qemu/hw/intc/xics_kvm.c
<<
>>
Prefs
   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 "trace.h"
  33#include "sysemu/kvm.h"
  34#include "hw/ppc/spapr.h"
  35#include "hw/ppc/spapr_cpu_core.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
  44static int kernel_xics_fd = -1;
  45
  46typedef struct KVMEnabledICP {
  47    unsigned long vcpu_id;
  48    QLIST_ENTRY(KVMEnabledICP) node;
  49} KVMEnabledICP;
  50
  51static QLIST_HEAD(, KVMEnabledICP)
  52    kvm_enabled_icps = QLIST_HEAD_INITIALIZER(&kvm_enabled_icps);
  53
  54static void kvm_disable_icps(void)
  55{
  56    KVMEnabledICP *enabled_icp, *next;
  57
  58    QLIST_FOREACH_SAFE(enabled_icp, &kvm_enabled_icps, node, next) {
  59        QLIST_REMOVE(enabled_icp, node);
  60        g_free(enabled_icp);
  61    }
  62}
  63
  64/*
  65 * ICP-KVM
  66 */
  67void icp_get_kvm_state(ICPState *icp)
  68{
  69    uint64_t state;
  70    int ret;
  71
  72    /* The KVM XICS device is not in use */
  73    if (kernel_xics_fd == -1) {
  74        return;
  75    }
  76
  77    /* ICP for this CPU thread is not in use, exiting */
  78    if (!icp->cs) {
  79        return;
  80    }
  81
  82    ret = kvm_get_one_reg(icp->cs, KVM_REG_PPC_ICP_STATE, &state);
  83    if (ret != 0) {
  84        error_report("Unable to retrieve KVM interrupt controller state"
  85                " for CPU %ld: %s", kvm_arch_vcpu_id(icp->cs), strerror(errno));
  86        exit(1);
  87    }
  88
  89    icp->xirr = state >> KVM_REG_PPC_ICP_XISR_SHIFT;
  90    icp->mfrr = (state >> KVM_REG_PPC_ICP_MFRR_SHIFT)
  91        & KVM_REG_PPC_ICP_MFRR_MASK;
  92    icp->pending_priority = (state >> KVM_REG_PPC_ICP_PPRI_SHIFT)
  93        & KVM_REG_PPC_ICP_PPRI_MASK;
  94}
  95
  96static void do_icp_synchronize_state(CPUState *cpu, run_on_cpu_data arg)
  97{
  98    icp_get_kvm_state(arg.host_ptr);
  99}
 100
 101void icp_synchronize_state(ICPState *icp)
 102{
 103    if (icp->cs) {
 104        run_on_cpu(icp->cs, do_icp_synchronize_state, RUN_ON_CPU_HOST_PTR(icp));
 105    }
 106}
 107
 108int icp_set_kvm_state(ICPState *icp, Error **errp)
 109{
 110    uint64_t state;
 111    int ret;
 112
 113    /* The KVM XICS device is not in use */
 114    if (kernel_xics_fd == -1) {
 115        return 0;
 116    }
 117
 118    /* ICP for this CPU thread is not in use, exiting */
 119    if (!icp->cs) {
 120        return 0;
 121    }
 122
 123    state = ((uint64_t)icp->xirr << KVM_REG_PPC_ICP_XISR_SHIFT)
 124        | ((uint64_t)icp->mfrr << KVM_REG_PPC_ICP_MFRR_SHIFT)
 125        | ((uint64_t)icp->pending_priority << KVM_REG_PPC_ICP_PPRI_SHIFT);
 126
 127    ret = kvm_set_one_reg(icp->cs, KVM_REG_PPC_ICP_STATE, &state);
 128    if (ret < 0) {
 129        error_setg_errno(errp, -ret,
 130                         "Unable to restore KVM interrupt controller state (0x%"
 131                         PRIx64 ") for CPU %ld", state,
 132                         kvm_arch_vcpu_id(icp->cs));
 133        return ret;
 134    }
 135
 136    return 0;
 137}
 138
 139void icp_kvm_realize(DeviceState *dev, Error **errp)
 140{
 141    ICPState *icp = ICP(dev);
 142    CPUState *cs;
 143    KVMEnabledICP *enabled_icp;
 144    unsigned long vcpu_id;
 145    int ret;
 146
 147    /* The KVM XICS device is not in use */
 148    if (kernel_xics_fd == -1) {
 149        return;
 150    }
 151
 152    cs = icp->cs;
 153    vcpu_id = kvm_arch_vcpu_id(cs);
 154
 155    /*
 156     * If we are reusing a parked vCPU fd corresponding to the CPU
 157     * which was hot-removed earlier we don't have to renable
 158     * KVM_CAP_IRQ_XICS capability again.
