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 unsigned int sigmask_len;
103 #ifdef KVM_CAP_IRQ_ROUTING
104 struct kvm_irq_routing *irq_routes;
105 int nr_allocated_irq_routes;
106 uint32_t *used_gsi_bitmap;
107 unsigned int gsi_count;
108 QTAILQ_HEAD(msi_hashtab, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE];
114 bool kvm_kernel_irqchip;
115 bool kvm_async_interrupts_allowed;
116 bool kvm_halt_in_kernel_allowed;
117 bool kvm_eventfds_allowed;
118 bool kvm_irqfds_allowed;
119 bool kvm_msi_via_irqfd_allowed;
120 bool kvm_gsi_routing_allowed;
121 bool kvm_gsi_direct_mapping;
123 bool kvm_readonly_mem_allowed;
125 static const KVMCapabilityInfo kvm_required_capabilites[] = {
126 KVM_CAP_INFO(USER_MEMORY),
127 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS),
131 static KVMSlot *kvm_alloc_slot(KVMState *s)
135 for (i = 0; i < s->nr_slots; i++) {
136 if (s->slots[i].memory_size == 0) {
141 fprintf(stderr, "%s: no free slot available\n", __func__);
145 static KVMSlot *kvm_lookup_matching_slot(KVMState *s,
151 for (i = 0; i < s->nr_slots; i++) {
152 KVMSlot *mem = &s->slots[i];
154 if (start_addr == mem->start_addr &&
155 end_addr == mem->start_addr + mem->memory_size) {
164 * Find overlapping slot with lowest start address
166 static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s,
170 KVMSlot *found = NULL;
173 for (i = 0; i < s->nr_slots; i++) {
174 KVMSlot *mem = &s->slots[i];
176 if (mem->memory_size == 0 ||
177 (found && found->start_addr < mem->start_addr)) {
181 if (end_addr > mem->start_addr &&
182 start_addr < mem->start_addr + mem->memory_size) {
190 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram,
195 for (i = 0; i < s->nr_slots; i++) {
196 KVMSlot *mem = &s->slots[i];
198 if (ram >= mem->ram && ram < mem->ram + mem->memory_size) {
199 *phys_addr = mem->start_addr + (ram - mem->ram);
207 static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot)
209 struct kvm_userspace_memory_region mem;
211 mem.slot = slot->slot;
212 mem.guest_phys_addr = slot->start_addr;
213 mem.userspace_addr = (unsigned long)slot->ram;
214 mem.flags = slot->flags;
215 if (s->migration_log) {
216 mem.flags |= KVM_MEM_LOG_DIRTY_PAGES;
219 if (slot->memory_size && mem.flags & KVM_MEM_READONLY) {
220 /* Set the slot size to 0 before setting the slot to the desired
221 * value. This is needed based on KVM commit 75d61fbc. */
223 kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
225 mem.memory_size = slot->memory_size;
226 return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
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);
273 * dirty pages logging control
276 static int kvm_mem_flags(KVMState *s, bool log_dirty, bool readonly)
279 flags = log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0;
280 if (readonly && kvm_readonly_mem_allowed) {
281 flags |= KVM_MEM_READONLY;
286 static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty)
288 KVMState *s = kvm_state;
289 int flags, mask = KVM_MEM_LOG_DIRTY_PAGES;
292 old_flags = mem->flags;
294 flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty, false);
297 /* If nothing changed effectively, no need to issue ioctl */
298 if (s->migration_log) {
299 flags |= KVM_MEM_LOG_DIRTY_PAGES;
302 if (flags == old_flags) {
306 return kvm_set_user_memory_region(s, mem);
309 static int kvm_dirty_pages_log_change(hwaddr phys_addr,
310 ram_addr_t size, bool log_dirty)
312 KVMState *s = kvm_state;
313 KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size);
316 fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-"
317 TARGET_FMT_plx "\n", __func__, phys_addr,
318 (hwaddr)(phys_addr + size - 1));
321 return kvm_slot_dirty_pages_log_change(mem, log_dirty);
324 static void kvm_log_start(MemoryListener *listener,
325 MemoryRegionSection *section)
329 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
330 int128_get64(section->size), true);
336 static void kvm_log_stop(MemoryListener *listener,
337 MemoryRegionSection *section)
341 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
342 int128_get64(section->size), false);
348 static int kvm_set_migration_log(int enable)
350 KVMState *s = kvm_state;
354 s->migration_log = enable;
356 for (i = 0; i < s->nr_slots; i++) {
359 if (!mem->memory_size) {
362 if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) {
365 err = kvm_set_user_memory_region(s, mem);
373 /* get kvm's dirty pages bitmap and update qemu's */
374 static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section,
375 unsigned long *bitmap)
377 ram_addr_t start = section->offset_within_region + section->mr->ram_addr;
378 ram_addr_t pages = int128_get64(section->size) / getpagesize();
380 cpu_physical_memory_set_dirty_lebitmap(bitmap, start, pages);
384 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
387 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
388 * This function updates qemu's dirty bitmap using
389 * memory_region_set_dirty(). This means all bits are set
392 * @start_add: start of logged region.
393 * @end_addr: end of logged region.
