVMState *s, unsigned int extension)
487 ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
495 static int kvm_set_ioeventfd_mmio(int fd, hwaddr addr, uint32_t val,
496 bool assign, uint32_t size, bool datamatch)
499 struct kvm_ioeventfd iofd;
501 iofd.datamatch = datamatch ? val : 0;
507 if (!kvm_enabled()) {
512 iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
515 iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
518 ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd);
527 static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val,
528 bool assign, uint32_t size, bool datamatch)
530 struct kvm_ioeventfd kick = {
531 .datamatch = datamatch ? val : 0,
533 .flags = KVM_IOEVENTFD_FLAG_PIO,
538 if (!kvm_enabled()) {
542 kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
545 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
547 r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
555 static int kvm_check_many_ioeventfds(void)
557 /* Userspace can use ioeventfd for io notification. This requires a host
558 * that supports eventfd(2) and an I/O thread; since eventfd does not
559 * support SIGIO it cannot interrupt the vcpu.
561 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
562 * can avoid creating too many ioeventfds.
564 #if defined(CONFIG_EVENTFD)
567 for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) {
568 ioeventfds[i] = eventfd(0, EFD_CLOEXEC);
569 if (ioeventfds[i] < 0) {
572 ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true);
574 close(ioeventfds[i]);
579 /* Decide whether many devices are supported or not */
580 ret = i == ARRAY_SIZE(ioeventfds);
583 kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true);
584 close(ioeventfds[i]);
592 static const KVMCapabilityInfo *
593 kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list)
596 if (!kvm_check_extension(s, list->value)) {
604 static void kvm_set_phys_mem(MemoryRegionSection *section, bool add)
606 KVMState *s = kvm_state;
609 MemoryRegion *mr = section->mr;
610 bool log_dirty = memory_region_is_logging(mr);
611 bool writeable = !mr->readonly && !mr->rom_device;
612 bool readonly_flag = mr->readonly || memory_region_is_romd(mr);
613 hwaddr start_addr = section->offset_within_address_space;
614 ram_addr_t size = int128_get64(section->size);
618 /* kvm works in page size chunks, but the function may be called
619 with sub-page size and unaligned start address. */
620 delta = TARGET_PAGE_ALIGN(size) - size;
626 size &= TARGET_PAGE_MASK;
627 if (!size || (start_addr & ~TARGET_PAGE_MASK)) {
631 if (!memory_region_is_ram(mr)) {
632 if (writeable || !kvm_readonly_mem_allowed) {
634 } else if (!mr->romd_mode) {
635 /* If the memory device is not in romd_mode, then we actually want
636 * to remove the kvm memory slot so all accesses will trap. */
641 ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta;
644 mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
649 if (add && start_addr >= mem->start_addr &&
650 (start_addr + size <= mem->start_addr + mem->memory_size) &&
651 (ram - start_addr == mem->ram - mem->start_addr)) {
652 /* The new slot fits into the existing one and comes with
653 * identical parameters - update flags and done. */
654 kvm_slot_dirty_pages_log_change(mem, log_dirty);
660 if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
661 kvm_physical_sync_dirty_bitmap(section);
664 /* unregister the overlapping slot */
665 mem->memory_size = 0;
666 err = kvm_set_user_memory_region(s, mem);
668 fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
669 __func__, strerror(-err));
673 /* Workaround for older KVM versions: we can't join slots, even not by
674 * unregistering the previous ones and then registering the larger
675 * slot. We have to maintain the existing fragmentation. Sigh.
