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(ICPState *icp)
 118 {
 119     icp_set_kvm_state(icp, 1);
 120 }
 121
 122 static void icp_kvm_realize(ICPState *icp, Error **errp)
 123 {
 124     CPUState *cs = icp->cs;
 125     KVMEnabledICP *enabled_icp;
 126     unsigned long vcpu_id = kvm_arch_vcpu_id(cs);
 127     int ret;
 128
 129     if (kernel_xics_fd == -1) {
 130         abort();
 131     }
 132
 133     /*
 134      * If we are reusing a parked vCPU fd corresponding to the CPU
 135      * which was hot-removed earlier we don't have to renable
 136      * KVM_CAP_IRQ_XICS capability again.
 137      */
 138     QLIST_FOREACH(enabled_icp, &kvm_enabled_icps, node) {
 139         if (enabled_icp->vcpu_id == vcpu_id) {
 140             return;
 141         }
 142     }
 143
 144     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_IRQ_XICS, 0, kernel_xics_fd, vcpu_id);
 145     if (ret < 0) {
 146         error_setg(errp, "Unable to connect CPU%ld to kernel XICS: %s", vcpu_id,
 147                    strerror(errno));
 148         return;
 149     }
 150     enabled_icp = g_malloc(sizeof(*enabled_icp));
 151     enabled_icp->vcpu_id = vcpu_id;
 152     QLIST_INSERT_HEAD(&kvm_enabled_icps, enabled_icp, node);
 153 }
 154
 155 static void icp_kvm_class_init(ObjectClass *klass, void *data)
 156 {
 157     ICPStateClass *icpc = ICP_CLASS(klass);
 158
 159     icpc->pre_save = icp_get_kvm_state;
 160     icpc->post_load = icp_set_kvm_state;
 161     icpc->realize = icp_kvm_realize;
 162     icpc->reset = icp_kvm_reset;
 163     icpc->synchronize_state = icp_synchronize_state;
 164 }
 165
 166 static const TypeInfo icp_kvm_info = {
 167     .name = TYPE_KVM_ICP,
 168     .parent = TYPE_ICP,
 169     .instance_size = sizeof(ICPState),
 170     .class_init = icp_kvm_class_init,
 171     .class_size = sizeof(ICPStateClass),
 172 };
 173
 174 /*
 175  * ICS-KVM
 176  */
 177 static void ics_get_kvm_state(ICSState *ics)
 178 {
 179     uint64_t state;
 180     int i;
 181     Error *local_err = NULL;
 182
 183     for (i = 0; i < ics->nr_irqs; i++) {
 184         ICSIRQState *irq = &ics->irqs[i];
 185
 186         kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
 187                           i + ics->offset, &state, false, &local_err);
 188         if (local_err) {
 189             error_report_err(local_err);
 190             exit(1);
 191         }
 192
 193         irq->server = state & KVM_XICS_DESTINATION_MASK;
 194         irq->saved_priority = (state >> KVM_XICS_PRIORITY_SHIFT)
 195             & KVM_XICS_PRIORITY_MASK;
 196         /*
 197          * To be consistent with the software emulation in xics.c, we
 198          * split out the masked state + priority that we get from the
 199          * kernel into 'current priority' (0xff if masked) and
 200          * 'saved priority' (if masked, this is the priority the
 201          * interrupt had before it was masked).  Masking and unmasking
 202          * are done with the ibm,int-off and ibm,int-on RTAS calls.
 203          */
 204         if (state & KVM_XICS_MASKED) {
 205             irq->priority = 0xff;
 206         } else {
 207             irq->priority = irq->saved_priority;
 208         }
 209
 210         irq->status = 0;
 211         if (state & KVM_XICS_PENDING) {
 212             if (state & KVM_XICS_LEVEL_SENSITIVE) {
 213                 irq->status |= XICS_STATUS_ASSERTED;
 214             } else {
 215                 /*
 216                  * A pending edge-triggered interrupt (or MSI)
 217                  * must have been rejected previously when we
 218                  * first detected it and tried to deliver it,
 219                  * so mark it as pending and previously rejected
 220                  * for consistency with how xics.c works.