395 static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection *section)
397 KVMState *s = kvm_state;
398 unsigned long size, allocated_size = 0;
402 hwaddr start_addr = section->offset_within_address_space;
403 hwaddr end_addr = start_addr + int128_get64(section->size);
405 d.dirty_bitmap = NULL;
406 while (start_addr < end_addr) {
407 mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr);
412 /* XXX bad kernel interface alert
413 * For dirty bitmap, kernel allocates array of size aligned to
414 * bits-per-long. But for case when the kernel is 64bits and
415 * the userspace is 32bits, userspace can't align to the same
416 * bits-per-long, since sizeof(long) is different between kernel
417 * and user space. This way, userspace will provide buffer which
418 * may be 4 bytes less than the kernel will use, resulting in
419 * userspace memory corruption (which is not detectable by valgrind
420 * too, in most cases).
421 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
422 * a hope that sizeof(long) wont become >8 any time soon.
424 size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS),
425 /*HOST_LONG_BITS*/ 64) / 8;
426 if (!d.dirty_bitmap) {
427 d.dirty_bitmap = g_malloc(size);
428 } else if (size > allocated_size) {
429 d.dirty_bitmap = g_realloc(d.dirty_bitmap, size);
431 allocated_size = size;
432 memset(d.dirty_bitmap, 0, allocated_size);
436 if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
437 DPRINTF("ioctl failed %d\n", errno);
442 kvm_get_dirty_pages_log_range(section, d.dirty_bitmap);
443 start_addr = mem->start_addr + mem->memory_size;
445 g_free(d.dirty_bitmap);
450 static void kvm_coalesce_mmio_region(MemoryListener *listener,
451 MemoryRegionSection *secion,
452 hwaddr start, hwaddr size)
454 KVMState *s = kvm_state;
456 if (s->coalesced_mmio) {
457 struct kvm_coalesced_mmio_zone zone;
463 (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
467 static void kvm_uncoalesce_mmio_region(MemoryListener *listener,
468 MemoryRegionSection *secion,
469 hwaddr start, hwaddr size)
471 KVMState *s = kvm_state;
473 if (s->coalesced_mmio) {
474 struct kvm_coalesced_mmio_zone zone;
480 (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
484 int kvm_check_extension(KVMState *s, unsigned int extension)
488 ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
496 int kvm_vm_check_extension(KVMState *s, unsigned int extension)
500 ret = kvm_vm_ioctl(s, KVM_CHECK_EXTENSION, extension);
502 /* VM wide version not implemented, use global one instead */
503 ret = kvm_check_extension(s, extension);
509 static int kvm_set_ioeventfd_mmio(int fd, hwaddr addr, uint32_t val,
510 bool assign, uint32_t size, bool datamatch)
513 struct kvm_ioeventfd iofd;
515 iofd.datamatch = datamatch ? val : 0;
521 if (!kvm_enabled()) {
526 iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
529 iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
532 ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd);
541 static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val,
542 bool assign, uint32_t size, bool datamatch)
544 struct kvm_ioeventfd kick = {
545 .datamatch = datamatch ? val : 0,
547 .flags = KVM_IOEVENTFD_FLAG_PIO,
552 if (!kvm_enabled()) {
556 kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
559 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
561 r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
569 static int kvm_check_many_ioeventfds(void)
571 /* Userspace can use ioeventfd for io notification. This requires a host
572 * that supports eventfd(2) and an I/O thread; since eventfd does not
573 * support SIGIO it cannot interrupt the vcpu.
575 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
576 * can avoid creating too many ioeventfds.
578 #if defined(CONFIG_EVENTFD)
581 for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) {
582 ioeventfds[i] = eventfd(0, EFD_CLOEXEC);
583 if (ioeventfds[i] < 0) {
586 ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true);
588 close(ioeventfds[i]);
593 /* Decide whether many devices are supported or not */
594 ret = i == ARRAY_SIZE(ioeventfds);
597 kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true);
598 close(ioeventfds[i]);
606 static const KVMCapabilityInfo *
607 kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list)
610 if (!kvm_check_extension(s, list->value)) {
618 static void kvm_set_phys_mem(MemoryRegionSection *section, bool add)
620 KVMState *s = kvm_state;
623 MemoryRegion *mr = section->mr;
624 bool log_dirty = memory_region_is_logging(mr);
625 bool writeable = !mr->readonly && !mr->rom_device;
626 bool readonly_flag = mr->readonly || memory_region_is_romd(mr);
627 hwaddr start_addr = section->offset_within_address_space;
628 ram_addr_t size = int128_get64(section->size);
632 /* kvm works in page size chunks, but the function may be called
633 with sub-page size and unaligned start address. */
634 delta = TARGET_PAGE_ALIGN(size) - size;
640 size &= TARGET_PAGE_MASK;
641 if (!size || (start_addr & ~TARGET_PAGE_MASK)) {
645 if (!memory_region_is_ram(mr)) {
646 if (writeable || !kvm_readonly_mem_allowed) {
648 } else if (!mr->romd_mode) {
649 /* If the memory device is not in romd_mode, then we actually want
650 * to remove the kvm memory slot so all accesses will trap. */
655 ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta;
658 mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
663 if (add && start_addr >= mem->start_addr &&
664 (start_addr + size <= mem->start_addr + mem->memory_size) &&
665 (ram - start_addr == mem->ram - mem->start_addr)) {
666 /* The new slot fits into the existing one and comes with
667 * identical parameters - update flags and done. */
668 kvm_slot_dirty_pages_log_change(mem, log_dirty);
674 if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
675 kvm_physical_sync_dirty_bitmap(section);
678 /* unregister the overlapping slot */
679 mem->memory_size = 0;
680 err = kvm_set_user_memory_region(s, mem);
682 fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
683 __func__, strerror(-err));
687 /* Workaround for older KVM versions: we can't join slots, even not by
688 * unregistering the previous ones and then registering the larger
689 * slot. We have to maintain the existing fragmentation. Sigh.