677 * This workaround assumes that the new slot starts at the same
678 * address as the first existing one. If not or if some overlapping
679 * slot comes around later, we will fail (not seen in practice so far)
680 * - and actually require a recent KVM version. */
681 if (s->broken_set_mem_region &&
682 old.start_addr == start_addr && old.memory_size < size && add) {
683 mem = kvm_alloc_slot(s);
684 mem->memory_size = old.memory_size;
685 mem->start_addr = old.start_addr;
687 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
689 err = kvm_set_user_memory_region(s, mem);
691 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
696 start_addr += old.memory_size;
697 ram += old.memory_size;
698 size -= old.memory_size;
702 /* register prefix slot */
703 if (old.start_addr < start_addr) {
704 mem = kvm_alloc_slot(s);
705 mem->memory_size = start_addr - old.start_addr;
706 mem->start_addr = old.start_addr;
708 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
710 err = kvm_set_user_memory_region(s, mem);
712 fprintf(stderr, "%s: error registering prefix slot: %s\n",
713 __func__, strerror(-err));
715 fprintf(stderr, "%s: This is probably because your kernel's " \
716 "PAGE_SIZE is too big. Please try to use 4k " \
717 "PAGE_SIZE!\n", __func__);
723 /* register suffix slot */
724 if (old.start_addr + old.memory_size > start_addr + size) {
725 ram_addr_t size_delta;
727 mem = kvm_alloc_slot(s);
728 mem->start_addr = start_addr + size;
729 size_delta = mem->start_addr - old.start_addr;
730 mem->memory_size = old.memory_size - size_delta;
731 mem->ram = old.ram + size_delta;
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 suffix slot: %s\n",
737 __func__, strerror(-err));
743 /* in case the KVM bug workaround already "consumed" the new slot */
750 mem = kvm_alloc_slot(s);
751 mem->memory_size = size;
752 mem->start_addr = start_addr;
754 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
756 err = kvm_set_user_memory_region(s, mem);
758 fprintf(stderr, "%s: error registering slot: %s\n", __func__,
764 static void kvm_region_add(MemoryListener *listener,
765 MemoryRegionSection *section)
767 memory_region_ref(section->mr);
768 kvm_set_phys_mem(section, true);
771 static void kvm_region_del(MemoryListener *listener,
772 MemoryRegionSection *section)
774 kvm_set_phys_mem(section, false);
775 memory_region_unref(section->mr);
778 static void kvm_log_sync(MemoryListener *listener,
779 MemoryRegionSection *section)
783 r = kvm_physical_sync_dirty_bitmap(section);
789 static void kvm_log_global_start(struct MemoryListener *listener)
793 r = kvm_set_migration_log(1);
797 static void kvm_log_global_stop(struct MemoryListener *listener)
801 r = kvm_set_migration_log(0);
805 static void kvm_mem_ioeventfd_add(MemoryListener *listener,
806 MemoryRegionSection *section,
807 bool match_data, uint64_t data,
810 int fd = event_notifier_get_fd(e);
813 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
814 data, true, int128_get64(section->size),
817 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
818 __func__, strerror(-r));
823 static void kvm_mem_ioeventfd_del(MemoryListener *listener,
824 MemoryRegionSection *section,
825 bool match_data, uint64_t data,
828 int fd = event_notifier_get_fd(e);
831 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
832 data, false, int128_get64(section->size),
839 static void kvm_io_ioeventfd_add(MemoryListener *listener,
840 MemoryRegionSection *section,
841 bool match_data, uint64_t data,
844 int fd = event_notifier_get_fd(e);
847 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
848 data, true, int128_get64(section->size),
851 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
852 __func__, strerror(-r));
857 static void kvm_io_ioeventfd_del(MemoryListener *listener,
858 MemoryRegionSection *section,
859 bool match_data, uint64_t data,
863 int fd = event_notifier_get_fd(e);
866 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
867 data, false, int128_get64(section->size),