 159     */
 160    QLIST_FOREACH(enabled_icp, &kvm_enabled_icps, node) {
 161        if (enabled_icp->vcpu_id == vcpu_id) {
 162            return;
 163        }
 164    }
 165
 166    ret = kvm_vcpu_enable_cap(cs, KVM_CAP_IRQ_XICS, 0, kernel_xics_fd, vcpu_id);
 167    if (ret < 0) {
 168        Error *local_err = NULL;
 169
 170        error_setg(&local_err, "Unable to connect CPU%ld to kernel XICS: %s",
 171                   vcpu_id, strerror(errno));
 172        if (errno == ENOSPC) {
 173            error_append_hint(&local_err, "Try -smp maxcpus=N with N < %u\n",
 174                              MACHINE(qdev_get_machine())->smp.max_cpus);
 175        }
 176        error_propagate(errp, local_err);
 177        return;
 178    }
 179    enabled_icp = g_malloc(sizeof(*enabled_icp));
 180    enabled_icp->vcpu_id = vcpu_id;
 181    QLIST_INSERT_HEAD(&kvm_enabled_icps, enabled_icp, node);
 182}
 183
 184/*
 185 * ICS-KVM
 186 */
 187void ics_get_kvm_state(ICSState *ics)
 188{
 189    uint64_t state;
 190    int i;
 191
 192    /* The KVM XICS device is not in use */
 193    if (kernel_xics_fd == -1) {
 194        return;
 195    }
 196
 197    for (i = 0; i < ics->nr_irqs; i++) {
 198        ICSIRQState *irq = &ics->irqs[i];
 199
 200        if (ics_irq_free(ics, i)) {
 201            continue;
 202        }
 203
 204        kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
 205                          i + ics->offset, &state, false, &error_fatal);
 206
 207        irq->server = state & KVM_XICS_DESTINATION_MASK;
 208        irq->saved_priority = (state >> KVM_XICS_PRIORITY_SHIFT)
 209            & KVM_XICS_PRIORITY_MASK;
 210        /*
 211         * To be consistent with the software emulation in xics.c, we
 212         * split out the masked state + priority that we get from the
 213         * kernel into 'current priority' (0xff if masked) and
 214         * 'saved priority' (if masked, this is the priority the
 215         * interrupt had before it was masked).  Masking and unmasking
 216         * are done with the ibm,int-off and ibm,int-on RTAS calls.
 217         */
 218        if (state & KVM_XICS_MASKED) {
 219            irq->priority = 0xff;
 220        } else {
 221            irq->priority = irq->saved_priority;
 222        }
 223
 224        irq->status = 0;
 225        if (state & KVM_XICS_PENDING) {
 226            if (state & KVM_XICS_LEVEL_SENSITIVE) {
 227                irq->status |= XICS_STATUS_ASSERTED;
 228            } else {
 229                /*
 230                 * A pending edge-triggered interrupt (or MSI)
 231                 * must have been rejected previously when we
 232                 * first detected it and tried to deliver it,
 233                 * so mark it as pending and previously rejected
 234                 * for consistency with how xics.c works.