691 * This workaround assumes that the new slot starts at the same
692 * address as the first existing one. If not or if some overlapping
693 * slot comes around later, we will fail (not seen in practice so far)
694 * - and actually require a recent KVM version. */
695 if (s->broken_set_mem_region &&
696 old.start_addr == start_addr && old.memory_size < size && add) {
697 mem = kvm_alloc_slot(s);
698 mem->memory_size = old.memory_size;
699 mem->start_addr = old.start_addr;
701 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
703 err = kvm_set_user_memory_region(s, mem);
705 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
710 start_addr += old.memory_size;
711 ram += old.memory_size;
712 size -= old.memory_size;
716 /* register prefix slot */
717 if (old.start_addr < start_addr) {
718 mem = kvm_alloc_slot(s);
719 mem->memory_size = start_addr - old.start_addr;
720 mem->start_addr = old.start_addr;
722 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
724 err = kvm_set_user_memory_region(s, mem);
726 fprintf(stderr, "%s: error registering prefix slot: %s\n",
727 __func__, strerror(-err));
729 fprintf(stderr, "%s: This is probably because your kernel's " \
730 "PAGE_SIZE is too big. Please try to use 4k " \
731 "PAGE_SIZE!\n", __func__);
737 /* register suffix slot */
738 if (old.start_addr + old.memory_size > start_addr + size) {
739 ram_addr_t size_delta;
741 mem = kvm_alloc_slot(s);
742 mem->start_addr = start_addr + size;
743 size_delta = mem->start_addr - old.start_addr;
744 mem->memory_size = old.memory_size - size_delta;
745 mem->ram = old.ram + size_delta;
746 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
748 err = kvm_set_user_memory_region(s, mem);
750 fprintf(stderr, "%s: error registering suffix slot: %s\n",
751 __func__, strerror(-err));
757 /* in case the KVM bug workaround already "consumed" the new slot */
764 mem = kvm_alloc_slot(s);
765 mem->memory_size = size;
766 mem->start_addr = start_addr;
768 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
770 err = kvm_set_user_memory_region(s, mem);
772 fprintf(stderr, "%s: error registering slot: %s\n", __func__,
778 static void kvm_region_add(MemoryListener *listener,
779 MemoryRegionSection *section)
781 memory_region_ref(section->mr);
782 kvm_set_phys_mem(section, true);
785 static void kvm_region_del(MemoryListener *listener,
786 MemoryRegionSection *section)
788 kvm_set_phys_mem(section, false);
789 memory_region_unref(section->mr);
792 static void kvm_log_sync(MemoryListener *listener,
793 MemoryRegionSection *section)
797 r = kvm_physical_sync_dirty_bitmap(section);
803 static void kvm_log_global_start(struct MemoryListener *listener)
807 r = kvm_set_migration_log(1);
811 static void kvm_log_global_stop(struct MemoryListener *listener)
815 r = kvm_set_migration_log(0);
819 static void kvm_mem_ioeventfd_add(MemoryListener *listener,
820 MemoryRegionSection *section,
821 bool match_data, uint64_t data,
824 int fd = event_notifier_get_fd(e);
827 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
828 data, true, int128_get64(section->size),
831 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
832 __func__, strerror(-r));
837 static void kvm_mem_ioeventfd_del(MemoryListener *listener,
838 MemoryRegionSection *section,
839 bool match_data, uint64_t data,
842 int fd = event_notifier_get_fd(e);
845 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
846 data, false, int128_get64(section->size),
853 static void kvm_io_ioeventfd_add(MemoryListener *listener,
854 MemoryRegionSection *section,
855 bool match_data, uint64_t data,
858 int fd = event_notifier_get_fd(e);
861 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
862 data, true, int128_get64(section->size),
865 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
866 __func__, strerror(-r));
871 static void kvm_io_ioeventfd_del(MemoryListener *listener,
872 MemoryRegionSection *section,
873 bool match_data, uint64_t data,
877 int fd = event_notifier_get_fd(e);
880 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
881 data, false, int128_get64(section->size),
888 static MemoryListener kvm_memory_listener = {
889 .region_add = kvm_region_add,
890 .region_del = kvm_region_del,
891 .log_start = kvm_log_start,
892 .log_stop = kvm_log_stop,
893 .log_sync = kvm_log_sync,
894 .log_global_start = kvm_log_global_start,
895 .log_global_stop = kvm_log_global_stop,
896 .eventfd_add = kvm_mem_ioeventfd_add,
897 .eventfd_del = kvm_mem_ioeventfd_del,
898 .coalesced_mmio_add = kvm_coalesce_mmio_region,
899 .coalesced_mmio_del = kvm_uncoalesce_mmio_region,
903 static MemoryListener kvm_io_listener = {
904 .eventfd_add = kvm_io_ioeventfd_add,
905 .eventfd_del = kvm_io_ioeventfd_del,
909 static void kvm_handle_interrupt(CPUState *cpu, int mask)
911 cpu->interrupt_request |= mask;
913 if (!qemu_cpu_is_self(cpu)) {
918 int kvm_set_irq(KVMState *s, int irq, int level)
