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 "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
  44 static int kernel_xics_fd = -1;
  45
  46 typedef struct KVMEnabledICP {
  47     unsigned long vcpu_id;
  48     QLIST_ENTRY(KVMEnabledICP) node;
  49 } KVMEnabledICP;
  50
  51 static QLIST_HEAD(, KVMEnabledICP)
  52     kvm_enabled_icps = QLIST_HEAD_INITIALIZER(&kvm_enabled_icps);
  53
  54 static 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  */
  67 void 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
  96 static void do_icp_synchronize_state(CPUState *cpu, run_on_cpu_data arg)
  97 {
  98     icp_get_kvm_state(arg.host_ptr);
  99 }
 100
 101 void 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
 108 int 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
 139 void 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_setg(errp, "Unable to connect CPU%ld to kernel XICS: %s", vcpu_id,
 169                    strerror(errno));
 170         return;
 171     }
 172     enabled_icp = g_malloc(sizeof(*enabled_icp));
 173     enabled_icp->vcpu_id = vcpu_id;
 174     QLIST_INSERT_HEAD(&kvm_enabled_icps, enabled_icp, node);
 175 }
 176
 177 /*
 178  * ICS-KVM
 179  */
 180 void ics_get_kvm_state(ICSState *ics)
 181 {
 182     uint64_t state;
 183     int i;
 184
 185     /* The KVM XICS device is not in use */
 186     if (kernel_xics_fd == -1) {
 187         return;
 188     }
 189
 190     for (i = 0; i < ics->nr_irqs; i++) {
 191         ICSIRQState *irq = &ics->irqs[i];
 192
 193         kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
 194                           i + ics->offset, &state, false, &error_fatal);
 195
 196         irq->server = state & KVM_XICS_DESTINATION_MASK;
 197         irq->saved_priority = (state >> KVM_XICS_PRIORITY_SHIFT)
 198             & KVM_XICS_PRIORITY_MASK;
 199         /*
 200          * To be consistent with the software emulation in xics.c, we
 201          * split out the masked state + priority that we get from the
 202          * kernel into 'current priority' (0xff if masked) and
 203          * 'saved priority' (if masked, this is the priority the
 204          * interrupt had before it was masked).  Masking and unmasking
 205          * are done with the ibm,int-off and ibm,int-on RTAS calls.
 206          */
 207         if (state & KVM_XICS_MASKED) {
 208             irq->priority = 0xff;
 209         } else {
 210             irq->priority = irq->saved_priority;
 211         }
 212
 213         irq->status = 0;
 214         if (state & KVM_XICS_PENDING) {
 215             if (state & KVM_XICS_LEVEL_SENSITIVE) {
 216                 irq->status |= XICS_STATUS_ASSERTED;
 217             } else {
 218                 /*
 219                  * A pending edge-triggered interrupt (or MSI)
 220                  * must have been rejected previously when we
 221                  * first detected it and tried to deliver it,
 222                  * so mark it as pending and previously rejected
 223                  * for consistency with how xics.c works.
