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 KVMEnabledICP {
  46     unsigned long vcpu_id;
  47     QLIST_ENTRY(KVMEnabledICP) node;
  48 } KVMEnabledICP;
  49
  50 static QLIST_HEAD(, KVMEnabledICP)
  51     kvm_enabled_icps = QLIST_HEAD_INITIALIZER(&kvm_enabled_icps);
  52
  53 /*
  54  * ICP-KVM
  55  */
  56 static void icp_get_kvm_state(ICPState *icp)
  57 {
  58     uint64_t state;
  59     int ret;
  60
  61     /* ICP for this CPU thread is not in use, exiting */
  62     if (!icp->cs) {
  63         return;
  64     }
  65
  66     ret = kvm_get_one_reg(icp->cs, KVM_REG_PPC_ICP_STATE, &state);
  67     if (ret != 0) {
  68         error_report("Unable to retrieve KVM interrupt controller state"
  69                 " for CPU %ld: %s", kvm_arch_vcpu_id(icp->cs), strerror(errno));
  70         exit(1);
  71     }
  72
  73     icp->xirr = state >> KVM_REG_PPC_ICP_XISR_SHIFT;
  74     icp->mfrr = (state >> KVM_REG_PPC_ICP_MFRR_SHIFT)
  75         & KVM_REG_PPC_ICP_MFRR_MASK;
  76     icp->pending_priority = (state >> KVM_REG_PPC_ICP_PPRI_SHIFT)
  77         & KVM_REG_PPC_ICP_PPRI_MASK;
  78 }
  79
  80 static void do_icp_synchronize_state(CPUState *cpu, run_on_cpu_data arg)
  81 {
  82     icp_get_kvm_state(arg.host_ptr);
  83 }
  84
  85 static void icp_synchronize_state(ICPState *icp)
  86 {
  87     if (icp->cs) {
  88         run_on_cpu(icp->cs, do_icp_synchronize_state, RUN_ON_CPU_HOST_PTR(icp));
  89     }
  90 }
  91
  92 static int icp_set_kvm_state(ICPState *icp, int version_id)
  93 {
  94     uint64_t state;
  95     int ret;
  96
  97     /* ICP for this CPU thread is not in use, exiting */
  98     if (!icp->cs) {
  99         return 0;
 100     }
 101
 102     state = ((uint64_t)icp->xirr << KVM_REG_PPC_ICP_XISR_SHIFT)
 103         | ((uint64_t)icp->mfrr << KVM_REG_PPC_ICP_MFRR_SHIFT)
 104         | ((uint64_t)icp->pending_priority << KVM_REG_PPC_ICP_PPRI_SHIFT);
 105
 106     ret = kvm_set_one_reg(icp->cs, KVM_REG_PPC_ICP_STATE, &state);
 107     if (ret != 0) {
 108         error_report("Unable to restore KVM interrupt controller state (0x%"
 109                 PRIx64 ") for CPU %ld: %s", state, kvm_arch_vcpu_id(icp->cs),
 110                 strerror(errno));
 111         return ret;
 112     }
 113
 114     return 0;
 115 }
 116
 117 static void icp_kvm_reset(DeviceState *dev)
 118 {
 119     ICPStateClass *icpc = ICP_GET_CLASS(dev);
 120
 121     icpc->parent_reset(dev);
 122
 123     icp_set_kvm_state(ICP(dev), 1);
 124 }
 125
 126 static void icp_kvm_realize(DeviceState *dev, Error **errp)
 127 {
 128     ICPState *icp = ICP(dev);
 129     ICPStateClass *icpc = ICP_GET_CLASS(icp);
 130     Error *local_err = NULL;
 131     CPUState *cs;
 132     KVMEnabledICP *enabled_icp;
 133     unsigned long vcpu_id;
 134     int ret;
 135
 136     if (kernel_xics_fd == -1) {
 137         abort();
 138     }
 139
 140     icpc->parent_realize(dev, &local_err);
 141     if (local_err) {
 142         error_propagate(errp, local_err);
 143         return;
 144     }
 145
 146     cs = icp->cs;
 147     vcpu_id = kvm_arch_vcpu_id(cs);
 148
 149     /*
 150      * If we are reusing a parked vCPU fd corresponding to the CPU
 151      * which was hot-removed earlier we don't have to renable
 152      * KVM_CAP_IRQ_XICS capability again.