920 struct kvm_irq_level event;
923 assert(kvm_async_interrupts_enabled());
927 ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event);
929 perror("kvm_set_irq");
933 return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status;
936 #ifdef KVM_CAP_IRQ_ROUTING
937 typedef struct KVMMSIRoute {
938 struct kvm_irq_routing_entry kroute;
939 QTAILQ_ENTRY(KVMMSIRoute) entry;
942 static void set_gsi(KVMState *s, unsigned int gsi)
944 s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32);
947 static void clear_gsi(KVMState *s, unsigned int gsi)
949 s->used_gsi_bitmap[gsi / 32] &= ~(1U << (gsi % 32));
952 void kvm_init_irq_routing(KVMState *s)
956 gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING) - 1;
958 unsigned int gsi_bits, i;
960 /* Round up so we can search ints using ffs */
961 gsi_bits = ALIGN(gsi_count, 32);
962 s->used_gsi_bitmap = g_malloc0(gsi_bits / 8);
963 s->gsi_count = gsi_count;
965 /* Mark any over-allocated bits as already in use */
966 for (i = gsi_count; i < gsi_bits; i++) {
971 s->irq_routes = g_malloc0(sizeof(*s->irq_routes));
972 s->nr_allocated_irq_routes = 0;
974 if (!s->direct_msi) {
975 for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) {
976 QTAILQ_INIT(&s->msi_hashtab[i]);
980 kvm_arch_init_irq_routing(s);
983 void kvm_irqchip_commit_routes(KVMState *s)
987 s->irq_routes->flags = 0;
988 ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes);
992 static void kvm_add_routing_entry(KVMState *s,
993 struct kvm_irq_routing_entry *entry)
995 struct kvm_irq_routing_entry *new;
998 if (s->irq_routes->nr == s->nr_allocated_irq_routes) {
999 n = s->nr_allocated_irq_routes * 2;
1003 size = sizeof(struct kvm_irq_routing);
1004 size += n * sizeof(*new);
1005 s->irq_routes = g_realloc(s->irq_routes, size);
1006 s->nr_allocated_irq_routes = n;
1008 n = s->irq_routes->nr++;
1009 new = &s->irq_routes->entries[n];
1013 set_gsi(s, entry->gsi);
1016 static int kvm_update_routing_entry(KVMState *s,
1017 struct kvm_irq_routing_entry *new_entry)
1019 struct kvm_irq_routing_entry *entry;
1022 for (n = 0; n < s->irq_routes->nr; n++) {
1023 entry = &s->irq_routes->entries[n];
1024 if (entry->gsi != new_entry->gsi) {
1028 if(!memcmp(entry, new_entry, sizeof *entry)) {
1032 *entry = *new_entry;
1034 kvm_irqchip_commit_routes(s);
1042 void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin)
1044 struct kvm_irq_routing_entry e = {};
1046 assert(pin < s->gsi_count);
1049 e.type = KVM_IRQ_ROUTING_IRQCHIP;
1051 e.u.irqchip.irqchip = irqchip;
1052 e.u.irqchip.pin = pin;
1053 kvm_add_routing_entry(s, &e);
1056 void kvm_irqchip_release_virq(KVMState *s, int virq)
1058 struct kvm_irq_routing_entry *e;
1061 if (kvm_gsi_direct_mapping()) {
1065 for (i = 0; i < s->irq_routes->nr; i++) {
1066 e = &s->irq_routes->entries[i];
1067 if (e->gsi == virq) {
1068 s->irq_routes->nr--;
1069 *e = s->irq_routes->entries[s->irq_routes->nr];
1075 static unsigned int kvm_hash_msi(uint32_t data)
1077 /* This is optimized for IA32 MSI layout. However, no other arch shall
1078 * repeat the mistake of not providing a direct MSI injection API. */
1082 static void kvm_flush_dynamic_msi_routes(KVMState *s)
1084 KVMMSIRoute *route, *next;
1087 for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) {
1088 QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) {
1089 kvm_irqchip_release_virq(s, route->kroute.gsi);
1090 QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry);
1096 static int kvm_irqchip_get_virq(KVMState *s)
1098 uint32_t *word = s->used_gsi_bitmap;
1099 int max_words = ALIGN(s->gsi_count, 32) / 32;
1104 /* Return the lowest unused GSI in the bitmap */
1105 for (i = 0; i < max_words; i++) {
1106 bit = ffs(~word[i]);
1111 return bit - 1 + i * 32;
1113 if (!s->direct_msi && retry) {
1115 kvm_flush_dynamic_msi_routes(s);
1122 static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg)
1124 unsigned int hash = kvm_hash_msi(msg.data);
1127 QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) {
1128 if (route->kroute.u.msi.address_lo == (uint32_t)msg.address &&
1129 route->kroute.u.msi.address_hi == (msg.address >> 32) &&
1130 route->kroute.u.msi.data == le32_to_cpu(msg.data)) {
1137 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1142 if (s->direct_msi) {
1143 msi.address_lo = (uint32_t)msg.address;
1144 msi.address_hi = msg.address >> 32;
1145 msi.data = le32_to_cpu(msg.data);
1147 memset(msi.pad, 0, sizeof(msi.pad));
1149 return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi);
1152 route = kvm_lookup_msi_route(s, msg);
1156 virq = kvm_irqchip_get_virq(s);
1161 route = g_malloc0(sizeof(KVMMSIRoute));
1162 route->kroute.gsi = virq;
1163 route->kroute.type = KVM_IRQ_ROUTING_MSI;
1164 route->kroute.flags = 0;
1165 route->kroute.u.msi.address_lo = (uint32_t)msg.address;
1166 route->kroute.u.msi.address_hi = msg.address >> 32;
1167 route->kroute.u.msi.data = le32_to_cpu(msg.data);