 221                  */
 222                 irq->status |= XICS_STATUS_MASKED_PENDING
 223                     | XICS_STATUS_REJECTED;
 224             }
 225         }
 226         if (state & KVM_XICS_PRESENTED) {
 227                 irq->status |= XICS_STATUS_PRESENTED;
 228         }
 229         if (state & KVM_XICS_QUEUED) {
 230                 irq->status |= XICS_STATUS_QUEUED;
 231         }
 232     }
 233 }
 234
 235 static void ics_synchronize_state(ICSState *ics)
 236 {
 237     ics_get_kvm_state(ics);
 238 }
 239
 240 static int ics_set_kvm_state(ICSState *ics, int version_id)
 241 {
 242     uint64_t state;
 243     int i;
 244     Error *local_err = NULL;
 245
 246     for (i = 0; i < ics->nr_irqs; i++) {
 247         ICSIRQState *irq = &ics->irqs[i];
 248         int ret;
 249
 250         state = irq->server;
 251         state |= (uint64_t)(irq->saved_priority & KVM_XICS_PRIORITY_MASK)
 252             << KVM_XICS_PRIORITY_SHIFT;
 253         if (irq->priority != irq->saved_priority) {
 254             assert(irq->priority == 0xff);
 255             state |= KVM_XICS_MASKED;
 256         }
 257
 258         if (ics->irqs[i].flags & XICS_FLAGS_IRQ_LSI) {
 259             state |= KVM_XICS_LEVEL_SENSITIVE;
 260             if (irq->status & XICS_STATUS_ASSERTED) {
 261                 state |= KVM_XICS_PENDING;
 262             }
 263         } else {
 264             if (irq->status & XICS_STATUS_MASKED_PENDING) {
 265                 state |= KVM_XICS_PENDING;
 266             }
 267         }
 268         if (irq->status & XICS_STATUS_PRESENTED) {
 269                 state |= KVM_XICS_PRESENTED;
 270         }
 271         if (irq->status & XICS_STATUS_QUEUED) {
 272                 state |= KVM_XICS_QUEUED;
 273         }
 274
 275         ret = kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
 276                                 i + ics->offset, &state, true, &local_err);
 277         if (local_err) {
 278             error_report_err(local_err);
 279             return ret;
 280         }
 281     }
 282
 283     return 0;
 284 }
 285
 286 static void ics_kvm_set_irq(void *opaque, int srcno, int val)
 287 {
 288     ICSState *ics = opaque;
 289     struct kvm_irq_level args;
 290     int rc;
 291
 292     args.irq = srcno + ics->offset;
 293     if (ics->irqs[srcno].flags & XICS_FLAGS_IRQ_MSI) {
 294         if (!val) {
 295             return;
 296         }
 297         args.level = KVM_INTERRUPT_SET;
 298     } else {
 299         args.level = val ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
 300     }
 301     rc = kvm_vm_ioctl(kvm_state, KVM_IRQ_LINE, &args);
 302     if (rc < 0) {
 303         perror("kvm_irq_line");
 304     }
 305 }
 306
 307 static void ics_kvm_reset(void *dev)
 308 {
 309     ICSState *ics = ICS_SIMPLE(dev);
 310     int i;
 311     uint8_t flags[ics->nr_irqs];
 312
 313     for (i = 0; i < ics->nr_irqs; i++) {
 314         flags[i] = ics->irqs[i].flags;
 315     }
 316
 317     memset(ics->irqs, 0, sizeof(ICSIRQState) * ics->nr_irqs);
 318
 319     for (i = 0; i < ics->nr_irqs; i++) {
 320         ics->irqs[i].priority = 0xff;
 321         ics->irqs[i].saved_priority = 0xff;
 322         ics->irqs[i].flags = flags[i];
 323     }
 324
 325     ics_set_kvm_state(ics, 1);
 326 }
 327
 328 static void ics_kvm_realize(ICSState *ics, Error **errp)
 329 {
 330     if (!ics->nr_irqs) {
 331         error_setg(errp, "Number of interrupts needs to be greater 0");
 332         return;
 333     }
 334     ics->irqs = g_malloc0(ics->nr_irqs * sizeof(ICSIRQState));
