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 "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)
 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_report("Unable to restore KVM interrupt controller state (0x%"
 130                 PRIx64 ") for CPU %ld: %s", state, kvm_arch_vcpu_id(icp->cs),
 131                 strerror(errno));
 132         return ret;
 133     }
 134
 135     return 0;
 136 }
 137
 138 void icp_kvm_realize(DeviceState *dev, Error **errp)
 139 {
 140     ICPState *icp = ICP(dev);
 141     CPUState *cs;
 142     KVMEnabledICP *enabled_icp;
 143     unsigned long vcpu_id;
 144     int ret;
 145
 146     /* The KVM XICS device is not in use */
 147     if (kernel_xics_fd == -1) {
 148         return;
 149     }
 150
 151     cs = icp->cs;
 152     vcpu_id = kvm_arch_vcpu_id(cs);
 153
 154     /*
 155      * If we are reusing a parked vCPU fd corresponding to the CPU
 156      * which was hot-removed earlier we don't have to renable
 157      * KVM_CAP_IRQ_XICS capability again.
 158      */
 159     QLIST_FOREACH(enabled_icp, &kvm_enabled_icps, node) {
 160         if (enabled_icp->vcpu_id == vcpu_id) {
 161             return;
 162         }
 163     }
 164
 165     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_IRQ_XICS, 0, kernel_xics_fd, vcpu_id);
 166     if (ret < 0) {
 167         error_setg(errp, "Unable to connect CPU%ld to kernel XICS: %s", vcpu_id,
 168                    strerror(errno));
 169         return;
 170     }
 171     enabled_icp = g_malloc(sizeof(*enabled_icp));
 172     enabled_icp->vcpu_id = vcpu_id;
 173     QLIST_INSERT_HEAD(&kvm_enabled_icps, enabled_icp, node);
 174 }
 175
 176 /*
 177  * ICS-KVM
 178  */
 179 void ics_get_kvm_state(ICSState *ics)
 180 {
 181     uint64_t state;
 182     int i;
 183
 184     /* The KVM XICS device is not in use */
 185     if (kernel_xics_fd == -1) {
 186         return;
 187     }
 188
 189     for (i = 0; i < ics->nr_irqs; i++) {
 190         ICSIRQState *irq = &ics->irqs[i];
 191
 192         kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
 193                           i + ics->offset, &state, false, &error_fatal);
 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 void ics_synchronize_state(ICSState *ics)
 238 {
 239     ics_get_kvm_state(ics);
 240 }
 241
 242 int ics_set_kvm_state_one(ICSState *ics, int srcno)
 243 {
 244     uint64_t state;
 245     Error *local_err = NULL;
 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         state |= KVM_XICS_MASKED;
 260     }
 261
 262     if (irq->flags & XICS_FLAGS_IRQ_LSI) {
 263         state |= KVM_XICS_LEVEL_SENSITIVE;
 264         if (irq->status & XICS_STATUS_ASSERTED) {
 265             state |= KVM_XICS_PENDING;
 266         }
 267     } else {
 268         if (irq->status & XICS_STATUS_MASKED_PENDING) {
 269             state |= KVM_XICS_PENDING;
 270         }
 271     }
 272     if (irq->status & XICS_STATUS_PRESENTED) {
 273         state |= KVM_XICS_PRESENTED;
 274     }
 275     if (irq->status & XICS_STATUS_QUEUED) {
 276         state |= KVM_XICS_QUEUED;
 277     }
 278
 279     ret = kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
 280                             srcno + ics->offset, &state, true, &local_err);
 281     if (local_err) {
 282         error_report_err(local_err);
 283         return ret;
 284     }
 285
 286     return 0;
 287 }
 288
 289 int ics_set_kvm_state(ICSState *ics)
 290 {
 291     int i;
 292
 293     /* The KVM XICS device is not in use */
 294     if (kernel_xics_fd == -1) {
 295         return 0;
 296     }
 297
 298     for (i = 0; i < ics->nr_irqs; i++) {
 299         int ret;
 300
 301         ret = ics_set_kvm_state_one(ics, i);
 302         if (ret) {
 303             return ret;
 304         }
 305     }
 306
 307     return 0;
 308 }
 309
 310 void ics_kvm_set_irq(ICSState *ics, int srcno, int val)
 311 {
 312     struct kvm_irq_level args;
 313     int rc;
 314
 315     /* The KVM XICS device should be in use */
 316     assert(kernel_xics_fd != -1);
 317
 318     args.irq = srcno + ics->offset;
 319     if (ics->irqs[srcno].flags & XICS_FLAGS_IRQ_MSI) {
 320         if (!val) {