874 static MemoryListener kvm_memory_listener = {
875 .region_add = kvm_region_add,
876 .region_del = kvm_region_del,
877 .log_start = kvm_log_start,
878 .log_stop = kvm_log_stop,
879 .log_sync = kvm_log_sync,
880 .log_global_start = kvm_log_global_start,
881 .log_global_stop = kvm_log_global_stop,
882 .eventfd_add = kvm_mem_ioeventfd_add,
883 .eventfd_del = kvm_mem_ioeventfd_del,
884 .coalesced_mmio_add = kvm_coalesce_mmio_region,
885 .coalesced_mmio_del = kvm_uncoalesce_mmio_region,
889 static MemoryListener kvm_io_listener = {
890 .eventfd_add = kvm_io_ioeventfd_add,
891 .eventfd_del = kvm_io_ioeventfd_del,
895 static void kvm_handle_interrupt(CPUState *cpu, int mask)
897 cpu->interrupt_request |= mask;
899 if (!qemu_cpu_is_self(cpu)) {
904 int kvm_set_irq(KVMState *s, int irq, int level)
906 struct kvm_irq_level event;
909 assert(kvm_async_interrupts_enabled());
913 ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event);
915 perror("kvm_set_irq");
919 return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status;
922 #ifdef KVM_CAP_IRQ_ROUTING
923 typedef struct KVMMSIRoute {
924 struct kvm_irq_routing_entry kroute;
925 QTAILQ_ENTRY(KVMMSIRoute) entry;
928 static void set_gsi(KVMState *s, unsigned int gsi)
930 s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32);
933 static void clear_gsi(KVMState *s, unsigned int gsi)
935 s->used_gsi_bitmap[gsi / 32] &= ~(1U << (gsi % 32));
938 void kvm_init_irq_routing(KVMState *s)
942 gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING);
944 unsigned int gsi_bits, i;
946 /* Round up so we can search ints using ffs */
947 gsi_bits = ALIGN(gsi_count, 32);
948 s->used_gsi_bitmap = g_malloc0(gsi_bits / 8);
949 s->gsi_count = gsi_count;
951 /* Mark any over-allocated bits as already in use */
952 for (i = gsi_count; i < gsi_bits; i++) {
957 s->irq_routes = g_malloc0(sizeof(*s->irq_routes));
958 s->nr_allocated_irq_routes = 0;
960 if (!s->direct_msi) {
961 for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) {
962 QTAILQ_INIT(&s->msi_hashtab[i]);
966 kvm_arch_init_irq_routing(s);
969 void kvm_irqchip_commit_routes(KVMState *s)
973 s->irq_routes->flags = 0;
974 ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes);
978 static void kvm_add_routing_entry(KVMState *s,
979 struct kvm_irq_routing_entry *entry)
981 struct kvm_irq_routing_entry *new;
984 if (s->irq_routes->nr == s->nr_allocated_irq_routes) {
985 n = s->nr_allocated_irq_routes * 2;
989 size = sizeof(struct kvm_irq_routing);
990 size += n * sizeof(*new);
991 s->irq_routes = g_realloc(s->irq_routes, size);
992 s->nr_allocated_irq_routes = n;
994 n = s->irq_routes->nr++;
995 new = &s->irq_routes->entries[n];
999 set_gsi(s, entry->gsi);
1002 static int kvm_update_routing_entry(KVMState *s,
1003 struct kvm_irq_routing_entry *new_entry)
1005 struct kvm_irq_routing_entry *entry;
1008 for (n = 0; n < s->irq_routes->nr; n++) {
1009 entry = &s->irq_routes->entries[n];
1010 if (entry->gsi != new_entry->gsi) {
1014 if(!memcmp(entry, new_entry, sizeof *entry)) {
1018 *entry = *new_entry;
1020 kvm_irqchip_commit_routes(s);
1028 void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin)
1030 struct kvm_irq_routing_entry e = {};
1032 assert(pin < s->gsi_count);
1035 e.type = KVM_IRQ_ROUTING_IRQCHIP;
1037 e.u.irqchip.irqchip = irqchip;
1038 e.u.irqchip.pin = pin;
1039 kvm_add_routing_entry(s, &e);
1042 void kvm_irqchip_release_virq(KVMState *s, int virq)
1044 struct kvm_irq_routing_entry *e;
1047 if (kvm_gsi_direct_mapping()) {
1051 for (i = 0; i < s->irq_routes->nr; i++) {
1052 e = &s->irq_routes->entries[i];
1053 if (e->gsi == virq) {
1054 s->irq_routes->nr--;
1055 *e = s->irq_routes->entries[s->irq_routes->nr];
1061 static unsigned int kvm_hash_msi(uint32_t data)
1063 /* This is optimized for IA32 MSI layout. However, no other arch shall
1064 * repeat the mistake of not providing a direct MSI injection API. */
1068 static void kvm_flush_dynamic_msi_routes(KVMState *s)
1070 KVMMSIRoute *route, *next;
1073 for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) {
