hw/intc/xics_kvm.c

   1 /*
   2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
   3  *
   4  * PAPR Virtualized Interrupt System, aka ICS/ICP aka xics, in-kernel emulation
   5  *
   6  * Copyright (c) 2013 David Gibson, IBM Corporation.
   7  *
   8  * Permission is hereby granted, free of charge, to any person obtaining a copy
   9  * of this software and associated documentation files (the "Software"), to deal
  10  * in the Software without restriction, including without limitation the rights
  11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  12  * copies of the Software, and to permit persons to whom the Software is
  13  * furnished to do so, subject to the following conditions:
  14  *
  15  * The above copyright notice and this permission notice shall be included in
  16  * all copies or substantial portions of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  24  * THE SOFTWARE.
  25  *
  26  */
  27
  28 #include "qemu/osdep.h"
  29 #include "qapi/error.h"
  30 #include "qemu-common.h"
  31 #include "cpu.h"
  32 #include "hw/hw.h"
  33 #include "trace.h"
  34 #include "sysemu/kvm.h"
  35 #include "hw/ppc/spapr.h"
  36 #include "hw/ppc/xics.h"
  37 #include "kvm_ppc.h"
  38 #include "qemu/config-file.h"
  39 #include "qemu/error-report.h"
  40
  41 #include <sys/ioctl.h>
  42
  43 static int kernel_xics_fd = -1;
  44
  45 typedef struct KVMXICSState {
  46     XICSState parent_obj;
  47
  48     int kernel_xics_fd;
  49 } KVMXICSState;
  50
  51 /*
  52  * ICP-KVM
  53  */
  54 static void icp_get_kvm_state(ICPState *ss)
  55 {
  56     uint64_t state;
  57     struct kvm_one_reg reg = {
  58         .id = KVM_REG_PPC_ICP_STATE,
  59         .addr = (uintptr_t)&state,
  60     };
  61     int ret;
  62
  63     /* ICP for this CPU thread is not in use, exiting */
  64     if (!ss->cs) {
  65         return;
  66     }
  67
  68     ret = kvm_vcpu_ioctl(ss->cs, KVM_GET_ONE_REG, &reg);
  69     if (ret != 0) {
  70         error_report("Unable to retrieve KVM interrupt controller state"
  71                 " for CPU %ld: %s", kvm_arch_vcpu_id(ss->cs), strerror(errno));
  72         exit(1);
  73     }
  74
  75     ss->xirr = state >> KVM_REG_PPC_ICP_XISR_SHIFT;
  76     ss->mfrr = (state >> KVM_REG_PPC_ICP_MFRR_SHIFT)
  77         & KVM_REG_PPC_ICP_MFRR_MASK;
  78     ss->pending_priority = (state >> KVM_REG_PPC_ICP_PPRI_SHIFT)
  79         & KVM_REG_PPC_ICP_PPRI_MASK;
  80 }
  81
  82 static int icp_set_kvm_state(ICPState *ss, int version_id)
  83 {
  84     uint64_t state;
  85     struct kvm_one_reg reg = {
  86         .id = KVM_REG_PPC_ICP_STATE,
  87         .addr = (uintptr_t)&state,
  88     };
  89     int ret;
  90
  91     /* ICP for this CPU thread is not in use, exiting */
  92     if (!ss->cs) {
  93         return 0;
  94     }
  95
  96     state = ((uint64_t)ss->xirr << KVM_REG_PPC_ICP_XISR_SHIFT)
  97         | ((uint64_t)ss->mfrr << KVM_REG_PPC_ICP_MFRR_SHIFT)
  98         | ((uint64_t)ss->pending_priority << KVM_REG_PPC_ICP_PPRI_SHIFT);
  99
 100     ret = kvm_vcpu_ioctl(ss->cs, KVM_SET_ONE_REG, &reg);
 101     if (ret != 0) {
 102         error_report("Unable to restore KVM interrupt controller state (0x%"
 103                 PRIx64 ") for CPU %ld: %s", state, kvm_arch_vcpu_id(ss->cs),