 335     ics->qirqs = qemu_allocate_irqs(ics_kvm_set_irq, ics, ics->nr_irqs);
 336
 337     qemu_register_reset(ics_kvm_reset, ics);
 338 }
 339
 340 static void ics_kvm_class_init(ObjectClass *klass, void *data)
 341 {
 342     ICSStateClass *icsc = ICS_BASE_CLASS(klass);
 343
 344     icsc->realize = ics_kvm_realize;
 345     icsc->pre_save = ics_get_kvm_state;
 346     icsc->post_load = ics_set_kvm_state;
 347     icsc->synchronize_state = ics_synchronize_state;
 348 }
 349
 350 static const TypeInfo ics_kvm_info = {
 351     .name = TYPE_ICS_KVM,
 352     .parent = TYPE_ICS_SIMPLE,
 353     .instance_size = sizeof(ICSState),
 354     .class_init = ics_kvm_class_init,
 355 };
 356
 357 /*
 358  * XICS-KVM
 359  */
 360
 361 static void rtas_dummy(PowerPCCPU *cpu, sPAPRMachineState *spapr,
 362                        uint32_t token,
 363                        uint32_t nargs, target_ulong args,
 364                        uint32_t nret, target_ulong rets)
 365 {
 366     error_report("pseries: %s must never be called for in-kernel XICS",
 367                  __func__);
 368 }
 369
 370 int xics_kvm_init(sPAPRMachineState *spapr, Error **errp)
 371 {
 372     int rc;
 373
 374     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
 375         error_setg(errp,
 376                    "KVM and IRQ_XICS capability must be present for in-kernel XICS");
 377         goto fail;
 378     }
 379
 380     spapr_rtas_register(RTAS_IBM_SET_XIVE, "ibm,set-xive", rtas_dummy);
 381     spapr_rtas_register(RTAS_IBM_GET_XIVE, "ibm,get-xive", rtas_dummy);
 382     spapr_rtas_register(RTAS_IBM_INT_OFF, "ibm,int-off", rtas_dummy);
 383     spapr_rtas_register(RTAS_IBM_INT_ON, "ibm,int-on", rtas_dummy);
 384
 385     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_SET_XIVE, "ibm,set-xive");
 386     if (rc < 0) {
 387         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,set-xive");
 388         goto fail;
 389     }
 390
 391     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_GET_XIVE, "ibm,get-xive");
 392     if (rc < 0) {
 393         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,get-xive");
 394         goto fail;
 395     }
 396
 397     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_ON, "ibm,int-on");
 398     if (rc < 0) {
 399         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-on");
 400         goto fail;
 401     }
 402
 403     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_OFF, "ibm,int-off");
 404     if (rc < 0) {
 405         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-off");
 406         goto fail;
 407     }
 408
 409     /* Create the KVM XICS device */
 410     rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false);
 411     if (rc < 0) {
 412         error_setg_errno(errp, -rc, "Error on KVM_CREATE_DEVICE for XICS");
 413         goto fail;
 414     }
 415
 416     kernel_xics_fd = rc;
 417     kvm_kernel_irqchip = true;
 418     kvm_msi_via_irqfd_allowed = true;
 419     kvm_gsi_direct_mapping = true;
 420
 421     return 0;
 422
 423 fail:
 424     kvmppc_define_rtas_kernel_token(0, "ibm,set-xive");
 425     kvmppc_define_rtas_kernel_token(0, "ibm,get-xive");
 426     kvmppc_define_rtas_kernel_token(0, "ibm,int-on");
 427     kvmppc_define_rtas_kernel_token(0, "ibm,int-off");
 428     return -1;
 429 }
 430
 431 static void xics_kvm_register_types(void)
 432 {
 433     type_register_static(&ics_kvm_info);
 434     type_register_static(&icp_kvm_info);
 435 }
 436
 437 type_init(xics_kvm_register_types)