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 /*
  46  * ICP-KVM
  47  */
  48 static void icp_get_kvm_state(ICPState *icp)
  49 {
  50     uint64_t state;
  51     struct kvm_one_reg reg = {
  52         .id = KVM_REG_PPC_ICP_STATE,
  53         .addr = (uintptr_t)&state,
  54     };
  55     int ret;
  56
  57     /* ICP for this CPU thread is not in use, exiting */
  58     if (!icp->cs) {
  59         return;
  60     }
  61
  62     ret = kvm_vcpu_ioctl(icp->cs, KVM_GET_ONE_REG, &reg);
  63     if (ret != 0) {
  64         error_report("Unable to retrieve KVM interrupt controller state"
  65                 " for CPU %ld: %s", kvm_arch_vcpu_id(icp->cs), strerror(errno));
  66         exit(1);
  67     }
  68
  69     icp->xirr = state >> KVM_REG_PPC_ICP_XISR_SHIFT;
  70     icp->mfrr = (state >> KVM_REG_PPC_ICP_MFRR_SHIFT)
  71         & KVM_REG_PPC_ICP_MFRR_MASK;
  72     icp->pending_priority = (state >> KVM_REG_PPC_ICP_PPRI_SHIFT)
  73         & KVM_REG_PPC_ICP_PPRI_MASK;
  74 }
  75
  76 static int icp_set_kvm_state(ICPState *icp, int version_id)
  77 {
  78     uint64_t state;
  79     struct kvm_one_reg reg = {
  80         .id = KVM_REG_PPC_ICP_STATE,
  81         .addr = (uintptr_t)&state,
  82     };
  83     int ret;
  84
  85     /* ICP for this CPU thread is not in use, exiting */
  86     if (!icp->cs) {
  87         return 0;
  88     }
  89
  90     state = ((uint64_t)icp->xirr << KVM_REG_PPC_ICP_XISR_SHIFT)
  91         | ((uint64_t)icp->mfrr << KVM_REG_PPC_ICP_MFRR_SHIFT)
  92         | ((uint64_t)icp->pending_priority << KVM_REG_PPC_ICP_PPRI_SHIFT);
  93
  94     ret = kvm_vcpu_ioctl(icp->cs, KVM_SET_ONE_REG, &reg);
  95     if (ret != 0) {
  96         error_report("Unable to restore KVM interrupt controller state (0x%"
  97                 PRIx64 ") for CPU %ld: %s", state, kvm_arch_vcpu_id(icp->cs),
  98                 strerror(errno));
  99         return ret;
 100     }
 101
 102     return 0;
 103 }
 104
 105 static void icp_kvm_reset(void *dev)
 106 {
 107     ICPState *icp = ICP(dev);
 108
 109     icp->xirr = 0;
 110     icp->pending_priority = 0xff;
 111     icp->mfrr = 0xff;
 112
 113     /* Make all outputs as deasserted only if the CPU thread is in use */
 114     if (icp->output) {
 115         qemu_set_irq(icp->output, 0);
 116     }
 117
 118     icp_set_kvm_state(icp, 1);
 119 }
 120
 121 static void icp_kvm_cpu_setup(ICPState *icp, PowerPCCPU *cpu)
 122 {
 123     CPUState *cs = CPU(cpu);
 124     int ret;
 125
 126     if (kernel_xics_fd == -1) {
 127         abort();
 128     }
 129
 130     /*
 131      * If we are reusing a parked vCPU fd corresponding to the CPU
 132      * which was hot-removed earlier we don't have to renable
 133      * KVM_CAP_IRQ_XICS capability again.