 104                 strerror(errno));
 105         return ret;
 106     }
 107
 108     return 0;
 109 }
 110
 111 static void icp_kvm_reset(DeviceState *dev)
 112 {
 113     ICPState *icp = ICP(dev);
 114
 115     icp->xirr = 0;
 116     icp->pending_priority = 0xff;
 117     icp->mfrr = 0xff;
 118
 119     /* Make all outputs as deasserted only if the CPU thread is in use */
 120     if (icp->output) {
 121         qemu_set_irq(icp->output, 0);
 122     }
 123
 124     icp_set_kvm_state(icp, 1);
 125 }
 126
 127 static void icp_kvm_class_init(ObjectClass *klass, void *data)
 128 {
 129     DeviceClass *dc = DEVICE_CLASS(klass);
 130     ICPStateClass *icpc = ICP_CLASS(klass);
 131
 132     dc->reset = icp_kvm_reset;
 133     icpc->pre_save = icp_get_kvm_state;
 134     icpc->post_load = icp_set_kvm_state;
 135 }
 136
 137 static const TypeInfo icp_kvm_info = {
 138     .name = TYPE_KVM_ICP,
 139     .parent = TYPE_ICP,
 140     .instance_size = sizeof(ICPState),
 141     .class_init = icp_kvm_class_init,
 142     .class_size = sizeof(ICPStateClass),
 143 };
 144
 145 /*
 146  * ICS-KVM
 147  */
 148 static void ics_get_kvm_state(ICSState *ics)
 149 {
 150     uint64_t state;
 151     struct kvm_device_attr attr = {
 152         .flags = 0,
 153         .group = KVM_DEV_XICS_GRP_SOURCES,
 154         .addr = (uint64_t)(uintptr_t)&state,
 155     };
 156     int i;
 157
 158     for (i = 0; i < ics->nr_irqs; i++) {
 159         ICSIRQState *irq = &ics->irqs[i];
 160         int ret;
 161
 162         attr.attr = i + ics->offset;
 163
 164         ret = ioctl(kernel_xics_fd, KVM_GET_DEVICE_ATTR, &attr);
 165         if (ret != 0) {
 166             error_report("Unable to retrieve KVM interrupt controller state"
 167                     " for IRQ %d: %s", i + ics->offset, strerror(errno));
 168             exit(1);
 169         }
 170
 171         irq->server = state & KVM_XICS_DESTINATION_MASK;
 172         irq->saved_priority = (state >> KVM_XICS_PRIORITY_SHIFT)
 173             & KVM_XICS_PRIORITY_MASK;
 174         /*
 175          * To be consistent with the software emulation in xics.c, we
 176          * split out the masked state + priority that we get from the
 177          * kernel into 'current priority' (0xff if masked) and
 178          * 'saved priority' (if masked, this is the priority the
 179          * interrupt had before it was masked).  Masking and unmasking
 180          * are done with the ibm,int-off and ibm,int-on RTAS calls.
 181          */
 182         if (state & KVM_XICS_MASKED) {
 183             irq->priority = 0xff;
 184         } else {
 185             irq->priority = irq->saved_priority;
 186         }
 187
 188         if (state & KVM_XICS_PENDING) {
 189             if (state & KVM_XICS_LEVEL_SENSITIVE) {
 190                 irq->status |= XICS_STATUS_ASSERTED;
 191             } else {
 192                 /*
 193                  * A pending edge-triggered interrupt (or MSI)
 194                  * must have been rejected previously when we
 195                  * first detected it and tried to deliver it,
 196                  * so mark it as pending and previously rejected
 197                  * for consistency with how xics.c works.