1074 QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) {
1075 kvm_irqchip_release_virq(s, route->kroute.gsi);
1076 QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry);
1082 static int kvm_irqchip_get_virq(KVMState *s)
1084 uint32_t *word = s->used_gsi_bitmap;
1085 int max_words = ALIGN(s->gsi_count, 32) / 32;
1090 /* Return the lowest unused GSI in the bitmap */
1091 for (i = 0; i < max_words; i++) {
1092 bit = ffs(~word[i]);
1097 return bit - 1 + i * 32;
1099 if (!s->direct_msi && retry) {
1101 kvm_flush_dynamic_msi_routes(s);
1108 static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg)
1110 unsigned int hash = kvm_hash_msi(msg.data);
1113 QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) {
1114 if (route->kroute.u.msi.address_lo == (uint32_t)msg.address &&
1115 route->kroute.u.msi.address_hi == (msg.address >> 32) &&
1116 route->kroute.u.msi.data == le32_to_cpu(msg.data)) {
1123 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1128 if (s->direct_msi) {
1129 msi.address_lo = (uint32_t)msg.address;
1130 msi.address_hi = msg.address >> 32;
1131 msi.data = le32_to_cpu(msg.data);
1133 memset(msi.pad, 0, sizeof(msi.pad));
1135 return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi);
1138 route = kvm_lookup_msi_route(s, msg);
1142 virq = kvm_irqchip_get_virq(s);
1147 route = g_malloc0(sizeof(KVMMSIRoute));
1148 route->kroute.gsi = virq;
1149 route->kroute.type = KVM_IRQ_ROUTING_MSI;
1150 route->kroute.flags = 0;
1151 route->kroute.u.msi.address_lo = (uint32_t)msg.address;
1152 route->kroute.u.msi.address_hi = msg.address >> 32;
1153 route->kroute.u.msi.data = le32_to_cpu(msg.data);
1155 kvm_add_routing_entry(s, &route->kroute);
1156 kvm_irqchip_commit_routes(s);
1158 QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route,
1162 assert(route->kroute.type == KVM_IRQ_ROUTING_MSI);
1164 return kvm_set_irq(s, route->kroute.gsi, 1);
1167 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1169 struct kvm_irq_routing_entry kroute = {};
1172 if (kvm_gsi_direct_mapping()) {
1173 return msg.data & 0xffff;
1176 if (!kvm_gsi_routing_enabled()) {
1180 virq = kvm_irqchip_get_virq(s);
1186 kroute.type = KVM_IRQ_ROUTING_MSI;
1188 kroute.u.msi.address_lo = (uint32_t)msg.address;
1189 kroute.u.msi.address_hi = msg.address >> 32;
1190 kroute.u.msi.data = le32_to_cpu(msg.data);
1192 kvm_add_routing_entry(s, &kroute);
1193 kvm_irqchip_commit_routes(s);
1198 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1200 struct kvm_irq_routing_entry kroute = {};
1202 if (kvm_gsi_direct_mapping()) {
1206 if (!kvm_irqchip_in_kernel()) {
1211 kroute.type = KVM_IRQ_ROUTING_MSI;
1213 kroute.u.msi.address_lo = (uint32_t)msg.address;
1214 kroute.u.msi.address_hi = msg.address >> 32;
1215 kroute.u.msi.data = le32_to_cpu(msg.data);
1217 return kvm_update_routing_entry(s, &kroute);
1220 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int rfd, int virq,
1223 struct kvm_irqfd irqfd = {
1226 .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN,
1230 irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE;
1231 irqfd.resamplefd = rfd;
1234 if (!kvm_irqfds_enabled()) {
1238 return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd);
1241 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter)
1243 struct kvm_irq_routing_entry kroute;
1246 if (!kvm_gsi_routing_enabled()) {
1250 virq = kvm_irqchip_get_virq(s);
1256 kroute.type = KVM_IRQ_ROUTING_S390_ADAPTER;
1258 kroute.u.adapter.summary_addr = adapter->summary_addr;
1259 kroute.u.adapter.ind_addr = adapter->ind_addr;
1260 kroute.u.adapter.summary_offset = adapter->summary_offset;
1261 kroute.u.adapter.ind_offset = adapter->ind_offset;
1262 kroute.u.adapter.adapter_id = adapter->adapter_id;
1264 kvm_add_routing_entry(s, &kroute);
1265 kvm_irqchip_commit_routes(s);
1270 #else /* !KVM_CAP_IRQ_ROUTING */
1272 void kvm_init_irq_routing(KVMState *s)
1276 void kvm_irqchip_release_virq(KVMState *s, int virq)