1169 kvm_add_routing_entry(s, &route->kroute);
1170 kvm_irqchip_commit_routes(s);
1172 QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route,
1176 assert(route->kroute.type == KVM_IRQ_ROUTING_MSI);
1178 return kvm_set_irq(s, route->kroute.gsi, 1);
1181 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1183 struct kvm_irq_routing_entry kroute = {};
1186 if (kvm_gsi_direct_mapping()) {
1187 return msg.data & 0xffff;
1190 if (!kvm_gsi_routing_enabled()) {
1194 virq = kvm_irqchip_get_virq(s);
1200 kroute.type = KVM_IRQ_ROUTING_MSI;
1202 kroute.u.msi.address_lo = (uint32_t)msg.address;
1203 kroute.u.msi.address_hi = msg.address >> 32;
1204 kroute.u.msi.data = le32_to_cpu(msg.data);
1206 kvm_add_routing_entry(s, &kroute);
1207 kvm_irqchip_commit_routes(s);
1212 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1214 struct kvm_irq_routing_entry kroute = {};
1216 if (kvm_gsi_direct_mapping()) {
1220 if (!kvm_irqchip_in_kernel()) {
1225 kroute.type = KVM_IRQ_ROUTING_MSI;
1227 kroute.u.msi.address_lo = (uint32_t)msg.address;
1228 kroute.u.msi.address_hi = msg.address >> 32;
1229 kroute.u.msi.data = le32_to_cpu(msg.data);
1231 return kvm_update_routing_entry(s, &kroute);
1234 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int rfd, int virq,
1237 struct kvm_irqfd irqfd = {
1240 .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN,
1244 irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE;
1245 irqfd.resamplefd = rfd;
1248 if (!kvm_irqfds_enabled()) {
1252 return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd);
1255 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter)
1257 struct kvm_irq_routing_entry kroute;
1260 if (!kvm_gsi_routing_enabled()) {
1264 virq = kvm_irqchip_get_virq(s);
1270 kroute.type = KVM_IRQ_ROUTING_S390_ADAPTER;
1272 kroute.u.adapter.summary_addr = adapter->summary_addr;
1273 kroute.u.adapter.ind_addr = adapter->ind_addr;
1274 kroute.u.adapter.summary_offset = adapter->summary_offset;
1275 kroute.u.adapter.ind_offset = adapter->ind_offset;
1276 kroute.u.adapter.adapter_id = adapter->adapter_id;
1278 kvm_add_routing_entry(s, &kroute);
1279 kvm_irqchip_commit_routes(s);
1284 #else /* !KVM_CAP_IRQ_ROUTING */
1286 void kvm_init_irq_routing(KVMState *s)
1290 void kvm_irqchip_release_virq(KVMState *s, int virq)
1294 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1299 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1304 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter)
1309 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign)
1314 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1318 #endif /* !KVM_CAP_IRQ_ROUTING */
1320 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n,
1321 EventNotifier *rn, int virq)
1323 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n),
1324 rn ? event_notifier_get_fd(rn) : -1, virq, true);
1327 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, int virq)
1329 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), -1, virq,
1333 static int kvm_irqchip_create(KVMState *s)
1337 if (!qemu_opt_get_bool(qemu_get_machine_opts(), "kernel_irqchip", true) ||
1338 (!kvm_check_extension(s, KVM_CAP_IRQCHIP) &&
1339 (kvm_vm_enable_cap(s, KVM_CAP_S390_IRQCHIP, 0) < 0))) {
1343 /* First probe and see if there's a arch-specific hook to create the
1344 * in-kernel irqchip for us */
1345 ret = kvm_arch_irqchip_create(s);
1348 } else if (ret == 0) {
1349 ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP);
1351 fprintf(stderr, "Create kernel irqchip failed\n");
1356 kvm_kernel_irqchip = true;
1357 /* If we have an in-kernel IRQ chip then we must have asynchronous
1358 * interrupt delivery (though the reverse is not necessarily true)
1360 kvm_async_interrupts_allowed = true;
1361 kvm_halt_in_kernel_allowed = true;
1363 kvm_init_irq_routing(s);
1368 /* Find number of supported CPUs using the recommended
1369 * procedure from the kernel API documentation to cope with
1370 * older kernels that may be missing capabilities.
1372 static int kvm_recommended_vcpus(KVMState *s)
1374 int ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS);
1375 return (ret) ? ret : 4;
1378 static int kvm_max_vcpus(KVMState *s)
1380 int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS);
1381 return (ret) ? ret : kvm_recommended_vcpus(s);
1384 int kvm_init(MachineClass *mc)
1386 static const char upgrade_note[] =
1387 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1388 "(see http://sourceforge.net/projects/kvm).\n";
1393 { "SMP", smp_cpus },
1394 { "hotpluggable", max_cpus },
1397 int soft_vcpus_limit, hard_vcpus_limit;
1399 const KVMCapabilityInfo *missing_cap;
1402 const char *kvm_type;
1404 s = g_malloc0(sizeof(KVMState));
1407 * On systems where the kernel can support different base page
1408 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1409 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1410 * page size for the system though.