 198                  */
 199                 irq->status |= XICS_STATUS_MASKED_PENDING
 200                     | XICS_STATUS_REJECTED;
 201             }
 202         }
 203     }
 204 }
 205
 206 static int ics_set_kvm_state(ICSState *ics, int version_id)
 207 {
 208     uint64_t state;
 209     struct kvm_device_attr attr = {
 210         .flags = 0,
 211         .group = KVM_DEV_XICS_GRP_SOURCES,
 212         .addr = (uint64_t)(uintptr_t)&state,
 213     };
 214     int i;
 215
 216     for (i = 0; i < ics->nr_irqs; i++) {
 217         ICSIRQState *irq = &ics->irqs[i];
 218         int ret;
 219
 220         attr.attr = i + ics->offset;
 221
 222         state = irq->server;
 223         state |= (uint64_t)(irq->saved_priority & KVM_XICS_PRIORITY_MASK)
 224             << KVM_XICS_PRIORITY_SHIFT;
 225         if (irq->priority != irq->saved_priority) {
 226             assert(irq->priority == 0xff);
 227             state |= KVM_XICS_MASKED;
 228         }
 229
 230         if (ics->irqs[i].flags & XICS_FLAGS_IRQ_LSI) {
 231             state |= KVM_XICS_LEVEL_SENSITIVE;
 232             if (irq->status & XICS_STATUS_ASSERTED) {
 233                 state |= KVM_XICS_PENDING;
 234             }
 235         } else {
 236             if (irq->status & XICS_STATUS_MASKED_PENDING) {
 237                 state |= KVM_XICS_PENDING;
 238             }
 239         }
 240
 241         ret = ioctl(kernel_xics_fd, KVM_SET_DEVICE_ATTR, &attr);
 242         if (ret != 0) {
 243             error_report("Unable to restore KVM interrupt controller state"
 244                     " for IRQs %d: %s", i + ics->offset, strerror(errno));
 245             return ret;
 246         }
 247     }
 248
 249     return 0;
 250 }
 251
 252 static void ics_kvm_set_irq(void *opaque, int srcno, int val)
 253 {
 254     ICSState *ics = opaque;
 255     struct kvm_irq_level args;
 256     int rc;
 257
 258     args.irq = srcno + ics->offset;
 259     if (ics->irqs[srcno].flags & XICS_FLAGS_IRQ_MSI) {
 260         if (!val) {
 261             return;
 262         }
 263         args.level = KVM_INTERRUPT_SET;
 264     } else {
 265         args.level = val ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
 266     }
 267     rc = kvm_vm_ioctl(kvm_state, KVM_IRQ_LINE, &args);
 268     if (rc < 0) {
 269         perror("kvm_irq_line");
 270     }
 271 }
 272
 273 static void ics_kvm_reset(DeviceState *dev)
 274 {
 275     ICSState *ics = ICS_SIMPLE(dev);
 276     int i;
 277     uint8_t flags[ics->nr_irqs];
 278
 279     for (i = 0; i < ics->nr_irqs; i++) {
 280         flags[i] = ics->irqs[i].flags;
 281     }
 282
 283     memset(ics->irqs, 0, sizeof(ICSIRQState) * ics->nr_irqs);
 284
 285     for (i = 0; i < ics->nr_irqs; i++) {
 286         ics->irqs[i].priority = 0xff;
 287         ics->irqs[i].saved_priority = 0xff;
 288         ics->irqs[i].flags = flags[i];
 289     }
 290
 291     ics_set_kvm_state(ics, 1);
 292 }
 293
 294 static void ics_kvm_realize(DeviceState *dev, Error **errp)
 295 {
 296     ICSState *ics = ICS_SIMPLE(dev);
 297
 298     if (!ics->nr_irqs) {
 299         error_setg(errp, "Number of interrupts needs to be greater 0");
 300         return;
 301     }
 302     ics->irqs = g_malloc0(ics->nr_irqs * sizeof(ICSIRQState));
 303     ics->qirqs = qemu_allocate_irqs(ics_kvm_set_irq, ics, ics->nr_irqs);
 304 }
 305
 306 static void ics_kvm_class_init(ObjectClass *klass, void *data)
 307 {
 308     DeviceClass *dc = DEVICE_CLASS(klass);
 309     ICSStateClass *icsc = ICS_BASE_CLASS(klass);
 310
 311     icsc->realize = ics_kvm_realize;
 312     dc->reset = ics_kvm_reset;
 313     icsc->pre_save = ics_get_kvm_state;
 314     icsc->post_load = ics_set_kvm_state;
 315 }
 316
 317 static const TypeInfo ics_kvm_info = {
 318     .name = TYPE_ICS_KVM,
 319     .parent = TYPE_ICS_SIMPLE,
 320     .instance_size = sizeof(ICSState),
 321     .class_init = ics_kvm_class_init,
 322 };
 323
 324 /*
 325  * XICS-KVM
 326  */
 327 static void xics_kvm_cpu_setup(XICSState *xics, PowerPCCPU *cpu)
 328 {
 329     CPUState *cs;
 330     ICPState *ss;
 331     int ret;
 332
 333     cs = CPU(cpu);
 334     ss = &xics->ss[cs->cpu_index];
 335
 336     assert(cs->cpu_index < xics->nr_servers);
 337     if (kernel_xics_fd == -1) {
 338         abort();
 339     }
 340
 341     /*
 342      * If we are reusing a parked vCPU fd corresponding to the CPU
 343      * which was hot-removed earlier we don't have to renable
 344      * KVM_CAP_IRQ_XICS capability again.
