1fcf5ef2aSThomas Huth /* 2fcf5ef2aSThomas Huth * ARM implementation of KVM hooks 3fcf5ef2aSThomas Huth * 4fcf5ef2aSThomas Huth * Copyright Christoffer Dall 2009-2010 5fcf5ef2aSThomas Huth * 6fcf5ef2aSThomas Huth * This work is licensed under the terms of the GNU GPL, version 2 or later. 7fcf5ef2aSThomas Huth * See the COPYING file in the top-level directory. 8fcf5ef2aSThomas Huth * 9fcf5ef2aSThomas Huth */ 10fcf5ef2aSThomas Huth 11fcf5ef2aSThomas Huth #include "qemu/osdep.h" 12fcf5ef2aSThomas Huth #include <sys/ioctl.h> 13fcf5ef2aSThomas Huth 14fcf5ef2aSThomas Huth #include <linux/kvm.h> 15fcf5ef2aSThomas Huth 16fcf5ef2aSThomas Huth #include "qemu-common.h" 17fcf5ef2aSThomas Huth #include "qemu/timer.h" 18fcf5ef2aSThomas Huth #include "qemu/error-report.h" 19db725815SMarkus Armbruster #include "qemu/main-loop.h" 20dea101a1SAndrew Jones #include "qom/object.h" 21dea101a1SAndrew Jones #include "qapi/error.h" 22fcf5ef2aSThomas Huth #include "sysemu/sysemu.h" 23fcf5ef2aSThomas Huth #include "sysemu/kvm.h" 24a27382e2SEric Auger #include "sysemu/kvm_int.h" 25fcf5ef2aSThomas Huth #include "kvm_arm.h" 26fcf5ef2aSThomas Huth #include "cpu.h" 27b05c81d2SEric Auger #include "trace.h" 28fcf5ef2aSThomas Huth #include "internals.h" 29b05c81d2SEric Auger #include "hw/pci/pci.h" 30fcf5ef2aSThomas Huth #include "exec/memattrs.h" 31fcf5ef2aSThomas Huth #include "exec/address-spaces.h" 32fcf5ef2aSThomas Huth #include "hw/boards.h" 3364552b6bSMarkus Armbruster #include "hw/irq.h" 34fcf5ef2aSThomas Huth #include "qemu/log.h" 35fcf5ef2aSThomas Huth 36fcf5ef2aSThomas Huth const KVMCapabilityInfo kvm_arch_required_capabilities[] = { 37fcf5ef2aSThomas Huth KVM_CAP_LAST_INFO 38fcf5ef2aSThomas Huth }; 39fcf5ef2aSThomas Huth 40fcf5ef2aSThomas Huth static bool cap_has_mp_state; 41202ccb6bSDongjiu Geng static bool cap_has_inject_serror_esr; 42694bcaa8SBeata Michalska static bool cap_has_inject_ext_dabt; 43fcf5ef2aSThomas Huth 44c4487d76SPeter Maydell static ARMHostCPUFeatures arm_host_cpu_features; 45c4487d76SPeter Maydell 46fcf5ef2aSThomas Huth int kvm_arm_vcpu_init(CPUState *cs) 47fcf5ef2aSThomas Huth { 48fcf5ef2aSThomas Huth ARMCPU *cpu = ARM_CPU(cs); 49fcf5ef2aSThomas Huth struct kvm_vcpu_init init; 50fcf5ef2aSThomas Huth 51fcf5ef2aSThomas Huth init.target = cpu->kvm_target; 52fcf5ef2aSThomas Huth memcpy(init.features, cpu->kvm_init_features, sizeof(init.features)); 53fcf5ef2aSThomas Huth 54fcf5ef2aSThomas Huth return kvm_vcpu_ioctl(cs, KVM_ARM_VCPU_INIT, &init); 55fcf5ef2aSThomas Huth } 56fcf5ef2aSThomas Huth 5714e99e0fSAndrew Jones int kvm_arm_vcpu_finalize(CPUState *cs, int feature) 5814e99e0fSAndrew Jones { 5914e99e0fSAndrew Jones return kvm_vcpu_ioctl(cs, KVM_ARM_VCPU_FINALIZE, &feature); 6014e99e0fSAndrew Jones } 6114e99e0fSAndrew Jones 62202ccb6bSDongjiu Geng void kvm_arm_init_serror_injection(CPUState *cs) 63202ccb6bSDongjiu Geng { 64202ccb6bSDongjiu Geng cap_has_inject_serror_esr = kvm_check_extension(cs->kvm_state, 65202ccb6bSDongjiu Geng KVM_CAP_ARM_INJECT_SERROR_ESR); 66202ccb6bSDongjiu Geng } 67202ccb6bSDongjiu Geng 68fcf5ef2aSThomas Huth bool kvm_arm_create_scratch_host_vcpu(const uint32_t *cpus_to_try, 69fcf5ef2aSThomas Huth int *fdarray, 70fcf5ef2aSThomas Huth struct kvm_vcpu_init *init) 71fcf5ef2aSThomas Huth { 720cdb4020SAndrew Jones int ret = 0, kvmfd = -1, vmfd = -1, cpufd = -1; 73fcf5ef2aSThomas Huth 74448058aaSDaniel P. Berrangé kvmfd = qemu_open_old("/dev/kvm", O_RDWR); 75fcf5ef2aSThomas Huth if (kvmfd < 0) { 76fcf5ef2aSThomas Huth goto err; 77fcf5ef2aSThomas Huth } 78fcf5ef2aSThomas Huth vmfd = ioctl(kvmfd, KVM_CREATE_VM, 0); 79fcf5ef2aSThomas Huth if (vmfd < 0) { 80fcf5ef2aSThomas Huth goto err; 81fcf5ef2aSThomas Huth } 82fcf5ef2aSThomas Huth cpufd = ioctl(vmfd, KVM_CREATE_VCPU, 0); 83fcf5ef2aSThomas Huth if (cpufd < 0) { 84fcf5ef2aSThomas Huth goto err; 85fcf5ef2aSThomas Huth } 86fcf5ef2aSThomas Huth 87fcf5ef2aSThomas Huth if (!init) { 88fcf5ef2aSThomas Huth /* Caller doesn't want the VCPU to be initialized, so skip it */ 89fcf5ef2aSThomas Huth goto finish; 90fcf5ef2aSThomas Huth } 91fcf5ef2aSThomas Huth 920cdb4020SAndrew Jones if (init->target == -1) { 930cdb4020SAndrew Jones struct kvm_vcpu_init preferred; 940cdb4020SAndrew Jones 950cdb4020SAndrew Jones ret = ioctl(vmfd, KVM_ARM_PREFERRED_TARGET, &preferred); 960cdb4020SAndrew Jones if (!ret) { 970cdb4020SAndrew Jones init->target = preferred.target; 980cdb4020SAndrew Jones } 990cdb4020SAndrew Jones } 100fcf5ef2aSThomas Huth if (ret >= 0) { 101fcf5ef2aSThomas Huth ret = ioctl(cpufd, KVM_ARM_VCPU_INIT, init); 102fcf5ef2aSThomas Huth if (ret < 0) { 103fcf5ef2aSThomas Huth goto err; 104fcf5ef2aSThomas Huth } 105fcf5ef2aSThomas Huth } else if (cpus_to_try) { 106fcf5ef2aSThomas Huth /* Old kernel which doesn't know about the 107fcf5ef2aSThomas Huth * PREFERRED_TARGET ioctl: we know it will only support 108fcf5ef2aSThomas Huth * creating one kind of guest CPU which is its preferred 109fcf5ef2aSThomas Huth * CPU type. 110fcf5ef2aSThomas Huth */ 1110cdb4020SAndrew Jones struct kvm_vcpu_init try; 1120cdb4020SAndrew Jones 113fcf5ef2aSThomas Huth while (*cpus_to_try != QEMU_KVM_ARM_TARGET_NONE) { 1140cdb4020SAndrew Jones try.target = *cpus_to_try++; 1150cdb4020SAndrew Jones memcpy(try.features, init->features, sizeof(init->features)); 1160cdb4020SAndrew Jones ret = ioctl(cpufd, KVM_ARM_VCPU_INIT, &try); 117fcf5ef2aSThomas Huth if (ret >= 0) { 118fcf5ef2aSThomas Huth break; 119fcf5ef2aSThomas Huth } 120fcf5ef2aSThomas Huth } 121fcf5ef2aSThomas Huth if (ret < 0) { 122fcf5ef2aSThomas Huth goto err; 123fcf5ef2aSThomas Huth } 1240cdb4020SAndrew Jones init->target = try.target; 125fcf5ef2aSThomas Huth } else { 126fcf5ef2aSThomas Huth /* Treat a NULL cpus_to_try argument the same as an empty 127fcf5ef2aSThomas Huth * list, which means we will fail the call since this must 128fcf5ef2aSThomas Huth * be an old kernel which doesn't support PREFERRED_TARGET. 