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"
28 #include "sysemu/accel.h"
30 #include "hw/pci/msi.h"
31 #include "hw/s390x/adapter.h"
32 #include "exec/gdbstub.h"
33 #include "sysemu/kvm.h"
34 #include "qemu/bswap.h"
35 #include "exec/memory.h"
36 #include "exec/ram_addr.h"
37 #include "exec/address-spaces.h"
38 #include "qemu/event_notifier.h"
41 #include "hw/boards.h"
43 /* This check must be after config-host.h is included */
45 #include <sys/eventfd.h>
48 /* KVM uses PAGE_SIZE in its definition of COALESCED_MMIO_MAX */
49 #define PAGE_SIZE TARGET_PAGE_SIZE
54 #define DPRINTF(fmt, ...) \
55 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
57 #define DPRINTF(fmt, ...) \
61 #define KVM_MSI_HASHTAB_SIZE 256
63 typedef struct KVMSlot
66 ram_addr_t memory_size;
72 typedef struct kvm_dirty_log KVMDirtyLog;
76 AccelState parent_obj;
83 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
84 bool coalesced_flush_in_progress;
85 int broken_set_mem_region;
87 int robust_singlestep;
89 #ifdef KVM_CAP_SET_GUEST_DEBUG
90 struct kvm_sw_breakpoint_head kvm_sw_breakpoints;
96 /* The man page (and posix) say ioctl numbers are signed int, but
97 * they're not. Linux, glibc and *BSD all treat ioctl numbers as
98 * unsigned, and treating them as signed here can break things */
99 unsigned irq_set_ioctl;
100 unsigned int sigmask_len;
101 #ifdef KVM_CAP_IRQ_ROUTING
102 struct kvm_irq_routing *irq_routes;
103 int nr_allocated_irq_routes;
104 uint32_t *used_gsi_bitmap;
105 unsigned int gsi_count;
106 QTAILQ_HEAD(msi_hashtab, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE];
111 #define TYPE_KVM_ACCEL ACCEL_CLASS_NAME("kvm")
113 #define KVM_STATE(obj) \
114 OBJECT_CHECK(KVMState, (obj), TYPE_KVM_ACCEL)
117 bool kvm_kernel_irqchip;
118 bool kvm_async_interrupts_allowed;
119 bool kvm_halt_in_kernel_allowed;
120 bool kvm_eventfds_allowed;
121 bool kvm_irqfds_allowed;
122 bool kvm_resamplefds_allowed;
123 bool kvm_msi_via_irqfd_allowed;
124 bool kvm_gsi_routing_allowed;
125 bool kvm_gsi_direct_mapping;
127 bool kvm_readonly_mem_allowed;
128 bool kvm_vm_attributes_allowed;
130 static const KVMCapabilityInfo kvm_required_capabilites[] = {
131 KVM_CAP_INFO(USER_MEMORY),
132 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS),
136 static KVMSlot *kvm_get_free_slot(KVMState *s)
140 for (i = 0; i < s->nr_slots; i++) {
141 if (s->slots[i].memory_size == 0) {
149 bool kvm_has_free_slot(MachineState *ms)
151 return kvm_get_free_slot(KVM_STATE(ms->accelerator));
154 static KVMSlot *kvm_alloc_slot(KVMState *s)
156 KVMSlot *slot = kvm_get_free_slot(s);
162 fprintf(stderr, "%s: no free slot available\n", __func__);
166 static KVMSlot *kvm_lookup_matching_slot(KVMState *s,
172 for (i = 0; i < s->nr_slots; i++) {
173 KVMSlot *mem = &s->slots[i];
175 if (start_addr == mem->start_addr &&
176 end_addr == mem->start_addr + mem->memory_size) {
185 * Find overlapping slot with lowest start address
187 static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s,
191 KVMSlot *found = NULL;
194 for (i = 0; i < s->nr_slots; i++) {
195 KVMSlot *mem = &s->slots[i];
197 if (mem->memory_size == 0 ||
198 (found && found->start_addr < mem->start_addr)) {
202 if (end_addr > mem->start_addr &&
203 start_addr < mem->start_addr + mem->memory_size) {
211 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram,
216 for (i = 0; i < s->nr_slots; i++) {
217 KVMSlot *mem = &s->slots[i];
219 if (ram >= mem->ram && ram < mem->ram + mem->memory_size) {
220 *phys_addr = mem->start_addr + (ram - mem->ram);
228 static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot)
230 struct kvm_userspace_memory_region mem;
232 mem.slot = slot->slot;
233 mem.guest_phys_addr = slot->start_addr;
234 mem.userspace_addr = (unsigned long)slot->ram;
235 mem.flags = slot->flags;
237 if (slot->memory_size && mem.flags & KVM_MEM_READONLY) {
238 /* Set the slot size to 0 before setting the slot to the desired
239 * value. This is needed based on KVM commit 75d61fbc. */
241 kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
243 mem.memory_size = slot->memory_size;
244 return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
247 int kvm_init_vcpu(CPUState *cpu)
249 KVMState *s = kvm_state;
253 DPRINTF("kvm_init_vcpu\n");
255 ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)kvm_arch_vcpu_id(cpu));
257 DPRINTF("kvm_create_vcpu failed\n");
263 cpu->kvm_vcpu_dirty = true;
265 mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
268 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
272 cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
274 if (cpu->kvm_run == MAP_FAILED) {
276 DPRINTF("mmap'ing vcpu state failed\n");
280 if (s->coalesced_mmio && !s->coalesced_mmio_ring) {
281 s->coalesced_mmio_ring =
282 (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE;
285 ret = kvm_arch_init_vcpu(cpu);
291 * dirty pages logging control
294 static int kvm_mem_flags(KVMState *s, bool log_dirty, bool readonly)
297 flags = log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0;
298 if (readonly && kvm_readonly_mem_allowed) {
299 flags |= KVM_MEM_READONLY;
304 static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty)
306 KVMState *s = kvm_state;
307 int flags, mask = KVM_MEM_LOG_DIRTY_PAGES;
310 old_flags = mem->flags;
312 flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty, false);
315 /* If nothing changed effectively, no need to issue ioctl */
316 if (flags == old_flags) {
320 return kvm_set_user_memory_region(s, mem);
323 static int kvm_dirty_pages_log_change(hwaddr phys_addr,
324 ram_addr_t size, bool log_dirty)
326 KVMState *s = kvm_state;
327 KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size);
332 return kvm_slot_dirty_pages_log_change(mem, log_dirty);
336 static void kvm_log_start(MemoryListener *listener,
337 MemoryRegionSection *section,
346 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
347 int128_get64(section->size), true);
353 static void kvm_log_stop(MemoryListener *listener,
354 MemoryRegionSection *section,
363 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
364 int128_get64(section->size), false);
370 /* get kvm's dirty pages bitmap and update qemu's */
371 static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section,
372 unsigned long *bitmap)
374 ram_addr_t start = section->offset_within_region + section->mr->ram_addr;
375 ram_addr_t pages = int128_get64(section->size) / getpagesize();
377 cpu_physical_memory_set_dirty_lebitmap(bitmap, start, pages);
381 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
384 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
385 * This function updates qemu's dirty bitmap using
386 * memory_region_set_dirty(). This means all bits are set
389 * @start_add: start of logged region.
