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 void do_icp_synchronize_state(CPUState *cpu, run_on_cpu_data arg)
  85 {
  86     icp_get_kvm_state(arg.host_ptr);
  87 }
  88
  89 static void icp_synchronize_state(ICPState *icp)
  90 {
  91     if (icp->cs) {
  92         run_on_cpu(icp->cs, do_icp_synchronize_state, RUN_ON_CPU_HOST_PTR(icp));
  93     }
  94 }
  95
  96 static int icp_set_kvm_state(ICPState *icp, int version_id)
  97 {
  98     uint64_t state;
  99     struct kvm_one_reg reg = {
 100         .id = KVM_REG_PPC_ICP_STATE,
 101         .addr = (uintptr_t)&state,
 102     };
 103     int ret;
 104
 105     /* ICP for this CPU thread is not in use, exiting */
 106     if (!icp->cs) {
 107         return 0;
 108     }
 109
 110     state = ((uint64_t)icp->xirr << KVM_REG_PPC_ICP_XISR_SHIFT)
 111         | ((uint64_t)icp->mfrr << KVM_REG_PPC_ICP_MFRR_SHIFT)
 112         | ((uint64_t)icp->pending_priority << KVM_REG_PPC_ICP_PPRI_SHIFT);
 113
 114     ret = kvm_vcpu_ioctl(icp->cs, KVM_SET_ONE_REG, &reg);
 115     if (ret != 0) {
 116         error_report("Unable to restore KVM interrupt controller state (0x%"
 117                 PRIx64 ") for CPU %ld: %s", state, kvm_arch_vcpu_id(icp->cs),
 118                 strerror(errno));
 119         return ret;
 120     }
 121
 122     return 0;
 123 }
 124
 125 static void icp_kvm_reset(ICPState *icp)
 126 {
 127     icp_set_kvm_state(icp, 1);
 128 }
 129
 130 static void icp_kvm_realize(ICPState *icp, Error **errp)
 131 {
 132     CPUState *cs = icp->cs;
 133     KVMEnabledICP *enabled_icp;
 134     unsigned long vcpu_id = kvm_arch_vcpu_id(cs);
 135     int ret;
 136
 137     if (kernel_xics_fd == -1) {
 138         abort();
 139     }
 140
 141     /*
 142      * If we are reusing a parked vCPU fd corresponding to the CPU
 143      * which was hot-removed earlier we don't have to renable
 144      * KVM_CAP_IRQ_XICS capability again.
 145      */
 146     QLIST_FOREACH(enabled_icp, &kvm_enabled_icps, node) {
 147         if (enabled_icp->vcpu_id == vcpu_id) {
 148             return;
 149         }
 150     }
 151
 152     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_IRQ_XICS, 0, kernel_xics_fd, vcpu_id);
 153     if (ret < 0) {
 154         error_setg(errp, "Unable to connect CPU%ld to kernel XICS: %s", vcpu_id,
 155                    strerror(errno));
 156         return;
 157     }
 158     enabled_icp = g_malloc(sizeof(*enabled_icp));
 159     enabled_icp->vcpu_id = vcpu_id;
 160     QLIST_INSERT_HEAD(&kvm_enabled_icps, enabled_icp, node);
 161 }
 162
 163 static void icp_kvm_class_init(ObjectClass *klass, void *data)
 164 {
 165     ICPStateClass *icpc = ICP_CLASS(klass);
 166
 167     icpc->pre_save = icp_get_kvm_state;
 168     icpc->post_load = icp_set_kvm_state;
 169     icpc->realize = icp_kvm_realize;
 170     icpc->reset = icp_kvm_reset;
 171     icpc->synchronize_state = icp_synchronize_state;
 172 }
 173
 174 static const TypeInfo icp_kvm_info = {
 175     .name = TYPE_KVM_ICP,
 176     .parent = TYPE_ICP,
 177     .instance_size = sizeof(ICPState),
 178     .class_init = icp_kvm_class_init,
 179     .class_size = sizeof(ICPStateClass),
 180 };
 181
 182 /*
 183  * ICS-KVM
 184  */
 185 static void ics_get_kvm_state(ICSState *ics)
 186 {
 187     uint64_t state;
 188     struct kvm_device_attr attr = {
 189         .flags = 0,
 190         .group = KVM_DEV_XICS_GRP_SOURCES,
 191         .addr = (uint64_t)(uintptr_t)&state,
 192     };
 193     int i;
 194
 195     for (i = 0; i < ics->nr_irqs; i++) {
 196         ICSIRQState *irq = &ics->irqs[i];
 197         int ret;
 198
 199         attr.attr = i + ics->offset;
 200
 201         ret = ioctl(kernel_xics_fd, KVM_GET_DEVICE_ATTR, &attr);
 202         if (ret != 0) {
 203             error_report("Unable to retrieve KVM interrupt controller state"
 204                     " for IRQ %d: %s", i + ics->offset, strerror(errno));
 205             exit(1);
 206         }
 207
 208         irq->server = state & KVM_XICS_DESTINATION_MASK;
 209         irq->saved_priority = (state >> KVM_XICS_PRIORITY_SHIFT)
 210             & KVM_XICS_PRIORITY_MASK;
 211         /*
 212          * To be consistent with the software emulation in xics.c, we
 213          * split out the masked state + priority that we get from the
 214          * kernel into 'current priority' (0xff if masked) and
 215          * 'saved priority' (if masked, this is the priority the
 216          * interrupt had before it was masked).  Masking and unmasking
 217          * are done with the ibm,int-off and ibm,int-on RTAS calls.
