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     struct kvm_one_reg reg = {
  60         .id = KVM_REG_PPC_ICP_STATE,
  61         .addr = (uintptr_t)&state,
  62     };
  63     int ret;
  64
  65     /* ICP for this CPU thread is not in use, exiting */
  66     if (!icp->cs) {
  67         return;
  68     }
  69
  70     ret = kvm_vcpu_ioctl(icp->cs, KVM_GET_ONE_REG, &reg);
  71     if (ret != 0) {
  72         error_report("Unable to retrieve KVM interrupt controller state"
  73                 " for CPU %ld: %s", kvm_arch_vcpu_id(icp->cs), strerror(errno));
  74         exit(1);
  75     }
  76
  77     icp->xirr = state >> KVM_REG_PPC_ICP_XISR_SHIFT;
  78     icp->mfrr = (state >> KVM_REG_PPC_ICP_MFRR_SHIFT)
  79         & KVM_REG_PPC_ICP_MFRR_MASK;
  80     icp->pending_priority = (state >> KVM_REG_PPC_ICP_PPRI_SHIFT)
  81         & KVM_REG_PPC_ICP_PPRI_MASK;
  82 }
  83
  84 static int icp_set_kvm_state(ICPState *icp, int version_id)
  85 {
  86     uint64_t state;
  87     struct kvm_one_reg reg = {
  88         .id = KVM_REG_PPC_ICP_STATE,
  89         .addr = (uintptr_t)&state,
  90     };
  91     int ret;
  92
  93     /* ICP for this CPU thread is not in use, exiting */
  94     if (!icp->cs) {
  95         return 0;
  96     }
  97
  98     state = ((uint64_t)icp->xirr << KVM_REG_PPC_ICP_XISR_SHIFT)
  99         | ((uint64_t)icp->mfrr << KVM_REG_PPC_ICP_MFRR_SHIFT)
 100         | ((uint64_t)icp->pending_priority << KVM_REG_PPC_ICP_PPRI_SHIFT);
 101
 102     ret = kvm_vcpu_ioctl(icp->cs, KVM_SET_ONE_REG, &reg);
 103     if (ret != 0) {
 104         error_report("Unable to restore KVM interrupt controller state (0x%"
 105                 PRIx64 ") for CPU %ld: %s", state, kvm_arch_vcpu_id(icp->cs),
 106                 strerror(errno));
 107         return ret;
 108     }
 109
 110     return 0;
 111 }
 112
 113 static void icp_kvm_reset(ICPState *icp)
 114 {
 115     icp_set_kvm_state(icp, 1);
 116 }
 117
 118 static void icp_kvm_realize(ICPState *icp, Error **errp)
 119 {
 120     CPUState *cs = icp->cs;
 121     KVMEnabledICP *enabled_icp;
 122     unsigned long vcpu_id = kvm_arch_vcpu_id(cs);
 123     int ret;
 124
 125     if (kernel_xics_fd == -1) {
 126         abort();
 127     }
 128
 129     /*
 130      * If we are reusing a parked vCPU fd corresponding to the CPU
 131      * which was hot-removed earlier we don't have to renable
 132      * KVM_CAP_IRQ_XICS capability again.
 133      */
 134     QLIST_FOREACH(enabled_icp, &kvm_enabled_icps, node) {
 135         if (enabled_icp->vcpu_id == vcpu_id) {
 136             return;
 137         }
 138     }
 139
 140     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_IRQ_XICS, 0, kernel_xics_fd, vcpu_id);
 141     if (ret < 0) {
 142         error_setg(errp, "Unable to connect CPU%ld to kernel XICS: %s", vcpu_id,
 143                    strerror(errno));
 144         return;
 145     }
 146     enabled_icp = g_malloc(sizeof(*enabled_icp));
 147     enabled_icp->vcpu_id = vcpu_id;
 148     QLIST_INSERT_HEAD(&kvm_enabled_icps, enabled_icp, node);
 149 }
 150
 151 static void icp_kvm_class_init(ObjectClass *klass, void *data)
 152 {
 153     ICPStateClass *icpc = ICP_CLASS(klass);
 154
 155     icpc->pre_save = icp_get_kvm_state;
 156     icpc->post_load = icp_set_kvm_state;
 157     icpc->realize = icp_kvm_realize;
 158     icpc->reset = icp_kvm_reset;
 159 }
 160
 161 static const TypeInfo icp_kvm_info = {
 162     .name = TYPE_KVM_ICP,
 163     .parent = TYPE_ICP,
 164     .instance_size = sizeof(ICPState),
 165     .class_init = icp_kvm_class_init,
 166     .class_size = sizeof(ICPStateClass),
 167 };
 168
 169 /*
 170  * ICS-KVM
 171  */
 172 static void ics_get_kvm_state(ICSState *ics)
 173 {
 174     uint64_t state;
 175     struct kvm_device_attr attr = {
 176         .flags = 0,
 177         .group = KVM_DEV_XICS_GRP_SOURCES,
 178         .addr = (uint64_t)(uintptr_t)&state,
 179     };
 180     int i;
 181
 182     for (i = 0; i < ics->nr_irqs; i++) {
 183         ICSIRQState *irq = &ics->irqs[i];
 184         int ret;
 185
 186         attr.attr = i + ics->offset;
 187
 188         ret = ioctl(kernel_xics_fd, KVM_GET_DEVICE_ATTR, &attr);
 189         if (ret != 0) {
 190             error_report("Unable to retrieve KVM interrupt controller state"
 191                     " for IRQ %d: %s", i + ics->offset, strerror(errno));
 192             exit(1);
 193         }
 194
 195         irq->server = state & KVM_XICS_DESTINATION_MASK;
 196         irq->saved_priority = (state >> KVM_XICS_PRIORITY_SHIFT)
 197             & KVM_XICS_PRIORITY_MASK;
 198         /*
 199          * To be consistent with the software emulation in xics.c, we
 200          * split out the masked state + priority that we get from the
 201          * kernel into 'current priority' (0xff if masked) and
 202          * 'saved priority' (if masked, this is the priority the
 203          * interrupt had before it was masked).  Masking and unmasking
 204          * are done with the ibm,int-off and ibm,int-on RTAS calls.