 321             return;
 322         }
 323         args.level = KVM_INTERRUPT_SET;
 324     } else {
 325         args.level = val ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
 326     }
 327     rc = kvm_vm_ioctl(kvm_state, KVM_IRQ_LINE, &args);
 328     if (rc < 0) {
 329         perror("kvm_irq_line");
 330     }
 331 }
 332
 333 static void rtas_dummy(PowerPCCPU *cpu, SpaprMachineState *spapr,
 334                        uint32_t token,
 335                        uint32_t nargs, target_ulong args,
 336                        uint32_t nret, target_ulong rets)
 337 {
 338     error_report("pseries: %s must never be called for in-kernel XICS",
 339                  __func__);
 340 }
 341
 342 int xics_kvm_init(SpaprMachineState *spapr, Error **errp)
 343 {
 344     int rc;
 345
 346     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
 347         error_setg(errp,
 348                    "KVM and IRQ_XICS capability must be present for in-kernel XICS");
 349         goto fail;
 350     }
 351
 352     spapr_rtas_register(RTAS_IBM_SET_XIVE, "ibm,set-xive", rtas_dummy);
 353     spapr_rtas_register(RTAS_IBM_GET_XIVE, "ibm,get-xive", rtas_dummy);
 354     spapr_rtas_register(RTAS_IBM_INT_OFF, "ibm,int-off", rtas_dummy);
 355     spapr_rtas_register(RTAS_IBM_INT_ON, "ibm,int-on", rtas_dummy);
 356
 357     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_SET_XIVE, "ibm,set-xive");
 358     if (rc < 0) {
 359         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,set-xive");
 360         goto fail;
 361     }
 362
 363     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_GET_XIVE, "ibm,get-xive");
 364     if (rc < 0) {
 365         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,get-xive");
 366         goto fail;
 367     }
 368
 369     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_ON, "ibm,int-on");
 370     if (rc < 0) {
 371         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-on");
 372         goto fail;
 373     }
 374
 375     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_OFF, "ibm,int-off");
 376     if (rc < 0) {
 377         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-off");
 378         goto fail;
 379     }
 380
 381     /* Create the KVM XICS device */
 382     rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false);
 383     if (rc < 0) {
 384         error_setg_errno(errp, -rc, "Error on KVM_CREATE_DEVICE for XICS");
 385         goto fail;
 386     }
 387
 388     kernel_xics_fd = rc;
 389     kvm_kernel_irqchip = true;
 390     kvm_msi_via_irqfd_allowed = true;
 391     kvm_gsi_direct_mapping = true;
 392
 393     return 0;
 394
 395 fail:
 396     kvmppc_define_rtas_kernel_token(0, "ibm,set-xive");
 397     kvmppc_define_rtas_kernel_token(0, "ibm,get-xive");
 398     kvmppc_define_rtas_kernel_token(0, "ibm,int-on");
 399     kvmppc_define_rtas_kernel_token(0, "ibm,int-off");
 400     return -1;
 401 }
 402
 403 void xics_kvm_disconnect(SpaprMachineState *spapr, Error **errp)
 404 {
 405     /* The KVM XICS device is not in use */
 406     if (kernel_xics_fd == -1) {
 407         return;
 408     }
 409
 410     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
 411         error_setg(errp,
 412                    "KVM and IRQ_XICS capability must be present for KVM XICS device");
 413         return;
 414     }
 415
 416     /*
 417      * Only on P9 using the XICS-on XIVE KVM device:
 418      *
 419      * When the KVM device fd is closed, the device is destroyed and
 420      * removed from the list of devices of the VM. The VCPU presenters
 421      * are also detached from the device.
 422      */
 423     close(kernel_xics_fd);
 424     kernel_xics_fd = -1;
 425
 426     spapr_rtas_unregister(RTAS_IBM_SET_XIVE);
 427     spapr_rtas_unregister(RTAS_IBM_GET_XIVE);
 428     spapr_rtas_unregister(RTAS_IBM_INT_OFF);
 429     spapr_rtas_unregister(RTAS_IBM_INT_ON);
 430
 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
 436     kvm_kernel_irqchip = false;
 437     kvm_msi_via_irqfd_allowed = false;
 438     kvm_gsi_direct_mapping = false;
 439
 440     /* Clear the presenter from the VCPUs */
 441     kvm_disable_icps();
 442 }