129fcf5ef2aSThomas Huth */ 130fcf5ef2aSThomas Huth goto err; 131fcf5ef2aSThomas Huth } 132fcf5ef2aSThomas Huth 133fcf5ef2aSThomas Huth finish: 134fcf5ef2aSThomas Huth fdarray[0] = kvmfd; 135fcf5ef2aSThomas Huth fdarray[1] = vmfd; 136fcf5ef2aSThomas Huth fdarray[2] = cpufd; 137fcf5ef2aSThomas Huth 138fcf5ef2aSThomas Huth return true; 139fcf5ef2aSThomas Huth 140fcf5ef2aSThomas Huth err: 141fcf5ef2aSThomas Huth if (cpufd >= 0) { 142fcf5ef2aSThomas Huth close(cpufd); 143fcf5ef2aSThomas Huth } 144fcf5ef2aSThomas Huth if (vmfd >= 0) { 145fcf5ef2aSThomas Huth close(vmfd); 146fcf5ef2aSThomas Huth } 147fcf5ef2aSThomas Huth if (kvmfd >= 0) { 148fcf5ef2aSThomas Huth close(kvmfd); 149fcf5ef2aSThomas Huth } 150fcf5ef2aSThomas Huth 151fcf5ef2aSThomas Huth return false; 152fcf5ef2aSThomas Huth } 153fcf5ef2aSThomas Huth 154fcf5ef2aSThomas Huth void kvm_arm_destroy_scratch_host_vcpu(int *fdarray) 155fcf5ef2aSThomas Huth { 156fcf5ef2aSThomas Huth int i; 157fcf5ef2aSThomas Huth 158fcf5ef2aSThomas Huth for (i = 2; i >= 0; i--) { 159fcf5ef2aSThomas Huth close(fdarray[i]); 160fcf5ef2aSThomas Huth } 161fcf5ef2aSThomas Huth } 162fcf5ef2aSThomas Huth 163c4487d76SPeter Maydell void kvm_arm_set_cpu_features_from_host(ARMCPU *cpu) 164fcf5ef2aSThomas Huth { 165c4487d76SPeter Maydell CPUARMState *env = &cpu->env; 166fcf5ef2aSThomas Huth 167c4487d76SPeter Maydell if (!arm_host_cpu_features.dtb_compatible) { 168c4487d76SPeter Maydell if (!kvm_enabled() || 169c4487d76SPeter Maydell !kvm_arm_get_host_cpu_features(&arm_host_cpu_features)) { 170c4487d76SPeter Maydell /* We can't report this error yet, so flag that we need to 171c4487d76SPeter Maydell * in arm_cpu_realizefn(). 172fcf5ef2aSThomas Huth */ 173c4487d76SPeter Maydell cpu->kvm_target = QEMU_KVM_ARM_TARGET_NONE; 174c4487d76SPeter Maydell cpu->host_cpu_probe_failed = true; 175c4487d76SPeter Maydell return; 176fcf5ef2aSThomas Huth } 177fcf5ef2aSThomas Huth } 178fcf5ef2aSThomas Huth 179c4487d76SPeter Maydell cpu->kvm_target = arm_host_cpu_features.target; 180c4487d76SPeter Maydell cpu->dtb_compatible = arm_host_cpu_features.dtb_compatible; 1814674097cSRichard Henderson cpu->isar = arm_host_cpu_features.isar; 182c4487d76SPeter Maydell env->features = arm_host_cpu_features.features; 183c4487d76SPeter Maydell } 184c4487d76SPeter Maydell 185dea101a1SAndrew Jones static bool kvm_no_adjvtime_get(Object *obj, Error **errp) 186dea101a1SAndrew Jones { 187dea101a1SAndrew Jones return !ARM_CPU(obj)->kvm_adjvtime; 188dea101a1SAndrew Jones } 189dea101a1SAndrew Jones 190dea101a1SAndrew Jones static void kvm_no_adjvtime_set(Object *obj, bool value, Error **errp) 191dea101a1SAndrew Jones { 192dea101a1SAndrew Jones ARM_CPU(obj)->kvm_adjvtime = !value; 193dea101a1SAndrew Jones } 194dea101a1SAndrew Jones 19568970d1eSAndrew Jones static bool kvm_steal_time_get(Object *obj, Error **errp) 19668970d1eSAndrew Jones { 19768970d1eSAndrew Jones return ARM_CPU(obj)->kvm_steal_time != ON_OFF_AUTO_OFF; 19868970d1eSAndrew Jones } 19968970d1eSAndrew Jones 20068970d1eSAndrew Jones static void kvm_steal_time_set(Object *obj, bool value, Error **errp) 20168970d1eSAndrew Jones { 20268970d1eSAndrew Jones ARM_CPU(obj)->kvm_steal_time = value ? ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF; 20368970d1eSAndrew Jones } 20468970d1eSAndrew Jones 205dea101a1SAndrew Jones /* KVM VCPU properties should be prefixed with "kvm-". */ 206dea101a1SAndrew Jones void kvm_arm_add_vcpu_properties(Object *obj) 207dea101a1SAndrew Jones { 2089e6f8d8aSfangying ARMCPU *cpu = ARM_CPU(obj); 2099e6f8d8aSfangying CPUARMState *env = &cpu->env; 210dea101a1SAndrew Jones 2119e6f8d8aSfangying if (arm_feature(env, ARM_FEATURE_GENERIC_TIMER)) { 2129e6f8d8aSfangying cpu->kvm_adjvtime = true; 213dea101a1SAndrew Jones object_property_add_bool(obj, "kvm-no-adjvtime", kvm_no_adjvtime_get, 214d2623129SMarkus Armbruster kvm_no_adjvtime_set); 215dea101a1SAndrew Jones object_property_set_description(obj, "kvm-no-adjvtime", 216dea101a1SAndrew Jones "Set on to disable the adjustment of " 217dea101a1SAndrew Jones "the virtual counter. VM stopped time " 2187eecec7dSMarkus Armbruster "will be counted."); 219dea101a1SAndrew Jones } 22068970d1eSAndrew Jones 22168970d1eSAndrew Jones cpu->kvm_steal_time = ON_OFF_AUTO_AUTO; 22268970d1eSAndrew Jones object_property_add_bool(obj, "kvm-steal-time", kvm_steal_time_get, 22368970d1eSAndrew Jones kvm_steal_time_set); 22468970d1eSAndrew Jones object_property_set_description(obj, "kvm-steal-time", 22568970d1eSAndrew Jones "Set off to disable KVM steal time."); 2269e6f8d8aSfangying } 227dea101a1SAndrew Jones 2287d20e681SPhilippe Mathieu-Daudé bool kvm_arm_pmu_supported(void) 229ae502508SAndrew Jones { 2307d20e681SPhilippe Mathieu-Daudé return kvm_check_extension(kvm_state, KVM_CAP_ARM_PMU_V3); 231ae502508SAndrew Jones } 232ae502508SAndrew Jones 233*bcb902a1SAndrew Jones int kvm_arm_get_max_vm_ipa_size(MachineState *ms, bool *fixed_ipa) 234a27382e2SEric Auger { 235a27382e2SEric Auger KVMState *s = KVM_STATE(ms->accelerator); 236a27382e2SEric Auger int ret; 237a27382e2SEric Auger 238a27382e2SEric Auger ret = kvm_check_extension(s, KVM_CAP_ARM_VM_IPA_SIZE); 239*bcb902a1SAndrew Jones *fixed_ipa = ret <= 0; 240*bcb902a1SAndrew Jones 241a27382e2SEric Auger return ret > 0 ? ret : 40; 242a27382e2SEric Auger } 243a27382e2SEric Auger 244fcf5ef2aSThomas Huth int kvm_arch_init(MachineState *ms, KVMState *s) 245fcf5ef2aSThomas Huth { 246fff9f555SEric Auger int ret = 0; 247fcf5ef2aSThomas Huth /* For ARM interrupt delivery is always asynchronous, 248fcf5ef2aSThomas Huth * whether we are using an in-kernel VGIC or not. 249fcf5ef2aSThomas Huth */ 250fcf5ef2aSThomas Huth kvm_async_interrupts_allowed = true; 251fcf5ef2aSThomas Huth 2525d721b78SAlexander Graf /* 2535d721b78SAlexander Graf * PSCI wakes up secondary cores, so we always need to 2545d721b78SAlexander Graf * have vCPUs waiting in kernel space 2555d721b78SAlexander Graf */ 2565d721b78SAlexander Graf kvm_halt_in_kernel_allowed = true; 2575d721b78SAlexander Graf 258fcf5ef2aSThomas Huth cap_has_mp_state = kvm_check_extension(s, KVM_CAP_MP_STATE); 259fcf5ef2aSThomas Huth 260fff9f555SEric Auger if (ms->smp.cpus > 256 && 261fff9f555SEric Auger !kvm_check_extension(s, KVM_CAP_ARM_IRQ_LINE_LAYOUT_2)) { 262fff9f555SEric Auger error_report("Using more than 256 vcpus requires a host kernel " 263fff9f555SEric Auger "with KVM_CAP_ARM_IRQ_LINE_LAYOUT_2"); 264fff9f555SEric Auger ret = -EINVAL; 265fff9f555SEric Auger } 266fff9f555SEric Auger 267694bcaa8SBeata Michalska if (kvm_check_extension(s, KVM_CAP_ARM_NISV_TO_USER)) { 268694bcaa8SBeata Michalska if (kvm_vm_enable_cap(s, KVM_CAP_ARM_NISV_TO_USER, 0)) { 269694bcaa8SBeata Michalska error_report("Failed to enable KVM_CAP_ARM_NISV_TO_USER cap"); 270694bcaa8SBeata Michalska } else { 271694bcaa8SBeata Michalska /* Set status for supporting the external dabt injection */ 272694bcaa8SBeata Michalska cap_has_inject_ext_dabt = kvm_check_extension(s, 273694bcaa8SBeata Michalska KVM_CAP_ARM_INJECT_EXT_DABT); 274694bcaa8SBeata Michalska } 275694bcaa8SBeata Michalska } 276694bcaa8SBeata Michalska 277fff9f555SEric Auger return ret; 278fcf5ef2aSThomas Huth } 279fcf5ef2aSThomas Huth 280fcf5ef2aSThomas Huth unsigned long kvm_arch_vcpu_id(CPUState *cpu) 281fcf5ef2aSThomas Huth { 282fcf5ef2aSThomas Huth return cpu->cpu_index; 283fcf5ef2aSThomas Huth } 284fcf5ef2aSThomas Huth 285fcf5ef2aSThomas Huth /* We track all the KVM devices which need their memory addresses 286fcf5ef2aSThomas Huth * passing to the kernel in a list of these structures. 