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 "hw/ppc/xics_spapr.h"
  38 #include "kvm_ppc.h"
  39 #include "qemu/config-file.h"
  40 #include "qemu/error-report.h"
  41
  42 #include <sys/ioctl.h>
  43
  44 static int kernel_xics_fd = -1;
  45
  46 typedef struct KVMEnabledICP {
  47     unsigned long vcpu_id;
  48     QLIST_ENTRY(KVMEnabledICP) node;
  49 } KVMEnabledICP;
  50
  51 static QLIST_HEAD(, KVMEnabledICP)
  52     kvm_enabled_icps = QLIST_HEAD_INITIALIZER(&kvm_enabled_icps);
  53
  54 /*
  55  * ICP-KVM
  56  */
  57 static void icp_get_kvm_state(ICPState *icp)
  58 {
  59     uint64_t state;
  60     int ret;
  61
  62     /* ICP for this CPU thread is not in use, exiting */
  63     if (!icp->cs) {
  64         return;
  65     }
  66
  67     ret = kvm_get_one_reg(icp->cs, KVM_REG_PPC_ICP_STATE, &state);
  68     if (ret != 0) {
  69         error_report("Unable to retrieve KVM interrupt controller state"
  70                 " for CPU %ld: %s", kvm_arch_vcpu_id(icp->cs), strerror(errno));
  71         exit(1);
  72     }
  73
  74     icp->xirr = state >> KVM_REG_PPC_ICP_XISR_SHIFT;
  75     icp->mfrr = (state >> KVM_REG_PPC_ICP_MFRR_SHIFT)
  76         & KVM_REG_PPC_ICP_MFRR_MASK;
  77     icp->pending_priority = (state >> KVM_REG_PPC_ICP_PPRI_SHIFT)
  78         & KVM_REG_PPC_ICP_PPRI_MASK;
  79 }
  80
  81 static void do_icp_synchronize_state(CPUState *cpu, run_on_cpu_data arg)
  82 {
  83     icp_get_kvm_state(arg.host_ptr);
  84 }
  85
  86 static void icp_synchronize_state(ICPState *icp)
  87 {
  88     if (icp->cs) {
  89         run_on_cpu(icp->cs, do_icp_synchronize_state, RUN_ON_CPU_HOST_PTR(icp));
  90     }
  91 }
  92
  93 static int icp_set_kvm_state(ICPState *icp, int version_id)
  94 {
  95     uint64_t state;
  96     int ret;
  97
  98     /* ICP for this CPU thread is not in use, exiting */
  99     if (!icp->cs) {
 100         return 0;
 101     }
 102
 103     state = ((uint64_t)icp->xirr << KVM_REG_PPC_ICP_XISR_SHIFT)
 104         | ((uint64_t)icp->mfrr << KVM_REG_PPC_ICP_MFRR_SHIFT)
 105         | ((uint64_t)icp->pending_priority << KVM_REG_PPC_ICP_PPRI_SHIFT);
 106
 107     ret = kvm_set_one_reg(icp->cs, KVM_REG_PPC_ICP_STATE, &state);
 108     if (ret != 0) {
 109         error_report("Unable to restore KVM interrupt controller state (0x%"
 110                 PRIx64 ") for CPU %ld: %s", state, kvm_arch_vcpu_id(icp->cs),
 111                 strerror(errno));
 112         return ret;
 113     }
 114
 115     return 0;
 116 }
 117
 118 static void icp_kvm_reset(DeviceState *dev)
 119 {
 120     ICPStateClass *icpc = ICP_GET_CLASS(dev);
 121
 122     icpc->parent_reset(dev);
 123
 124     icp_set_kvm_state(ICP(dev), 1);
 125 }
 126
 127 static void icp_kvm_realize(DeviceState *dev, Error **errp)
 128 {
 129     ICPState *icp = ICP(dev);
 130     ICPStateClass *icpc = ICP_GET_CLASS(icp);
 131     Error *local_err = NULL;
 132     CPUState *cs;
 133     KVMEnabledICP *enabled_icp;
 134     unsigned long vcpu_id;
 135     int ret;
 136
 137     if (kernel_xics_fd == -1) {
 138         abort();
 139     }
 140
 141     icpc->parent_realize(dev, &local_err);
 142     if (local_err) {
 143         error_propagate(errp, local_err);
 144         return;
 145     }
 146
 147     cs = icp->cs;
 148     vcpu_id = kvm_arch_vcpu_id(cs);
 149
 150     /*
 151      * If we are reusing a parked vCPU fd corresponding to the CPU
 152      * which was hot-removed earlier we don't have to renable
 153      * KVM_CAP_IRQ_XICS capability again.