1280 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1285 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1290 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter)
1295 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign)
1300 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1304 #endif /* !KVM_CAP_IRQ_ROUTING */
1306 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n,
1307 EventNotifier *rn, int virq)
1309 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n),
1310 rn ? event_notifier_get_fd(rn) : -1, virq, true);
1313 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, int virq)
1315 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), -1, virq,
1319 static int kvm_irqchip_create(KVMState *s)
1323 if (!qemu_opt_get_bool(qemu_get_machine_opts(), "kernel_irqchip", true) ||
1324 (!kvm_check_extension(s, KVM_CAP_IRQCHIP) &&
1325 (kvm_vm_enable_cap(s, KVM_CAP_S390_IRQCHIP, 0) < 0))) {
1329 /* First probe and see if there's a arch-specific hook to create the
1330 * in-kernel irqchip for us */
1331 ret = kvm_arch_irqchip_create(s);
1334 } else if (ret == 0) {
1335 ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP);
1337 fprintf(stderr, "Create kernel irqchip failed\n");
1342 kvm_kernel_irqchip = true;
1343 /* If we have an in-kernel IRQ chip then we must have asynchronous
1344 * interrupt delivery (though the reverse is not necessarily true)
1346 kvm_async_interrupts_allowed = true;
1347 kvm_halt_in_kernel_allowed = true;
1349 kvm_init_irq_routing(s);
1354 /* Find number of supported CPUs using the recommended
1355 * procedure from the kernel API documentation to cope with
1356 * older kernels that may be missing capabilities.
1358 static int kvm_recommended_vcpus(KVMState *s)
1360 int ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS);
1361 return (ret) ? ret : 4;
1364 static int kvm_max_vcpus(KVMState *s)
1366 int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS);
1367 return (ret) ? ret : kvm_recommended_vcpus(s);
1370 int kvm_init(MachineClass *mc)
1372 static const char upgrade_note[] =
1373 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1374 "(see http://sourceforge.net/projects/kvm).\n";
1379 { "SMP", smp_cpus },
1380 { "hotpluggable", max_cpus },
1383 int soft_vcpus_limit, hard_vcpus_limit;
1385 const KVMCapabilityInfo *missing_cap;
1388 const char *kvm_type;
1390 s = g_malloc0(sizeof(KVMState));
1393 * On systems where the kernel can support different base page
1394 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1395 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1396 * page size for the system though.
1398 assert(TARGET_PAGE_SIZE <= getpagesize());
1401 #ifdef KVM_CAP_SET_GUEST_DEBUG
1402 QTAILQ_INIT(&s->kvm_sw_breakpoints);
1405 s->fd = qemu_open("/dev/kvm", O_RDWR);
1407 fprintf(stderr, "Could not access KVM kernel module: %m\n");
1412 ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
1413 if (ret < KVM_API_VERSION) {
1417 fprintf(stderr, "kvm version too old\n");
1421 if (ret > KVM_API_VERSION) {
1423 fprintf(stderr, "kvm version not supported\n");
1427 s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
1429 /* If unspecified, use the default value */
1434 s->slots = g_malloc0(s->nr_slots * sizeof(KVMSlot));
1436 for (i = 0; i < s->nr_slots; i++) {
1437 s->slots[i].slot = i;
1440 /* check the vcpu limits */
1441 soft_vcpus_limit = kvm_recommended_vcpus(s);
1442 hard_vcpus_limit = kvm_max_vcpus(s);
1445 if (nc->num > soft_vcpus_limit) {
1447 "Warning: Number of %s cpus requested (%d) exceeds "
1448 "the recommended cpus supported by KVM (%d)\n",
1449 nc->name, nc->num, soft_vcpus_limit);
1451 if (nc->num > hard_vcpus_limit) {
1452 fprintf(stderr, "Number of %s cpus requested (%d) exceeds "
1453 "the maximum cpus supported by KVM (%d)\n",
1454 nc->name, nc->num, hard_vcpus_limit);
1461 kvm_type = qemu_opt_get(qemu_get_machine_opts(), "kvm-type");
1463 type = mc->kvm_type(kvm_type);
1464 } else if (kvm_type) {
1466 fprintf(stderr, "Invalid argument kvm-type=%s\n", kvm_type);
1471 ret = kvm_ioctl(s, KVM_CREATE_VM, type);