 218          */
 219         if (state & KVM_XICS_MASKED) {
 220             irq->priority = 0xff;
 221         } else {
 222             irq->priority = irq->saved_priority;
 223         }
 224
 225         irq->status = 0;
 226         if (state & KVM_XICS_PENDING) {
 227             if (state & KVM_XICS_LEVEL_SENSITIVE) {
 228                 irq->status |= XICS_STATUS_ASSERTED;
 229             } else {
 230                 /*
 231                  * A pending edge-triggered interrupt (or MSI)
 232                  * must have been rejected previously when we
 233                  * first detected it and tried to deliver it,
 234                  * so mark it as pending and previously rejected
 235                  * for consistency with how xics.c works.
 236                  */
 237                 irq->status |= XICS_STATUS_MASKED_PENDING
 238                     | XICS_STATUS_REJECTED;
 239             }
 240         }
 241         if (state & KVM_XICS_PRESENTED) {
 242                 irq->status |= XICS_STATUS_PRESENTED;
 243         }
 244         if (state & KVM_XICS_QUEUED) {
 245                 irq->status |= XICS_STATUS_QUEUED;
 246         }
 247     }
 248 }
 249
 250 static void ics_synchronize_state(ICSState *ics)
 251 {
 252     ics_get_kvm_state(ics);
 253 }
 254
 255 static int ics_set_kvm_state(ICSState *ics, int version_id)
 256 {
 257     uint64_t state;
 258     struct kvm_device_attr attr = {
 259         .flags = 0,
 260         .group = KVM_DEV_XICS_GRP_SOURCES,
 261         .addr = (uint64_t)(uintptr_t)&state,
 262     };
 263     int i;
 264
 265     for (i = 0; i < ics->nr_irqs; i++) {
 266         ICSIRQState *irq = &ics->irqs[i];
 267         int ret;
 268
 269         attr.attr = i + ics->offset;
 270
 271         state = irq->server;
 272         state |= (uint64_t)(irq->saved_priority & KVM_XICS_PRIORITY_MASK)
 273             << KVM_XICS_PRIORITY_SHIFT;
 274         if (irq->priority != irq->saved_priority) {
 275             assert(irq->priority == 0xff);
 276             state |= KVM_XICS_MASKED;
 277         }
 278
 279         if (ics->irqs[i].flags & XICS_FLAGS_IRQ_LSI) {
 280             state |= KVM_XICS_LEVEL_SENSITIVE;
 281             if (irq->status & XICS_STATUS_ASSERTED) {
 282                 state |= KVM_XICS_PENDING;
 283             }
 284         } else {
 285             if (irq->status & XICS_STATUS_MASKED_PENDING) {
 286                 state |= KVM_XICS_PENDING;
 287             }
 288         }
 289         if (irq->status & XICS_STATUS_PRESENTED) {
 290                 state |= KVM_XICS_PRESENTED;
 291         }
 292         if (irq->status & XICS_STATUS_QUEUED) {
 293                 state |= KVM_XICS_QUEUED;
 294         }
 295
 296         ret = ioctl(kernel_xics_fd, KVM_SET_DEVICE_ATTR, &attr);
 297         if (ret != 0) {
 298             error_report("Unable to restore KVM interrupt controller state"
 299                     " for IRQs %d: %s", i + ics->offset, strerror(errno));
 300             return ret;
 301         }
 302     }
 303
 304     return 0;
 305 }
 306
 307 static void ics_kvm_set_irq(void *opaque, int srcno, int val)
 308 {
 309     ICSState *ics = opaque;
 310     struct kvm_irq_level args;
 311     int rc;
 312
 313     args.irq = srcno + ics->offset;
 314     if (ics->irqs[srcno].flags & XICS_FLAGS_IRQ_MSI) {
 315         if (!val) {
 316             return;
 317         }
 318         args.level = KVM_INTERRUPT_SET;
 319     } else {
 320         args.level = val ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
 321     }
 322     rc = kvm_vm_ioctl(kvm_state, KVM_IRQ_LINE, &args);
 323     if (rc < 0) {
 324         perror("kvm_irq_line");
 325     }
 326 }
 327
 328 static void ics_kvm_reset(void *dev)
 329 {
 330     ICSState *ics = ICS_SIMPLE(dev);
 331     int i;
 332     uint8_t flags[ics->nr_irqs];
 333
 334     for (i = 0; i < ics->nr_irqs; i++) {
 335         flags[i] = ics->irqs[i].flags;
 336     }
 337
 338     memset(ics->irqs, 0, sizeof(ICSIRQState) * ics->nr_irqs);
 339
 340     for (i = 0; i < ics->nr_irqs; i++) {
 341         ics->irqs[i].priority = 0xff;