 154      */
 155     QLIST_FOREACH(enabled_icp, &kvm_enabled_icps, node) {
 156         if (enabled_icp->vcpu_id == vcpu_id) {
 157             return;
 158         }
 159     }
 160
 161     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_IRQ_XICS, 0, kernel_xics_fd, vcpu_id);
 162     if (ret < 0) {
 163         error_setg(errp, "Unable to connect CPU%ld to kernel XICS: %s", vcpu_id,
 164                    strerror(errno));
 165         return;
 166     }
 167     enabled_icp = g_malloc(sizeof(*enabled_icp));
 168     enabled_icp->vcpu_id = vcpu_id;
 169     QLIST_INSERT_HEAD(&kvm_enabled_icps, enabled_icp, node);
 170 }
 171
 172 static void icp_kvm_class_init(ObjectClass *klass, void *data)
 173 {
 174     DeviceClass *dc = DEVICE_CLASS(klass);
 175     ICPStateClass *icpc = ICP_CLASS(klass);
 176
 177     device_class_set_parent_realize(dc, icp_kvm_realize,
 178                                     &icpc->parent_realize);
 179     device_class_set_parent_reset(dc, icp_kvm_reset,
 180                                   &icpc->parent_reset);
 181
 182     icpc->pre_save = icp_get_kvm_state;
 183     icpc->post_load = icp_set_kvm_state;
 184     icpc->synchronize_state = icp_synchronize_state;
 185 }
 186
 187 static const TypeInfo icp_kvm_info = {
 188     .name = TYPE_KVM_ICP,
 189     .parent = TYPE_ICP,
 190     .instance_size = sizeof(ICPState),
 191     .class_init = icp_kvm_class_init,
 192     .class_size = sizeof(ICPStateClass),
 193 };
 194
 195 /*
 196  * ICS-KVM
 197  */
 198 static void ics_get_kvm_state(ICSState *ics)
 199 {
 200     uint64_t state;
 201     int i;
 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, &error_fatal);
 208
 209         irq->server = state & KVM_XICS_DESTINATION_MASK;
 210         irq->saved_priority = (state >> KVM_XICS_PRIORITY_SHIFT)
 211             & KVM_XICS_PRIORITY_MASK;
 212         /*
 213          * To be consistent with the software emulation in xics.c, we
 214          * split out the masked state + priority that we get from the
 215          * kernel into 'current priority' (0xff if masked) and
 216          * 'saved priority' (if masked, this is the priority the
 217          * interrupt had before it was masked).  Masking and unmasking
 218          * are done with the ibm,int-off and ibm,int-on RTAS calls.
 219          */
 220         if (state & KVM_XICS_MASKED) {
 221             irq->priority = 0xff;
 222         } else {
 223             irq->priority = irq->saved_priority;
 224         }
 225
 226         irq->status = 0;
 227         if (state & KVM_XICS_PENDING) {
 228             if (state & KVM_XICS_LEVEL_SENSITIVE) {
 229                 irq->status |= XICS_STATUS_ASSERTED;
 230             } else {
 231                 /*
 232                  * A pending edge-triggered interrupt (or MSI)
 233                  * must have been rejected previously when we
 234                  * first detected it and tried to deliver it,
 235                  * so mark it as pending and previously rejected
 236                  * for consistency with how xics.c works.