1412 assert(TARGET_PAGE_SIZE <= getpagesize());
1417 #ifdef KVM_CAP_SET_GUEST_DEBUG
1418 QTAILQ_INIT(&s->kvm_sw_breakpoints);
1421 s->fd = qemu_open("/dev/kvm", O_RDWR);
1423 fprintf(stderr, "Could not access KVM kernel module: %m\n");
1428 ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
1429 if (ret < KVM_API_VERSION) {
1433 fprintf(stderr, "kvm version too old\n");
1437 if (ret > KVM_API_VERSION) {
1439 fprintf(stderr, "kvm version not supported\n");
1443 s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
1445 /* If unspecified, use the default value */
1450 s->slots = g_malloc0(s->nr_slots * sizeof(KVMSlot));
1452 for (i = 0; i < s->nr_slots; i++) {
1453 s->slots[i].slot = i;
1456 /* check the vcpu limits */
1457 soft_vcpus_limit = kvm_recommended_vcpus(s);
1458 hard_vcpus_limit = kvm_max_vcpus(s);
1461 if (nc->num > soft_vcpus_limit) {
1463 "Warning: Number of %s cpus requested (%d) exceeds "
1464 "the recommended cpus supported by KVM (%d)\n",
1465 nc->name, nc->num, soft_vcpus_limit);
1467 if (nc->num > hard_vcpus_limit) {
1468 fprintf(stderr, "Number of %s cpus requested (%d) exceeds "
1469 "the maximum cpus supported by KVM (%d)\n",
1470 nc->name, nc->num, hard_vcpus_limit);
1477 kvm_type = qemu_opt_get(qemu_get_machine_opts(), "kvm-type");
1479 type = mc->kvm_type(kvm_type);
1480 } else if (kvm_type) {
1482 fprintf(stderr, "Invalid argument kvm-type=%s\n", kvm_type);
1487 ret = kvm_ioctl(s, KVM_CREATE_VM, type);
1488 } while (ret == -EINTR);
1491 fprintf(stderr, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret,
1495 fprintf(stderr, "Please add the 'switch_amode' kernel parameter to "
1496 "your host kernel command line\n");
1502 missing_cap = kvm_check_extension_list(s, kvm_required_capabilites);
1505 kvm_check_extension_list(s, kvm_arch_required_capabilities);
1509 fprintf(stderr, "kvm does not support %s\n%s",
1510 missing_cap->name, upgrade_note);
1514 s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
1516 s->broken_set_mem_region = 1;
1517 ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
1519 s->broken_set_mem_region = 0;
1522 #ifdef KVM_CAP_VCPU_EVENTS
1523 s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS);
1526 s->robust_singlestep =
1527 kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP);
1529 #ifdef KVM_CAP_DEBUGREGS
1530 s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS);
1533 #ifdef KVM_CAP_XSAVE
1534 s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE);
1538 s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS);
1541 #ifdef KVM_CAP_PIT_STATE2
1542 s->pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2);
1545 #ifdef KVM_CAP_IRQ_ROUTING
1546 s->direct_msi = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0);
1549 s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3);
1551 s->irq_set_ioctl = KVM_IRQ_LINE;
1552 if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) {
1553 s->irq_set_ioctl = KVM_IRQ_LINE_STATUS;
1556 #ifdef KVM_CAP_READONLY_MEM
1557 kvm_readonly_mem_allowed =
1558 (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0);
1561 kvm_eventfds_allowed =
1562 (kvm_check_extension(s, KVM_CAP_IOEVENTFD) > 0);
1564 ret = kvm_arch_init(s);
1569 ret = kvm_irqchip_create(s);
1575 memory_listener_register(&kvm_memory_listener, &address_space_memory);
1576 memory_listener_register(&kvm_io_listener, &address_space_io);
1578 s->many_ioeventfds = kvm_check_many_ioeventfds();
1580 cpu_interrupt_handler = kvm_handle_interrupt;
1598 void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len)
1600 s->sigmask_len = sigmask_len;
1603 static void kvm_handle_io(uint16_t port, void *data, int direction, int size,
1607 uint8_t *ptr = data;
1609 for (i = 0; i < count; i++) {
1610 address_space_rw(&address_space_io, port, ptr, size,
1611 direction == KVM_EXIT_IO_OUT);
1616 static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run)
1618 fprintf(stderr, "KVM internal error. Suberror: %d\n",
1619 run->internal.suberror);
1621 if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) {
1624 for (i = 0; i < run->internal.ndata; ++i) {
1625 fprintf(stderr, "extra data[%d]: %"PRIx64"\n",
1626 i, (uint64_t)run->internal.data[i]);
1629 if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) {
1630 fprintf(stderr, "emulation failure\n");
1631 if (!kvm_arch_stop_on_emulation_error(cpu)) {
1632 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1633 return EXCP_INTERRUPT;
1636 /* FIXME: Should trigger a qmp message to let management know
1637 * something went wrong.