 224                  */
 225                 irq->status |= XICS_STATUS_MASKED_PENDING
 226                     | XICS_STATUS_REJECTED;
 227             }
 228         }
 229         if (state & KVM_XICS_PRESENTED) {
 230                 irq->status |= XICS_STATUS_PRESENTED;
 231         }
 232         if (state & KVM_XICS_QUEUED) {
 233                 irq->status |= XICS_STATUS_QUEUED;
 234         }
 235     }
 236 }
 237
 238 void ics_synchronize_state(ICSState *ics)
 239 {
 240     ics_get_kvm_state(ics);
 241 }
 242
 243 int ics_set_kvm_state_one(ICSState *ics, int srcno, Error **errp)
 244 {
 245     uint64_t state;
 246     ICSIRQState *irq = &ics->irqs[srcno];
 247     int ret;
 248
 249     /* The KVM XICS device is not in use */
 250     if (kernel_xics_fd == -1) {
 251         return 0;
 252     }
 253
 254     state = irq->server;
 255     state |= (uint64_t)(irq->saved_priority & KVM_XICS_PRIORITY_MASK)
 256         << KVM_XICS_PRIORITY_SHIFT;
 257     if (irq->priority != irq->saved_priority) {
 258         assert(irq->priority == 0xff);
 259     }
 260
 261     if (irq->priority == 0xff) {
 262         state |= KVM_XICS_MASKED;
 263     }
 264
 265     if (irq->flags & XICS_FLAGS_IRQ_LSI) {
 266         state |= KVM_XICS_LEVEL_SENSITIVE;
 267         if (irq->status & XICS_STATUS_ASSERTED) {
 268             state |= KVM_XICS_PENDING;
 269         }
 270     } else {
 271         if (irq->status & XICS_STATUS_MASKED_PENDING) {
 272             state |= KVM_XICS_PENDING;
 273         }
 274     }
 275     if (irq->status & XICS_STATUS_PRESENTED) {
 276         state |= KVM_XICS_PRESENTED;
 277     }
 278     if (irq->status & XICS_STATUS_QUEUED) {
 279         state |= KVM_XICS_QUEUED;
 280     }
 281
 282     ret = kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
 283                             srcno + ics->offset, &state, true, errp);
 284     if (ret < 0) {
 285         return ret;
 286     }
 287
 288     return 0;
 289 }
 290
 291 int ics_set_kvm_state(ICSState *ics, Error **errp)
 292 {
 293     int i;
 294
 295     /* The KVM XICS device is not in use */
 296     if (kernel_xics_fd == -1) {
 297         return 0;
 298     }
 299
 300     for (i = 0; i < ics->nr_irqs; i++) {
 301         Error *local_err = NULL;
 302         int ret;
 303
 304         ret = ics_set_kvm_state_one(ics, i, &local_err);
 305         if (ret < 0) {
 306             error_propagate(errp, local_err);
 307             return ret;
 308         }
 309     }
 310
 311     return 0;
 312 }
 313
 314 void ics_kvm_set_irq(ICSState *ics, int srcno, int val)
 315 {
 316     struct kvm_irq_level args;
 317     int rc;
 318
 319     /* The KVM XICS device should be in use */
 320     assert(kernel_xics_fd != -1);
 321
 322     args.irq = srcno + ics->offset;
 323     if (ics->irqs[srcno].flags & XICS_FLAGS_IRQ_MSI) {
 324         if (!val) {
 325             return;
 326         }
 327         args.level = KVM_INTERRUPT_SET;
 328     } else {
 329         args.level = val ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
 330     }
 331     rc = kvm_vm_ioctl(kvm_state, KVM_IRQ_LINE, &args);
 332     if (rc < 0) {
 333         perror("kvm_irq_line");
 334     }
 335 }
 336
 337 int xics_kvm_connect(SpaprMachineState *spapr, Error **errp)
 338 {
 339     int rc;
 340     CPUState *cs;
 341     Error *local_err = NULL;
 342
 343     /*
 344      * The KVM XICS device already in use. This is the case when
 345      * rebooting under the XICS-only interrupt mode.