 205          */
 206         if (state & KVM_XICS_MASKED) {
 207             irq->priority = 0xff;
 208         } else {
 209             irq->priority = irq->saved_priority;
 210         }
 211
 212         irq->status = 0;
 213         if (state & KVM_XICS_PENDING) {
 214             if (state & KVM_XICS_LEVEL_SENSITIVE) {
 215                 irq->status |= XICS_STATUS_ASSERTED;
 216             } else {
 217                 /*
 218                  * A pending edge-triggered interrupt (or MSI)
 219                  * must have been rejected previously when we
 220                  * first detected it and tried to deliver it,
 221                  * so mark it as pending and previously rejected
 222                  * for consistency with how xics.c works.
 223                  */
 224                 irq->status |= XICS_STATUS_MASKED_PENDING
 225                     | XICS_STATUS_REJECTED;
 226             }
 227         }
 228         if (state & KVM_XICS_PRESENTED) {
 229                 irq->status |= XICS_STATUS_PRESENTED;
 230         }
 231         if (state & KVM_XICS_QUEUED) {
 232                 irq->status |= XICS_STATUS_QUEUED;
 233         }
 234     }
 235 }
 236
 237 static int ics_set_kvm_state(ICSState *ics, int version_id)
 238 {
 239     uint64_t state;
 240     struct kvm_device_attr attr = {
 241         .flags = 0,
 242         .group = KVM_DEV_XICS_GRP_SOURCES,
 243         .addr = (uint64_t)(uintptr_t)&state,
 244     };
 245     int i;
 246
 247     for (i = 0; i < ics->nr_irqs; i++) {
 248         ICSIRQState *irq = &ics->irqs[i];
 249         int ret;
 250
 251         attr.attr = i + ics->offset;
 252
 253         state = irq->server;
 254         state |= (uint64_t)(irq->saved_priority & KVM_XICS_PRIORITY_MASK)
 255             << KVM_XICS_PRIORITY_SHIFT;
 256         if (irq->priority != irq->saved_priority) {
 257             assert(irq->priority == 0xff);
 258             state |= KVM_XICS_MASKED;
 259         }
 260
 261         if (ics->irqs[i].flags & XICS_FLAGS_IRQ_LSI) {
 262             state |= KVM_XICS_LEVEL_SENSITIVE;
 263             if (irq->status & XICS_STATUS_ASSERTED) {
 264                 state |= KVM_XICS_PENDING;
 265             }
 266         } else {
 267             if (irq->status & XICS_STATUS_MASKED_PENDING) {
 268                 state |= KVM_XICS_PENDING;
 269             }
 270         }
 271         if (irq->status & XICS_STATUS_PRESENTED) {
 272                 state |= KVM_XICS_PRESENTED;
 273         }
 274         if (irq->status & XICS_STATUS_QUEUED) {
 275                 state |= KVM_XICS_QUEUED;
 276         }
 277
 278         ret = ioctl(kernel_xics_fd, KVM_SET_DEVICE_ATTR, &attr);
 279         if (ret != 0) {
 280             error_report("Unable to restore KVM interrupt controller state"
 281                     " for IRQs %d: %s", i + ics->offset, strerror(errno));
 282             return ret;
 283         }
 284     }
 285
 286     return 0;
 287 }
 288
 289 static void ics_kvm_set_irq(void *opaque, int srcno, int val)
 290 {
 291     ICSState *ics = opaque;
 292     struct kvm_irq_level args;
 293     int rc;
 294
 295     args.irq = srcno + ics->offset;
 296     if (ics->irqs[srcno].flags & XICS_FLAGS_IRQ_MSI) {
 297         if (!val) {
 298             return;
 299         }
 300         args.level = KVM_INTERRUPT_SET;
 301     } else {
 302         args.level = val ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
 303     }
 304     rc = kvm_vm_ioctl(kvm_state, KVM_IRQ_LINE, &args);
 305     if (rc < 0) {
 306         perror("kvm_irq_line");
 307     }
 308 }
 309
 310 static void ics_kvm_reset(void *dev)
 311 {
 312     ICSState *ics = ICS_SIMPLE(dev);
 313     int i;
 314     uint8_t flags[ics->nr_irqs];
 315
 316     for (i = 0; i < ics->nr_irqs; i++) {
 317         flags[i] = ics->irqs[i].flags;
 318     }
 319
 320     memset(ics->irqs, 0, sizeof(ICSIRQState) * ics->nr_irqs);
 321
 322     for (i = 0; i < ics->nr_irqs; i++) {
 323         ics->irqs[i].priority = 0xff;