1472 } while (ret == -EINTR);
1475 fprintf(stderr, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret,
1479 fprintf(stderr, "Please add the 'switch_amode' kernel parameter to "
1480 "your host kernel command line\n");
1486 missing_cap = kvm_check_extension_list(s, kvm_required_capabilites);
1489 kvm_check_extension_list(s, kvm_arch_required_capabilities);
1493 fprintf(stderr, "kvm does not support %s\n%s",
1494 missing_cap->name, upgrade_note);
1498 s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
1500 s->broken_set_mem_region = 1;
1501 ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
1503 s->broken_set_mem_region = 0;
1506 #ifdef KVM_CAP_VCPU_EVENTS
1507 s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS);
1510 s->robust_singlestep =
1511 kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP);
1513 #ifdef KVM_CAP_DEBUGREGS
1514 s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS);
1517 #ifdef KVM_CAP_XSAVE
1518 s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE);
1522 s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS);
1525 #ifdef KVM_CAP_PIT_STATE2
1526 s->pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2);
1529 #ifdef KVM_CAP_IRQ_ROUTING
1530 s->direct_msi = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0);
1533 s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3);
1535 s->irq_set_ioctl = KVM_IRQ_LINE;
1536 if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) {
1537 s->irq_set_ioctl = KVM_IRQ_LINE_STATUS;
1540 #ifdef KVM_CAP_READONLY_MEM
1541 kvm_readonly_mem_allowed =
1542 (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0);
1545 kvm_eventfds_allowed =
1546 (kvm_check_extension(s, KVM_CAP_IOEVENTFD) > 0);
1548 ret = kvm_arch_init(s);
1553 ret = kvm_irqchip_create(s);
1559 memory_listener_register(&kvm_memory_listener, &address_space_memory);
1560 memory_listener_register(&kvm_io_listener, &address_space_io);
1562 s->many_ioeventfds = kvm_check_many_ioeventfds();
1564 cpu_interrupt_handler = kvm_handle_interrupt;
1582 static void kvm_handle_io(uint16_t port, void *data, int direction, int size,
1586 uint8_t *ptr = data;
1588 for (i = 0; i < count; i++) {
1589 address_space_rw(&address_space_io, port, ptr, size,
1590 direction == KVM_EXIT_IO_OUT);
1595 static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run)
1597 fprintf(stderr, "KVM internal error. Suberror: %d\n",
1598 run->internal.suberror);
1600 if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) {
1603 for (i = 0; i < run->internal.ndata; ++i) {
1604 fprintf(stderr, "extra data[%d]: %"PRIx64"\n",
1605 i, (uint64_t)run->internal.data[i]);
1608 if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) {
1609 fprintf(stderr, "emulation failure\n");
1610 if (!kvm_arch_stop_on_emulation_error(cpu)) {
1611 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1612 return EXCP_INTERRUPT;
1615 /* FIXME: Should trigger a qmp message to let management know
1616 * something went wrong.
1621 void kvm_flush_coalesced_mmio_buffer(void)
1623 KVMState *s = kvm_state;
1625 if (s->coalesced_flush_in_progress) {
1629 s->coalesced_flush_in_progress = true;
1631 if (s->coalesced_mmio_ring) {
1632 struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring;
1633 while (ring->first != ring->last) {
1634 struct kvm_coalesced_mmio *ent;
1636 ent = &ring->coalesced_mmio[ring->first];
1638 cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
1640 ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
1644 s->coalesced_flush_in_progress = false;
1647 static void do_kvm_cpu_synchronize_state(void *arg)
1649 CPUState *cpu = arg;
1651 if (!cpu->kvm_vcpu_dirty) {
1652 kvm_arch_get_registers(cpu);
1653 cpu->kvm_vcpu_dirty = true;
1657 void kvm_cpu_synchronize_state(CPUState *cpu)
1659 if (!cpu->kvm_vcpu_dirty) {
1660 run_on_cpu(cpu, do_kvm_cpu_synchronize_state, cpu);
1664 void kvm_cpu_synchronize_post_reset(CPUState *cpu)
1666 kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE);
1667 cpu->kvm_vcpu_dirty = false;