1642 void kvm_flush_coalesced_mmio_buffer(void)
1644 KVMState *s = kvm_state;
1646 if (s->coalesced_flush_in_progress) {
1650 s->coalesced_flush_in_progress = true;
1652 if (s->coalesced_mmio_ring) {
1653 struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring;
1654 while (ring->first != ring->last) {
1655 struct kvm_coalesced_mmio *ent;
1657 ent = &ring->coalesced_mmio[ring->first];
1659 cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
1661 ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
1665 s->coalesced_flush_in_progress = false;
1668 static void do_kvm_cpu_synchronize_state(void *arg)
1670 CPUState *cpu = arg;
1672 if (!cpu->kvm_vcpu_dirty) {
1673 kvm_arch_get_registers(cpu);
1674 cpu->kvm_vcpu_dirty = true;
1678 void kvm_cpu_synchronize_state(CPUState *cpu)
1680 if (!cpu->kvm_vcpu_dirty) {
1681 run_on_cpu(cpu, do_kvm_cpu_synchronize_state, cpu);
1685 static void do_kvm_cpu_synchronize_post_reset(void *arg)
1687 CPUState *cpu = arg;
1689 kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE);
1690 cpu->kvm_vcpu_dirty = false;
1693 void kvm_cpu_synchronize_post_reset(CPUState *cpu)
1695 run_on_cpu(cpu, do_kvm_cpu_synchronize_post_reset, cpu);
1698 static void do_kvm_cpu_synchronize_post_init(void *arg)
1700 CPUState *cpu = arg;
1702 kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE);
1703 cpu->kvm_vcpu_dirty = false;
1706 void kvm_cpu_synchronize_post_init(CPUState *cpu)
1708 run_on_cpu(cpu, do_kvm_cpu_synchronize_post_init, cpu);
1711 int kvm_cpu_exec(CPUState *cpu)
1713 struct kvm_run *run = cpu->kvm_run;
1716 DPRINTF("kvm_cpu_exec()\n");
1718 if (kvm_arch_process_async_events(cpu)) {
1719 cpu->exit_request = 0;
1724 if (cpu->kvm_vcpu_dirty) {
1725 kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE);
1726 cpu->kvm_vcpu_dirty = false;
1729 kvm_arch_pre_run(cpu, run);
1730 if (cpu->exit_request) {
1731 DPRINTF("interrupt exit requested\n");
1733 * KVM requires us to reenter the kernel after IO exits to complete
1734 * instruction emulation. This self-signal will ensure that we
1737 qemu_cpu_kick_self();
1739 qemu_mutex_unlock_iothread();
1741 run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0);
1743 qemu_mutex_lock_iothread();
1744 kvm_arch_post_run(cpu, run);
1747 if (run_ret == -EINTR || run_ret == -EAGAIN) {
1748 DPRINTF("io window exit\n");
1749 ret = EXCP_INTERRUPT;
1752 fprintf(stderr, "error: kvm run failed %s\n",
1753 strerror(-run_ret));
1757 trace_kvm_run_exit(cpu->cpu_index, run->exit_reason);
1758 switch (run->exit_reason) {
1760 DPRINTF("handle_io\n");
1761 kvm_handle_io(run->io.port,
1762 (uint8_t *)run + run->io.data_offset,
1769 DPRINTF("handle_mmio\n");
1770 cpu_physical_memory_rw(run->mmio.phys_addr,
1773 run->mmio.is_write);
1776 case KVM_EXIT_IRQ_WINDOW_OPEN:
1777 DPRINTF("irq_window_open\n");
1778 ret = EXCP_INTERRUPT;
1780 case KVM_EXIT_SHUTDOWN:
1781 DPRINTF("shutdown\n");
1782 qemu_system_reset_request();
1783 ret = EXCP_INTERRUPT;
1785 case KVM_EXIT_UNKNOWN:
1786 fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n",
1787 (uint64_t)run->hw.hardware_exit_reason);
1790 case KVM_EXIT_INTERNAL_ERROR:
1791 ret = kvm_handle_internal_error(cpu, run);
1793 case KVM_EXIT_SYSTEM_EVENT:
1794 switch (run->system_event.type) {
1795 case KVM_SYSTEM_EVENT_SHUTDOWN:
1796 qemu_system_shutdown_request();
1797 ret = EXCP_INTERRUPT;
1799 case KVM_SYSTEM_EVENT_RESET:
1800 qemu_system_reset_request();
1801 ret = EXCP_INTERRUPT;
1804 DPRINTF("kvm_arch_handle_exit\n");
1805 ret = kvm_arch_handle_exit(cpu, run);
1810 DPRINTF("kvm_arch_handle_exit\n");
1811 ret = kvm_arch_handle_exit(cpu, run);
1817 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1818 vm_stop(RUN_STATE_INTERNAL_ERROR);
1821 cpu->exit_request = 0;
1825 int kvm_ioctl(KVMState *s, int type, ...)
1832 arg = va_arg(ap, void *);
1835 trace_kvm_ioctl(type, arg);
1836 ret = ioctl(s->fd, type, arg);
1843 int kvm_vm_ioctl(KVMState *s, int type, ...)
1850 arg = va_arg(ap, void *);
1853 trace_kvm_vm_ioctl(type, arg);
1854 ret = ioctl(s->vmfd, type, arg);
1861 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...)
1868 arg = va_arg(ap, void *);
1871 trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg);
1872 ret = ioctl(cpu->kvm_fd, type, arg);
1879 int kvm_device_ioctl(int fd, int type, ...)