 237                  */
 238                 irq->status |= XICS_STATUS_MASKED_PENDING
 239                     | XICS_STATUS_REJECTED;
 240             }
 241         }
 242         if (state & KVM_XICS_PRESENTED) {
 243                 irq->status |= XICS_STATUS_PRESENTED;
 244         }
 245         if (state & KVM_XICS_QUEUED) {
 246                 irq->status |= XICS_STATUS_QUEUED;
 247         }
 248     }
 249 }
 250
 251 static void ics_synchronize_state(ICSState *ics)
 252 {
 253     ics_get_kvm_state(ics);
 254 }
 255
 256 static int ics_set_kvm_state(ICSState *ics, int version_id)
 257 {
 258     uint64_t state;
 259     int i;
 260     Error *local_err = NULL;
 261
 262     for (i = 0; i < ics->nr_irqs; i++) {
 263         ICSIRQState *irq = &ics->irqs[i];
 264         int ret;
 265
 266         state = irq->server;
 267         state |= (uint64_t)(irq->saved_priority & KVM_XICS_PRIORITY_MASK)
 268             << KVM_XICS_PRIORITY_SHIFT;
 269         if (irq->priority != irq->saved_priority) {
 270             assert(irq->priority == 0xff);
 271             state |= KVM_XICS_MASKED;
 272         }
 273
 274         if (ics->irqs[i].flags & XICS_FLAGS_IRQ_LSI) {
 275             state |= KVM_XICS_LEVEL_SENSITIVE;
 276             if (irq->status & XICS_STATUS_ASSERTED) {
 277                 state |= KVM_XICS_PENDING;
 278             }
 279         } else {
 280             if (irq->status & XICS_STATUS_MASKED_PENDING) {
 281                 state |= KVM_XICS_PENDING;
 282             }
 283         }
 284         if (irq->status & XICS_STATUS_PRESENTED) {
 285                 state |= KVM_XICS_PRESENTED;
 286         }
 287         if (irq->status & XICS_STATUS_QUEUED) {
 288                 state |= KVM_XICS_QUEUED;
 289         }
 290
 291         ret = kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
 292                                 i + ics->offset, &state, true, &local_err);
 293         if (local_err) {
 294             error_report_err(local_err);
 295             return ret;
 296         }
 297     }
 298
 299     return 0;
 300 }
 301
 302 void ics_kvm_set_irq(void *opaque, int srcno, int val)
 303 {
 304     ICSState *ics = opaque;
 305     struct kvm_irq_level args;
 306     int rc;
 307
 308     args.irq = srcno + ics->offset;
 309     if (ics->irqs[srcno].flags & XICS_FLAGS_IRQ_MSI) {
 310         if (!val) {
 311             return;
 312         }
 313         args.level = KVM_INTERRUPT_SET;
 314     } else {
 315         args.level = val ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
 316     }
 317     rc = kvm_vm_ioctl(kvm_state, KVM_IRQ_LINE, &args);
 318     if (rc < 0) {
 319         perror("kvm_irq_line");
 320     }
 321 }
 322
 323 static void ics_kvm_reset(DeviceState *dev)
 324 {
 325     ICSStateClass *icsc = ICS_BASE_GET_CLASS(dev);
 326
 327     icsc->parent_reset(dev);
 328
 329     ics_set_kvm_state(ICS_KVM(dev), 1);
 330 }
 331
 332 static void ics_kvm_reset_handler(void *dev)
 333 {
 334     ics_kvm_reset(dev);
 335 }
 336
 337 static void ics_kvm_realize(DeviceState *dev, Error **errp)
 338 {
 339     ICSState *ics = ICS_KVM(dev);
 340     ICSStateClass *icsc = ICS_BASE_GET_CLASS(ics);
 341     Error *local_err = NULL;
 342
 343     icsc->parent_realize(dev, &local_err);
 344     if (local_err) {
 345         error_propagate(errp, local_err);
 346         return;
 347     }
 348
 349     qemu_register_reset(ics_kvm_reset_handler, ics);
 350 }
 351
 352 static void ics_kvm_class_init(ObjectClass *klass, void *data)
 353 {