287fcf5ef2aSThomas Huth * When board init is complete we run through the list and 288fcf5ef2aSThomas Huth * tell the kernel the base addresses of the memory regions. 289fcf5ef2aSThomas Huth * We use a MemoryListener to track mapping and unmapping of 290fcf5ef2aSThomas Huth * the regions during board creation, so the board models don't 291fcf5ef2aSThomas Huth * need to do anything special for the KVM case. 29219d1bd0bSEric Auger * 29319d1bd0bSEric Auger * Sometimes the address must be OR'ed with some other fields 29419d1bd0bSEric Auger * (for example for KVM_VGIC_V3_ADDR_TYPE_REDIST_REGION). 29519d1bd0bSEric Auger * @kda_addr_ormask aims at storing the value of those fields. 296fcf5ef2aSThomas Huth */ 297fcf5ef2aSThomas Huth typedef struct KVMDevice { 298fcf5ef2aSThomas Huth struct kvm_arm_device_addr kda; 299fcf5ef2aSThomas Huth struct kvm_device_attr kdattr; 30019d1bd0bSEric Auger uint64_t kda_addr_ormask; 301fcf5ef2aSThomas Huth MemoryRegion *mr; 302fcf5ef2aSThomas Huth QSLIST_ENTRY(KVMDevice) entries; 303fcf5ef2aSThomas Huth int dev_fd; 304fcf5ef2aSThomas Huth } KVMDevice; 305fcf5ef2aSThomas Huth 306b58deb34SPaolo Bonzini static QSLIST_HEAD(, KVMDevice) kvm_devices_head; 307fcf5ef2aSThomas Huth 308fcf5ef2aSThomas Huth static void kvm_arm_devlistener_add(MemoryListener *listener, 309fcf5ef2aSThomas Huth MemoryRegionSection *section) 310fcf5ef2aSThomas Huth { 311fcf5ef2aSThomas Huth KVMDevice *kd; 312fcf5ef2aSThomas Huth 313fcf5ef2aSThomas Huth QSLIST_FOREACH(kd, &kvm_devices_head, entries) { 314fcf5ef2aSThomas Huth if (section->mr == kd->mr) { 315fcf5ef2aSThomas Huth kd->kda.addr = section->offset_within_address_space; 316fcf5ef2aSThomas Huth } 317fcf5ef2aSThomas Huth } 318fcf5ef2aSThomas Huth } 319fcf5ef2aSThomas Huth 320fcf5ef2aSThomas Huth static void kvm_arm_devlistener_del(MemoryListener *listener, 321fcf5ef2aSThomas Huth MemoryRegionSection *section) 322fcf5ef2aSThomas Huth { 323fcf5ef2aSThomas Huth KVMDevice *kd; 324fcf5ef2aSThomas Huth 325fcf5ef2aSThomas Huth QSLIST_FOREACH(kd, &kvm_devices_head, entries) { 326fcf5ef2aSThomas Huth if (section->mr == kd->mr) { 327fcf5ef2aSThomas Huth kd->kda.addr = -1; 328fcf5ef2aSThomas Huth } 329fcf5ef2aSThomas Huth } 330fcf5ef2aSThomas Huth } 331fcf5ef2aSThomas Huth 332fcf5ef2aSThomas Huth static MemoryListener devlistener = { 333fcf5ef2aSThomas Huth .region_add = kvm_arm_devlistener_add, 334fcf5ef2aSThomas Huth .region_del = kvm_arm_devlistener_del, 335fcf5ef2aSThomas Huth }; 336fcf5ef2aSThomas Huth 337fcf5ef2aSThomas Huth static void kvm_arm_set_device_addr(KVMDevice *kd) 338fcf5ef2aSThomas Huth { 339fcf5ef2aSThomas Huth struct kvm_device_attr *attr = &kd->kdattr; 340fcf5ef2aSThomas Huth int ret; 341fcf5ef2aSThomas Huth 342fcf5ef2aSThomas Huth /* If the device control API is available and we have a device fd on the 343fcf5ef2aSThomas Huth * KVMDevice struct, let's use the newer API 344fcf5ef2aSThomas Huth */ 345fcf5ef2aSThomas Huth if (kd->dev_fd >= 0) { 346fcf5ef2aSThomas Huth uint64_t addr = kd->kda.addr; 34719d1bd0bSEric Auger 34819d1bd0bSEric Auger addr |= kd->kda_addr_ormask; 349fcf5ef2aSThomas Huth attr->addr = (uintptr_t)&addr; 350fcf5ef2aSThomas Huth ret = kvm_device_ioctl(kd->dev_fd, KVM_SET_DEVICE_ATTR, attr); 351fcf5ef2aSThomas Huth } else { 352fcf5ef2aSThomas Huth ret = kvm_vm_ioctl(kvm_state, KVM_ARM_SET_DEVICE_ADDR, &kd->kda); 353fcf5ef2aSThomas Huth } 354fcf5ef2aSThomas Huth 355fcf5ef2aSThomas Huth if (ret < 0) { 356fcf5ef2aSThomas Huth fprintf(stderr, "Failed to set device address: %s\n", 357fcf5ef2aSThomas Huth strerror(-ret)); 358fcf5ef2aSThomas Huth abort(); 359fcf5ef2aSThomas Huth } 360fcf5ef2aSThomas Huth } 361fcf5ef2aSThomas Huth 362fcf5ef2aSThomas Huth static void kvm_arm_machine_init_done(Notifier *notifier, void *data) 363fcf5ef2aSThomas Huth { 364fcf5ef2aSThomas Huth KVMDevice *kd, *tkd; 365fcf5ef2aSThomas Huth 366fcf5ef2aSThomas Huth QSLIST_FOREACH_SAFE(kd, &kvm_devices_head, entries, tkd) { 367fcf5ef2aSThomas Huth if (kd->kda.addr != -1) { 368fcf5ef2aSThomas Huth kvm_arm_set_device_addr(kd); 369fcf5ef2aSThomas Huth } 370fcf5ef2aSThomas Huth memory_region_unref(kd->mr); 3715ff9aaabSZheng Xiang QSLIST_REMOVE_HEAD(&kvm_devices_head, entries); 372fcf5ef2aSThomas Huth g_free(kd); 373fcf5ef2aSThomas Huth } 3740bbe4354SPeter Xu memory_listener_unregister(&devlistener); 375fcf5ef2aSThomas Huth } 376fcf5ef2aSThomas Huth 377fcf5ef2aSThomas Huth static Notifier notify = { 378fcf5ef2aSThomas Huth .notify = kvm_arm_machine_init_done, 379fcf5ef2aSThomas Huth }; 380fcf5ef2aSThomas Huth 381fcf5ef2aSThomas Huth void kvm_arm_register_device(MemoryRegion *mr, uint64_t devid, uint64_t group, 38219d1bd0bSEric Auger uint64_t attr, int dev_fd, uint64_t addr_ormask) 383fcf5ef2aSThomas Huth { 384fcf5ef2aSThomas Huth KVMDevice *kd; 385fcf5ef2aSThomas Huth 386fcf5ef2aSThomas Huth if (!kvm_irqchip_in_kernel()) { 387fcf5ef2aSThomas Huth return; 388fcf5ef2aSThomas Huth } 389fcf5ef2aSThomas Huth 390fcf5ef2aSThomas Huth if (QSLIST_EMPTY(&kvm_devices_head)) { 391fcf5ef2aSThomas Huth memory_listener_register(&devlistener, &address_space_memory); 392fcf5ef2aSThomas Huth qemu_add_machine_init_done_notifier(¬ify); 393fcf5ef2aSThomas Huth } 394fcf5ef2aSThomas Huth kd = g_new0(KVMDevice, 1); 395fcf5ef2aSThomas Huth kd->mr = mr; 396fcf5ef2aSThomas Huth kd->kda.id = devid; 397fcf5ef2aSThomas Huth kd->kda.addr = -1; 398fcf5ef2aSThomas Huth kd->kdattr.flags = 0; 399fcf5ef2aSThomas Huth kd->kdattr.group = group; 400fcf5ef2aSThomas Huth kd->kdattr.attr = attr; 401fcf5ef2aSThomas Huth kd->dev_fd = dev_fd; 40219d1bd0bSEric Auger kd->kda_addr_ormask = addr_ormask; 403fcf5ef2aSThomas Huth QSLIST_INSERT_HEAD(&kvm_devices_head, kd, entries); 404fcf5ef2aSThomas Huth memory_region_ref(kd->mr); 405fcf5ef2aSThomas Huth } 406fcf5ef2aSThomas Huth 407fcf5ef2aSThomas Huth static int compare_u64(const void *a, const void *b) 408fcf5ef2aSThomas Huth { 409fcf5ef2aSThomas Huth if (*(uint64_t *)a > *(uint64_t *)b) { 410fcf5ef2aSThomas Huth return 1; 411fcf5ef2aSThomas Huth } 412fcf5ef2aSThomas Huth if (*(uint64_t *)a < *(uint64_t *)b) { 413fcf5ef2aSThomas Huth return -1; 414fcf5ef2aSThomas Huth } 415fcf5ef2aSThomas Huth return 0; 416fcf5ef2aSThomas Huth } 417fcf5ef2aSThomas Huth 418e5ac4200SAndrew Jones /* 419e5ac4200SAndrew Jones * cpreg_values are sorted in ascending order by KVM register ID 420e5ac4200SAndrew Jones * (see kvm_arm_init_cpreg_list). This allows us to cheaply find 421e5ac4200SAndrew Jones * the storage for a KVM register by ID with a binary search. 422e5ac4200SAndrew Jones */ 423e5ac4200SAndrew Jones static uint64_t *kvm_arm_get_cpreg_ptr(ARMCPU *cpu, uint64_t regidx) 424e5ac4200SAndrew Jones { 425e5ac4200SAndrew Jones uint64_t *res; 426e5ac4200SAndrew Jones 427e5ac4200SAndrew Jones res = bsearch(®idx, cpu->cpreg_indexes, cpu->cpreg_array_len, 428e5ac4200SAndrew Jones sizeof(uint64_t), compare_u64); 429e5ac4200SAndrew Jones assert(res); 430e5ac4200SAndrew Jones 431e5ac4200SAndrew Jones return &cpu->cpreg_values[res - cpu->cpreg_indexes]; 432e5ac4200SAndrew Jones } 433e5ac4200SAndrew Jones 434c8a44709SDongjiu Geng /* Initialize the ARMCPU cpreg list according to the kernel's 435fcf5ef2aSThomas Huth * definition of what CPU registers it knows about (and throw away 436fcf5ef2aSThomas Huth * the previous TCG-created cpreg list). 437fcf5ef2aSThomas Huth */ 438fcf5ef2aSThomas Huth int kvm_arm_init_cpreg_list(ARMCPU *cpu) 439fcf5ef2aSThomas Huth { 440fcf5ef2aSThomas Huth struct kvm_reg_list rl; 441fcf5ef2aSThomas Huth struct kvm_reg_list *rlp; 442fcf5ef2aSThomas Huth int i, ret, arraylen; 443fcf5ef2aSThomas Huth CPUState *cs = CPU(cpu); 444fcf5ef2aSThomas Huth 445fcf5ef2aSThomas Huth rl.n = 0; 446fcf5ef2aSThomas Huth ret = kvm_vcpu_ioctl(cs, KVM_GET_REG_LIST, &rl); 447fcf5ef2aSThomas Huth if (ret != -E2BIG) { 448fcf5ef2aSThomas Huth return ret; 449fcf5ef2aSThomas Huth } 450fcf5ef2aSThomas Huth rlp = g_malloc(sizeof(struct kvm_reg_list) + rl.n * sizeof(uint64_t)); 451fcf5ef2aSThomas Huth rlp->n = rl.n; 452fcf5ef2aSThomas Huth ret = kvm_vcpu_ioctl(cs, KVM_GET_REG_LIST, rlp); 453fcf5ef2aSThomas Huth if (ret) { 454fcf5ef2aSThomas Huth goto out; 455fcf5ef2aSThomas Huth } 456fcf5ef2aSThomas Huth /* Sort the list we get back from the kernel, since cpreg_tuples 457fcf5ef2aSThomas Huth * must be in strictly ascending order. 458fcf5ef2aSThomas Huth */ 459fcf5ef2aSThomas Huth qsort(&rlp->reg, rlp->n, sizeof(rlp->reg[0]), compare_u64); 460fcf5ef2aSThomas Huth 461fcf5ef2aSThomas Huth for (i = 0, arraylen = 0; i < rlp->n; i++) { 462fcf5ef2aSThomas Huth if (!kvm_arm_reg_syncs_via_cpreg_list(rlp->reg[i])) { 463fcf5ef2aSThomas Huth continue; 464fcf5ef2aSThomas Huth } 465fcf5ef2aSThomas Huth switch (rlp->reg[i] & KVM_REG_SIZE_MASK) { 466fcf5ef2aSThomas Huth case KVM_REG_SIZE_U32: 467fcf5ef2aSThomas Huth case KVM_REG_SIZE_U64: 468fcf5ef2aSThomas Huth break; 469fcf5ef2aSThomas Huth default: 470fcf5ef2aSThomas Huth fprintf(stderr, "Can't handle size of register in kernel list\n"); 471fcf5ef2aSThomas Huth ret = -EINVAL; 472fcf5ef2aSThomas Huth goto out; 473fcf5ef2aSThomas Huth } 474fcf5ef2aSThomas Huth 475fcf5ef2aSThomas Huth arraylen++; 476fcf5ef2aSThomas Huth } 477fcf5ef2aSThomas Huth 478fcf5ef2aSThomas Huth cpu->cpreg_indexes = g_renew(uint64_t, cpu->cpreg_indexes, arraylen); 479fcf5ef2aSThomas Huth cpu->cpreg_values = g_renew(uint64_t, cpu->cpreg_values, arraylen); 480fcf5ef2aSThomas Huth cpu->cpreg_vmstate_indexes = g_renew(uint64_t, cpu->cpreg_vmstate_indexes, 481fcf5ef2aSThomas Huth arraylen); 482fcf5ef2aSThomas Huth cpu->cpreg_vmstate_values = g_renew(uint64_t, cpu->cpreg_vmstate_values, 483fcf5ef2aSThomas Huth arraylen); 484fcf5ef2aSThomas Huth cpu->cpreg_array_len = arraylen; 485fcf5ef2aSThomas Huth cpu->cpreg_vmstate_array_len = arraylen; 486fcf5ef2aSThomas Huth 487fcf5ef2aSThomas Huth for (i = 0, arraylen = 0; i < rlp->n; i++) { 488fcf5ef2aSThomas Huth uint64_t regidx = rlp->reg[i]; 489fcf5ef2aSThomas Huth if (!kvm_arm_reg_syncs_via_cpreg_list(regidx)) { 490fcf5ef2aSThomas Huth continue; 491fcf5ef2aSThomas Huth } 492fcf5ef2aSThomas Huth cpu->cpreg_indexes[arraylen] = regidx; 493fcf5ef2aSThomas Huth arraylen++; 494fcf5ef2aSThomas Huth } 495fcf5ef2aSThomas Huth assert(cpu->cpreg_array_len == arraylen); 496fcf5ef2aSThomas Huth 497fcf5ef2aSThomas Huth if (!write_kvmstate_to_list(cpu)) { 498fcf5ef2aSThomas Huth /* Shouldn't happen unless kernel is inconsistent about 499fcf5ef2aSThomas Huth * what registers exist. 500fcf5ef2aSThomas Huth */ 501fcf5ef2aSThomas Huth fprintf(stderr, "Initial read of kernel register state failed\n"); 502fcf5ef2aSThomas Huth ret = -EINVAL; 503fcf5ef2aSThomas Huth goto out; 504fcf5ef2aSThomas Huth } 505fcf5ef2aSThomas Huth 506fcf5ef2aSThomas Huth out: 507fcf5ef2aSThomas Huth g_free(rlp); 508fcf5ef2aSThomas Huth return ret; 509fcf5ef2aSThomas Huth } 510fcf5ef2aSThomas Huth 511fcf5ef2aSThomas Huth bool write_kvmstate_to_list(ARMCPU *cpu) 512fcf5ef2aSThomas Huth { 513fcf5ef2aSThomas Huth CPUState *cs = CPU(cpu); 514fcf5ef2aSThomas Huth int i; 515fcf5ef2aSThomas Huth bool ok = true; 516fcf5ef2aSThomas Huth 517fcf5ef2aSThomas Huth for (i = 0; i < cpu->cpreg_array_len; i++) { 518fcf5ef2aSThomas Huth struct kvm_one_reg r; 519fcf5ef2aSThomas Huth uint64_t regidx = cpu->cpreg_indexes[i]; 520fcf5ef2aSThomas Huth uint32_t v32; 521fcf5ef2aSThomas Huth int ret; 522fcf5ef2aSThomas Huth 523fcf5ef2aSThomas Huth r.id = regidx; 524fcf5ef2aSThomas Huth 525fcf5ef2aSThomas Huth switch (regidx & KVM_REG_SIZE_MASK) { 526fcf5ef2aSThomas Huth case KVM_REG_SIZE_U32: 527fcf5ef2aSThomas Huth r.addr = (uintptr_t)&v32; 528fcf5ef2aSThomas Huth ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &r); 529fcf5ef2aSThomas Huth if (!ret) { 530fcf5ef2aSThomas Huth cpu->cpreg_values[i] = v32; 531fcf5ef2aSThomas Huth } 532fcf5ef2aSThomas Huth break; 533fcf5ef2aSThomas Huth case KVM_REG_SIZE_U64: 534fcf5ef2aSThomas Huth r.addr = (uintptr_t)(cpu->cpreg_values + i); 535fcf5ef2aSThomas Huth ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &r); 536fcf5ef2aSThomas Huth break; 537fcf5ef2aSThomas Huth default: 538fcf5ef2aSThomas Huth abort(); 539fcf5ef2aSThomas Huth } 540fcf5ef2aSThomas Huth if (ret) { 541fcf5ef2aSThomas Huth ok = false; 542fcf5ef2aSThomas Huth } 543fcf5ef2aSThomas Huth } 544fcf5ef2aSThomas Huth return ok; 545fcf5ef2aSThomas Huth } 546fcf5ef2aSThomas Huth 547fcf5ef2aSThomas Huth bool write_list_to_kvmstate(ARMCPU *cpu, int level) 548fcf5ef2aSThomas Huth { 549fcf5ef2aSThomas Huth CPUState *cs = CPU(cpu); 550fcf5ef2aSThomas Huth int i; 551fcf5ef2aSThomas Huth bool ok = true; 552fcf5ef2aSThomas Huth 553fcf5ef2aSThomas Huth for (i = 0; i < cpu->cpreg_array_len; i++) { 554fcf5ef2aSThomas Huth struct kvm_one_reg r; 555fcf5ef2aSThomas Huth uint64_t regidx = cpu->cpreg_indexes[i]; 556fcf5ef2aSThomas Huth uint32_t v32; 557fcf5ef2aSThomas Huth int ret; 558fcf5ef2aSThomas Huth 559fcf5ef2aSThomas Huth if (kvm_arm_cpreg_level(regidx) > level) { 560fcf5ef2aSThomas Huth continue; 561fcf5ef2aSThomas Huth } 562fcf5ef2aSThomas Huth 563fcf5ef2aSThomas Huth r.id = regidx; 564fcf5ef2aSThomas Huth switch (regidx & KVM_REG_SIZE_MASK) { 565fcf5ef2aSThomas Huth case KVM_REG_SIZE_U32: 566fcf5ef2aSThomas Huth v32 = cpu->cpreg_values[i]; 567fcf5ef2aSThomas Huth r.addr = (uintptr_t)&v32; 568fcf5ef2aSThomas Huth break; 569fcf5ef2aSThomas Huth case KVM_REG_SIZE_U64: 570fcf5ef2aSThomas Huth r.addr = (uintptr_t)(cpu->cpreg_values + i); 571fcf5ef2aSThomas Huth break; 572fcf5ef2aSThomas Huth default: 573fcf5ef2aSThomas Huth abort(); 574fcf5ef2aSThomas Huth } 575fcf5ef2aSThomas Huth ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &r); 576fcf5ef2aSThomas Huth if (ret) { 577fcf5ef2aSThomas Huth /* We might fail for "unknown register" and also for 578fcf5ef2aSThomas Huth * "you tried to set a register which is constant with 579fcf5ef2aSThomas Huth * a different value from what it actually contains". 580fcf5ef2aSThomas Huth */ 581fcf5ef2aSThomas Huth ok = false; 582fcf5ef2aSThomas Huth } 583fcf5ef2aSThomas Huth } 584fcf5ef2aSThomas Huth return ok; 585fcf5ef2aSThomas Huth } 586fcf5ef2aSThomas Huth 587e5ac4200SAndrew Jones void kvm_arm_cpu_pre_save(ARMCPU *cpu) 588e5ac4200SAndrew Jones { 589e5ac4200SAndrew Jones /* KVM virtual time adjustment */ 590e5ac4200SAndrew Jones if (cpu->kvm_vtime_dirty) { 591e5ac4200SAndrew Jones *kvm_arm_get_cpreg_ptr(cpu, KVM_REG_ARM_TIMER_CNT) = cpu->kvm_vtime; 592e5ac4200SAndrew Jones } 593e5ac4200SAndrew Jones } 594e5ac4200SAndrew Jones 595e5ac4200SAndrew Jones void kvm_arm_cpu_post_load(ARMCPU *cpu) 596e5ac4200SAndrew Jones { 597e5ac4200SAndrew Jones /* KVM virtual time adjustment */ 598e5ac4200SAndrew Jones if (cpu->kvm_adjvtime) { 599e5ac4200SAndrew Jones cpu->kvm_vtime = *kvm_arm_get_cpreg_ptr(cpu, KVM_REG_ARM_TIMER_CNT); 600e5ac4200SAndrew Jones cpu->kvm_vtime_dirty = true; 601e5ac4200SAndrew Jones } 602e5ac4200SAndrew Jones } 603e5ac4200SAndrew Jones 604fcf5ef2aSThomas Huth void kvm_arm_reset_vcpu(ARMCPU *cpu) 605fcf5ef2aSThomas Huth { 606fcf5ef2aSThomas Huth int ret; 607fcf5ef2aSThomas Huth 608fcf5ef2aSThomas Huth /* Re-init VCPU so that all registers are set to 609fcf5ef2aSThomas Huth * their respective reset values. 610fcf5ef2aSThomas Huth */ 611fcf5ef2aSThomas Huth ret = kvm_arm_vcpu_init(CPU(cpu)); 612fcf5ef2aSThomas Huth if (ret < 0) { 613fcf5ef2aSThomas Huth fprintf(stderr, "kvm_arm_vcpu_init failed: %s\n", strerror(-ret)); 614fcf5ef2aSThomas Huth abort(); 615fcf5ef2aSThomas Huth } 616fcf5ef2aSThomas Huth if (!write_kvmstate_to_list(cpu)) { 617fcf5ef2aSThomas Huth fprintf(stderr, "write_kvmstate_to_list failed\n"); 618fcf5ef2aSThomas Huth abort(); 619fcf5ef2aSThomas Huth } 620b698e4eeSPeter Maydell /* 621b698e4eeSPeter Maydell * Sync the reset values also into the CPUState. This is necessary 622b698e4eeSPeter Maydell * because the next thing we do will be a kvm_arch_put_registers() 623b698e4eeSPeter Maydell * which will update the list values from the CPUState before copying 624b698e4eeSPeter Maydell * the list values back to KVM. It's OK to ignore failure returns here 625b698e4eeSPeter Maydell * for the same reason we do so in kvm_arch_get_registers(). 626b698e4eeSPeter Maydell */ 627b698e4eeSPeter Maydell write_list_to_cpustate(cpu); 628fcf5ef2aSThomas Huth } 629fcf5ef2aSThomas Huth 630fcf5ef2aSThomas Huth /* 631fcf5ef2aSThomas Huth * Update KVM's MP_STATE based on what QEMU thinks it is 632fcf5ef2aSThomas Huth */ 633fcf5ef2aSThomas Huth int kvm_arm_sync_mpstate_to_kvm(ARMCPU *cpu) 634fcf5ef2aSThomas Huth { 635fcf5ef2aSThomas Huth if (cap_has_mp_state) { 636fcf5ef2aSThomas Huth struct kvm_mp_state mp_state = { 637062ba099SAlex Bennée .mp_state = (cpu->power_state == PSCI_OFF) ? 638062ba099SAlex Bennée KVM_MP_STATE_STOPPED : KVM_MP_STATE_RUNNABLE 639fcf5ef2aSThomas Huth }; 640fcf5ef2aSThomas Huth int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state); 641fcf5ef2aSThomas Huth if (ret) { 642fcf5ef2aSThomas Huth fprintf(stderr, "%s: failed to set MP_STATE %d/%s\n", 643fcf5ef2aSThomas Huth __func__, ret, strerror(-ret)); 644fcf5ef2aSThomas Huth return -1; 645fcf5ef2aSThomas Huth } 646fcf5ef2aSThomas Huth } 647fcf5ef2aSThomas Huth 648fcf5ef2aSThomas Huth return 0; 649fcf5ef2aSThomas Huth } 650fcf5ef2aSThomas Huth 651fcf5ef2aSThomas Huth /* 652fcf5ef2aSThomas Huth * Sync the KVM MP_STATE into QEMU 653fcf5ef2aSThomas Huth */ 654fcf5ef2aSThomas Huth int kvm_arm_sync_mpstate_to_qemu(ARMCPU *cpu) 655fcf5ef2aSThomas Huth { 656fcf5ef2aSThomas Huth if (cap_has_mp_state) { 657fcf5ef2aSThomas Huth struct kvm_mp_state mp_state; 658fcf5ef2aSThomas Huth int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MP_STATE, &mp_state); 659fcf5ef2aSThomas Huth if (ret) { 660fcf5ef2aSThomas Huth fprintf(stderr, "%s: failed to get MP_STATE %d/%s\n", 661fcf5ef2aSThomas Huth __func__, ret, strerror(-ret)); 662fcf5ef2aSThomas Huth abort(); 663fcf5ef2aSThomas Huth } 664062ba099SAlex Bennée cpu->power_state = (mp_state.mp_state == KVM_MP_STATE_STOPPED) ? 