390 * @end_addr: end of logged region.
392 static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection *section)
394 KVMState *s = kvm_state;
395 unsigned long size, allocated_size = 0;
399 hwaddr start_addr = section->offset_within_address_space;
400 hwaddr end_addr = start_addr + int128_get64(section->size);
402 d.dirty_bitmap = NULL;
403 while (start_addr < end_addr) {
404 mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr);
409 /* XXX bad kernel interface alert
410 * For dirty bitmap, kernel allocates array of size aligned to
411 * bits-per-long. But for case when the kernel is 64bits and
412 * the userspace is 32bits, userspace can't align to the same
413 * bits-per-long, since sizeof(long) is different between kernel
414 * and user space. This way, userspace will provide buffer which
415 * may be 4 bytes less than the kernel will use, resulting in
416 * userspace memory corruption (which is not detectable by valgrind
417 * too, in most cases).
418 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
419 * a hope that sizeof(long) wont become >8 any time soon.
421 size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS),
422 /*HOST_LONG_BITS*/ 64) / 8;
423 if (!d.dirty_bitmap) {
424 d.dirty_bitmap = g_malloc(size);
425 } else if (size > allocated_size) {
426 d.dirty_bitmap = g_realloc(d.dirty_bitmap, size);
428 allocated_size = size;
429 memset(d.dirty_bitmap, 0, allocated_size);
433 if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
434 DPRINTF("ioctl failed %d\n", errno);
439 kvm_get_dirty_pages_log_range(section, d.dirty_bitmap);
440 start_addr = mem->start_addr + mem->memory_size;
442 g_free(d.dirty_bitmap);
447 static void kvm_coalesce_mmio_region(MemoryListener *listener,
448 MemoryRegionSection *secion,
449 hwaddr start, hwaddr size)
451 KVMState *s = kvm_state;
453 if (s->coalesced_mmio) {
454 struct kvm_coalesced_mmio_zone zone;
460 (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
464 static void kvm_uncoalesce_mmio_region(MemoryListener *listener,
465 MemoryRegionSection *secion,
466 hwaddr start, hwaddr size)
468 KVMState *s = kvm_state;
470 if (s->coalesced_mmio) {
471 struct kvm_coalesced_mmio_zone zone;
477 (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
481 int kvm_check_extension(KVMState *s, unsigned int extension)
485 ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
493 int kvm_vm_check_extension(KVMState *s, unsigned int extension)
497 ret = kvm_vm_ioctl(s, KVM_CHECK_EXTENSION, extension);
499 /* VM wide version not implemented, use global one instead */
500 ret = kvm_check_extension(s, extension);
506 static uint32_t adjust_ioeventfd_endianness(uint32_t val, uint32_t size)
508 #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN)
509 /* The kernel expects ioeventfd values in HOST_WORDS_BIGENDIAN
510 * endianness, but the memory core hands them in target endianness.
511 * For example, PPC is always treated as big-endian even if running
512 * on KVM and on PPC64LE. Correct here.
526 static int kvm_set_ioeventfd_mmio(int fd, hwaddr addr, uint32_t val,
527 bool assign, uint32_t size, bool datamatch)
530 struct kvm_ioeventfd iofd = {
531 .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0,
538 if (!kvm_enabled()) {
543 iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
546 iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
549 ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd);
558 static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val,
559 bool assign, uint32_t size, bool datamatch)
561 struct kvm_ioeventfd kick = {
562 .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0,
564 .flags = KVM_IOEVENTFD_FLAG_PIO,
569 if (!kvm_enabled()) {
573 kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
576 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
578 r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
586 static int kvm_check_many_ioeventfds(void)
588 /* Userspace can use ioeventfd for io notification. This requires a host
589 * that supports eventfd(2) and an I/O thread; since eventfd does not
590 * support SIGIO it cannot interrupt the vcpu.
592 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
593 * can avoid creating too many ioeventfds.