 345      */
 346     if (ss->cap_irq_xics_enabled) {
 347         return;
 348     }
 349
 350     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_IRQ_XICS, 0, kernel_xics_fd,
 351                               kvm_arch_vcpu_id(cs));
 352     if (ret < 0) {
 353         error_report("Unable to connect CPU%ld to kernel XICS: %s",
 354                      kvm_arch_vcpu_id(cs), strerror(errno));
 355         exit(1);
 356     }
 357     ss->cap_irq_xics_enabled = true;
 358 }
 359
 360 static void rtas_dummy(PowerPCCPU *cpu, sPAPRMachineState *spapr,
 361                        uint32_t token,
 362                        uint32_t nargs, target_ulong args,
 363                        uint32_t nret, target_ulong rets)
 364 {
 365     error_report("pseries: %s must never be called for in-kernel XICS",
 366                  __func__);
 367 }
 368
 369 static void xics_kvm_realize(DeviceState *dev, Error **errp)
 370 {
 371     int rc;
 372     struct kvm_create_device xics_create_device = {
 373         .type = KVM_DEV_TYPE_XICS,
 374         .flags = 0,
 375     };
 376
 377     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
 378         error_setg(errp,
 379                    "KVM and IRQ_XICS capability must be present for in-kernel XICS");
 380         goto fail;
 381     }
 382
 383     spapr_rtas_register(RTAS_IBM_SET_XIVE, "ibm,set-xive", rtas_dummy);
 384     spapr_rtas_register(RTAS_IBM_GET_XIVE, "ibm,get-xive", rtas_dummy);
 385     spapr_rtas_register(RTAS_IBM_INT_OFF, "ibm,int-off", rtas_dummy);
 386     spapr_rtas_register(RTAS_IBM_INT_ON, "ibm,int-on", rtas_dummy);
 387
 388     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_SET_XIVE, "ibm,set-xive");
 389     if (rc < 0) {
 390         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,set-xive");
 391         goto fail;
 392     }
 393
 394     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_GET_XIVE, "ibm,get-xive");
 395     if (rc < 0) {
 396         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,get-xive");
 397         goto fail;
 398     }
 399
 400     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_ON, "ibm,int-on");
 401     if (rc < 0) {
 402         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-on");
 403         goto fail;
 404     }
 405
 406     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_OFF, "ibm,int-off");
 407     if (rc < 0) {
 408         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-off");
 409         goto fail;
 410     }
 411
 412     /* Create the kernel ICP */
 413     rc = kvm_vm_ioctl(kvm_state, KVM_CREATE_DEVICE, &xics_create_device);
 414     if (rc < 0) {
 415         error_setg_errno(errp, -rc, "Error on KVM_CREATE_DEVICE for XICS");
 416         goto fail;
 417     }
 418
 419     kernel_xics_fd = xics_create_device.fd;
 420
 421     kvm_kernel_irqchip = true;
 422     kvm_msi_via_irqfd_allowed = true;
 423     kvm_gsi_direct_mapping = true;
 424
 425     return;
 426
 427 fail:
 428     kvmppc_define_rtas_kernel_token(0, "ibm,set-xive");
 429     kvmppc_define_rtas_kernel_token(0, "ibm,get-xive");
 430     kvmppc_define_rtas_kernel_token(0, "ibm,int-on");
 431     kvmppc_define_rtas_kernel_token(0, "ibm,int-off");
 432 }
 433
 434 static void xics_kvm_class_init(ObjectClass *oc, void *data)
 435 {
 436     DeviceClass *dc = DEVICE_CLASS(oc);
 437     XICSStateClass *xsc = XICS_COMMON_CLASS(oc);
 438
 439     dc->realize = xics_kvm_realize;
 440     xsc->cpu_setup = xics_kvm_cpu_setup;
 441 }
 442
 443 static const TypeInfo xics_spapr_kvm_info = {
 444     .name          = TYPE_XICS_SPAPR_KVM,
 445     .parent        = TYPE_XICS_COMMON,
 446     .instance_size = sizeof(KVMXICSState),
 447     .class_init    = xics_kvm_class_init,
 448 };
 449
 450 static void xics_kvm_register_types(void)
 451 {
 452     type_register_static(&xics_spapr_kvm_info);
 453     type_register_static(&ics_kvm_info);
 454     type_register_static(&icp_kvm_info);
 455 }
 456
 457 type_init(xics_kvm_register_types)