| 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 "trace.h" |
| 31 | #include "system/kvm.h" |
| 32 | #include "hw/ppc/spapr.h" |
| 33 | #include "hw/ppc/spapr_cpu_core.h" |
| 34 | #include "hw/ppc/xics.h" |
| 35 | #include "hw/ppc/xics_spapr.h" |
| 36 | #include "kvm_ppc.h" |
| 37 | #include "qemu/config-file.h" |
| 38 | #include "qemu/error-report.h" |
| 39 | |
| 40 | #include <sys/ioctl.h> |
| 41 | |
| 42 | static int kernel_xics_fd = -1; |
| 43 | |
| 44 | typedef struct KVMEnabledICP { |
| 45 | unsigned long vcpu_id; |
| 46 | QLIST_ENTRY(KVMEnabledICP) node; |
| 47 | } KVMEnabledICP; |
| 48 | |
| 49 | static QLIST_HEAD(, KVMEnabledICP) |
| 50 | kvm_enabled_icps = QLIST_HEAD_INITIALIZER(&kvm_enabled_icps); |
| 51 | |
| 52 | static void kvm_disable_icps(void) |
| 53 | { |
| 54 | KVMEnabledICP *enabled_icp, *next; |
| 55 | |
| 56 | QLIST_FOREACH_SAFE(enabled_icp, &kvm_enabled_icps, node, next) { |
| 57 | QLIST_REMOVE(enabled_icp, node); |
| 58 | g_free(enabled_icp); |
| 59 | } |
| 60 | } |
| 61 | |
| 62 | /* |
| 63 | * ICP-KVM |
| 64 | */ |
| 65 | void icp_get_kvm_state(ICPState *icp) |
| 66 | { |
| 67 | uint64_t state; |
| 68 | int ret; |
| 69 | |
| 70 | /* The KVM XICS device is not in use */ |
| 71 | if (kernel_xics_fd == -1) { |
| 72 | return; |
| 73 | } |
| 74 | |
| 75 | /* ICP for this CPU thread is not in use, exiting */ |
| 76 | if (!icp->cs) { |
| 77 | return; |
| 78 | } |
| 79 | |
| 80 | ret = kvm_get_one_reg(icp->cs, KVM_REG_PPC_ICP_STATE, &state); |
| 81 | if (ret != 0) { |
| 82 | error_report("Unable to retrieve KVM interrupt controller state" |
| 83 | " for CPU %ld: %s", kvm_arch_vcpu_id(icp->cs), strerror(errno)); |
| 84 | exit(1); |
| 85 | } |
| 86 | |
| 87 | icp->xirr = state >> KVM_REG_PPC_ICP_XISR_SHIFT; |
| 88 | icp->mfrr = (state >> KVM_REG_PPC_ICP_MFRR_SHIFT) |
| 89 | & KVM_REG_PPC_ICP_MFRR_MASK; |
| 90 | icp->pending_priority = (state >> KVM_REG_PPC_ICP_PPRI_SHIFT) |
| 91 | & KVM_REG_PPC_ICP_PPRI_MASK; |
| 92 | } |
| 93 | |
| 94 | static void do_icp_synchronize_state(CPUState *cpu, run_on_cpu_data arg) |
| 95 | { |
| 96 | icp_get_kvm_state(arg.host_ptr); |
| 97 | } |
| 98 | |
| 99 | void icp_synchronize_state(ICPState *icp) |
| 100 | { |
| 101 | if (icp->cs) { |
| 102 | run_on_cpu(icp->cs, do_icp_synchronize_state, RUN_ON_CPU_HOST_PTR(icp)); |
| 103 | } |
| 104 | } |
| 105 | |
| 106 | int icp_set_kvm_state(ICPState *icp, Error **errp) |
| 107 | { |
| 108 | uint64_t state; |
| 109 | int ret; |
| 110 | |
| 111 | /* The KVM XICS device is not in use */ |
| 112 | if (kernel_xics_fd == -1) { |
| 113 | return 0; |
| 114 | } |
| 115 | |
| 116 | /* ICP for this CPU thread is not in use, exiting */ |
| 117 | if (!icp->cs) { |
| 118 | return 0; |
| 119 | } |
| 120 | |
| 121 | state = ((uint64_t)icp->xirr << KVM_REG_PPC_ICP_XISR_SHIFT) |
| 122 | | ((uint64_t)icp->mfrr << KVM_REG_PPC_ICP_MFRR_SHIFT) |
| 123 | | ((uint64_t)icp->pending_priority << KVM_REG_PPC_ICP_PPRI_SHIFT); |
| 124 | |
| 125 | ret = kvm_set_one_reg(icp->cs, KVM_REG_PPC_ICP_STATE, &state); |
| 126 | if (ret < 0) { |