595 #if defined(CONFIG_EVENTFD)
598 for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) {
599 ioeventfds[i] = eventfd(0, EFD_CLOEXEC);
600 if (ioeventfds[i] < 0) {
603 ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true);
605 close(ioeventfds[i]);
610 /* Decide whether many devices are supported or not */
611 ret = i == ARRAY_SIZE(ioeventfds);
614 kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true);
615 close(ioeventfds[i]);
623 static const KVMCapabilityInfo *
624 kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list)
627 if (!kvm_check_extension(s, list->value)) {
635 static void kvm_set_phys_mem(MemoryRegionSection *section, bool add)
637 KVMState *s = kvm_state;
640 MemoryRegion *mr = section->mr;
641 bool log_dirty = memory_region_get_dirty_log_mask(mr) != 0;
642 bool writeable = !mr->readonly && !mr->rom_device;
643 bool readonly_flag = mr->readonly || memory_region_is_romd(mr);
644 hwaddr start_addr = section->offset_within_address_space;
645 ram_addr_t size = int128_get64(section->size);
649 /* kvm works in page size chunks, but the function may be called
650 with sub-page size and unaligned start address. Pad the start
651 address to next and truncate size to previous page boundary. */
652 delta = (TARGET_PAGE_SIZE - (start_addr & ~TARGET_PAGE_MASK));
653 delta &= ~TARGET_PAGE_MASK;
659 size &= TARGET_PAGE_MASK;
660 if (!size || (start_addr & ~TARGET_PAGE_MASK)) {
664 if (!memory_region_is_ram(mr)) {
665 if (writeable || !kvm_readonly_mem_allowed) {
667 } else if (!mr->romd_mode) {
668 /* If the memory device is not in romd_mode, then we actually want
669 * to remove the kvm memory slot so all accesses will trap. */
674 ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta;
677 mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
682 if (add && start_addr >= mem->start_addr &&
683 (start_addr + size <= mem->start_addr + mem->memory_size) &&
684 (ram - start_addr == mem->ram - mem->start_addr)) {
685 /* The new slot fits into the existing one and comes with
686 * identical parameters - update flags and done. */
687 kvm_slot_dirty_pages_log_change(mem, log_dirty);
693 if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
694 kvm_physical_sync_dirty_bitmap(section);
697 /* unregister the overlapping slot */
698 mem->memory_size = 0;
699 err = kvm_set_user_memory_region(s, mem);
701 fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
702 __func__, strerror(-err));
706 /* Workaround for older KVM versions: we can't join slots, even not by
707 * unregistering the previous ones and then registering the larger
708 * slot. We have to maintain the existing fragmentation. Sigh.
710 * This workaround assumes that the new slot starts at the same
711 * address as the first existing one. If not or if some overlapping
712 * slot comes around later, we will fail (not seen in practice so far)
713 * - and actually require a recent KVM version. */
714 if (s->broken_set_mem_region &&
715 old.start_addr == start_addr && old.memory_size < size && add) {
716 mem = kvm_alloc_slot(s);
717 mem->memory_size = old.memory_size;
718 mem->start_addr = old.start_addr;
720 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
722 err = kvm_set_user_memory_region(s, mem);
724 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
729 start_addr += old.memory_size;
730 ram += old.memory_size;
731 size -= old.memory_size;
735 /* register prefix slot */
736 if (old.start_addr < start_addr) {
737 mem = kvm_alloc_slot(s);
738 mem->memory_size = start_addr - old.start_addr;
739 mem->start_addr = old.start_addr;
741 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
743 err = kvm_set_user_memory_region(s, mem);
745 fprintf(stderr, "%s: error registering prefix slot: %s\n",
746 __func__, strerror(-err));
748 fprintf(stderr, "%s: This is probably because your kernel's " \
749 "PAGE_SIZE is too big. Please try to use 4k " \
750 "PAGE_SIZE!\n", __func__);
756 /* register suffix slot */
757 if (old.start_addr + old.memory_size > start_addr + size) {
758 ram_addr_t size_delta;
760 mem = kvm_alloc_slot(s);
761 mem->start_addr = start_addr + size;
762 size_delta = mem->start_addr - old.start_addr;
763 mem->memory_size = old.memory_size - size_delta;
764 mem->ram = old.ram + size_delta;
765 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
767 err = kvm_set_user_memory_region(s, mem);
769 fprintf(stderr, "%s: error registering suffix slot: %s\n",
770 __func__, strerror(-err));
776 /* in case the KVM bug workaround already "consumed" the new slot */
783 mem = kvm_alloc_slot(s);
784 mem->memory_size = size;
785 mem->start_addr = start_addr;
787 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
789 err = kvm_set_user_memory_region(s, mem);
791 fprintf(stderr, "%s: error registering slot: %s\n", __func__,
797 static void kvm_region_add(MemoryListener *listener,
798 MemoryRegionSection *section)
800 memory_region_ref(section->mr);
801 kvm_set_phys_mem(section, true);
804 static void kvm_region_del(MemoryListener *listener,
805 MemoryRegionSection *section)
807 kvm_set_phys_mem(section, false);
808 memory_region_unref(section->mr);
811 static void kvm_log_sync(MemoryListener *listener,
812 MemoryRegionSection *section)
816 r = kvm_physical_sync_dirty_bitmap(section);
822 static void kvm_mem_ioeventfd_add(MemoryListener *listener,
823 MemoryRegionSection *section,
824 bool match_data, uint64_t data,
827 int fd = event_notifier_get_fd(e);
830 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
831 data, true, int128_get64(section->size),
834 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
835 __func__, strerror(-r));
840 static void kvm_mem_ioeventfd_del(MemoryListener *listener,
841 MemoryRegionSection *section,
842 bool match_data, uint64_t data,
