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 typedef struct KVMXICSState {
  44     XICSState parent_obj;
  45
  46     int kernel_xics_fd;
  47 } KVMXICSState;
  48
  49 /*
  50  * ICP-KVM
  51  */
  52 static void icp_get_kvm_state(ICPState *ss)
  53 {
  54     uint64_t state;
  55     struct kvm_one_reg reg = {
  56         .id = KVM_REG_PPC_ICP_STATE,
  57         .addr = (uintptr_t)&state,
  58     };
  59     int ret;
  60
  61     /* ICP for this CPU thread is not in use, exiting */
  62     if (!ss->cs) {
  63         return;
  64     }
  65
  66     ret = kvm_vcpu_ioctl(ss->cs, KVM_GET_ONE_REG, &reg);
  67     if (ret != 0) {
  68         error_report("Unable to retrieve KVM interrupt controller state"
  69                 " for CPU %ld: %s", kvm_arch_vcpu_id(ss->cs), strerror(errno));
  70         exit(1);
  71     }
  72
  73     ss->xirr = state >> KVM_REG_PPC_ICP_XISR_SHIFT;
  74     ss->mfrr = (state >> KVM_REG_PPC_ICP_MFRR_SHIFT)
  75         & KVM_REG_PPC_ICP_MFRR_MASK;
  76     ss->pending_priority = (state >> KVM_REG_PPC_ICP_PPRI_SHIFT)
  77         & KVM_REG_PPC_ICP_PPRI_MASK;
  78 }
  79
  80 static int icp_set_kvm_state(ICPState *ss, int version_id)
  81 {
  82     uint64_t state;
  83     struct kvm_one_reg reg = {
  84         .id = KVM_REG_PPC_ICP_STATE,
  85         .addr = (uintptr_t)&state,
  86     };
  87     int ret;
  88
  89     /* ICP for this CPU thread is not in use, exiting */
  90     if (!ss->cs) {
  91         return 0;
  92     }
  93
  94     state = ((uint64_t)ss->xirr << KVM_REG_PPC_ICP_XISR_SHIFT)
  95         | ((uint64_t)ss->mfrr << KVM_REG_PPC_ICP_MFRR_SHIFT)
  96         | ((uint64_t)ss->pending_priority << KVM_REG_PPC_ICP_PPRI_SHIFT);
  97
  98     ret = kvm_vcpu_ioctl(ss->cs, KVM_SET_ONE_REG, &reg);
  99     if (ret != 0) {
 100         error_report("Unable to restore KVM interrupt controller state (0x%"
 101                 PRIx64 ") for CPU %ld: %s", state, kvm_arch_vcpu_id(ss->cs),
 102                 strerror(errno));
 103         return ret;
 104     }
 105
 106     return 0;
 107 }
 108
 109 static void icp_kvm_reset(DeviceState *dev)
 110 {
 111     ICPState *icp = ICP(dev);
 112
 113     icp->xirr = 0;
 114     icp->pending_priority = 0xff;
 115     icp->mfrr = 0xff;
 116
 117     /* Make all outputs as deasserted only if the CPU thread is in use */
 118     if (icp->output) {
 119         qemu_set_irq(icp->output, 0);
 120     }
 121
 122     icp_set_kvm_state(icp, 1);
 123 }
 124
 125 static void icp_kvm_class_init(ObjectClass *klass, void *data)
 126 {
 127     DeviceClass *dc = DEVICE_CLASS(klass);
 128     ICPStateClass *icpc = ICP_CLASS(klass);
 129
 130     dc->reset = icp_kvm_reset;
 131     icpc->pre_save = icp_get_kvm_state;
 132     icpc->post_load = icp_set_kvm_state;
 133 }
 134
 135 static const TypeInfo icp_kvm_info = {
 136     .name = TYPE_KVM_ICP,
 137     .parent = TYPE_ICP,
 138     .instance_size = sizeof(ICPState),
 139     .class_init = icp_kvm_class_init,
 140     .class_size = sizeof(ICPStateClass),
 141 };
 142
 143 /*
 144  * ICS-KVM
 145  */
 146 static void ics_get_kvm_state(ICSState *ics)
 147 {
 148     KVMXICSState *xicskvm = XICS_SPAPR_KVM(ics->xics);
 149     uint64_t state;
 150     struct kvm_device_attr attr = {
 151         .flags = 0,
 152         .group = KVM_DEV_XICS_GRP_SOURCES,
 153         .addr = (uint64_t)(uintptr_t)&state,
 154     };
 155     int i;
 156
 157     for (i = 0; i < ics->nr_irqs; i++) {
 158         ICSIRQState *irq = &ics->irqs[i];
 159         int ret;
 160
 161         attr.attr = i + ics->offset;
 162
 163         ret = ioctl(xicskvm->kernel_xics_fd, KVM_GET_DEVICE_ATTR, &attr);
 164         if (ret != 0) {
 165             error_report("Unable to retrieve KVM interrupt controller state"
 166                     " for IRQ %d: %s", i + ics->offset, strerror(errno));
 167             exit(1);
 168         }
 169
 170         irq->server = state & KVM_XICS_DESTINATION_MASK;
 171         irq->saved_priority = (state >> KVM_XICS_PRIORITY_SHIFT)
 172             & KVM_XICS_PRIORITY_MASK;
 173         /*
 174          * To be consistent with the software emulation in xics.c, we
 175          * split out the masked state + priority that we get from the
 176          * kernel into 'current priority' (0xff if masked) and
 177          * 'saved priority' (if masked, this is the priority the
 178          * interrupt had before it was masked).  Masking and unmasking
 179          * are done with the ibm,int-off and ibm,int-on RTAS calls.