| 127 | error_setg_errno(errp, -ret, |
| 128 | "Unable to restore KVM interrupt controller state (0x%" |
| 129 | PRIx64 ") for CPU %ld", state, |
| 130 | kvm_arch_vcpu_id(icp->cs)); |
| 131 | return ret; |
| 132 | } |
| 133 | |
| 134 | return 0; |
| 135 | } |
| 136 | |
| 137 | void icp_kvm_realize(DeviceState *dev, Error **errp) |
| 138 | { |
| 139 | ICPState *icp = ICP(dev); |
| 140 | CPUState *cs; |
| 141 | KVMEnabledICP *enabled_icp; |
| 142 | unsigned long vcpu_id; |
| 143 | int ret; |
| 144 | |
| 145 | /* The KVM XICS device is not in use */ |
| 146 | if (kernel_xics_fd == -1) { |
| 147 | return; |
| 148 | } |
| 149 | |
| 150 | cs = icp->cs; |
| 151 | vcpu_id = kvm_arch_vcpu_id(cs); |
| 152 | |
| 153 | /* |
| 154 | * If we are reusing a parked vCPU fd corresponding to the CPU |
| 155 | * which was hot-removed earlier we don't have to renable |
| 156 | * KVM_CAP_IRQ_XICS capability again. |
| 157 | */ |
| 158 | QLIST_FOREACH(enabled_icp, &kvm_enabled_icps, node) { |
| 159 | if (enabled_icp->vcpu_id == vcpu_id) { |
| 160 | return; |
| 161 | } |
| 162 | } |
| 163 | |
| 164 | ret = kvm_vcpu_enable_cap(cs, KVM_CAP_IRQ_XICS, 0, kernel_xics_fd, vcpu_id); |
| 165 | if (ret < 0) { |
| 166 | Error *local_err = NULL; |
| 167 | |
| 168 | error_setg_errno(&local_err, errno, |
| 169 | "Unable to connect CPU%ld to kernel XICS", |
| 170 | vcpu_id); |
| 171 | if (errno == ENOSPC) { |
| 172 | error_append_hint(&local_err, "Try -smp maxcpus=N with N < %u\n", |
| 173 | MACHINE(qdev_get_machine())->smp.max_cpus); |
| 174 | } |
| 175 | error_propagate(errp, local_err); |
| 176 | return; |
| 177 | } |
| 178 | enabled_icp = g_malloc(sizeof(*enabled_icp)); |
| 179 | enabled_icp->vcpu_id = vcpu_id; |
| 180 | QLIST_INSERT_HEAD(&kvm_enabled_icps, enabled_icp, node); |
| 181 | } |
| 182 | |
| 183 | /* |
| 184 | * ICS-KVM |
| 185 | */ |
| 186 | void ics_get_kvm_state(ICSState *ics) |
| 187 | { |
| 188 | uint64_t state; |
| 189 | int i; |
| 190 | |
| 191 | /* The KVM XICS device is not in use */ |
| 192 | if (kernel_xics_fd == -1) { |
| 193 | return; |
| 194 | } |
| 195 | |
| 196 | for (i = 0; i < ics->nr_irqs; i++) { |
| 197 | ICSIRQState *irq = &ics->irqs[i]; |
| 198 | |
| 199 | if (ics_irq_free(ics, i)) { |
| 200 | continue; |
| 201 | } |
| 202 | |
| 203 | kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES, |
| 204 | i + ics->offset, &state, false, &error_fatal); |
| 205 | |
| 206 | irq->server = state & KVM_XICS_DESTINATION_MASK; |
| 207 | irq->saved_priority = (state >> KVM_XICS_PRIORITY_SHIFT) |
| 208 | & KVM_XICS_PRIORITY_MASK; |
| 209 | /* |
| 210 | * To be consistent with the software emulation in xics.c, we |
| 211 | * split out the masked state + priority that we get from the |
| 212 | * kernel into 'current priority' (0xff if masked) and |
| 213 | * 'saved priority' (if masked, this is the priority the |
| 214 | * interrupt had before it was masked). Masking and unmasking |
| 215 | * are done with the ibm,int-off and ibm,int-on RTAS calls. |
| 216 | */ |
| 217 | if (state & KVM_XICS_MASKED) { |
| 218 | irq->priority = 0xff; |