845 int fd = event_notifier_get_fd(e);
848 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
849 data, false, int128_get64(section->size),
856 static void kvm_io_ioeventfd_add(MemoryListener *listener,
857 MemoryRegionSection *section,
858 bool match_data, uint64_t data,
861 int fd = event_notifier_get_fd(e);
864 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
865 data, true, int128_get64(section->size),
868 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
869 __func__, strerror(-r));
874 static void kvm_io_ioeventfd_del(MemoryListener *listener,
875 MemoryRegionSection *section,
876 bool match_data, uint64_t data,
880 int fd = event_notifier_get_fd(e);
883 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
884 data, false, int128_get64(section->size),
891 static MemoryListener kvm_memory_listener = {
892 .region_add = kvm_region_add,
893 .region_del = kvm_region_del,
894 .log_start = kvm_log_start,
895 .log_stop = kvm_log_stop,
896 .log_sync = kvm_log_sync,
897 .eventfd_add = kvm_mem_ioeventfd_add,
898 .eventfd_del = kvm_mem_ioeventfd_del,
899 .coalesced_mmio_add = kvm_coalesce_mmio_region,
900 .coalesced_mmio_del = kvm_uncoalesce_mmio_region,
904 static MemoryListener kvm_io_listener = {
905 .eventfd_add = kvm_io_ioeventfd_add,
906 .eventfd_del = kvm_io_ioeventfd_del,
910 static void kvm_handle_interrupt(CPUState *cpu, int mask)
912 cpu->interrupt_request |= mask;
914 if (!qemu_cpu_is_self(cpu)) {
919 int kvm_set_irq(KVMState *s, int irq, int level)
921 struct kvm_irq_level event;
924 assert(kvm_async_interrupts_enabled());
928 ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event);
930 perror("kvm_set_irq");
934 return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status;
937 #ifdef KVM_CAP_IRQ_ROUTING
938 typedef struct KVMMSIRoute {
939 struct kvm_irq_routing_entry kroute;
940 QTAILQ_ENTRY(KVMMSIRoute) entry;
943 static void set_gsi(KVMState *s, unsigned int gsi)
945 s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32);
948 static void clear_gsi(KVMState *s, unsigned int gsi)
950 s->used_gsi_bitmap[gsi / 32] &= ~(1U << (gsi % 32));
953 void kvm_init_irq_routing(KVMState *s)
957 gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING) - 1;
959 unsigned int gsi_bits, i;
961 /* Round up so we can search ints using ffs */
962 gsi_bits = ALIGN(gsi_count, 32);
963 s->used_gsi_bitmap = g_malloc0(gsi_bits / 8);
964 s->gsi_count = gsi_count;
966 /* Mark any over-allocated bits as already in use */
967 for (i = gsi_count; i < gsi_bits; i++) {
972 s->irq_routes = g_malloc0(sizeof(*s->irq_routes));
973 s->nr_allocated_irq_routes = 0;
975 if (!s->direct_msi) {
976 for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) {
977 QTAILQ_INIT(&s->msi_hashtab[i]);
981 kvm_arch_init_irq_routing(s);
984 void kvm_irqchip_commit_routes(KVMState *s)
988 s->irq_routes->flags = 0;
989 ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes);
993 static void kvm_add_routing_entry(KVMState *s,
994 struct kvm_irq_routing_entry *entry)
996 struct kvm_irq_routing_entry *new;
999 if (s->irq_routes->nr == s->nr_allocated_irq_routes) {
1000 n = s->nr_allocated_irq_routes * 2;
1004 size = sizeof(struct kvm_irq_routing);
1005 size += n * sizeof(*new);
1006 s->irq_routes = g_realloc(s->irq_routes, size);
1007 s->nr_allocated_irq_routes = n;
1009 n = s->irq_routes->nr++;
1010 new = &s->irq_routes->entries[n];
1014 set_gsi(s, entry->gsi);
1017 static int kvm_update_routing_entry(KVMState *s,
1018 struct kvm_irq_routing_entry *new_entry)
1020 struct kvm_irq_routing_entry *entry;
1023 for (n = 0; n < s->irq_routes->nr; n++) {
1024 entry = &s->irq_routes->entries[n];
1025 if (entry->gsi != new_entry->gsi) {
1029 if(!memcmp(entry, new_entry, sizeof *entry)) {
1033 *entry = *new_entry;
1035 kvm_irqchip_commit_routes(s);
1043 void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin)
1045 struct kvm_irq_routing_entry e = {};
1047 assert(pin < s->gsi_count);
1050 e.type = KVM_IRQ_ROUTING_IRQCHIP;
1052 e.u.irqchip.irqchip = irqchip;
1053 e.u.irqchip.pin = pin;
1054 kvm_add_routing_entry(s, &e);
1057 void kvm_irqchip_release_virq(KVMState *s, int virq)
1059 struct kvm_irq_routing_entry *e;
1062 if (kvm_gsi_direct_mapping()) {
1066 for (i = 0; i < s->irq_routes->nr; i++) {
1067 e = &s->irq_routes->entries[i];
1068 if (e->gsi == virq) {
1069 s->irq_routes->nr--;
1070 *e = s->irq_routes->entries[s->irq_routes->nr];
1076 static unsigned int kvm_hash_msi(uint32_t data)
1078 /* This is optimized for IA32 MSI layout. However, no other arch shall
1079 * repeat the mistake of not providing a direct MSI injection API. */
1083 static void kvm_flush_dynamic_msi_routes(KVMState *s)
1085 KVMMSIRoute *route, *next;
1088 for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) {
1089 QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) {
1090 kvm_irqchip_release_virq(s, route->kroute.gsi);
1091 QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry);
1097 static int kvm_irqchip_get_virq(KVMState *s)
1099 uint32_t *word = s->used_gsi_bitmap;
1100 int max_words = ALIGN(s->gsi_count, 32) / 32;
1105 /* Return the lowest unused GSI in the bitmap */
1106 for (i = 0; i < max_words; i++) {
1107 zeroes = ctz32(~word[i]);
1112 return zeroes + i * 32;
1114 if (!s->direct_msi && retry) {
1116 kvm_flush_dynamic_msi_routes(s);
1123 static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg)
1125 unsigned int hash = kvm_hash_msi(msg.data);
1128 QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) {
1129 if (route->kroute.u.msi.address_lo == (uint32_t)msg.address &&
1130 route->kroute.u.msi.address_hi == (msg.address >> 32) &&
1131 route->kroute.u.msi.data == le32_to_cpu(msg.data)) {
1138 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1143 if (s->direct_msi) {
1144 msi.address_lo = (uint32_t)msg.address;