 346      */
 347     if (kernel_xics_fd != -1) {
 348         return 0;
 349     }
 350
 351     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
 352         error_setg(errp,
 353                    "KVM and IRQ_XICS capability must be present for in-kernel XICS");
 354         return -1;
 355     }
 356
 357     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_SET_XIVE, "ibm,set-xive");
 358     if (rc < 0) {
 359         error_setg_errno(&local_err, -rc,
 360                          "kvmppc_define_rtas_kernel_token: ibm,set-xive");
 361         goto fail;
 362     }
 363
 364     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_GET_XIVE, "ibm,get-xive");
 365     if (rc < 0) {
 366         error_setg_errno(&local_err, -rc,
 367                          "kvmppc_define_rtas_kernel_token: ibm,get-xive");
 368         goto fail;
 369     }
 370
 371     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_ON, "ibm,int-on");
 372     if (rc < 0) {
 373         error_setg_errno(&local_err, -rc,
 374                          "kvmppc_define_rtas_kernel_token: ibm,int-on");
 375         goto fail;
 376     }
 377
 378     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_OFF, "ibm,int-off");
 379     if (rc < 0) {
 380         error_setg_errno(&local_err, -rc,
 381                          "kvmppc_define_rtas_kernel_token: ibm,int-off");
 382         goto fail;
 383     }
 384
 385     /* Create the KVM XICS device */
 386     rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false);
 387     if (rc < 0) {
 388         error_setg_errno(&local_err, -rc, "Error on KVM_CREATE_DEVICE for XICS");
 389         goto fail;
 390     }
 391
 392     kernel_xics_fd = rc;
 393     kvm_kernel_irqchip = true;
 394     kvm_msi_via_irqfd_allowed = true;
 395     kvm_gsi_direct_mapping = true;
 396
 397     /* Create the presenters */
 398     CPU_FOREACH(cs) {
 399         PowerPCCPU *cpu = POWERPC_CPU(cs);
 400
 401         icp_kvm_realize(DEVICE(spapr_cpu_state(cpu)->icp), &local_err);
 402         if (local_err) {
 403             goto fail;
 404         }
 405     }
 406
 407     /* Update the KVM sources */
 408     ics_set_kvm_state(spapr->ics, &local_err);
 409     if (local_err) {
 410         goto fail;
 411     }
 412
 413     /* Connect the presenters to the initial VCPUs of the machine */
 414     CPU_FOREACH(cs) {
 415         PowerPCCPU *cpu = POWERPC_CPU(cs);
 416         icp_set_kvm_state(spapr_cpu_state(cpu)->icp, &local_err);
 417         if (local_err) {
 418             goto fail;
 419         }
 420     }
 421
 422     return 0;
 423
 424 fail:
 425     error_propagate(errp, local_err);
 426     xics_kvm_disconnect(spapr, NULL);
 427     return -1;
 428 }
 429
 430 void xics_kvm_disconnect(SpaprMachineState *spapr, Error **errp)
 431 {
 432     /*
 433      * Only on P9 using the XICS-on XIVE KVM device:
 434      *
 435      * When the KVM device fd is closed, the device is destroyed and
 436      * removed from the list of devices of the VM. The VCPU presenters
 437      * are also detached from the device.
 438      */
 439     if (kernel_xics_fd != -1) {
 440         close(kernel_xics_fd);
 441         kernel_xics_fd = -1;
 442     }
 443
 444     kvmppc_define_rtas_kernel_token(0, "ibm,set-xive");
 445     kvmppc_define_rtas_kernel_token(0, "ibm,get-xive");
 446     kvmppc_define_rtas_kernel_token(0, "ibm,int-on");
 447     kvmppc_define_rtas_kernel_token(0, "ibm,int-off");
 448
 449     kvm_kernel_irqchip = false;
 450     kvm_msi_via_irqfd_allowed = false;
 451     kvm_gsi_direct_mapping = false;
 452
 453     /* Clear the presenter from the VCPUs */
 454     kvm_disable_icps();
 455 }
 456
 457 /*
 458  * This is a heuristic to detect older KVMs on POWER9 hosts that don't
 459  * support destruction of a KVM XICS device while the VM is running.
 460  * Required to start a spapr machine with ic-mode=dual,kernel-irqchip=on.
 461  */
 462 bool xics_kvm_has_broken_disconnect(SpaprMachineState *spapr)
 463 {
 464     int rc;
 465
 466     rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false);
 467     if (rc < 0) {
 468         /*
 469          * The error is ignored on purpose. The KVM XICS setup code
 470          * will catch it again anyway. The goal here is to see if
 471          * close() actually destroys the device or not.
 472          */
 473         return false;
 474     }
 475
 476     close(rc);
 477
 478     rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false);
 479     if (rc >= 0) {
 480         close(rc);
 481         return false;
 482     }
 483
 484     return errno == EEXIST;
 485 }