 134      */
 135     if (icp->cap_irq_xics_enabled) {
 136         return;
 137     }
 138
 139     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_IRQ_XICS, 0, kernel_xics_fd,
 140                               kvm_arch_vcpu_id(cs));
 141     if (ret < 0) {
 142         error_report("Unable to connect CPU%ld to kernel XICS: %s",
 143                      kvm_arch_vcpu_id(cs), strerror(errno));
 144         exit(1);
 145     }
 146     icp->cap_irq_xics_enabled = true;
 147 }
 148
 149 static void icp_kvm_realize(DeviceState *dev, Error **errp)
 150 {
 151     qemu_register_reset(icp_kvm_reset, dev);
 152 }
 153
 154 static void icp_kvm_class_init(ObjectClass *klass, void *data)
 155 {
 156     DeviceClass *dc = DEVICE_CLASS(klass);
 157     ICPStateClass *icpc = ICP_CLASS(klass);
 158
 159     dc->realize = icp_kvm_realize;
 160     icpc->pre_save = icp_get_kvm_state;
 161     icpc->post_load = icp_set_kvm_state;
 162     icpc->cpu_setup = icp_kvm_cpu_setup;
 163 }
 164
 165 static const TypeInfo icp_kvm_info = {
 166     .name = TYPE_KVM_ICP,
 167     .parent = TYPE_ICP,
 168     .instance_size = sizeof(ICPState),
 169     .class_init = icp_kvm_class_init,
 170     .class_size = sizeof(ICPStateClass),
 171 };
 172
 173 /*
 174  * ICS-KVM
 175  */
 176 static void ics_get_kvm_state(ICSState *ics)
 177 {
 178     uint64_t state;
 179     struct kvm_device_attr attr = {
 180         .flags = 0,
 181         .group = KVM_DEV_XICS_GRP_SOURCES,
 182         .addr = (uint64_t)(uintptr_t)&state,
 183     };
 184     int i;
 185
 186     for (i = 0; i < ics->nr_irqs; i++) {
 187         ICSIRQState *irq = &ics->irqs[i];
 188         int ret;
 189
 190         attr.attr = i + ics->offset;
 191
 192         ret = ioctl(kernel_xics_fd, KVM_GET_DEVICE_ATTR, &attr);
 193         if (ret != 0) {
 194             error_report("Unable to retrieve KVM interrupt controller state"
 195                     " for IRQ %d: %s", i + ics->offset, strerror(errno));
 196             exit(1);
 197         }
 198
 199         irq->server = state & KVM_XICS_DESTINATION_MASK;
 200         irq->saved_priority = (state >> KVM_XICS_PRIORITY_SHIFT)
 201             & KVM_XICS_PRIORITY_MASK;
 202         /*
 203          * To be consistent with the software emulation in xics.c, we
 204          * split out the masked state + priority that we get from the
 205          * kernel into 'current priority' (0xff if masked) and
 206          * 'saved priority' (if masked, this is the priority the
 207          * interrupt had before it was masked).  Masking and unmasking
 208          * are done with the ibm,int-off and ibm,int-on RTAS calls.
 209          */
 210         if (state & KVM_XICS_MASKED) {
 211             irq->priority = 0xff;
 212         } else {
 213             irq->priority = irq->saved_priority;
 214         }
 215
 216         if (state & KVM_XICS_PENDING) {
 217             if (state & KVM_XICS_LEVEL_SENSITIVE) {
 218                 irq->status |= XICS_STATUS_ASSERTED;
 219             } else {
 220                 /*
 221                  * A pending edge-triggered interrupt (or MSI)
 222                  * must have been rejected previously when we
 223                  * first detected it and tried to deliver it,
 224                  * so mark it as pending and previously rejected
 225                  * for consistency with how xics.c works.
 226                  */
 227                 irq->status |= XICS_STATUS_MASKED_PENDING
 228                     | XICS_STATUS_REJECTED;
 229             }
 230         }
 231     }
 232 }
 233
 234 static int ics_set_kvm_state(ICSState *ics, int version_id)
 235 {
 236     uint64_t state;
 237     struct kvm_device_attr attr = {
 238         .flags = 0,
 239         .group = KVM_DEV_XICS_GRP_SOURCES,
 240         .addr = (uint64_t)(uintptr_t)&state,
 241     };
 242     int i;
 243
 244     for (i = 0; i < ics->nr_irqs; i++) {
 245         ICSIRQState *irq = &ics->irqs[i];
 246         int ret;
 247
 248         attr.attr = i + ics->offset;
 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
 269         ret = ioctl(kernel_xics_fd, KVM_SET_DEVICE_ATTR, &attr);
 270         if (ret != 0) {
 271             error_report("Unable to restore KVM interrupt controller state"
 272                     " for IRQs %d: %s", i + ics->offset, strerror(errno));
 273             return ret;
 274         }
 275     }
 276
 277     return 0;
 278 }
 279
 280 static void ics_kvm_set_irq(void *opaque, int srcno, int val)
 281 {
 282     ICSState *ics = opaque;
 283     struct kvm_irq_level args;