 153      */
 154     QLIST_FOREACH(enabled_icp, &kvm_enabled_icps, node) {
 155         if (enabled_icp->vcpu_id == vcpu_id) {
 156             return;
 157         }
 158     }
 159
 160     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_IRQ_XICS, 0, kernel_xics_fd, vcpu_id);
 161     if (ret < 0) {
 162         error_setg(errp, "Unable to connect CPU%ld to kernel XICS: %s", vcpu_id,
 163                    strerror(errno));
 164         return;
 165     }
 166     enabled_icp = g_malloc(sizeof(*enabled_icp));
 167     enabled_icp->vcpu_id = vcpu_id;
 168     QLIST_INSERT_HEAD(&kvm_enabled_icps, enabled_icp, node);
 169 }
 170
 171 static void icp_kvm_class_init(ObjectClass *klass, void *data)
 172 {
 173     DeviceClass *dc = DEVICE_CLASS(klass);
 174     ICPStateClass *icpc = ICP_CLASS(klass);
 175
 176     device_class_set_parent_realize(dc, icp_kvm_realize,
 177                                     &icpc->parent_realize);
 178     device_class_set_parent_reset(dc, icp_kvm_reset,
 179                                   &icpc->parent_reset);
 180
 181     icpc->pre_save = icp_get_kvm_state;
 182     icpc->post_load = icp_set_kvm_state;
 183     icpc->synchronize_state = icp_synchronize_state;
 184 }
 185
 186 static const TypeInfo icp_kvm_info = {
 187     .name = TYPE_KVM_ICP,
 188     .parent = TYPE_ICP,
 189     .instance_size = sizeof(ICPState),
 190     .class_init = icp_kvm_class_init,
 191     .class_size = sizeof(ICPStateClass),
 192 };
 193
 194 /*
 195  * ICS-KVM
 196  */
 197 static void ics_get_kvm_state(ICSState *ics)
 198 {
 199     uint64_t state;
 200     int i;
 201     Error *local_err = NULL;
 202
 203     for (i = 0; i < ics->nr_irqs; i++) {
 204         ICSIRQState *irq = &ics->irqs[i];
 205
 206         kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
 207                           i + ics->offset, &state, false, &local_err);
 208         if (local_err) {
 209             error_report_err(local_err);
 210             exit(1);
 211         }
 212
 213         irq->server = state & KVM_XICS_DESTINATION_MASK;
 214         irq->saved_priority = (state >> KVM_XICS_PRIORITY_SHIFT)
 215             & KVM_XICS_PRIORITY_MASK;
 216         /*
 217          * To be consistent with the software emulation in xics.c, we
 218          * split out the masked state + priority that we get from the
 219          * kernel into 'current priority' (0xff if masked) and
 220          * 'saved priority' (if masked, this is the priority the
 221          * interrupt had before it was masked).  Masking and unmasking
 222          * are done with the ibm,int-off and ibm,int-on RTAS calls.
 223          */
 224         if (state & KVM_XICS_MASKED) {
 225             irq->priority = 0xff;
 226         } else {
 227             irq->priority = irq->saved_priority;
 228         }
 229
 230         irq->status = 0;
 231         if (state & KVM_XICS_PENDING) {
 232             if (state & KVM_XICS_LEVEL_SENSITIVE) {
 233                 irq->status |= XICS_STATUS_ASSERTED;
 234             } else {
 235                 /*
 236                  * A pending edge-triggered interrupt (or MSI)
 237                  * must have been rejected previously when we
 238                  * first detected it and tried to deliver it,
 239                  * so mark it as pending and previously rejected
 240                  * for consistency with how xics.c works.
