2
* ARM implementation of KVM hooks
4
* Copyright Christoffer Dall 2009-2010
6
* This work is licensed under the terms of the GNU GPL, version 2 or later.
7
* See the COPYING file in the top-level directory.
11
#include "qemu/osdep.h"
12
#include <sys/ioctl.h>
14
#include <linux/kvm.h>
16
#include "qemu-common.h"
17
#include "qemu/timer.h"
18
#include "qemu/error-report.h"
19
#include "sysemu/sysemu.h"
20
#include "sysemu/kvm.h"
23
#include "internals.h"
24
#include "hw/arm/arm.h"
25
#include "exec/memattrs.h"
26
#include "exec/address-spaces.h"
27
#include "hw/boards.h"
30
const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
34
static bool cap_has_mp_state;
36
int kvm_arm_vcpu_init(CPUState *cs)
38
ARMCPU *cpu = ARM_CPU(cs);
39
struct kvm_vcpu_init init;
41
init.target = cpu->kvm_target;
42
memcpy(init.features, cpu->kvm_init_features, sizeof(init.features));
44
return kvm_vcpu_ioctl(cs, KVM_ARM_VCPU_INIT, &init);
47
bool kvm_arm_create_scratch_host_vcpu(const uint32_t *cpus_to_try,
49
struct kvm_vcpu_init *init)
51
int ret, kvmfd = -1, vmfd = -1, cpufd = -1;
53
kvmfd = qemu_open("/dev/kvm", O_RDWR);
57
vmfd = ioctl(kvmfd, KVM_CREATE_VM, 0);
61
cpufd = ioctl(vmfd, KVM_CREATE_VCPU, 0);
67
/* Caller doesn't want the VCPU to be initialized, so skip it */
71
ret = ioctl(vmfd, KVM_ARM_PREFERRED_TARGET, init);
73
ret = ioctl(cpufd, KVM_ARM_VCPU_INIT, init);
77
} else if (cpus_to_try) {
78
/* Old kernel which doesn't know about the
79
* PREFERRED_TARGET ioctl: we know it will only support
80
* creating one kind of guest CPU which is its preferred
83
while (*cpus_to_try != QEMU_KVM_ARM_TARGET_NONE) {
84
init->target = *cpus_to_try++;
85
memset(init->features, 0, sizeof(init->features));
86
ret = ioctl(cpufd, KVM_ARM_VCPU_INIT, init);
95
/* Treat a NULL cpus_to_try argument the same as an empty
96
* list, which means we will fail the call since this must
97
* be an old kernel which doesn't support PREFERRED_TARGET.
123
void kvm_arm_destroy_scratch_host_vcpu(int *fdarray)
127
for (i = 2; i >= 0; i--) {
132
static void kvm_arm_host_cpu_class_init(ObjectClass *oc, void *data)
134
ARMHostCPUClass *ahcc = ARM_HOST_CPU_CLASS(oc);
136
/* All we really need to set up for the 'host' CPU
137
* is the feature bits -- we rely on the fact that the
138
* various ID register values in ARMCPU are only used for
141
if (!kvm_arm_get_host_cpu_features(ahcc)) {
142
fprintf(stderr, "Failed to retrieve host CPU features!\n");
147
static void kvm_arm_host_cpu_initfn(Object *obj)
149
ARMHostCPUClass *ahcc = ARM_HOST_CPU_GET_CLASS(obj);
150
ARMCPU *cpu = ARM_CPU(obj);
151
CPUARMState *env = &cpu->env;
153
cpu->kvm_target = ahcc->target;
154
cpu->dtb_compatible = ahcc->dtb_compatible;
155
env->features = ahcc->features;
158
static const TypeInfo host_arm_cpu_type_info = {
159
.name = TYPE_ARM_HOST_CPU,
160
#ifdef TARGET_AARCH64
161
.parent = TYPE_AARCH64_CPU,
163
.parent = TYPE_ARM_CPU,
165
.instance_init = kvm_arm_host_cpu_initfn,
166
.class_init = kvm_arm_host_cpu_class_init,
167
.class_size = sizeof(ARMHostCPUClass),
170
int kvm_arch_init(MachineState *ms, KVMState *s)
172
/* For ARM interrupt delivery is always asynchronous,
173
* whether we are using an in-kernel VGIC or not.