1145 msi.address_hi = msg.address >> 32;
1146 msi.data = le32_to_cpu(msg.data);
1148 memset(msi.pad, 0, sizeof(msi.pad));
1150 return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi);
1153 route = kvm_lookup_msi_route(s, msg);
1157 virq = kvm_irqchip_get_virq(s);
1162 route = g_malloc0(sizeof(KVMMSIRoute));
1163 route->kroute.gsi = virq;
1164 route->kroute.type = KVM_IRQ_ROUTING_MSI;
1165 route->kroute.flags = 0;
1166 route->kroute.u.msi.address_lo = (uint32_t)msg.address;
1167 route->kroute.u.msi.address_hi = msg.address >> 32;
1168 route->kroute.u.msi.data = le32_to_cpu(msg.data);
1170 kvm_add_routing_entry(s, &route->kroute);
1171 kvm_irqchip_commit_routes(s);
1173 QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route,
1177 assert(route->kroute.type == KVM_IRQ_ROUTING_MSI);
1179 return kvm_set_irq(s, route->kroute.gsi, 1);
1182 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1184 struct kvm_irq_routing_entry kroute = {};
1187 if (kvm_gsi_direct_mapping()) {
1188 return kvm_arch_msi_data_to_gsi(msg.data);
1191 if (!kvm_gsi_routing_enabled()) {
1195 virq = kvm_irqchip_get_virq(s);
1201 kroute.type = KVM_IRQ_ROUTING_MSI;
1203 kroute.u.msi.address_lo = (uint32_t)msg.address;
1204 kroute.u.msi.address_hi = msg.address >> 32;
1205 kroute.u.msi.data = le32_to_cpu(msg.data);
1206 if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data)) {
1207 kvm_irqchip_release_virq(s, virq);
1211 kvm_add_routing_entry(s, &kroute);
1212 kvm_irqchip_commit_routes(s);
1217 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1219 struct kvm_irq_routing_entry kroute = {};
1221 if (kvm_gsi_direct_mapping()) {
1225 if (!kvm_irqchip_in_kernel()) {
1230 kroute.type = KVM_IRQ_ROUTING_MSI;
1232 kroute.u.msi.address_lo = (uint32_t)msg.address;
1233 kroute.u.msi.address_hi = msg.address >> 32;
1234 kroute.u.msi.data = le32_to_cpu(msg.data);
1235 if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data)) {
1239 return kvm_update_routing_entry(s, &kroute);
1242 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int rfd, int virq,
1245 struct kvm_irqfd irqfd = {
1248 .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN,
1252 irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE;
1253 irqfd.resamplefd = rfd;
1256 if (!kvm_irqfds_enabled()) {
1260 return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd);
1263 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter)
1265 struct kvm_irq_routing_entry kroute = {};
1268 if (!kvm_gsi_routing_enabled()) {
1272 virq = kvm_irqchip_get_virq(s);
1278 kroute.type = KVM_IRQ_ROUTING_S390_ADAPTER;
1280 kroute.u.adapter.summary_addr = adapter->summary_addr;
1281 kroute.u.adapter.ind_addr = adapter->ind_addr;
1282 kroute.u.adapter.summary_offset = adapter->summary_offset;
1283 kroute.u.adapter.ind_offset = adapter->ind_offset;
1284 kroute.u.adapter.adapter_id = adapter->adapter_id;
1286 kvm_add_routing_entry(s, &kroute);
1287 kvm_irqchip_commit_routes(s);
1292 #else /* !KVM_CAP_IRQ_ROUTING */
1294 void kvm_init_irq_routing(KVMState *s)
1298 void kvm_irqchip_release_virq(KVMState *s, int virq)
1302 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1307 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1312 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter)
1317 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign)
1322 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1326 #endif /* !KVM_CAP_IRQ_ROUTING */
1328 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n,
1329 EventNotifier *rn, int virq)
1331 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n),
1332 rn ? event_notifier_get_fd(rn) : -1, virq, true);
1335 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, int virq)
1337 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), -1, virq,
1341 static int kvm_irqchip_create(MachineState *machine, KVMState *s)
1345 if (!machine_kernel_irqchip_allowed(machine) ||
1346 (!kvm_check_extension(s, KVM_CAP_IRQCHIP) &&
1347 (kvm_vm_enable_cap(s, KVM_CAP_S390_IRQCHIP, 0) < 0))) {
1351 /* First probe and see if there's a arch-specific hook to create the
1352 * in-kernel irqchip for us */
1353 ret = kvm_arch_irqchip_create(s);
1356 } else if (ret == 0) {
1357 ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP);
1359 fprintf(stderr, "Create kernel irqchip failed\n");
1364 kvm_kernel_irqchip = true;
1365 /* If we have an in-kernel IRQ chip then we must have asynchronous
1366 * interrupt delivery (though the reverse is not necessarily true)
1368 kvm_async_interrupts_allowed = true;
1369 kvm_halt_in_kernel_allowed = true;
1371 kvm_init_irq_routing(s);
1376 /* Find number of supported CPUs using the recommended
1377 * procedure from the kernel API documentation to cope with
1378 * older kernels that may be missing capabilities.
1380 static int kvm_recommended_vcpus(KVMState *s)
1382 int ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS);
1383 return (ret) ? ret : 4;
1386 static int kvm_max_vcpus(KVMState *s)
1388 int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS);
1389 return (ret) ? ret : kvm_recommended_vcpus(s);
1392 static int kvm_init(MachineState *ms)
1394 MachineClass *mc = MACHINE_GET_CLASS(ms);
1395 static const char upgrade_note[] =
1396 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1397 "(see http://sourceforge.net/projects/kvm).\n";
1402 { "SMP", smp_cpus },
1403 { "hotpluggable", max_cpus },
1406 int soft_vcpus_limit, hard_vcpus_limit;
1408 const KVMCapabilityInfo *missing_cap;
1411 const char *kvm_type;
1413 s = KVM_STATE(ms->accelerator);
1416 * On systems where the kernel can support different base page
1417 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1418 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1419 * page size for the system though.