 241                  */
 242                 irq->status |= XICS_STATUS_MASKED_PENDING
 243                     | XICS_STATUS_REJECTED;
 244             }
 245         }
 246         if (state & KVM_XICS_PRESENTED) {
 247                 irq->status |= XICS_STATUS_PRESENTED;
 248         }
 249         if (state & KVM_XICS_QUEUED) {
 250                 irq->status |= XICS_STATUS_QUEUED;
 251         }
 252     }
 253 }
 254
 255 static void ics_synchronize_state(ICSState *ics)
 256 {
 257     ics_get_kvm_state(ics);
 258 }
 259
 260 static int ics_set_kvm_state(ICSState *ics, int version_id)
 261 {
 262     uint64_t state;
 263     int i;
 264     Error *local_err = NULL;
 265
 266     for (i = 0; i < ics->nr_irqs; i++) {
 267         ICSIRQState *irq = &ics->irqs[i];
 268         int ret;
 269
 270         state = irq->server;
 271         state |= (uint64_t)(irq->saved_priority & KVM_XICS_PRIORITY_MASK)
 272             << KVM_XICS_PRIORITY_SHIFT;
 273         if (irq->priority != irq->saved_priority) {
 274             assert(irq->priority == 0xff);
 275             state |= KVM_XICS_MASKED;
 276         }
 277
 278         if (ics->irqs[i].flags & XICS_FLAGS_IRQ_LSI) {
 279             state |= KVM_XICS_LEVEL_SENSITIVE;
 280             if (irq->status & XICS_STATUS_ASSERTED) {
 281                 state |= KVM_XICS_PENDING;
 282             }
 283         } else {
 284             if (irq->status & XICS_STATUS_MASKED_PENDING) {
 285                 state |= KVM_XICS_PENDING;
 286             }
 287         }
 288         if (irq->status & XICS_STATUS_PRESENTED) {
 289                 state |= KVM_XICS_PRESENTED;
 290         }
 291         if (irq->status & XICS_STATUS_QUEUED) {
 292                 state |= KVM_XICS_QUEUED;
 293         }
 294
 295         ret = kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
 296                                 i + ics->offset, &state, true, &local_err);
 297         if (local_err) {
 298             error_report_err(local_err);
 299             return ret;
 300         }
 301     }
 302
 303     return 0;
 304 }
 305
 306 static void ics_kvm_set_irq(void *opaque, int srcno, int val)
 307 {
 308     ICSState *ics = opaque;
 309     struct kvm_irq_level args;
 310     int rc;
 311
 312     args.irq = srcno + ics->offset;
 313     if (ics->irqs[srcno].flags & XICS_FLAGS_IRQ_MSI) {
 314         if (!val) {
 315             return;
 316         }
 317         args.level = KVM_INTERRUPT_SET;
 318     } else {
 319         args.level = val ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
 320     }
 321     rc = kvm_vm_ioctl(kvm_state, KVM_IRQ_LINE, &args);
 322     if (rc < 0) {
 323         perror("kvm_irq_line");
 324     }
 325 }
 326
 327 static void ics_kvm_reset(DeviceState *dev)
 328 {
 329     ICSStateClass *icsc = ICS_BASE_GET_CLASS(dev);
 330
 331     icsc->parent_reset(dev);
 332
 333     ics_set_kvm_state(ICS_KVM(dev), 1);
 334 }
 335
 336 static void ics_kvm_reset_handler(void *dev)
 337 {
 338     ics_kvm_reset(dev);
 339 }
 340
 341 static void ics_kvm_realize(DeviceState *dev, Error **errp)
 342 {
 343     ICSState *ics = ICS_KVM(dev);
 344     ICSStateClass *icsc = ICS_BASE_GET_CLASS(ics);
 345     Error *local_err = NULL;
 346
 347     icsc->parent_realize(dev, &local_err);
 348     if (local_err) {
 349         error_propagate(errp, local_err);
 350         return;
 351     }
 352     ics->qirqs = qemu_allocate_irqs(ics_kvm_set_irq, ics, ics->nr_irqs);
 353
 354     qemu_register_reset(ics_kvm_reset_handler, ics);
 355 }
 356
 357 static void ics_kvm_class_init(ObjectClass *klass, void *data)