 180          */
 181         if (state & KVM_XICS_MASKED) {
 182             irq->priority = 0xff;
 183         } else {
 184             irq->priority = irq->saved_priority;
 185         }
 186
 187         if (state & KVM_XICS_PENDING) {
 188             if (state & KVM_XICS_LEVEL_SENSITIVE) {
 189                 irq->status |= XICS_STATUS_ASSERTED;
 190             } else {
 191                 /*
 192                  * A pending edge-triggered interrupt (or MSI)
 193                  * must have been rejected previously when we
 194                  * first detected it and tried to deliver it,
 195                  * so mark it as pending and previously rejected
 196                  * for consistency with how xics.c works.
 197                  */
 198                 irq->status |= XICS_STATUS_MASKED_PENDING
 199                     | XICS_STATUS_REJECTED;
 200             }
 201         }
 202     }
 203 }
 204
 205 static int ics_set_kvm_state(ICSState *ics, int version_id)
 206 {
 207     KVMXICSState *xicskvm = XICS_SPAPR_KVM(ics->xics);
 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(xicskvm->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(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(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_CLASS(klass);
 310
 311     dc->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_KVM_ICS,
 319     .parent = TYPE_ICS,
 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     KVMXICSState *xicskvm = XICS_SPAPR_KVM(xics);
 332
 333     cs = CPU(cpu);
 334     ss = &xics->ss[cs->cpu_index];
 335
 336     assert(cs->cpu_index < xics->nr_servers);
 337     if (xicskvm->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     if (xicskvm->kernel_xics_fd != -1) {
 351         int ret;
 352
 353         ret = kvm_vcpu_enable_cap(cs, KVM_CAP_IRQ_XICS, 0,
 354                                   xicskvm->kernel_xics_fd,
 355                                   kvm_arch_vcpu_id(cs));
 356         if (ret < 0) {
 357             error_report("Unable to connect CPU%ld to kernel XICS: %s",
 358                     kvm_arch_vcpu_id(cs), strerror(errno));
 359             exit(1);
 360         }
 361         ss->cap_irq_xics_enabled = true;
 362     }
 363 }
 364
 365 static void xics_kvm_set_nr_irqs(XICSState *xics, uint32_t nr_irqs,
 366                                  Error **errp)
 367 {
 368     xics->nr_irqs = xics->ics->nr_irqs = nr_irqs;
 369 }
 370
 371 static void xics_kvm_set_nr_servers(XICSState *xics, uint32_t nr_servers,
 372                                     Error **errp)
 373 {
 374     int i;
 375
 376     xics->nr_servers = nr_servers;
 377
 378     xics->ss = g_malloc0(xics->nr_servers * sizeof(ICPState));
 379     for (i = 0; i < xics->nr_servers; i++) {
 380         char buffer[32];
 381         object_initialize(&xics->ss[i], sizeof(xics->ss[i]), TYPE_KVM_ICP);
 382         snprintf(buffer, sizeof(buffer), "icp[%d]", i);
 383         object_property_add_child(OBJECT(xics), buffer, OBJECT(&xics->ss[i]),
 384                                   errp);
 385     }
 386 }
 387
 388 static void rtas_dummy(PowerPCCPU *cpu, sPAPRMachineState *spapr,
 389                        uint32_t token,
 390                        uint32_t nargs, target_ulong args,
 391                        uint32_t nret, target_ulong rets)
 392 {
 393     error_report("pseries: %s must never be called for in-kernel XICS",
 394                  __func__);
 395 }
 396
 397 static void xics_kvm_realize(DeviceState *dev, Error **errp)
 398 {
 399     KVMXICSState *xicskvm = XICS_SPAPR_KVM(dev);
 400     XICSState *xics = XICS_COMMON(dev);
 401     int i, rc;
 402     Error *error = NULL;
 403     struct kvm_create_device xics_create_device = {
 404         .type = KVM_DEV_TYPE_XICS,
 405         .flags = 0,
 406     };
 407