1421 assert(TARGET_PAGE_SIZE <= getpagesize());
1426 #ifdef KVM_CAP_SET_GUEST_DEBUG
1427 QTAILQ_INIT(&s->kvm_sw_breakpoints);
1430 s->fd = qemu_open("/dev/kvm", O_RDWR);
1432 fprintf(stderr, "Could not access KVM kernel module: %m\n");
1437 ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
1438 if (ret < KVM_API_VERSION) {
1442 fprintf(stderr, "kvm version too old\n");
1446 if (ret > KVM_API_VERSION) {
1448 fprintf(stderr, "kvm version not supported\n");
1452 s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
1454 /* If unspecified, use the default value */
1459 s->slots = g_malloc0(s->nr_slots * sizeof(KVMSlot));
1461 for (i = 0; i < s->nr_slots; i++) {
1462 s->slots[i].slot = i;
1465 /* check the vcpu limits */
1466 soft_vcpus_limit = kvm_recommended_vcpus(s);
1467 hard_vcpus_limit = kvm_max_vcpus(s);
1470 if (nc->num > soft_vcpus_limit) {
1472 "Warning: Number of %s cpus requested (%d) exceeds "
1473 "the recommended cpus supported by KVM (%d)\n",
1474 nc->name, nc->num, soft_vcpus_limit);
1476 if (nc->num > hard_vcpus_limit) {
1477 fprintf(stderr, "Number of %s cpus requested (%d) exceeds "
1478 "the maximum cpus supported by KVM (%d)\n",
1479 nc->name, nc->num, hard_vcpus_limit);
1486 kvm_type = qemu_opt_get(qemu_get_machine_opts(), "kvm-type");
1488 type = mc->kvm_type(kvm_type);
1489 } else if (kvm_type) {
1491 fprintf(stderr, "Invalid argument kvm-type=%s\n", kvm_type);
1496 ret = kvm_ioctl(s, KVM_CREATE_VM, type);
1497 } while (ret == -EINTR);
1500 fprintf(stderr, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret,
1504 if (ret == -EINVAL) {
1506 "Host kernel setup problem detected. Please verify:\n");
1507 fprintf(stderr, "- for kernels supporting the switch_amode or"
1508 " user_mode parameters, whether\n");
1510 " user space is running in primary address space\n");
1512 "- for kernels supporting the vm.allocate_pgste sysctl, "
1513 "whether it is enabled\n");
1520 missing_cap = kvm_check_extension_list(s, kvm_required_capabilites);
1523 kvm_check_extension_list(s, kvm_arch_required_capabilities);
1527 fprintf(stderr, "kvm does not support %s\n%s",
1528 missing_cap->name, upgrade_note);
1532 s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
1534 s->broken_set_mem_region = 1;
1535 ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
1537 s->broken_set_mem_region = 0;
1540 #ifdef KVM_CAP_VCPU_EVENTS
1541 s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS);
1544 s->robust_singlestep =
1545 kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP);
1547 #ifdef KVM_CAP_DEBUGREGS
1548 s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS);
1551 #ifdef KVM_CAP_XSAVE
1552 s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE);
1556 s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS);
1559 #ifdef KVM_CAP_PIT_STATE2
1560 s->pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2);
1563 #ifdef KVM_CAP_IRQ_ROUTING
1564 s->direct_msi = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0);
1567 s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3);
1569 s->irq_set_ioctl = KVM_IRQ_LINE;
1570 if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) {
1571 s->irq_set_ioctl = KVM_IRQ_LINE_STATUS;
1574 #ifdef KVM_CAP_READONLY_MEM
1575 kvm_readonly_mem_allowed =
1576 (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0);
1579 kvm_eventfds_allowed =
1580 (kvm_check_extension(s, KVM_CAP_IOEVENTFD) > 0);
1582 kvm_irqfds_allowed =
1583 (kvm_check_extension(s, KVM_CAP_IRQFD) > 0);
1585 kvm_resamplefds_allowed =
1586 (kvm_check_extension(s, KVM_CAP_IRQFD_RESAMPLE) > 0);
1588 kvm_vm_attributes_allowed =
1589 (kvm_check_extension(s, KVM_CAP_VM_ATTRIBUTES) > 0);
1591 ret = kvm_arch_init(ms, s);
1596 ret = kvm_irqchip_create(ms, s);
1602 memory_listener_register(&kvm_memory_listener, &address_space_memory);
1603 memory_listener_register(&kvm_io_listener, &address_space_io);
1605 s->many_ioeventfds = kvm_check_many_ioeventfds();
1607 cpu_interrupt_handler = kvm_handle_interrupt;
1624 void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len)
1626 s->sigmask_len = sigmask_len;
1629 static void kvm_handle_io(uint16_t port, MemTxAttrs attrs, void *data, int direction,
1630 int size, uint32_t count)
1633 uint8_t *ptr = data;
1635 for (i = 0; i < count; i++) {
1636 address_space_rw(&address_space_io, port, attrs,
1638 direction == KVM_EXIT_IO_OUT);
1643 static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run)
1645 fprintf(stderr, "KVM internal error. Suberror: %d\n",
1646 run->internal.suberror);
1648 if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) {
1651 for (i = 0; i < run->internal.ndata; ++i) {
1652 fprintf(stderr, "extra data[%d]: %"PRIx64"\n",
1653 i, (uint64_t)run->internal.data[i]);
1656 if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) {
1657 fprintf(stderr, "emulation failure\n");
1658 if (!kvm_arch_stop_on_emulation_error(cpu)) {
1659 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1660 return EXCP_INTERRUPT;
1663 /* FIXME: Should trigger a qmp message to let management know
1664 * something went wrong.