665062ba099SAlex Bennée PSCI_OFF : PSCI_ON; 666fcf5ef2aSThomas Huth } 667fcf5ef2aSThomas Huth 668fcf5ef2aSThomas Huth return 0; 669fcf5ef2aSThomas Huth } 670fcf5ef2aSThomas Huth 671e5ac4200SAndrew Jones void kvm_arm_get_virtual_time(CPUState *cs) 672e5ac4200SAndrew Jones { 673e5ac4200SAndrew Jones ARMCPU *cpu = ARM_CPU(cs); 674e5ac4200SAndrew Jones struct kvm_one_reg reg = { 675e5ac4200SAndrew Jones .id = KVM_REG_ARM_TIMER_CNT, 676e5ac4200SAndrew Jones .addr = (uintptr_t)&cpu->kvm_vtime, 677e5ac4200SAndrew Jones }; 678e5ac4200SAndrew Jones int ret; 679e5ac4200SAndrew Jones 680e5ac4200SAndrew Jones if (cpu->kvm_vtime_dirty) { 681e5ac4200SAndrew Jones return; 682e5ac4200SAndrew Jones } 683e5ac4200SAndrew Jones 684e5ac4200SAndrew Jones ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); 685e5ac4200SAndrew Jones if (ret) { 686e5ac4200SAndrew Jones error_report("Failed to get KVM_REG_ARM_TIMER_CNT"); 687e5ac4200SAndrew Jones abort(); 688e5ac4200SAndrew Jones } 689e5ac4200SAndrew Jones 690e5ac4200SAndrew Jones cpu->kvm_vtime_dirty = true; 691e5ac4200SAndrew Jones } 692e5ac4200SAndrew Jones 693e5ac4200SAndrew Jones void kvm_arm_put_virtual_time(CPUState *cs) 694e5ac4200SAndrew Jones { 695e5ac4200SAndrew Jones ARMCPU *cpu = ARM_CPU(cs); 696e5ac4200SAndrew Jones struct kvm_one_reg reg = { 697e5ac4200SAndrew Jones .id = KVM_REG_ARM_TIMER_CNT, 698e5ac4200SAndrew Jones .addr = (uintptr_t)&cpu->kvm_vtime, 699e5ac4200SAndrew Jones }; 700e5ac4200SAndrew Jones int ret; 701e5ac4200SAndrew Jones 702e5ac4200SAndrew Jones if (!cpu->kvm_vtime_dirty) { 703e5ac4200SAndrew Jones return; 704e5ac4200SAndrew Jones } 705e5ac4200SAndrew Jones 706e5ac4200SAndrew Jones ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); 707e5ac4200SAndrew Jones if (ret) { 708e5ac4200SAndrew Jones error_report("Failed to set KVM_REG_ARM_TIMER_CNT"); 709e5ac4200SAndrew Jones abort(); 710e5ac4200SAndrew Jones } 711e5ac4200SAndrew Jones 712e5ac4200SAndrew Jones cpu->kvm_vtime_dirty = false; 713e5ac4200SAndrew Jones } 714e5ac4200SAndrew Jones 715202ccb6bSDongjiu Geng int kvm_put_vcpu_events(ARMCPU *cpu) 716202ccb6bSDongjiu Geng { 717202ccb6bSDongjiu Geng CPUARMState *env = &cpu->env; 718202ccb6bSDongjiu Geng struct kvm_vcpu_events events; 719202ccb6bSDongjiu Geng int ret; 720202ccb6bSDongjiu Geng 721202ccb6bSDongjiu Geng if (!kvm_has_vcpu_events()) { 722202ccb6bSDongjiu Geng return 0; 723202ccb6bSDongjiu Geng } 724202ccb6bSDongjiu Geng 725202ccb6bSDongjiu Geng memset(&events, 0, sizeof(events)); 726202ccb6bSDongjiu Geng events.exception.serror_pending = env->serror.pending; 727202ccb6bSDongjiu Geng 728202ccb6bSDongjiu Geng /* Inject SError to guest with specified syndrome if host kernel 729202ccb6bSDongjiu Geng * supports it, otherwise inject SError without syndrome. 730202ccb6bSDongjiu Geng */ 731202ccb6bSDongjiu Geng if (cap_has_inject_serror_esr) { 732202ccb6bSDongjiu Geng events.exception.serror_has_esr = env->serror.has_esr; 733202ccb6bSDongjiu Geng events.exception.serror_esr = env->serror.esr; 734202ccb6bSDongjiu Geng } 735202ccb6bSDongjiu Geng 736202ccb6bSDongjiu Geng ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_VCPU_EVENTS, &events); 737202ccb6bSDongjiu Geng if (ret) { 738202ccb6bSDongjiu Geng error_report("failed to put vcpu events"); 739202ccb6bSDongjiu Geng } 740202ccb6bSDongjiu Geng 741202ccb6bSDongjiu Geng return ret; 742202ccb6bSDongjiu Geng } 743202ccb6bSDongjiu Geng 744202ccb6bSDongjiu Geng int kvm_get_vcpu_events(ARMCPU *cpu) 745202ccb6bSDongjiu Geng { 746202ccb6bSDongjiu Geng CPUARMState *env = &cpu->env; 747202ccb6bSDongjiu Geng struct kvm_vcpu_events events; 748202ccb6bSDongjiu Geng int ret; 749202ccb6bSDongjiu Geng 750202ccb6bSDongjiu Geng if (!kvm_has_vcpu_events()) { 751202ccb6bSDongjiu Geng return 0; 752202ccb6bSDongjiu Geng } 753202ccb6bSDongjiu Geng 754202ccb6bSDongjiu Geng memset(&events, 0, sizeof(events)); 755202ccb6bSDongjiu Geng ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_VCPU_EVENTS, &events); 756202ccb6bSDongjiu Geng if (ret) { 757202ccb6bSDongjiu Geng error_report("failed to get vcpu events"); 758202ccb6bSDongjiu Geng return ret; 759202ccb6bSDongjiu Geng } 760202ccb6bSDongjiu Geng 761202ccb6bSDongjiu Geng env->serror.pending = events.exception.serror_pending; 762202ccb6bSDongjiu Geng env->serror.has_esr = events.exception.serror_has_esr; 763202ccb6bSDongjiu Geng env->serror.esr = events.exception.serror_esr; 764202ccb6bSDongjiu Geng 765202ccb6bSDongjiu Geng return 0; 766202ccb6bSDongjiu Geng } 767202ccb6bSDongjiu Geng 768fcf5ef2aSThomas Huth void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run) 769fcf5ef2aSThomas Huth { 7701711bfa5SBeata Michalska ARMCPU *cpu = ARM_CPU(cs); 7711711bfa5SBeata Michalska CPUARMState *env = &cpu->env; 7721711bfa5SBeata Michalska 7731711bfa5SBeata Michalska if (unlikely(env->ext_dabt_raised)) { 7741711bfa5SBeata Michalska /* 7751711bfa5SBeata Michalska * Verifying that the ext DABT has been properly injected, 7761711bfa5SBeata Michalska * otherwise risking indefinitely re-running the faulting instruction 7771711bfa5SBeata Michalska * Covering a very narrow case for kernels 5.5..5.5.4 7781711bfa5SBeata Michalska * when injected abort was misconfigured to be 7791711bfa5SBeata Michalska * an IMPLEMENTATION DEFINED exception (for 32-bit EL1) 7801711bfa5SBeata Michalska */ 7811711bfa5SBeata Michalska if (!arm_feature(env, ARM_FEATURE_AARCH64) && 7821711bfa5SBeata Michalska unlikely(!kvm_arm_verify_ext_dabt_pending(cs))) { 7831711bfa5SBeata Michalska 7841711bfa5SBeata Michalska error_report("Data abort exception with no valid ISS generated by " 7851711bfa5SBeata Michalska "guest memory access. KVM unable to emulate faulting " 7861711bfa5SBeata Michalska "instruction. Failed to inject an external data abort " 7871711bfa5SBeata Michalska "into the guest."); 7881711bfa5SBeata Michalska abort(); 7891711bfa5SBeata Michalska } 7901711bfa5SBeata Michalska /* Clear the status */ 7911711bfa5SBeata Michalska env->ext_dabt_raised = 0; 7921711bfa5SBeata Michalska } 793fcf5ef2aSThomas Huth } 794fcf5ef2aSThomas Huth 795fcf5ef2aSThomas Huth MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run) 796fcf5ef2aSThomas Huth { 7975d721b78SAlexander Graf ARMCPU *cpu; 