 235                 */
 236                irq->status |= XICS_STATUS_MASKED_PENDING
 237                    | XICS_STATUS_REJECTED;
 238            }
 239        }
 240        if (state & KVM_XICS_PRESENTED) {
 241                irq->status |= XICS_STATUS_PRESENTED;
 242        }
 243        if (state & KVM_XICS_QUEUED) {
 244                irq->status |= XICS_STATUS_QUEUED;
 245        }
 246    }
 247}
 248
 249void ics_synchronize_state(ICSState *ics)
 250{
 251    ics_get_kvm_state(ics);
 252}
 253
 254int ics_set_kvm_state_one(ICSState *ics, int srcno, Error **errp)
 255{
 256    uint64_t state;
 257    ICSIRQState *irq = &ics->irqs[srcno];
 258    int ret;
 259
 260    /* The KVM XICS device is not in use */
 261    if (kernel_xics_fd == -1) {
 262        return 0;
 263    }
 264
 265    state = irq->server;
 266    state |= (uint64_t)(irq->saved_priority & KVM_XICS_PRIORITY_MASK)
 267        << KVM_XICS_PRIORITY_SHIFT;
 268    if (irq->priority != irq->saved_priority) {
 269        assert(irq->priority == 0xff);
 270    }
 271
 272    if (irq->priority == 0xff) {
 273        state |= KVM_XICS_MASKED;
 274    }
 275
 276    if (irq->flags & XICS_FLAGS_IRQ_LSI) {
 277        state |= KVM_XICS_LEVEL_SENSITIVE;
 278        if (irq->status & XICS_STATUS_ASSERTED) {
 279            state |= KVM_XICS_PENDING;
 280        }
 281    } else {
 282        if (irq->status & XICS_STATUS_MASKED_PENDING) {
 283            state |= KVM_XICS_PENDING;
 284        }
 285    }
 286    if (irq->status & XICS_STATUS_PRESENTED) {
 287        state |= KVM_XICS_PRESENTED;
 288    }
 289    if (irq->status & XICS_STATUS_QUEUED) {
 290        state |= KVM_XICS_QUEUED;
 291    }
 292
 293    ret = kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
 294                            srcno + ics->offset, &state, true, errp);
 295    if (ret < 0) {
 296        return ret;
 297    }
 298
 299    return 0;
 300}
 301
 302int ics_set_kvm_state(ICSState *ics, Error **errp)
 303{
 304    int i;
 305
 306    /* The KVM XICS device is not in use */
 307    if (kernel_xics_fd == -1) {
 308        return 0;
 309    }
 310
 311    for (i = 0; i < ics->nr_irqs; i++) {
 312        int ret;
 313
 314        if (ics_irq_free(ics, i)) {
 315            continue;
 316        }
 317
 318        ret = ics_set_kvm_state_one(ics, i, errp);
 319        if (ret < 0) {
 320            return ret;
 321        }
 322    }
 323
 324    return 0;
 325}
 326
 327void ics_kvm_set_irq(ICSState *ics, int srcno, int val)
 328{
 329    struct kvm_irq_level args;
 330    int rc;
 331
 332    /* The KVM XICS device should be in use */
 333    assert(kernel_xics_fd != -1);
 334
 335    args.irq = srcno + ics->offset;
 336    if (ics->irqs[srcno].flags & XICS_FLAGS_IRQ_MSI) {
 337        if (!val) {
 338            return;
 339        }
 340        args.level = KVM_INTERRUPT_SET;
 341    } else {
 342        args.level = val ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
 343    }
 344    rc = kvm_vm_ioctl(kvm_state, KVM_IRQ_LINE, &args);
 345    if (rc < 0) {
 346        perror("kvm_irq_line");
 347    }
 348}
 349
 350int xics_kvm_connect(SpaprInterruptController *intc, uint32_t nr_servers,
 351                     Error **errp)
 352{
 353    ICSState *ics = ICS_SPAPR(intc);
 354    int rc;
 355    CPUState *cs;
 356    Error *local_err = NULL;
 357
 358    /*
 359     * The KVM XICS device already in use. This is the case when
 360     * rebooting under the XICS-only interrupt mode.