1670 void kvm_cpu_synchronize_post_init(CPUState *cpu)
1672 kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE);
1673 cpu->kvm_vcpu_dirty = false;
1676 int kvm_cpu_exec(CPUState *cpu)
1678 struct kvm_run *run = cpu->kvm_run;
1681 DPRINTF("kvm_cpu_exec()\n");
1683 if (kvm_arch_process_async_events(cpu)) {
1684 cpu->exit_request = 0;
1689 if (cpu->kvm_vcpu_dirty) {
1690 kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE);
1691 cpu->kvm_vcpu_dirty = false;
1694 kvm_arch_pre_run(cpu, run);
1695 if (cpu->exit_request) {
1696 DPRINTF("interrupt exit requested\n");
1698 * KVM requires us to reenter the kernel after IO exits to complete
1699 * instruction emulation. This self-signal will ensure that we
1702 qemu_cpu_kick_self();
1704 qemu_mutex_unlock_iothread();
1706 run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0);
1708 qemu_mutex_lock_iothread();
1709 kvm_arch_post_run(cpu, run);
1712 if (run_ret == -EINTR || run_ret == -EAGAIN) {
1713 DPRINTF("io window exit\n");
1714 ret = EXCP_INTERRUPT;
1717 fprintf(stderr, "error: kvm run failed %s\n",
1718 strerror(-run_ret));
1722 trace_kvm_run_exit(cpu->cpu_index, run->exit_reason);
1723 switch (run->exit_reason) {
1725 DPRINTF("handle_io\n");
1726 kvm_handle_io(run->io.port,
1727 (uint8_t *)run + run->io.data_offset,
1734 DPRINTF("handle_mmio\n");
1735 cpu_physical_memory_rw(run->mmio.phys_addr,
1738 run->mmio.is_write);
1741 case KVM_EXIT_IRQ_WINDOW_OPEN:
1742 DPRINTF("irq_window_open\n");
1743 ret = EXCP_INTERRUPT;
1745 case KVM_EXIT_SHUTDOWN:
1746 DPRINTF("shutdown\n");
1747 qemu_system_reset_request();
1748 ret = EXCP_INTERRUPT;
1750 case KVM_EXIT_UNKNOWN:
1751 fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n",
1752 (uint64_t)run->hw.hardware_exit_reason);
1755 case KVM_EXIT_INTERNAL_ERROR:
1756 ret = kvm_handle_internal_error(cpu, run);
1759 DPRINTF("kvm_arch_handle_exit\n");
1760 ret = kvm_arch_handle_exit(cpu, run);
1766 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1767 vm_stop(RUN_STATE_INTERNAL_ERROR);
1770 cpu->exit_request = 0;
1774 int kvm_ioctl(KVMState *s, int type, ...)
1781 arg = va_arg(ap, void *);
1784 trace_kvm_ioctl(type, arg);
1785 ret = ioctl(s->fd, type, arg);
1792 int kvm_vm_ioctl(KVMState *s, int type, ...)
1799 arg = va_arg(ap, void *);
1802 trace_kvm_vm_ioctl(type, arg);
1803 ret = ioctl(s->vmfd, type, arg);
1810 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...)
1817 arg = va_arg(ap, void *);
1820 trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg);
1821 ret = ioctl(cpu->kvm_fd, type, arg);
1828 int kvm_device_ioctl(int fd, int type, ...)
1835 arg = va_arg(ap, void *);
1838 trace_kvm_device_ioctl(fd, type, arg);
1839 ret = ioctl(fd, type, arg);
1846 int kvm_has_sync_mmu(void)
1848 return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU);
1851 int kvm_has_vcpu_events(void)
1853 return kvm_state->vcpu_events;
1856 int kvm_has_robust_singlestep(void)
1858 return kvm_state->robust_singlestep;
1861 int kvm_has_debugregs(void)
1863 return kvm_state->debugregs;
1866 int kvm_has_xsave(void)
1868 return kvm_state->xsave;
1871 int kvm_has_xcrs(void)
1873 return kvm_state->xcrs;
1876 int kvm_has_pit_state2(void)
1878 return kvm_state->pit_state2;
1881 int kvm_has_many_ioeventfds(void)
1883 if (!kvm_enabled()) {
1886 return kvm_state->many_ioeventfds;
1889 int kvm_has_gsi_routing(void)
1891 #ifdef KVM_CAP_IRQ_ROUTING
1892 return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING);
1898 int kvm_has_intx_set_mask(void)
1900 return kvm_state->intx_set_mask;
1903 void kvm_setup_guest_memory(void *start, size_t size)
1905 #ifdef CONFIG_VALGRIND_H
1906 VALGRIND_MAKE_MEM_DEFINED(start, size);
1908 if (!kvm_has_sync_mmu()) {
1909 int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK);
1912 perror("qemu_madvise");
1914 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1920 #ifdef KVM_CAP_SET_GUEST_DEBUG
1921 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu,
1924 struct kvm_sw_breakpoint *bp;
1926 QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) {