7985d721b78SAlexander Graf uint32_t switched_level; 7995d721b78SAlexander Graf 8005d721b78SAlexander Graf if (kvm_irqchip_in_kernel()) { 8015d721b78SAlexander Graf /* 8025d721b78SAlexander Graf * We only need to sync timer states with user-space interrupt 8035d721b78SAlexander Graf * controllers, so return early and save cycles if we don't. 8045d721b78SAlexander Graf */ 8055d721b78SAlexander Graf return MEMTXATTRS_UNSPECIFIED; 8065d721b78SAlexander Graf } 8075d721b78SAlexander Graf 8085d721b78SAlexander Graf cpu = ARM_CPU(cs); 8095d721b78SAlexander Graf 8105d721b78SAlexander Graf /* Synchronize our shadowed in-kernel device irq lines with the kvm ones */ 8115d721b78SAlexander Graf if (run->s.regs.device_irq_level != cpu->device_irq_level) { 8125d721b78SAlexander Graf switched_level = cpu->device_irq_level ^ run->s.regs.device_irq_level; 8135d721b78SAlexander Graf 8145d721b78SAlexander Graf qemu_mutex_lock_iothread(); 8155d721b78SAlexander Graf 8165d721b78SAlexander Graf if (switched_level & KVM_ARM_DEV_EL1_VTIMER) { 8175d721b78SAlexander Graf qemu_set_irq(cpu->gt_timer_outputs[GTIMER_VIRT], 8185d721b78SAlexander Graf !!(run->s.regs.device_irq_level & 8195d721b78SAlexander Graf KVM_ARM_DEV_EL1_VTIMER)); 8205d721b78SAlexander Graf switched_level &= ~KVM_ARM_DEV_EL1_VTIMER; 8215d721b78SAlexander Graf } 8225d721b78SAlexander Graf 8235d721b78SAlexander Graf if (switched_level & KVM_ARM_DEV_EL1_PTIMER) { 8245d721b78SAlexander Graf qemu_set_irq(cpu->gt_timer_outputs[GTIMER_PHYS], 8255d721b78SAlexander Graf !!(run->s.regs.device_irq_level & 8265d721b78SAlexander Graf KVM_ARM_DEV_EL1_PTIMER)); 8275d721b78SAlexander Graf switched_level &= ~KVM_ARM_DEV_EL1_PTIMER; 8285d721b78SAlexander Graf } 8295d721b78SAlexander Graf 830b1659527SAndrew Jones if (switched_level & KVM_ARM_DEV_PMU) { 831b1659527SAndrew Jones qemu_set_irq(cpu->pmu_interrupt, 832b1659527SAndrew Jones !!(run->s.regs.device_irq_level & KVM_ARM_DEV_PMU)); 833b1659527SAndrew Jones switched_level &= ~KVM_ARM_DEV_PMU; 834b1659527SAndrew Jones } 8355d721b78SAlexander Graf 8365d721b78SAlexander Graf if (switched_level) { 8375d721b78SAlexander Graf qemu_log_mask(LOG_UNIMP, "%s: unhandled in-kernel device IRQ %x\n", 8385d721b78SAlexander Graf __func__, switched_level); 8395d721b78SAlexander Graf } 8405d721b78SAlexander Graf 8415d721b78SAlexander Graf /* We also mark unknown levels as processed to not waste cycles */ 8425d721b78SAlexander Graf cpu->device_irq_level = run->s.regs.device_irq_level; 8435d721b78SAlexander Graf qemu_mutex_unlock_iothread(); 8445d721b78SAlexander Graf } 8455d721b78SAlexander Graf 846fcf5ef2aSThomas Huth return MEMTXATTRS_UNSPECIFIED; 847fcf5ef2aSThomas Huth } 848fcf5ef2aSThomas Huth 849538f0497SPhilippe Mathieu-Daudé void kvm_arm_vm_state_change(void *opaque, bool running, RunState state) 850e5ac4200SAndrew Jones { 851e5ac4200SAndrew Jones CPUState *cs = opaque; 852e5ac4200SAndrew Jones ARMCPU *cpu = ARM_CPU(cs); 853e5ac4200SAndrew Jones 854e5ac4200SAndrew Jones if (running) { 855e5ac4200SAndrew Jones if (cpu->kvm_adjvtime) { 856e5ac4200SAndrew Jones kvm_arm_put_virtual_time(cs); 857e5ac4200SAndrew Jones } 858e5ac4200SAndrew Jones } else { 859e5ac4200SAndrew Jones if (cpu->kvm_adjvtime) { 860e5ac4200SAndrew Jones kvm_arm_get_virtual_time(cs); 861e5ac4200SAndrew Jones } 862e5ac4200SAndrew Jones } 863e5ac4200SAndrew Jones } 864fcf5ef2aSThomas Huth 865694bcaa8SBeata Michalska /** 866694bcaa8SBeata Michalska * kvm_arm_handle_dabt_nisv: 867694bcaa8SBeata Michalska * @cs: CPUState 868694bcaa8SBeata Michalska * @esr_iss: ISS encoding (limited) for the exception from Data Abort 869694bcaa8SBeata Michalska * ISV bit set to '0b0' -> no valid instruction syndrome 870694bcaa8SBeata Michalska * @fault_ipa: faulting address for the synchronous data abort 871694bcaa8SBeata Michalska * 872694bcaa8SBeata Michalska * Returns: 0 if the exception has been handled, < 0 otherwise 873694bcaa8SBeata Michalska */ 874694bcaa8SBeata Michalska static int kvm_arm_handle_dabt_nisv(CPUState *cs, uint64_t esr_iss, 875694bcaa8SBeata Michalska uint64_t fault_ipa) 876694bcaa8SBeata Michalska { 8771711bfa5SBeata Michalska ARMCPU *cpu = ARM_CPU(cs); 8781711bfa5SBeata Michalska CPUARMState *env = &cpu->env; 879694bcaa8SBeata Michalska /* 880694bcaa8SBeata Michalska * Request KVM to inject the external data abort into the guest 881694bcaa8SBeata Michalska */ 882694bcaa8SBeata Michalska if (cap_has_inject_ext_dabt) { 883694bcaa8SBeata Michalska struct kvm_vcpu_events events = { }; 884694bcaa8SBeata Michalska /* 885694bcaa8SBeata Michalska * The external data abort event will be handled immediately by KVM 886694bcaa8SBeata Michalska * using the address fault that triggered the exit on given VCPU. 887694bcaa8SBeata Michalska * Requesting injection of the external data abort does not rely 888694bcaa8SBeata Michalska * on any other VCPU state. Therefore, in this particular case, the VCPU 889694bcaa8SBeata Michalska * synchronization can be exceptionally skipped. 890694bcaa8SBeata Michalska */ 891694bcaa8SBeata Michalska events.exception.ext_dabt_pending = 1; 892694bcaa8SBeata Michalska /* KVM_CAP_ARM_INJECT_EXT_DABT implies KVM_CAP_VCPU_EVENTS */ 8931711bfa5SBeata Michalska if (!kvm_vcpu_ioctl(cs, KVM_SET_VCPU_EVENTS, &events)) { 8941711bfa5SBeata Michalska env->ext_dabt_raised = 1; 8951711bfa5SBeata Michalska return 0; 8961711bfa5SBeata Michalska } 897694bcaa8SBeata Michalska } else { 898694bcaa8SBeata Michalska error_report("Data abort exception triggered by guest memory access " 899694bcaa8SBeata Michalska "at physical address: 0x" TARGET_FMT_lx, 900694bcaa8SBeata Michalska (target_ulong)fault_ipa); 901694bcaa8SBeata Michalska error_printf("KVM unable to emulate faulting instruction.