1669 void kvm_flush_coalesced_mmio_buffer(void)
1671 KVMState *s = kvm_state;
1673 if (s->coalesced_flush_in_progress) {
1677 s->coalesced_flush_in_progress = true;
1679 if (s->coalesced_mmio_ring) {
1680 struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring;
1681 while (ring->first != ring->last) {
1682 struct kvm_coalesced_mmio *ent;
1684 ent = &ring->coalesced_mmio[ring->first];
1686 cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
1688 ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
1692 s->coalesced_flush_in_progress = false;
1695 static void do_kvm_cpu_synchronize_state(void *arg)
1697 CPUState *cpu = arg;
1699 if (!cpu->kvm_vcpu_dirty) {
1700 kvm_arch_get_registers(cpu);
1701 cpu->kvm_vcpu_dirty = true;
1705 void kvm_cpu_synchronize_state(CPUState *cpu)
1707 if (!cpu->kvm_vcpu_dirty) {
1708 run_on_cpu(cpu, do_kvm_cpu_synchronize_state, cpu);
1712 static void do_kvm_cpu_synchronize_post_reset(void *arg)
1714 CPUState *cpu = arg;
1716 kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE);
1717 cpu->kvm_vcpu_dirty = false;
1720 void kvm_cpu_synchronize_post_reset(CPUState *cpu)
1722 run_on_cpu(cpu, do_kvm_cpu_synchronize_post_reset, cpu);
1725 static void do_kvm_cpu_synchronize_post_init(void *arg)
1727 CPUState *cpu = arg;
1729 kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE);
1730 cpu->kvm_vcpu_dirty = false;
1733 void kvm_cpu_synchronize_post_init(CPUState *cpu)
1735 run_on_cpu(cpu, do_kvm_cpu_synchronize_post_init, cpu);
1738 void kvm_cpu_clean_state(CPUState *cpu)
1740 cpu->kvm_vcpu_dirty = false;
1743 int kvm_cpu_exec(CPUState *cpu)
1745 struct kvm_run *run = cpu->kvm_run;
1748 DPRINTF("kvm_cpu_exec()\n");
1750 if (kvm_arch_process_async_events(cpu)) {
1751 cpu->exit_request = 0;
1758 if (cpu->kvm_vcpu_dirty) {
1759 kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE);
1760 cpu->kvm_vcpu_dirty = false;
1763 kvm_arch_pre_run(cpu, run);
1764 if (cpu->exit_request) {
1765 DPRINTF("interrupt exit requested\n");
1767 * KVM requires us to reenter the kernel after IO exits to complete
1768 * instruction emulation. This self-signal will ensure that we
1771 qemu_cpu_kick_self();
1773 qemu_mutex_unlock_iothread();
1775 run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0);
1777 qemu_mutex_lock_iothread();
1778 attrs = kvm_arch_post_run(cpu, run);
1781 if (run_ret == -EINTR || run_ret == -EAGAIN) {
1782 DPRINTF("io window exit\n");
1783 ret = EXCP_INTERRUPT;
1786 fprintf(stderr, "error: kvm run failed %s\n",
1787 strerror(-run_ret));
1789 if (run_ret == -EBUSY) {
1791 "This is probably because your SMT is enabled.\n"
1792 "VCPU can only run on primary threads with all "
1793 "secondary threads offline.\n");
1800 trace_kvm_run_exit(cpu->cpu_index, run->exit_reason);
1801 switch (run->exit_reason) {
1803 DPRINTF("handle_io\n");
1804 kvm_handle_io(run->io.port, attrs,
1805 (uint8_t *)run + run->io.data_offset,
1812 DPRINTF("handle_mmio\n");
1813 address_space_rw(&address_space_memory,
1814 run->mmio.phys_addr, attrs,
1817 run->mmio.is_write);
1820 case KVM_EXIT_IRQ_WINDOW_OPEN:
1821 DPRINTF("irq_window_open\n");
1822 ret = EXCP_INTERRUPT;
1824 case KVM_EXIT_SHUTDOWN:
1825 DPRINTF("shutdown\n");
1826 qemu_system_reset_request();
1827 ret = EXCP_INTERRUPT;
1829 case KVM_EXIT_UNKNOWN:
1830 fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n",
1831 (uint64_t)run->hw.hardware_exit_reason);
1834 case KVM_EXIT_INTERNAL_ERROR:
1835 ret = kvm_handle_internal_error(cpu, run);
1837 case KVM_EXIT_SYSTEM_EVENT:
1838 switch (run->system_event.type) {
1839 case KVM_SYSTEM_EVENT_SHUTDOWN:
1840 qemu_system_shutdown_request();
1841 ret = EXCP_INTERRUPT;
1843 case KVM_SYSTEM_EVENT_RESET:
1844 qemu_system_reset_request();
1845 ret = EXCP_INTERRUPT;
1848 DPRINTF("kvm_arch_handle_exit\n");
1849 ret = kvm_arch_handle_exit(cpu, run);
1854 DPRINTF("kvm_arch_handle_exit\n");
1855 ret = kvm_arch_handle_exit(cpu, run);
1861 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1862 vm_stop(RUN_STATE_INTERNAL_ERROR);
1865 cpu->exit_request = 0;
1869 int kvm_ioctl(KVMState *s, int type, ...)
1876 arg = va_arg(ap, void *);
1879 trace_kvm_ioctl(type, arg);
1880 ret = ioctl(s->fd, type, arg);
1887 int kvm_vm_ioctl(KVMState *s, int type, ...)
1894 arg = va_arg(ap, void *);
1897 trace_kvm_vm_ioctl(type, arg);
1898 ret = ioctl(s->vmfd, type, arg);
1905 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...)
1912 arg = va_arg(ap, void *);
1915 trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg);
1916 ret = ioctl(cpu->kvm_fd, type, arg);
1923 int kvm_device_ioctl(int fd, int type, ...)