175
kvm_async_interrupts_allowed = true;
177
cap_has_mp_state = kvm_check_extension(s, KVM_CAP_MP_STATE);
179
type_register_static(&host_arm_cpu_type_info);
184
unsigned long kvm_arch_vcpu_id(CPUState *cpu)
186
return cpu->cpu_index;
189
/* We track all the KVM devices which need their memory addresses
190
* passing to the kernel in a list of these structures.
191
* When board init is complete we run through the list and
192
* tell the kernel the base addresses of the memory regions.
193
* We use a MemoryListener to track mapping and unmapping of
194
* the regions during board creation, so the board models don't
195
* need to do anything special for the KVM case.
197
typedef struct KVMDevice {
198
struct kvm_arm_device_addr kda;
199
struct kvm_device_attr kdattr;
201
QSLIST_ENTRY(KVMDevice) entries;
205
static QSLIST_HEAD(kvm_devices_head, KVMDevice) kvm_devices_head;
207
static void kvm_arm_devlistener_add(MemoryListener *listener,
208
MemoryRegionSection *section)
212
QSLIST_FOREACH(kd, &kvm_devices_head, entries) {
213
if (section->mr == kd->mr) {
214
kd->kda.addr = section->offset_within_address_space;
219
static void kvm_arm_devlistener_del(MemoryListener *listener,
220
MemoryRegionSection *section)
224
QSLIST_FOREACH(kd, &kvm_devices_head, entries) {
225
if (section->mr == kd->mr) {
231
static MemoryListener devlistener = {
232
.region_add = kvm_arm_devlistener_add,
233
.region_del = kvm_arm_devlistener_del,
236
static void kvm_arm_set_device_addr(KVMDevice *kd)
238
struct kvm_device_attr *attr = &kd->kdattr;
241
/* If the device control API is available and we have a device fd on the
242
* KVMDevice struct, let's use the newer API
244
if (kd->dev_fd >= 0) {
245
uint64_t addr = kd->kda.addr;
246
attr->addr = (uintptr_t)&addr;
247
ret = kvm_device_ioctl(kd->dev_fd, KVM_SET_DEVICE_ATTR, attr);
249
ret = kvm_vm_ioctl(kvm_state, KVM_ARM_SET_DEVICE_ADDR, &kd->kda);
253
fprintf(stderr, "Failed to set device address: %s\n",
259
static void kvm_arm_machine_init_done(Notifier *notifier, void *data)
263
memory_listener_unregister(&devlistener);
264
QSLIST_FOREACH_SAFE(kd, &kvm_devices_head, entries, tkd) {
265
if (kd->kda.addr != -1) {
266
kvm_arm_set_device_addr(kd);
268
memory_region_unref(kd->mr);
273
static Notifier notify = {
274
.notify = kvm_arm_machine_init_done,
277
void kvm_arm_register_device(MemoryRegion *mr, uint64_t devid, uint64_t group,
278
uint64_t attr, int dev_fd)
282
if (!kvm_irqchip_in_kernel()) {
286
if (QSLIST_EMPTY(&kvm_devices_head)) {
287
memory_listener_register(&devlistener, &address_space_memory);
288
qemu_add_machine_init_done_notifier(¬ify);
290
kd = g_new0(KVMDevice, 1);
294
kd->kdattr.flags = 0;
295
kd->kdattr.group = group;
296
kd->kdattr.attr = attr;
298
QSLIST_INSERT_HEAD(&kvm_devices_head, kd, entries);
299
memory_region_ref(kd->mr);
302
static int compare_u64(const void *a, const void *b)
304
if (*(uint64_t *)a > *(uint64_t *)b) {
307
if (*(uint64_t *)a < *(uint64_t *)b) {
313
/* Initialize the CPUState's cpreg list according to the kernel's
314
* definition of what CPU registers it knows about (and throw away
315
* the previous TCG-created cpreg list).