 408     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
 409         error_setg(errp,
 410                    "KVM and IRQ_XICS capability must be present for in-kernel XICS");
 411         goto fail;
 412     }
 413
 414     spapr_rtas_register(RTAS_IBM_SET_XIVE, "ibm,set-xive", rtas_dummy);
 415     spapr_rtas_register(RTAS_IBM_GET_XIVE, "ibm,get-xive", rtas_dummy);
 416     spapr_rtas_register(RTAS_IBM_INT_OFF, "ibm,int-off", rtas_dummy);
 417     spapr_rtas_register(RTAS_IBM_INT_ON, "ibm,int-on", rtas_dummy);
 418
 419     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_SET_XIVE, "ibm,set-xive");
 420     if (rc < 0) {
 421         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,set-xive");
 422         goto fail;
 423     }
 424
 425     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_GET_XIVE, "ibm,get-xive");
 426     if (rc < 0) {
 427         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,get-xive");
 428         goto fail;
 429     }
 430
 431     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_ON, "ibm,int-on");
 432     if (rc < 0) {
 433         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-on");
 434         goto fail;
 435     }
 436
 437     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_OFF, "ibm,int-off");
 438     if (rc < 0) {
 439         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-off");
 440         goto fail;
 441     }
 442
 443     /* Create the kernel ICP */
 444     rc = kvm_vm_ioctl(kvm_state, KVM_CREATE_DEVICE, &xics_create_device);
 445     if (rc < 0) {
 446         error_setg_errno(errp, -rc, "Error on KVM_CREATE_DEVICE for XICS");
 447         goto fail;
 448     }
 449
 450     xicskvm->kernel_xics_fd = xics_create_device.fd;
 451
 452     object_property_set_bool(OBJECT(xics->ics), true, "realized", &error);
 453     if (error) {
 454         error_propagate(errp, error);
 455         goto fail;
 456     }
 457
 458     assert(xics->nr_servers);
 459     for (i = 0; i < xics->nr_servers; i++) {
 460         object_property_set_bool(OBJECT(&xics->ss[i]), true, "realized",
 461                                  &error);
 462         if (error) {
 463             error_propagate(errp, error);
 464             goto fail;
 465         }
 466     }
 467
 468     kvm_kernel_irqchip = true;
 469     kvm_msi_via_irqfd_allowed = true;
 470     kvm_gsi_direct_mapping = true;
 471
 472     return;
 473
 474 fail:
 475     kvmppc_define_rtas_kernel_token(0, "ibm,set-xive");
 476     kvmppc_define_rtas_kernel_token(0, "ibm,get-xive");
 477     kvmppc_define_rtas_kernel_token(0, "ibm,int-on");
 478     kvmppc_define_rtas_kernel_token(0, "ibm,int-off");
 479 }
 480
 481 static void xics_kvm_initfn(Object *obj)
 482 {
 483     XICSState *xics = XICS_COMMON(obj);
 484
 485     xics->ics = ICS(object_new(TYPE_KVM_ICS));
 486     object_property_add_child(obj, "ics", OBJECT(xics->ics), NULL);
 487     xics->ics->xics = xics;
 488 }
 489
 490 static void xics_kvm_class_init(ObjectClass *oc, void *data)
 491 {
 492     DeviceClass *dc = DEVICE_CLASS(oc);
 493     XICSStateClass *xsc = XICS_COMMON_CLASS(oc);
 494
 495     dc->realize = xics_kvm_realize;
 496     xsc->cpu_setup = xics_kvm_cpu_setup;
 497     xsc->set_nr_irqs = xics_kvm_set_nr_irqs;
 498     xsc->set_nr_servers = xics_kvm_set_nr_servers;
 499 }
 500
 501 static const TypeInfo xics_spapr_kvm_info = {
 502     .name          = TYPE_XICS_SPAPR_KVM,
 503     .parent        = TYPE_XICS_COMMON,
 504     .instance_size = sizeof(KVMXICSState),
 505     .class_init    = xics_kvm_class_init,
 506     .instance_init = xics_kvm_initfn,
 507 };
 508
 509 static void xics_kvm_register_types(void)
 510 {
 511     type_register_static(&xics_spapr_kvm_info);
 512     type_register_static(&ics_kvm_info);
 513     type_register_static(&icp_kvm_info);
 514 }
 515
 516 type_init(xics_kvm_register_types)