 358 {
 359     ICSStateClass *icsc = ICS_BASE_CLASS(klass);
 360     DeviceClass *dc = DEVICE_CLASS(klass);
 361
 362     device_class_set_parent_realize(dc, ics_kvm_realize,
 363                                     &icsc->parent_realize);
 364     device_class_set_parent_reset(dc, ics_kvm_reset,
 365                                   &icsc->parent_reset);
 366
 367     icsc->pre_save = ics_get_kvm_state;
 368     icsc->post_load = ics_set_kvm_state;
 369     icsc->synchronize_state = ics_synchronize_state;
 370 }
 371
 372 static const TypeInfo ics_kvm_info = {
 373     .name = TYPE_ICS_KVM,
 374     .parent = TYPE_ICS_BASE,
 375     .instance_size = sizeof(ICSState),
 376     .class_init = ics_kvm_class_init,
 377 };
 378
 379 /*
 380  * XICS-KVM
 381  */
 382
 383 static void rtas_dummy(PowerPCCPU *cpu, sPAPRMachineState *spapr,
 384                        uint32_t token,
 385                        uint32_t nargs, target_ulong args,
 386                        uint32_t nret, target_ulong rets)
 387 {
 388     error_report("pseries: %s must never be called for in-kernel XICS",
 389                  __func__);
 390 }
 391
 392 int xics_kvm_init(sPAPRMachineState *spapr, Error **errp)
 393 {
 394     int rc;
 395
 396     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
 397         error_setg(errp,
 398                    "KVM and IRQ_XICS capability must be present for in-kernel XICS");
 399         goto fail;
 400     }
 401
 402     spapr_rtas_register(RTAS_IBM_SET_XIVE, "ibm,set-xive", rtas_dummy);
 403     spapr_rtas_register(RTAS_IBM_GET_XIVE, "ibm,get-xive", rtas_dummy);
 404     spapr_rtas_register(RTAS_IBM_INT_OFF, "ibm,int-off", rtas_dummy);
 405     spapr_rtas_register(RTAS_IBM_INT_ON, "ibm,int-on", rtas_dummy);
 406
 407     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_SET_XIVE, "ibm,set-xive");
 408     if (rc < 0) {
 409         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,set-xive");
 410         goto fail;
 411     }
 412
 413     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_GET_XIVE, "ibm,get-xive");
 414     if (rc < 0) {
 415         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,get-xive");
 416         goto fail;
 417     }
 418
 419     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_ON, "ibm,int-on");
 420     if (rc < 0) {
 421         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-on");
 422         goto fail;
 423     }
 424
 425     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_OFF, "ibm,int-off");
 426     if (rc < 0) {
 427         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-off");
 428         goto fail;
 429     }
 430
 431     /* Create the KVM XICS device */
 432     rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false);
 433     if (rc < 0) {
 434         error_setg_errno(errp, -rc, "Error on KVM_CREATE_DEVICE for XICS");
 435         goto fail;
 436     }
 437
 438     kernel_xics_fd = rc;
 439     kvm_kernel_irqchip = true;
 440     kvm_msi_via_irqfd_allowed = true;
 441     kvm_gsi_direct_mapping = true;
 442
 443     return 0;
 444
 445 fail:
 446     kvmppc_define_rtas_kernel_token(0, "ibm,set-xive");
 447     kvmppc_define_rtas_kernel_token(0, "ibm,get-xive");
 448     kvmppc_define_rtas_kernel_token(0, "ibm,int-on");
 449     kvmppc_define_rtas_kernel_token(0, "ibm,int-off");
 450     return -1;
 451 }
 452
 453 static void xics_kvm_register_types(void)
 454 {
 455     type_register_static(&ics_kvm_info);
 456     type_register_static(&icp_kvm_info);
 457 }
 458
 459 type_init(xics_kvm_register_types)