317
int kvm_arm_init_cpreg_list(ARMCPU *cpu)
319
struct kvm_reg_list rl;
320
struct kvm_reg_list *rlp;
321
int i, ret, arraylen;
322
CPUState *cs = CPU(cpu);
325
ret = kvm_vcpu_ioctl(cs, KVM_GET_REG_LIST, &rl);
329
rlp = g_malloc(sizeof(struct kvm_reg_list) + rl.n * sizeof(uint64_t));
331
ret = kvm_vcpu_ioctl(cs, KVM_GET_REG_LIST, rlp);
335
/* Sort the list we get back from the kernel, since cpreg_tuples
336
* must be in strictly ascending order.
338
qsort(&rlp->reg, rlp->n, sizeof(rlp->reg[0]), compare_u64);
340
for (i = 0, arraylen = 0; i < rlp->n; i++) {
341
if (!kvm_arm_reg_syncs_via_cpreg_list(rlp->reg[i])) {
344
switch (rlp->reg[i] & KVM_REG_SIZE_MASK) {
345
case KVM_REG_SIZE_U32:
346
case KVM_REG_SIZE_U64:
349
fprintf(stderr, "Can't handle size of register in kernel list\n");
357
cpu->cpreg_indexes = g_renew(uint64_t, cpu->cpreg_indexes, arraylen);
358
cpu->cpreg_values = g_renew(uint64_t, cpu->cpreg_values, arraylen);
359
cpu->cpreg_vmstate_indexes = g_renew(uint64_t, cpu->cpreg_vmstate_indexes,
361
cpu->cpreg_vmstate_values = g_renew(uint64_t, cpu->cpreg_vmstate_values,
363
cpu->cpreg_array_len = arraylen;
364
cpu->cpreg_vmstate_array_len = arraylen;
366
for (i = 0, arraylen = 0; i < rlp->n; i++) {
367
uint64_t regidx = rlp->reg[i];
368
if (!kvm_arm_reg_syncs_via_cpreg_list(regidx)) {
371
cpu->cpreg_indexes[arraylen] = regidx;
374
assert(cpu->cpreg_array_len == arraylen);
376
if (!write_kvmstate_to_list(cpu)) {
377
/* Shouldn't happen unless kernel is inconsistent about
378
* what registers exist.
380
fprintf(stderr, "Initial read of kernel register state failed\n");
390
bool write_kvmstate_to_list(ARMCPU *cpu)
392
CPUState *cs = CPU(cpu);
396
for (i = 0; i < cpu->cpreg_array_len; i++) {
397
struct kvm_one_reg r;
398
uint64_t regidx = cpu->cpreg_indexes[i];
404
switch (regidx & KVM_REG_SIZE_MASK) {
405
case KVM_REG_SIZE_U32:
406
r.addr = (uintptr_t)&v32;
407
ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &r);
409
cpu->cpreg_values[i] = v32;
412
case KVM_REG_SIZE_U64:
413
r.addr = (uintptr_t)(cpu->cpreg_values + i);
414
ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &r);
426
bool write_list_to_kvmstate(ARMCPU *cpu, int level)
428
CPUState *cs = CPU(cpu);
432
for (i = 0; i < cpu->cpreg_array_len; i++) {
433
struct kvm_one_reg r;
434
uint64_t regidx = cpu->cpreg_indexes[i];
438
if (kvm_arm_cpreg_level(regidx) > level) {
443
switch (regidx & KVM_REG_SIZE_MASK) {
444
case KVM_REG_SIZE_U32:
445
v32 = cpu->cpreg_values[i];
446
r.addr = (uintptr_t)&v32;
448
case KVM_REG_SIZE_U64:
449
r.addr = (uintptr_t)(cpu->cpreg_values + i);
454
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &r);
456
/* We might fail for "unknown register" and also for
457
* "you tried to set a register which is constant with
458
* a different value from what it actually contains".