| 219 | } else { |
| 220 | irq->priority = irq->saved_priority; |
| 221 | } |
| 222 | |
| 223 | irq->status = 0; |
| 224 | if (state & KVM_XICS_PENDING) { |
| 225 | if (state & KVM_XICS_LEVEL_SENSITIVE) { |
| 226 | irq->status |= XICS_STATUS_ASSERTED; |
| 227 | } else { |
| 228 | /* |
| 229 | * A pending edge-triggered interrupt (or MSI) |
| 230 | * must have been rejected previously when we |
| 231 | * first detected it and tried to deliver it, |
| 232 | * so mark it as pending and previously rejected |
| 233 | * for consistency with how xics.c works. |
| 234 | */ |
| 235 | irq->status |= XICS_STATUS_MASKED_PENDING |
| 236 | | XICS_STATUS_REJECTED; |
| 237 | } |
| 238 | } |
| 239 | if (state & KVM_XICS_PRESENTED) { |
| 240 | irq->status |= XICS_STATUS_PRESENTED; |
| 241 | } |
| 242 | if (state & KVM_XICS_QUEUED) { |
| 243 | irq->status |= XICS_STATUS_QUEUED; |
| 244 | } |
| 245 | } |
| 246 | } |
| 247 | |
| 248 | void ics_synchronize_state(ICSState *ics) |
| 249 | { |
| 250 | ics_get_kvm_state(ics); |
| 251 | } |
| 252 | |
| 253 | int ics_set_kvm_state_one(ICSState *ics, int srcno, Error **errp) |
| 254 | { |
| 255 | uint64_t state; |
| 256 | ICSIRQState *irq = &ics->irqs[srcno]; |
| 257 | int ret; |
| 258 | |
| 259 | /* The KVM XICS device is not in use */ |
| 260 | if (kernel_xics_fd == -1) { |
| 261 | return 0; |
| 262 | } |
| 263 | |
| 264 | state = irq->server; |
| 265 | state |= (uint64_t)(irq->saved_priority & KVM_XICS_PRIORITY_MASK) |
| 266 | << KVM_XICS_PRIORITY_SHIFT; |
| 267 | if (irq->priority != irq->saved_priority) { |
| 268 | assert(irq->priority == 0xff); |
| 269 | } |
| 270 | |
| 271 | if (irq->priority == 0xff) { |
| 272 | state |= KVM_XICS_MASKED; |
| 273 | } |
| 274 | |
| 275 | if (irq->flags & XICS_FLAGS_IRQ_LSI) { |
| 276 | state |= KVM_XICS_LEVEL_SENSITIVE; |
| 277 | if (irq->status & XICS_STATUS_ASSERTED) { |
| 278 | state |= KVM_XICS_PENDING; |
| 279 | } |
| 280 | } else { |
| 281 | if (irq->status & XICS_STATUS_MASKED_PENDING) { |
| 282 | state |= KVM_XICS_PENDING; |
| 283 | } |
| 284 | } |
| 285 | if (irq->status & XICS_STATUS_PRESENTED) { |
| 286 | state |= KVM_XICS_PRESENTED; |
| 287 | } |
| 288 | if (irq->status & XICS_STATUS_QUEUED) { |
| 289 | state |= KVM_XICS_QUEUED; |
| 290 | } |
| 291 | |
| 292 | ret = kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES, |
| 293 | srcno + ics->offset, &state, true, errp); |
| 294 | if (ret < 0) { |
| 295 | return ret; |
| 296 | } |
| 297 | |
| 298 | return 0; |
| 299 | } |
| 300 | |
| 301 | int ics_set_kvm_state(ICSState *ics, Error **errp) |
| 302 | { |
| 303 | int i; |
| 304 | |
| 305 | /* The KVM XICS device is not in use */ |
| 306 | if (kernel_xics_fd == -1) { |
| 307 | return 0; |
| 308 | } |
| 309 | |
| 310 | for (i = 0; i < ics->nr_irqs; i++) { |
| 311 | int ret; |
| 312 | |
| 313 | if (ics_irq_free(ics, i)) { |
| 314 | continue; |
| 315 | } |
| 316 | |
| 317 | ret = ics_set_kvm_state_one(ics, i, errp); |
| 318 | if (ret < 0) { |
| 319 | return ret; |
| 320 | } |
| 321 | } |
| 322 | |
| 323 | return 0; |