1886 arg = va_arg(ap, void *);
1889 trace_kvm_device_ioctl(fd, type, arg);
1890 ret = ioctl(fd, type, arg);
1897 int kvm_has_sync_mmu(void)
1899 return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU);
1902 int kvm_has_vcpu_events(void)
1904 return kvm_state->vcpu_events;
1907 int kvm_has_robust_singlestep(void)
1909 return kvm_state->robust_singlestep;
1912 int kvm_has_debugregs(void)
1914 return kvm_state->debugregs;
1917 int kvm_has_xsave(void)
1919 return kvm_state->xsave;
1922 int kvm_has_xcrs(void)
1924 return kvm_state->xcrs;
1927 int kvm_has_pit_state2(void)
1929 return kvm_state->pit_state2;
1932 int kvm_has_many_ioeventfds(void)
1934 if (!kvm_enabled()) {
1937 return kvm_state->many_ioeventfds;
1940 int kvm_has_gsi_routing(void)
1942 #ifdef KVM_CAP_IRQ_ROUTING
1943 return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING);
1949 int kvm_has_intx_set_mask(void)
1951 return kvm_state->intx_set_mask;
1954 void kvm_setup_guest_memory(void *start, size_t size)
1956 #ifdef CONFIG_VALGRIND_H
1957 VALGRIND_MAKE_MEM_DEFINED(start, size);
1959 if (!kvm_has_sync_mmu()) {
1960 int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK);
1963 perror("qemu_madvise");
1965 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1971 #ifdef KVM_CAP_SET_GUEST_DEBUG
1972 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu,
1975 struct kvm_sw_breakpoint *bp;
1977 QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) {
1985 int kvm_sw_breakpoints_active(CPUState *cpu)
1987 return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints);
1990 struct kvm_set_guest_debug_data {
1991 struct kvm_guest_debug dbg;
1996 static void kvm_invoke_set_guest_debug(void *data)
1998 struct kvm_set_guest_debug_data *dbg_data = data;
2000 dbg_data->err = kvm_vcpu_ioctl(dbg_data->cpu, KVM_SET_GUEST_DEBUG,
2004 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2006 struct kvm_set_guest_debug_data data;
2008 data.dbg.control = reinject_trap;
2010 if (cpu->singlestep_enabled) {
2011 data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
2013 kvm_arch_update_guest_debug(cpu, &data.dbg);
2016 run_on_cpu(cpu, kvm_invoke_set_guest_debug, &data);
2020 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2021 target_ulong len, int type)
2023 struct kvm_sw_breakpoint *bp;
2026 if (type == GDB_BREAKPOINT_SW) {
2027 bp = kvm_find_sw_breakpoint(cpu, addr);
2033 bp = g_malloc(sizeof(struct kvm_sw_breakpoint));
2040 err = kvm_arch_insert_sw_breakpoint(cpu, bp);
2046 QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
2048 err = kvm_arch_insert_hw_breakpoint(addr, len, type);
2055 err = kvm_update_guest_debug(cpu, 0);
2063 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2064 target_ulong len, int type)
2066 struct kvm_sw_breakpoint *bp;
2069 if (type == GDB_BREAKPOINT_SW) {
2070 bp = kvm_find_sw_breakpoint(cpu, addr);
2075 if (bp->use_count > 1) {
2080 err = kvm_arch_remove_sw_breakpoint(cpu, bp);
2085 QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
2088 err = kvm_arch_remove_hw_breakpoint(addr, len, type);
2095 err = kvm_update_guest_debug(cpu, 0);
2103 void kvm_remove_all_breakpoints(CPUState *cpu)
2105 struct kvm_sw_breakpoint *bp, *next;
2106 KVMState *s = cpu->kvm_state;
2109 QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
2110 if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) {
2111 /* Try harder to find a CPU that currently sees the breakpoint. */
2112 CPU_FOREACH(tmpcpu) {
2113 if (kvm_arch_remove_sw_breakpoint(tmpcpu, bp) == 0) {
2118 QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry);
2121 kvm_arch_remove_all_hw_breakpoints();
2124 kvm_update_guest_debug(cpu, 0);
2128 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2130 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2135 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2136 target_ulong len, int type)
2141 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2142 target_ulong len, int type)
2147 void kvm_remove_all_breakpoints(CPUState *cpu)
2150 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2152 int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset)
2154 KVMState *s = kvm_state;
2155 struct kvm_signal_mask *sigmask;
2159 return kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, NULL);
2162 sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset));
2164 sigmask->len = s->sigmask_len;
2165 memcpy(sigmask->sigset, sigset, sizeof(*sigset));
2166 r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask);
2171 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2173 return kvm_arch_on_sigbus_vcpu(cpu, code, addr);
2176 int kvm_on_sigbus(int code, void *addr)
2178 return kvm_arch_on_sigbus(code, addr);
2181 int kvm_create_device(KVMState *s, uint64_t type, bool test)
2184 struct kvm_create_device create_dev;
2186 create_dev.type = type;
2188 create_dev.flags = test ? KVM_CREATE_DEVICE_TEST : 0;
2190 if (!kvm_check_extension(s, KVM_CAP_DEVICE_CTRL)) {
2194 ret = kvm_vm_ioctl(s, KVM_CREATE_DEVICE, &create_dev);
2199 return test ? 0 : create_dev.fd;
2202 int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source)
2204 struct kvm_one_reg reg;
2208 reg.addr = (uintptr_t) source;
2209 r = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
2211 trace_kvm_failed_reg_set(id, strerror(r));
2216 int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target)
2218 struct kvm_one_reg reg;
2222 reg.addr = (uintptr_t) target;
2223 r = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
2225 trace_kvm_failed_reg_get(id, strerror(r));