 361     */
 362    if (kernel_xics_fd != -1) {
 363        return 0;
 364    }
 365
 366    if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
 367        error_setg(errp,
 368                   "KVM and IRQ_XICS capability must be present for in-kernel XICS");
 369        return -1;
 370    }
 371
 372    rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_SET_XIVE, "ibm,set-xive");
 373    if (rc < 0) {
 374        error_setg_errno(&local_err, -rc,
 375                         "kvmppc_define_rtas_kernel_token: ibm,set-xive");
 376        goto fail;
 377    }
 378
 379    rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_GET_XIVE, "ibm,get-xive");
 380    if (rc < 0) {
 381        error_setg_errno(&local_err, -rc,
 382                         "kvmppc_define_rtas_kernel_token: ibm,get-xive");
 383        goto fail;
 384    }
 385
 386    rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_ON, "ibm,int-on");
 387    if (rc < 0) {
 388        error_setg_errno(&local_err, -rc,
 389                         "kvmppc_define_rtas_kernel_token: ibm,int-on");
 390        goto fail;
 391    }
 392
 393    rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_OFF, "ibm,int-off");
 394    if (rc < 0) {
 395        error_setg_errno(&local_err, -rc,
 396                         "kvmppc_define_rtas_kernel_token: ibm,int-off");
 397        goto fail;
 398    }
 399
 400    /* Create the KVM XICS device */
 401    rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false);
 402    if (rc < 0) {
 403        error_setg_errno(&local_err, -rc, "Error on KVM_CREATE_DEVICE for XICS");
 404        goto fail;
 405    }
 406
 407    /* Tell KVM about the # of VCPUs we may have (POWER9 and newer only) */
 408    if (kvm_device_check_attr(rc, KVM_DEV_XICS_GRP_CTRL,
 409                              KVM_DEV_XICS_NR_SERVERS)) {
 410        if (kvm_device_access(rc, KVM_DEV_XICS_GRP_CTRL,
 411                              KVM_DEV_XICS_NR_SERVERS, &nr_servers, true,
 412                              &local_err)) {
 413            goto fail;
 414        }
 415    }
 416
 417    kernel_xics_fd = rc;
 418    kvm_kernel_irqchip = true;
 419    kvm_msi_via_irqfd_allowed = true;
 420    kvm_gsi_direct_mapping = true;
 421
 422    /* Create the presenters */
 423    CPU_FOREACH(cs) {
 424        PowerPCCPU *cpu = POWERPC_CPU(cs);
 425
 426        icp_kvm_realize(DEVICE(spapr_cpu_state(cpu)->icp), &local_err);
 427        if (local_err) {
 428            goto fail;
 429        }
 430    }
 431
 432    /* Update the KVM sources */
 433    ics_set_kvm_state(ics, &local_err);
 434    if (local_err) {
 435        goto fail;
 436    }
 437
 438    /* Connect the presenters to the initial VCPUs of the machine */
 439    CPU_FOREACH(cs) {
 440        PowerPCCPU *cpu = POWERPC_CPU(cs);
 441        icp_set_kvm_state(spapr_cpu_state(cpu)->icp, &local_err);
 442        if (local_err) {
 443            goto fail;
 444        }
 445    }
 446
 447    return 0;
 448
 449fail:
 450    error_propagate(errp, local_err);
 451    xics_kvm_disconnect(intc);
 452    return -1;
 453}
 454
 455void xics_kvm_disconnect(SpaprInterruptController *intc)
 456{
 457    /*
 458     * Only on P9 using the XICS-on XIVE KVM device:
 459     *
 460     * When the KVM device fd is closed, the device is destroyed and
 461     * removed from the list of devices of the VM. The VCPU presenters
 462     * are also detached from the device.
 463     */
 464    if (kernel_xics_fd != -1) {
 465        close(kernel_xics_fd);
 466        kernel_xics_fd = -1;
 467    }
 468
 469    kvmppc_define_rtas_kernel_token(0, "ibm,set-xive");
 470    kvmppc_define_rtas_kernel_token(0, "ibm,get-xive");
 471    kvmppc_define_rtas_kernel_token(0, "ibm,int-on");
 472    kvmppc_define_rtas_kernel_token(0, "ibm,int-off");
 473
 474    kvm_kernel_irqchip = false;
 475    kvm_msi_via_irqfd_allowed = false;
 476    kvm_gsi_direct_mapping = false;
 477
 478    /* Clear the presenter from the VCPUs */
 479    kvm_disable_icps();
 480}
 481
 482/*
 483 * This is a heuristic to detect older KVMs on POWER9 hosts that don't
 484 * support destruction of a KVM XICS device while the VM is running.
 485 * Required to start a spapr machine with ic-mode=dual,kernel-irqchip=on.
 486 */
 487bool xics_kvm_has_broken_disconnect(void)
 488{
 489    int rc;
 490
 491    rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false);
 492    if (rc < 0) {
 493        /*
 494         * The error is ignored on purpose. The KVM XICS setup code
 495         * will catch it again anyway. The goal here is to see if
 496         * close() actually destroys the device or not.
 497         */
 498        return false;
 499    }
 500
 501    close(rc);
 502
 503    rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false);
 504    if (rc >= 0) {
 505        close(rc);
 506        return false;
 507    }
 508
 509    return errno == EEXIST;
 510}
 511