466
void kvm_arm_reset_vcpu(ARMCPU *cpu)
470
/* Re-init VCPU so that all registers are set to
471
* their respective reset values.
473
ret = kvm_arm_vcpu_init(CPU(cpu));
475
fprintf(stderr, "kvm_arm_vcpu_init failed: %s\n", strerror(-ret));
478
if (!write_kvmstate_to_list(cpu)) {
479
fprintf(stderr, "write_kvmstate_to_list failed\n");
485
* Update KVM's MP_STATE based on what QEMU thinks it is
487
int kvm_arm_sync_mpstate_to_kvm(ARMCPU *cpu)
489
if (cap_has_mp_state) {
490
struct kvm_mp_state mp_state = {
491
.mp_state = (cpu->power_state == PSCI_OFF) ?
492
KVM_MP_STATE_STOPPED : KVM_MP_STATE_RUNNABLE
494
int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
496
fprintf(stderr, "%s: failed to set MP_STATE %d/%s\n",
497
__func__, ret, strerror(-ret));
506
* Sync the KVM MP_STATE into QEMU
508
int kvm_arm_sync_mpstate_to_qemu(ARMCPU *cpu)
510
if (cap_has_mp_state) {
511
struct kvm_mp_state mp_state;
512
int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MP_STATE, &mp_state);
514
fprintf(stderr, "%s: failed to get MP_STATE %d/%s\n",
515
__func__, ret, strerror(-ret));
518
cpu->power_state = (mp_state.mp_state == KVM_MP_STATE_STOPPED) ?
525
void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
529
MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
531
return MEMTXATTRS_UNSPECIFIED;
535
int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
539
switch (run->exit_reason) {
541
if (kvm_arm_handle_debug(cs, &run->debug.arch)) {
543
} /* otherwise return to guest */
546
qemu_log_mask(LOG_UNIMP, "%s: un-handled exit reason %d\n",
547
__func__, run->exit_reason);
553
bool kvm_arch_stop_on_emulation_error(CPUState *cs)
558
int kvm_arch_process_async_events(CPUState *cs)
563
/* The #ifdef protections are until 32bit headers are imported and can
564
* be removed once both 32 and 64 bit reach feature parity.
566
void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg)
568
#ifdef KVM_GUESTDBG_USE_SW_BP
569
if (kvm_sw_breakpoints_active(cs)) {
570
dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
573
#ifdef KVM_GUESTDBG_USE_HW
574
if (kvm_arm_hw_debug_active(cs)) {
575
dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW;
576
kvm_arm_copy_hw_debug_data(&dbg->arch);
581
void kvm_arch_init_irq_routing(KVMState *s)
585
int kvm_arch_irqchip_create(MachineState *ms, KVMState *s)
587
if (machine_kernel_irqchip_split(ms)) {
588
perror("-machine kernel_irqchip=split is not supported on ARM.");
592
/* If we can create the VGIC using the newer device control API, we
593
* let the device do this when it initializes itself, otherwise we
594
* fall back to the old API */
595
return kvm_check_extension(s, KVM_CAP_DEVICE_CTRL);
598
int kvm_arm_vgic_probe(void)
600
if (kvm_create_device(kvm_state,
601
KVM_DEV_TYPE_ARM_VGIC_V3, true) == 0) {
603
} else if (kvm_create_device(kvm_state,
604
KVM_DEV_TYPE_ARM_VGIC_V2, true) == 0) {
611
int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
612
uint64_t address, uint32_t data, PCIDevice *dev)
617
int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
618
int vector, PCIDevice *dev)
623
int kvm_arch_release_virq_post(int virq)
628
int kvm_arch_msi_data_to_gsi(uint32_t data)
630
return (data - 32) & 0xffff;