 342         ics->irqs[i].saved_priority = 0xff;
 343         ics->irqs[i].flags = flags[i];
 344     }
 345
 346     ics_set_kvm_state(ics, 1);
 347 }
 348
 349 static void ics_kvm_realize(ICSState *ics, Error **errp)
 350 {
 351     if (!ics->nr_irqs) {
 352         error_setg(errp, "Number of interrupts needs to be greater 0");
 353         return;
 354     }
 355     ics->irqs = g_malloc0(ics->nr_irqs * sizeof(ICSIRQState));
 356     ics->qirqs = qemu_allocate_irqs(ics_kvm_set_irq, ics, ics->nr_irqs);
 357
 358     qemu_register_reset(ics_kvm_reset, ics);
 359 }
 360
 361 static void ics_kvm_class_init(ObjectClass *klass, void *data)
 362 {
 363     ICSStateClass *icsc = ICS_BASE_CLASS(klass);
 364
 365     icsc->realize = ics_kvm_realize;
 366     icsc->pre_save = ics_get_kvm_state;
 367     icsc->post_load = ics_set_kvm_state;
 368     icsc->synchronize_state = ics_synchronize_state;
 369 }
 370
 371 static const TypeInfo ics_kvm_info = {
 372     .name = TYPE_ICS_KVM,
 373     .parent = TYPE_ICS_SIMPLE,
 374     .instance_size = sizeof(ICSState),
 375     .class_init = ics_kvm_class_init,
 376 };
 377
 378 /*
 379  * XICS-KVM
 380  */
 381
 382 static void rtas_dummy(PowerPCCPU *cpu, sPAPRMachineState *spapr,
 383                        uint32_t token,
 384                        uint32_t nargs, target_ulong args,
 385                        uint32_t nret, target_ulong rets)
 386 {
 387     error_report("pseries: %s must never be called for in-kernel XICS",
 388                  __func__);
 389 }
 390
 391 int xics_kvm_init(sPAPRMachineState *spapr, Error **errp)
 392 {
 393     int rc;
 394     struct kvm_create_device xics_create_device = {
 395         .type = KVM_DEV_TYPE_XICS,
 396         .flags = 0,
 397     };
 398
 399     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
 400         error_setg(errp,
 401                    "KVM and IRQ_XICS capability must be present for in-kernel XICS");
 402         goto fail;
 403     }
 404
 405     spapr_rtas_register(RTAS_IBM_SET_XIVE, "ibm,set-xive", rtas_dummy);
 406     spapr_rtas_register(RTAS_IBM_GET_XIVE, "ibm,get-xive", rtas_dummy);
 407     spapr_rtas_register(RTAS_IBM_INT_OFF, "ibm,int-off", rtas_dummy);
 408     spapr_rtas_register(RTAS_IBM_INT_ON, "ibm,int-on", rtas_dummy);
 409
 410     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_SET_XIVE, "ibm,set-xive");
 411     if (rc < 0) {
 412         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,set-xive");
 413         goto fail;
 414     }
 415
 416     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_GET_XIVE, "ibm,get-xive");
 417     if (rc < 0) {
 418         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,get-xive");
 419         goto fail;
 420     }
 421
 422     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_ON, "ibm,int-on");
 423     if (rc < 0) {
 424         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-on");
 425         goto fail;
 426     }
 427
 428     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_OFF, "ibm,int-off");
 429     if (rc < 0) {
 430         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-off");
 431         goto fail;
 432     }
 433
 434     /* Create the kernel ICP */
 435     rc = kvm_vm_ioctl(kvm_state, KVM_CREATE_DEVICE, &xics_create_device);
 436     if (rc < 0) {
 437         error_setg_errno(errp, -rc, "Error on KVM_CREATE_DEVICE for XICS");
 438         goto fail;
 439     }
 440
 441     kernel_xics_fd = xics_create_device.fd;
 442
 443     kvm_kernel_irqchip = true;
 444     kvm_msi_via_irqfd_allowed = true;
 445     kvm_gsi_direct_mapping = true;
 446
 447     return rc;
 448
 449 fail:
 450     kvmppc_define_rtas_kernel_token(0, "ibm,set-xive");
 451     kvmppc_define_rtas_kernel_token(0, "ibm,get-xive");
 452     kvmppc_define_rtas_kernel_token(0, "ibm,int-on");
 453     kvmppc_define_rtas_kernel_token(0, "ibm,int-off");
 454     return -1;
 455 }
 456
 457 static void xics_kvm_register_types(void)
 458 {
 459     type_register_static(&ics_kvm_info);
 460     type_register_static(&icp_kvm_info);
 461 }
 462
 463 type_init(xics_kvm_register_types)