| 324 | } |
| 325 | |
| 326 | void ics_kvm_set_irq(ICSState *ics, int srcno, int val) |
| 327 | { |
| 328 | struct kvm_irq_level args; |
| 329 | int rc; |
| 330 | |
| 331 | /* The KVM XICS device should be in use */ |
| 332 | assert(kernel_xics_fd != -1); |
| 333 | |
| 334 | args.irq = srcno + ics->offset; |
| 335 | if (ics->irqs[srcno].flags & XICS_FLAGS_IRQ_MSI) { |
| 336 | if (!val) { |
| 337 | return; |
| 338 | } |
| 339 | args.level = KVM_INTERRUPT_SET; |
| 340 | } else { |
| 341 | args.level = val ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET; |
| 342 | } |
| 343 | rc = kvm_vm_ioctl(kvm_state, KVM_IRQ_LINE, &args); |
| 344 | if (rc < 0) { |
| 345 | perror("kvm_irq_line"); |
| 346 | } |
| 347 | } |
| 348 | |
| 349 | int xics_kvm_connect(SpaprInterruptController *intc, uint32_t nr_servers, |
| 350 | Error **errp) |
| 351 | { |
| 352 | ICSState *ics = ICS_SPAPR(intc); |
| 353 | int rc; |
| 354 | CPUState *cs; |
| 355 | Error *local_err = NULL; |
| 356 | |
| 357 | /* |
| 358 | * The KVM XICS device already in use. This is the case when |
| 359 | * rebooting under the XICS-only interrupt mode. |
| 360 | */ |
| 361 | if (kernel_xics_fd != -1) { |
| 362 | return 0; |
| 363 | } |
| 364 | |
| 365 | if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) { |
| 366 | error_setg(errp, |
| 367 | "KVM and IRQ_XICS capability must be present for in-kernel XICS"); |
| 368 | return -1; |
| 369 | } |
| 370 | |
| 371 | rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_SET_XIVE, "ibm,set-xive"); |
| 372 | if (rc < 0) { |
| 373 | error_setg_errno(&local_err, -rc, |
| 374 | "kvmppc_define_rtas_kernel_token: ibm,set-xive"); |
| 375 | goto fail; |
| 376 | } |
| 377 | |
| 378 | rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_GET_XIVE, "ibm,get-xive"); |
| 379 | if (rc < 0) { |
| 380 | error_setg_errno(&local_err, -rc, |
| 381 | "kvmppc_define_rtas_kernel_token: ibm,get-xive"); |
| 382 | goto fail; |
| 383 | } |
| 384 | |
| 385 | rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_ON, "ibm,int-on"); |
| 386 | if (rc < 0) { |
| 387 | error_setg_errno(&local_err, -rc, |
| 388 | "kvmppc_define_rtas_kernel_token: ibm,int-on"); |
| 389 | goto fail; |
| 390 | } |
| 391 | |
| 392 | rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_OFF, "ibm,int-off"); |
| 393 | if (rc < 0) { |
| 394 | error_setg_errno(&local_err, -rc, |
| 395 | "kvmppc_define_rtas_kernel_token: ibm,int-off"); |
| 396 | goto fail; |
| 397 | } |
| 398 | |
| 399 | /* Create the KVM XICS device */ |
| 400 | rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false); |
| 401 | if (rc < 0) { |
| 402 | error_setg_errno(&local_err, -rc, "Error on KVM_CREATE_DEVICE for XICS"); |
| 403 | goto fail; |
| 404 | } |
| 405 | |
| 406 | /* Tell KVM about the # of VCPUs we may have (POWER9 and newer only) */ |
| 407 | if (kvm_device_check_attr(rc, KVM_DEV_XICS_GRP_CTRL, |
| 408 | KVM_DEV_XICS_NR_SERVERS)) { |
| 409 | if (kvm_device_access(rc, KVM_DEV_XICS_GRP_CTRL, |
| 410 | KVM_DEV_XICS_NR_SERVERS, &nr_servers, true, |
| 411 | &local_err)) { |