1934 int kvm_sw_breakpoints_active(CPUState *cpu)
1936 return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints);
1939 struct kvm_set_guest_debug_data {
1940 struct kvm_guest_debug dbg;
1945 static void kvm_invoke_set_guest_debug(void *data)
1947 struct kvm_set_guest_debug_data *dbg_data = data;
1949 dbg_data->err = kvm_vcpu_ioctl(dbg_data->cpu, KVM_SET_GUEST_DEBUG,
1953 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
1955 struct kvm_set_guest_debug_data data;
1957 data.dbg.control = reinject_trap;
1959 if (cpu->singlestep_enabled) {
1960 data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
1962 kvm_arch_update_guest_debug(cpu, &data.dbg);
1965 run_on_cpu(cpu, kvm_invoke_set_guest_debug, &data);
1969 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
1970 target_ulong len, int type)
1972 struct kvm_sw_breakpoint *bp;
1975 if (type == GDB_BREAKPOINT_SW) {
1976 bp = kvm_find_sw_breakpoint(cpu, addr);
1982 bp = g_malloc(sizeof(struct kvm_sw_breakpoint));
1989 err = kvm_arch_insert_sw_breakpoint(cpu, bp);
1995 QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
1997 err = kvm_arch_insert_hw_breakpoint(addr, len, type);
2004 err = kvm_update_guest_debug(cpu, 0);
2012 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2013 target_ulong len, int type)
2015 struct kvm_sw_breakpoint *bp;
2018 if (type == GDB_BREAKPOINT_SW) {
2019 bp = kvm_find_sw_breakpoint(cpu, addr);
2024 if (bp->use_count > 1) {
2029 err = kvm_arch_remove_sw_breakpoint(cpu, bp);
2034 QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
2037 err = kvm_arch_remove_hw_breakpoint(addr, len, type);
2044 err = kvm_update_guest_debug(cpu, 0);
2052 void kvm_remove_all_breakpoints(CPUState *cpu)
2054 struct kvm_sw_breakpoint *bp, *next;
2055 KVMState *s = cpu->kvm_state;
2057 QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
2058 if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) {
2059 /* Try harder to find a CPU that currently sees the breakpoint. */
2061 if (kvm_arch_remove_sw_breakpoint(cpu, bp) == 0) {
2066 QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry);
2069 kvm_arch_remove_all_hw_breakpoints();
2072 kvm_update_guest_debug(cpu, 0);
2076 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2078 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2083 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2084 target_ulong len, int type)
2089 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2090 target_ulong len, int type)
2095 void kvm_remove_all_breakpoints(CPUState *cpu)
2098 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2100 int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset)
2102 struct kvm_signal_mask *sigmask;
2106 return kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, NULL);
2109 sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset));
2112 memcpy(sigmask->sigset, sigset, sizeof(*sigset));
2113 r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask);
2118 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2120 return kvm_arch_on_sigbus_vcpu(cpu, code, addr);
2123 int kvm_on_sigbus(int code, void *addr)
2125 return kvm_arch_on_sigbus(code, addr);
2128 int kvm_create_device(KVMState *s, uint64_t type, bool test)
2131 struct kvm_create_device create_dev;
2133 create_dev.type = type;
2135 create_dev.flags = test ? KVM_CREATE_DEVICE_TEST : 0;
2137 if (!kvm_check_extension(s, KVM_CAP_DEVICE_CTRL)) {
2141 ret = kvm_vm_ioctl(s, KVM_CREATE_DEVICE, &create_dev);
2146 return test ? 0 : create_dev.fd;
2149 int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source)
2151 struct kvm_one_reg reg;
2155 reg.addr = (uintptr_t) source;
2156 r = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
2158 trace_kvm_failed_reg_set(id, strerror(r));
2163 int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target)
2165 struct kvm_one_reg reg;
2169 reg.addr = (uintptr_t) target;
2170 r = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
2172 trace_kvm_failed_reg_get(id, strerror(r));