1930 arg = va_arg(ap, void *);
1933 trace_kvm_device_ioctl(fd, type, arg);
1934 ret = ioctl(fd, type, arg);
1941 int kvm_vm_check_attr(KVMState *s, uint32_t group, uint64_t attr)
1944 struct kvm_device_attr attribute = {
1949 if (!kvm_vm_attributes_allowed) {
1953 ret = kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attribute);
1954 /* kvm returns 0 on success for HAS_DEVICE_ATTR */
1958 int kvm_has_sync_mmu(void)
1960 return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU);
1963 int kvm_has_vcpu_events(void)
1965 return kvm_state->vcpu_events;
1968 int kvm_has_robust_singlestep(void)
1970 return kvm_state->robust_singlestep;
1973 int kvm_has_debugregs(void)
1975 return kvm_state->debugregs;
1978 int kvm_has_xsave(void)
1980 return kvm_state->xsave;
1983 int kvm_has_xcrs(void)
1985 return kvm_state->xcrs;
1988 int kvm_has_pit_state2(void)
1990 return kvm_state->pit_state2;
1993 int kvm_has_many_ioeventfds(void)
1995 if (!kvm_enabled()) {
1998 return kvm_state->many_ioeventfds;
2001 int kvm_has_gsi_routing(void)
2003 #ifdef KVM_CAP_IRQ_ROUTING
2004 return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING);
2010 int kvm_has_intx_set_mask(void)
2012 return kvm_state->intx_set_mask;
2015 void kvm_setup_guest_memory(void *start, size_t size)
2017 if (!kvm_has_sync_mmu()) {
2018 int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK);
2021 perror("qemu_madvise");
2023 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
2029 #ifdef KVM_CAP_SET_GUEST_DEBUG
2030 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu,
2033 struct kvm_sw_breakpoint *bp;
2035 QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) {
2043 int kvm_sw_breakpoints_active(CPUState *cpu)
2045 return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints);
2048 struct kvm_set_guest_debug_data {
2049 struct kvm_guest_debug dbg;
2054 static void kvm_invoke_set_guest_debug(void *data)
2056 struct kvm_set_guest_debug_data *dbg_data = data;
2058 dbg_data->err = kvm_vcpu_ioctl(dbg_data->cpu, KVM_SET_GUEST_DEBUG,
2062 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2064 struct kvm_set_guest_debug_data data;
2066 data.dbg.control = reinject_trap;
2068 if (cpu->singlestep_enabled) {
2069 data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
2071 kvm_arch_update_guest_debug(cpu, &data.dbg);
2074 run_on_cpu(cpu, kvm_invoke_set_guest_debug, &data);
2078 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2079 target_ulong len, int type)
2081 struct kvm_sw_breakpoint *bp;
2084 if (type == GDB_BREAKPOINT_SW) {
2085 bp = kvm_find_sw_breakpoint(cpu, addr);
2091 bp = g_malloc(sizeof(struct kvm_sw_breakpoint));
2094 err = kvm_arch_insert_sw_breakpoint(cpu, bp);
2100 QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
2102 err = kvm_arch_insert_hw_breakpoint(addr, len, type);
2109 err = kvm_update_guest_debug(cpu, 0);
2117 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2118 target_ulong len, int type)
2120 struct kvm_sw_breakpoint *bp;
2123 if (type == GDB_BREAKPOINT_SW) {
2124 bp = kvm_find_sw_breakpoint(cpu, addr);
2129 if (bp->use_count > 1) {
2134 err = kvm_arch_remove_sw_breakpoint(cpu, bp);
2139 QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
2142 err = kvm_arch_remove_hw_breakpoint(addr, len, type);
2149 err = kvm_update_guest_debug(cpu, 0);
2157 void kvm_remove_all_breakpoints(CPUState *cpu)
2159 struct kvm_sw_breakpoint *bp, *next;
2160 KVMState *s = cpu->kvm_state;
2163 QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
2164 if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) {
2165 /* Try harder to find a CPU that currently sees the breakpoint. */
2166 CPU_FOREACH(tmpcpu) {
2167 if (kvm_arch_remove_sw_breakpoint(tmpcpu, bp) == 0) {
2172 QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry);
2175 kvm_arch_remove_all_hw_breakpoints();
2178 kvm_update_guest_debug(cpu, 0);
2182 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2184 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2189 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2190 target_ulong len, int type)
2195 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2196 target_ulong len, int type)
2201 void kvm_remove_all_breakpoints(CPUState *cpu)
2204 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2206 int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset)
2208 KVMState *s = kvm_state;
2209 struct kvm_signal_mask *sigmask;
2213 return kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, NULL);
2216 sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset));
2218 sigmask->len = s->sigmask_len;
2219 memcpy(sigmask->sigset, sigset, sizeof(*sigset));
2220 r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask);
2225 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2227 return kvm_arch_on_sigbus_vcpu(cpu, code, addr);
2230 int kvm_on_sigbus(int code, void *addr)
2232 return kvm_arch_on_sigbus(code, addr);
2235 int kvm_create_device(KVMState *s, uint64_t type, bool test)
2238 struct kvm_create_device create_dev;
2240 create_dev.type = type;
2242 create_dev.flags = test ? KVM_CREATE_DEVICE_TEST : 0;
2244 if (!kvm_check_extension(s, KVM_CAP_DEVICE_CTRL)) {
2248 ret = kvm_vm_ioctl(s, KVM_CREATE_DEVICE, &create_dev);
2253 return test ? 0 : create_dev.fd;
2256 int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source)
2258 struct kvm_one_reg reg;
2262 reg.addr = (uintptr_t) source;
2263 r = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
2265 trace_kvm_failed_reg_set(id, strerror(r));
2270 int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target)
2272 struct kvm_one_reg reg;
2276 reg.addr = (uintptr_t) target;
2277 r = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
2279 trace_kvm_failed_reg_get(id, strerror(r));
2284 static void kvm_accel_class_init(ObjectClass *oc, void *data)
2286 AccelClass *ac = ACCEL_CLASS(oc);
2288 ac->init_machine = kvm_init;
2289 ac->allowed = &kvm_allowed;
2292 static const TypeInfo kvm_accel_type = {
2293 .name = TYPE_KVM_ACCEL,
2294 .parent = TYPE_ACCEL,
2295 .class_init = kvm_accel_class_init,
2296 .instance_size = sizeof(KVMState),
2299 static void kvm_type_init(void)
2301 type_register_static(&kvm_accel_type);
2304 type_init(kvm_type_init);