\n"); 902694bcaa8SBeata Michalska } 903694bcaa8SBeata Michalska return -1; 904694bcaa8SBeata Michalska } 905694bcaa8SBeata Michalska 906fcf5ef2aSThomas Huth int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run) 907fcf5ef2aSThomas Huth { 908fcf5ef2aSThomas Huth int ret = 0; 909fcf5ef2aSThomas Huth 910fcf5ef2aSThomas Huth switch (run->exit_reason) { 911fcf5ef2aSThomas Huth case KVM_EXIT_DEBUG: 912fcf5ef2aSThomas Huth if (kvm_arm_handle_debug(cs, &run->debug.arch)) { 913fcf5ef2aSThomas Huth ret = EXCP_DEBUG; 914fcf5ef2aSThomas Huth } /* otherwise return to guest */ 915fcf5ef2aSThomas Huth break; 916694bcaa8SBeata Michalska case KVM_EXIT_ARM_NISV: 917694bcaa8SBeata Michalska /* External DABT with no valid iss to decode */ 918694bcaa8SBeata Michalska ret = kvm_arm_handle_dabt_nisv(cs, run->arm_nisv.esr_iss, 919694bcaa8SBeata Michalska run->arm_nisv.fault_ipa); 920694bcaa8SBeata Michalska break; 921fcf5ef2aSThomas Huth default: 922fcf5ef2aSThomas Huth qemu_log_mask(LOG_UNIMP, "%s: un-handled exit reason %d\n", 923fcf5ef2aSThomas Huth __func__, run->exit_reason); 924fcf5ef2aSThomas Huth break; 925fcf5ef2aSThomas Huth } 926fcf5ef2aSThomas Huth return ret; 927fcf5ef2aSThomas Huth } 928fcf5ef2aSThomas Huth 929fcf5ef2aSThomas Huth bool kvm_arch_stop_on_emulation_error(CPUState *cs) 930fcf5ef2aSThomas Huth { 931fcf5ef2aSThomas Huth return true; 932fcf5ef2aSThomas Huth } 933fcf5ef2aSThomas Huth 934fcf5ef2aSThomas Huth int kvm_arch_process_async_events(CPUState *cs) 935fcf5ef2aSThomas Huth { 936fcf5ef2aSThomas Huth return 0; 937fcf5ef2aSThomas Huth } 938fcf5ef2aSThomas Huth 939fcf5ef2aSThomas Huth void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg) 940fcf5ef2aSThomas Huth { 941fcf5ef2aSThomas Huth if (kvm_sw_breakpoints_active(cs)) { 942fcf5ef2aSThomas Huth dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP; 943fcf5ef2aSThomas Huth } 944fcf5ef2aSThomas Huth if (kvm_arm_hw_debug_active(cs)) { 945fcf5ef2aSThomas Huth dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW; 946fcf5ef2aSThomas Huth kvm_arm_copy_hw_debug_data(&dbg->arch); 947fcf5ef2aSThomas Huth } 948fcf5ef2aSThomas Huth } 949fcf5ef2aSThomas Huth 950fcf5ef2aSThomas Huth void kvm_arch_init_irq_routing(KVMState *s) 951fcf5ef2aSThomas Huth { 952fcf5ef2aSThomas Huth } 953fcf5ef2aSThomas Huth 9544376c40dSPaolo Bonzini int kvm_arch_irqchip_create(KVMState *s) 955fcf5ef2aSThomas Huth { 9564376c40dSPaolo Bonzini if (kvm_kernel_irqchip_split()) { 957fcf5ef2aSThomas Huth perror("-machine kernel_irqchip=split is not supported on ARM."); 958fcf5ef2aSThomas Huth exit(1); 959fcf5ef2aSThomas Huth } 960fcf5ef2aSThomas Huth 961fcf5ef2aSThomas Huth /* If we can create the VGIC using the newer device control API, we 962fcf5ef2aSThomas Huth * let the device do this when it initializes itself, otherwise we 963fcf5ef2aSThomas Huth * fall back to the old API */ 964fcf5ef2aSThomas Huth return kvm_check_extension(s, KVM_CAP_DEVICE_CTRL); 965fcf5ef2aSThomas Huth } 966fcf5ef2aSThomas Huth 967fcf5ef2aSThomas Huth int kvm_arm_vgic_probe(void) 968fcf5ef2aSThomas Huth { 969d45efe47SEric Auger int val = 0; 970d45efe47SEric Auger 971fcf5ef2aSThomas Huth if (kvm_create_device(kvm_state, 972fcf5ef2aSThomas Huth KVM_DEV_TYPE_ARM_VGIC_V3, true) == 0) { 973d45efe47SEric Auger val |= KVM_ARM_VGIC_V3; 974fcf5ef2aSThomas Huth } 975d45efe47SEric Auger if (kvm_create_device(kvm_state, 976d45efe47SEric Auger KVM_DEV_TYPE_ARM_VGIC_V2, true) == 0) { 977d45efe47SEric Auger val |= KVM_ARM_VGIC_V2; 978d45efe47SEric Auger } 979d45efe47SEric Auger return val; 980fcf5ef2aSThomas Huth } 981fcf5ef2aSThomas Huth 982f6530926SEric Auger int kvm_arm_set_irq(int cpu, int irqtype, int irq, int level) 983f6530926SEric Auger { 984f6530926SEric Auger int kvm_irq = (irqtype << KVM_ARM_IRQ_TYPE_SHIFT) | irq; 985f6530926SEric Auger int cpu_idx1 = cpu % 256; 986f6530926SEric Auger int cpu_idx2 = cpu / 256; 987f6530926SEric Auger 988f6530926SEric Auger kvm_irq |= (cpu_idx1 << KVM_ARM_IRQ_VCPU_SHIFT) | 989f6530926SEric Auger (cpu_idx2 << KVM_ARM_IRQ_VCPU2_SHIFT); 990f6530926SEric Auger 991f6530926SEric Auger return kvm_set_irq(kvm_state, kvm_irq, !!level); 992f6530926SEric Auger } 993f6530926SEric Auger 994fcf5ef2aSThomas Huth int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route, 995fcf5ef2aSThomas Huth uint64_t address, uint32_t data, PCIDevice *dev) 996fcf5ef2aSThomas Huth { 997b05c81d2SEric Auger AddressSpace *as = pci_device_iommu_address_space(dev); 998b05c81d2SEric Auger hwaddr xlat, len, doorbell_gpa; 999b05c81d2SEric Auger MemoryRegionSection mrs; 1000b05c81d2SEric Auger MemoryRegion *mr; 1001b05c81d2SEric Auger int ret = 1; 1002b05c81d2SEric Auger 1003b05c81d2SEric Auger if (as == &address_space_memory) { 1004fcf5ef2aSThomas Huth return 0; 1005fcf5ef2aSThomas Huth } 1006fcf5ef2aSThomas Huth 1007b05c81d2SEric Auger /* MSI doorbell address is translated by an IOMMU */ 1008b05c81d2SEric Auger 1009b05c81d2SEric Auger rcu_read_lock(); 1010bc6b1cecSPeter Maydell mr = address_space_translate(as, address, &xlat, &len, true, 1011bc6b1cecSPeter Maydell MEMTXATTRS_UNSPECIFIED); 1012b05c81d2SEric Auger if (!mr) { 1013b05c81d2SEric Auger goto unlock; 1014b05c81d2SEric Auger } 1015b05c81d2SEric Auger mrs = memory_region_find(mr, xlat, 1); 1016b05c81d2SEric Auger if (!mrs.mr) { 1017b05c81d2SEric Auger goto unlock; 1018b05c81d2SEric Auger } 1019b05c81d2SEric Auger 1020b05c81d2SEric Auger doorbell_gpa = mrs.offset_within_address_space; 1021b05c81d2SEric Auger memory_region_unref(mrs.mr); 1022b05c81d2SEric Auger 1023b05c81d2SEric Auger route->u.msi.address_lo = doorbell_gpa; 1024b05c81d2SEric Auger route->u.msi.address_hi = doorbell_gpa >> 32; 1025b05c81d2SEric Auger 1026b05c81d2SEric Auger trace_kvm_arm_fixup_msi_route(address, doorbell_gpa); 1027b05c81d2SEric Auger 1028b05c81d2SEric Auger ret = 0; 1029b05c81d2SEric Auger 1030b05c81d2SEric Auger unlock: 1031b05c81d2SEric Auger rcu_read_unlock(); 1032b05c81d2SEric Auger return ret; 1033b05c81d2SEric Auger } 1034b05c81d2SEric Auger 1035fcf5ef2aSThomas Huth int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route, 1036fcf5ef2aSThomas Huth int vector, PCIDevice *dev) 1037fcf5ef2aSThomas Huth { 1038fcf5ef2aSThomas Huth return 0; 1039fcf5ef2aSThomas Huth } 1040fcf5ef2aSThomas Huth 1041fcf5ef2aSThomas Huth int kvm_arch_release_virq_post(int virq) 1042fcf5ef2aSThomas Huth { 1043fcf5ef2aSThomas Huth return 0; 1044fcf5ef2aSThomas Huth } 1045fcf5ef2aSThomas Huth 1046fcf5ef2aSThomas Huth int kvm_arch_msi_data_to_gsi(uint32_t data) 1047fcf5ef2aSThomas Huth { 1048fcf5ef2aSThomas Huth return (data - 32) & 0xffff; 1049fcf5ef2aSThomas Huth } 105092a5199bSTom Lendacky 105192a5199bSTom Lendacky bool kvm_arch_cpu_check_are_resettable(void) 105292a5199bSTom Lendacky { 105392a5199bSTom Lendacky return true; 105492a5199bSTom Lendacky } 1055