 284     int rc;
 285
 286     args.irq = srcno + ics->offset;
 287     if (ics->irqs[srcno].flags & XICS_FLAGS_IRQ_MSI) {
 288         if (!val) {
 289             return;
 290         }
 291         args.level = KVM_INTERRUPT_SET;
 292     } else {
 293         args.level = val ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
 294     }
 295     rc = kvm_vm_ioctl(kvm_state, KVM_IRQ_LINE, &args);
 296     if (rc < 0) {
 297         perror("kvm_irq_line");
 298     }
 299 }
 300
 301 static void ics_kvm_reset(void *dev)
 302 {
 303     ICSState *ics = ICS_SIMPLE(dev);
 304     int i;
 305     uint8_t flags[ics->nr_irqs];
 306
 307     for (i = 0; i < ics->nr_irqs; i++) {
 308         flags[i] = ics->irqs[i].flags;
 309     }
 310
 311     memset(ics->irqs, 0, sizeof(ICSIRQState) * ics->nr_irqs);
 312
 313     for (i = 0; i < ics->nr_irqs; i++) {
 314         ics->irqs[i].priority = 0xff;
 315         ics->irqs[i].saved_priority = 0xff;
 316         ics->irqs[i].flags = flags[i];
 317     }
 318
 319     ics_set_kvm_state(ics, 1);
 320 }
 321
 322 static void ics_kvm_realize(DeviceState *dev, Error **errp)
 323 {
 324     ICSState *ics = ICS_SIMPLE(dev);
 325
 326     if (!ics->nr_irqs) {
 327         error_setg(errp, "Number of interrupts needs to be greater 0");
 328         return;
 329     }
 330     ics->irqs = g_malloc0(ics->nr_irqs * sizeof(ICSIRQState));
 331     ics->qirqs = qemu_allocate_irqs(ics_kvm_set_irq, ics, ics->nr_irqs);
 332
 333     qemu_register_reset(ics_kvm_reset, dev);
 334 }
 335
 336 static void ics_kvm_class_init(ObjectClass *klass, void *data)
 337 {
 338     ICSStateClass *icsc = ICS_BASE_CLASS(klass);
 339
 340     icsc->realize = ics_kvm_realize;
 341     icsc->pre_save = ics_get_kvm_state;
 342     icsc->post_load = ics_set_kvm_state;
 343 }
 344
 345 static const TypeInfo ics_kvm_info = {
 346     .name = TYPE_ICS_KVM,
 347     .parent = TYPE_ICS_SIMPLE,
 348     .instance_size = sizeof(ICSState),
 349     .class_init = ics_kvm_class_init,
 350 };
 351
 352 /*
 353  * XICS-KVM
 354  */
 355
 356 static void rtas_dummy(PowerPCCPU *cpu, sPAPRMachineState *spapr,
 357                        uint32_t token,
 358                        uint32_t nargs, target_ulong args,
 359                        uint32_t nret, target_ulong rets)
 360 {
 361     error_report("pseries: %s must never be called for in-kernel XICS",
 362                  __func__);
 363 }
 364
 365 int xics_kvm_init(sPAPRMachineState *spapr, Error **errp)
 366 {
 367     int rc;
 368     struct kvm_create_device xics_create_device = {
 369         .type = KVM_DEV_TYPE_XICS,
 370         .flags = 0,
 371     };
 372
 373     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
 374         error_setg(errp,
 375                    "KVM and IRQ_XICS capability must be present for in-kernel XICS");
 376         goto fail;
 377     }
 378
 379     spapr_rtas_register(RTAS_IBM_SET_XIVE, "ibm,set-xive", rtas_dummy);
 380     spapr_rtas_register(RTAS_IBM_GET_XIVE, "ibm,get-xive", rtas_dummy);
 381     spapr_rtas_register(RTAS_IBM_INT_OFF, "ibm,int-off", rtas_dummy);
 382     spapr_rtas_register(RTAS_IBM_INT_ON, "ibm,int-on", rtas_dummy);
 383
 384     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_SET_XIVE, "ibm,set-xive");
 385     if (rc < 0) {
 386         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,set-xive");
 387         goto fail;
 388     }
 389
 390     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_GET_XIVE, "ibm,get-xive");
 391     if (rc < 0) {
 392         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,get-xive");
 393         goto fail;
 394     }
 395
 396     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_ON, "ibm,int-on");
 397     if (rc < 0) {
 398         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-on");
 399         goto fail;
 400     }
 401
 402     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_OFF, "ibm,int-off");
 403     if (rc < 0) {
 404         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-off");
 405         goto fail;
 406     }
 407
 408     /* Create the kernel ICP */
 409     rc = kvm_vm_ioctl(kvm_state, KVM_CREATE_DEVICE, &xics_create_device);
 410     if (rc < 0) {
 411         error_setg_errno(errp, -rc, "Error on KVM_CREATE_DEVICE for XICS");
 412         goto fail;
 413     }
 414
 415     kernel_xics_fd = xics_create_device.fd;
 416
 417     kvm_kernel_irqchip = true;
 418     kvm_msi_via_irqfd_allowed = true;
 419     kvm_gsi_direct_mapping = true;
 420
 421     return rc;
 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)