| 412 | goto fail; |
| 413 | } |
| 414 | } |
| 415 | |
| 416 | kernel_xics_fd = rc; |
| 417 | kvm_kernel_irqchip = true; |
| 418 | kvm_msi_via_irqfd_allowed = true; |
| 419 | kvm_gsi_direct_mapping = true; |
| 420 | |
| 421 | /* Create the presenters */ |
| 422 | CPU_FOREACH(cs) { |
| 423 | PowerPCCPU *cpu = POWERPC_CPU(cs); |
| 424 | |
| 425 | icp_kvm_realize(DEVICE(spapr_cpu_state(cpu)->icp), &local_err); |
| 426 | if (local_err) { |
| 427 | goto fail; |
| 428 | } |
| 429 | } |
| 430 | |
| 431 | /* Update the KVM sources */ |
| 432 | ics_set_kvm_state(ics, &local_err); |
| 433 | if (local_err) { |
| 434 | goto fail; |
| 435 | } |
| 436 | |
| 437 | /* Connect the presenters to the initial VCPUs of the machine */ |
| 438 | CPU_FOREACH(cs) { |
| 439 | PowerPCCPU *cpu = POWERPC_CPU(cs); |
| 440 | icp_set_kvm_state(spapr_cpu_state(cpu)->icp, &local_err); |
| 441 | if (local_err) { |
| 442 | goto fail; |
| 443 | } |
| 444 | } |
| 445 | |
| 446 | return 0; |
| 447 | |
| 448 | fail: |
| 449 | error_propagate(errp, local_err); |
| 450 | xics_kvm_disconnect(intc); |
| 451 | return -1; |
| 452 | } |
| 453 | |
| 454 | void xics_kvm_disconnect(SpaprInterruptController *intc) |
| 455 | { |
| 456 | /* |
| 457 | * Only on P9 using the XICS-on XIVE KVM device: |
| 458 | * |
| 459 | * When the KVM device fd is closed, the device is destroyed and |
| 460 | * removed from the list of devices of the VM. The VCPU presenters |
| 461 | * are also detached from the device. |
| 462 | */ |
| 463 | if (kernel_xics_fd != -1) { |
| 464 | close(kernel_xics_fd); |
| 465 | kernel_xics_fd = -1; |
| 466 | } |
| 467 | |
| 468 | kvmppc_define_rtas_kernel_token(0, "ibm,set-xive"); |
| 469 | kvmppc_define_rtas_kernel_token(0, "ibm,get-xive"); |
| 470 | kvmppc_define_rtas_kernel_token(0, "ibm,int-on"); |
| 471 | kvmppc_define_rtas_kernel_token(0, "ibm,int-off"); |
| 472 | |
| 473 | kvm_kernel_irqchip = false; |
| 474 | kvm_msi_via_irqfd_allowed = false; |
| 475 | kvm_gsi_direct_mapping = false; |
| 476 | |
| 477 | /* Clear the presenter from the VCPUs */ |
| 478 | kvm_disable_icps(); |
| 479 | } |
| 480 | |
| 481 | /* |
| 482 | * This is a heuristic to detect older KVMs on POWER9 hosts that don't |
| 483 | * support destruction of a KVM XICS device while the VM is running. |
| 484 | * Required to start a spapr machine with ic-mode=dual,kernel-irqchip=on. |
| 485 | */ |
| 486 | bool xics_kvm_has_broken_disconnect(void) |
| 487 | { |
| 488 | int rc; |
| 489 | |
| 490 | rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false); |
| 491 | if (rc < 0) { |
| 492 | /* |
| 493 | * The error is ignored on purpose. The KVM XICS setup code |
| 494 | * will catch it again anyway. The goal here is to see if |
| 495 | * close() actually destroys the device or not. |
| 496 | */ |
| 497 | return false; |
| 498 | } |
| 499 | |
| 500 | close(rc); |
| 501 | |
| 502 | rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false); |
| 503 | if (rc >= 0) { |
| 504 | close(rc); |
| 505 | return false; |
| 506 | } |
| 507 | |
| 508 | return errno == EEXIST; |
| 509 | } |