 324         ics->irqs[i].saved_priority = 0xff;
 325         ics->irqs[i].flags = flags[i];
 326     }
 327
 328     ics_set_kvm_state(ics, 1);
 329 }
 330
 331 static void ics_kvm_realize(ICSState *ics, Error **errp)
 332 {
 333     if (!ics->nr_irqs) {
 334         error_setg(errp, "Number of interrupts needs to be greater 0");
 335         return;
 336     }
 337     ics->irqs = g_malloc0(ics->nr_irqs * sizeof(ICSIRQState));
 338     ics->qirqs = qemu_allocate_irqs(ics_kvm_set_irq, ics, ics->nr_irqs);
 339
 340     qemu_register_reset(ics_kvm_reset, ics);
 341 }
 342
 343 static void ics_kvm_class_init(ObjectClass *klass, void *data)
 344 {
 345     ICSStateClass *icsc = ICS_BASE_CLASS(klass);
 346
 347     icsc->realize = ics_kvm_realize;
 348     icsc->pre_save = ics_get_kvm_state;
 349     icsc->post_load = ics_set_kvm_state;
 350 }
 351
 352 static const TypeInfo ics_kvm_info = {
 353     .name = TYPE_ICS_KVM,
 354     .parent = TYPE_ICS_SIMPLE,
 355     .instance_size = sizeof(ICSState),
 356     .class_init = ics_kvm_class_init,
 357 };
 358
 359 /*
 360  * XICS-KVM
 361  */
 362
 363 static void rtas_dummy(PowerPCCPU *cpu, sPAPRMachineState *spapr,
 364                        uint32_t token,
 365                        uint32_t nargs, target_ulong args,
 366                        uint32_t nret, target_ulong rets)
 367 {
 368     error_report("pseries: %s must never be called for in-kernel XICS",
 369                  __func__);
 370 }
 371
 372 int xics_kvm_init(sPAPRMachineState *spapr, Error **errp)
 373 {
 374     int rc;
 375     struct kvm_create_device xics_create_device = {
 376         .type = KVM_DEV_TYPE_XICS,
 377         .flags = 0,
 378     };
 379
 380     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
 381         error_setg(errp,
 382                    "KVM and IRQ_XICS capability must be present for in-kernel XICS");
 383         goto fail;
 384     }
 385
 386     spapr_rtas_register(RTAS_IBM_SET_XIVE, "ibm,set-xive", rtas_dummy);
 387     spapr_rtas_register(RTAS_IBM_GET_XIVE, "ibm,get-xive", rtas_dummy);
 388     spapr_rtas_register(RTAS_IBM_INT_OFF, "ibm,int-off", rtas_dummy);
 389     spapr_rtas_register(RTAS_IBM_INT_ON, "ibm,int-on", rtas_dummy);
 390
 391     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_SET_XIVE, "ibm,set-xive");
 392     if (rc < 0) {
 393         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,set-xive");
 394         goto fail;
 395     }
 396
 397     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_GET_XIVE, "ibm,get-xive");
 398     if (rc < 0) {
 399         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,get-xive");
 400         goto fail;
 401     }
 402
 403     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_ON, "ibm,int-on");
 404     if (rc < 0) {
 405         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-on");
 406         goto fail;
 407     }
 408
 409     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_OFF, "ibm,int-off");
 410     if (rc < 0) {
 411         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-off");
 412         goto fail;
 413     }
 414
 415     /* Create the kernel ICP */
 416     rc = kvm_vm_ioctl(kvm_state, KVM_CREATE_DEVICE, &xics_create_device);
 417     if (rc < 0) {
 418         error_setg_errno(errp, -rc, "Error on KVM_CREATE_DEVICE for XICS");
 419         goto fail;
 420     }
 421
 422     kernel_xics_fd = xics_create_device.fd;
 423
 424     kvm_kernel_irqchip = true;
 425     kvm_msi_via_irqfd_allowed = true;
 426     kvm_gsi_direct_mapping = true;
 427
 428     return rc;
 429
 430 fail:
 431     kvmppc_define_rtas_kernel_token(0, "ibm,set-xive");
 432     kvmppc_define_rtas_kernel_token(0, "ibm,get-xive");
 433     kvmppc_define_rtas_kernel_token(0, "ibm,int-on");
 434     kvmppc_define_rtas_kernel_token(0, "ibm,int-off");
 435     return -1;
 436 }
 437
 438 static void xics_kvm_register_types(void)
 439 {
 440     type_register_static(&ics_kvm_info);
 441     type_register_static(&icp_kvm_info);
 442 }
 443
 444 type_init(xics_kvm_register_types)