 354     ICSStateClass *icsc = ICS_BASE_CLASS(klass);
 355     DeviceClass *dc = DEVICE_CLASS(klass);
 356
 357     device_class_set_parent_realize(dc, ics_kvm_realize,
 358                                     &icsc->parent_realize);
 359     device_class_set_parent_reset(dc, ics_kvm_reset,
 360                                   &icsc->parent_reset);
 361
 362     icsc->pre_save = ics_get_kvm_state;
 363     icsc->post_load = ics_set_kvm_state;
 364     icsc->synchronize_state = ics_synchronize_state;
 365 }
 366
 367 static const TypeInfo ics_kvm_info = {
 368     .name = TYPE_ICS_KVM,
 369     .parent = TYPE_ICS_BASE,
 370     .instance_size = sizeof(ICSState),
 371     .class_init = ics_kvm_class_init,
 372 };
 373
 374 /*
 375  * XICS-KVM
 376  */
 377
 378 static void rtas_dummy(PowerPCCPU *cpu, sPAPRMachineState *spapr,
 379                        uint32_t token,
 380                        uint32_t nargs, target_ulong args,
 381                        uint32_t nret, target_ulong rets)
 382 {
 383     error_report("pseries: %s must never be called for in-kernel XICS",
 384                  __func__);
 385 }
 386
 387 int xics_kvm_init(sPAPRMachineState *spapr, Error **errp)
 388 {
 389     int rc;
 390
 391     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
 392         error_setg(errp,
 393                    "KVM and IRQ_XICS capability must be present for in-kernel XICS");
 394         goto fail;
 395     }
 396
 397     spapr_rtas_register(RTAS_IBM_SET_XIVE, "ibm,set-xive", rtas_dummy);
 398     spapr_rtas_register(RTAS_IBM_GET_XIVE, "ibm,get-xive", rtas_dummy);
 399     spapr_rtas_register(RTAS_IBM_INT_OFF, "ibm,int-off", rtas_dummy);
 400     spapr_rtas_register(RTAS_IBM_INT_ON, "ibm,int-on", rtas_dummy);
 401
 402     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_SET_XIVE, "ibm,set-xive");
 403     if (rc < 0) {
 404         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,set-xive");
 405         goto fail;
 406     }
 407
 408     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_GET_XIVE, "ibm,get-xive");
 409     if (rc < 0) {
 410         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,get-xive");
 411         goto fail;
 412     }
 413
 414     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_ON, "ibm,int-on");
 415     if (rc < 0) {
 416         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-on");
 417         goto fail;
 418     }
 419
 420     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_OFF, "ibm,int-off");
 421     if (rc < 0) {
 422         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-off");
 423         goto fail;
 424     }
 425
 426     /* Create the KVM XICS device */
 427     rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false);
 428     if (rc < 0) {
 429         error_setg_errno(errp, -rc, "Error on KVM_CREATE_DEVICE for XICS");
 430         goto fail;
 431     }
 432
 433     kernel_xics_fd = rc;
 434     kvm_kernel_irqchip = true;
 435     kvm_msi_via_irqfd_allowed = true;
 436     kvm_gsi_direct_mapping = true;
 437
 438     return 0;
 439
 440 fail:
 441     kvmppc_define_rtas_kernel_token(0, "ibm,set-xive");
 442     kvmppc_define_rtas_kernel_token(0, "ibm,get-xive");
 443     kvmppc_define_rtas_kernel_token(0, "ibm,int-on");
 444     kvmppc_define_rtas_kernel_token(0, "ibm,int-off");
 445     return -1;
 446 }
 447
 448 static void xics_kvm_register_types(void)
 449 {
 450     type_register_static(&ics_kvm_info);
 451     type_register_static(&icp_kvm_info);
 452 }
 453
 454 type_init(xics_kvm_register_types)