arch/powerpc/kvm/powerpc.c

   1 /*
   2  * This program is free software; you can redistribute it and/or modify
   3  * it under the terms of the GNU General Public License, version 2, as
   4  * published by the Free Software Foundation.
   5  *
   6  * This program is distributed in the hope that it will be useful,
   7  * but WITHOUT ANY WARRANTY; without even the implied warranty of
   8  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   9  * GNU General Public License for more details.
  10  *
  11  * You should have received a copy of the GNU General Public License
  12  * along with this program; if not, write to the Free Software
  13  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  14  *
  15  * Copyright IBM Corp. 2007
  16  *
  17  * Authors: Hollis Blanchard <hollisb@us.ibm.com>
  18  *          Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
  19  */
  20
  21 #include <linux/errno.h>
  22 #include <linux/err.h>
  23 #include <linux/kvm_host.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/hrtimer.h>
  26 #include <linux/fs.h>
  27 #include <linux/slab.h>
  28 #include <linux/file.h>
  29 #include <asm/cputable.h>
  30 #include <asm/uaccess.h>
  31 #include <asm/kvm_ppc.h>
  32 #include <asm/tlbflush.h>
  33 #include <asm/cputhreads.h>
  34 #include <asm/irqflags.h>
  35 #include "timing.h"
  36 #include "irq.h"
  37 #include "../mm/mmu_decl.h"
  38
  39 #define CREATE_TRACE_POINTS
  40 #include "trace.h"
  41
  42 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
  43 {
  44         return !!(v->arch.pending_exceptions) ||
  45                v->requests;
  46 }
  47
  48 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  49 {
  50         return 1;
  51 }
  52
  53 #ifndef CONFIG_KVM_BOOK3S_64_HV
  54 /*
  55  * Common checks before entering the guest world.  Call with interrupts
  56  * disabled.
  57  *
  58  * returns:
  59  *
  60  * == 1 if we're ready to go into guest state
  61  * <= 0 if we need to go back to the host with return value
  62  */
  63 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
  64 {
  65         int r = 1;
  66
  67         WARN_ON_ONCE(!irqs_disabled());
  68         while (true) {
  69                 if (need_resched()) {
  70                         local_irq_enable();
  71                         cond_resched();
  72                         local_irq_disable();
  73                         continue;
  74                 }
  75
  76                 if (signal_pending(current)) {
  77                         kvmppc_account_exit(vcpu, SIGNAL_EXITS);
  78                         vcpu->run->exit_reason = KVM_EXIT_INTR;
  79                         r = -EINTR;
  80                         break;
  81                 }
  82
  83                 vcpu->mode = IN_GUEST_MODE;
  84
  85                 /*
  86                  * Reading vcpu->requests must happen after setting vcpu->mode,
  87                  * so we don't miss a request because the requester sees
  88                  * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
  89                  * before next entering the guest (and thus doesn't IPI).
  90                  */
  91                 smp_mb();
  92
  93                 if (vcpu->requests) {
  94                         /* Make sure we process requests preemptable */
  95                         local_irq_enable();
  96                         trace_kvm_check_requests(vcpu);
  97                         r = kvmppc_core_check_requests(vcpu);
  98                         local_irq_disable();
  99                         if (r > 0)
 100                                 continue;
 101                         break;
 102                 }
 103
 104                 if (kvmppc_core_prepare_to_enter(vcpu)) {
 105                         /* interrupts got enabled in between, so we
 106                            are back at square 1 */
 107                         continue;
 108                 }
 109
 110 #ifdef CONFIG_PPC64
 111                 /* lazy EE magic */
 112                 hard_irq_disable();
 113                 if (lazy_irq_pending()) {
 114                         /* Got an interrupt in between, try again */
 115                         local_irq_enable();
 116                         local_irq_disable();
 117                         kvm_guest_exit();
 118                         continue;
 119                 }
 120
 121                 trace_hardirqs_on();
 122 #endif
 123
 124                 kvm_guest_enter();
 125                 break;
 126         }
 127
 128         return r;
 129 }
 130 #endif /* CONFIG_KVM_BOOK3S_64_HV */
 131
 132 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
 133 {
 134         int nr = kvmppc_get_gpr(vcpu, 11);
 135         int r;
 136         unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
 137         unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
 138         unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
 139         unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
 140         unsigned long r2 = 0;
 141
 142         if (!(vcpu->arch.shared->msr & MSR_SF)) {
 143                 /* 32 bit mode */
 144                 param1 &= 0xffffffff;
 145                 param2 &= 0xffffffff;
 146                 param3 &= 0xffffffff;
 147                 param4 &= 0xffffffff;
 148         }
 149
 150         switch (nr) {
 151         case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
 152         {
 153                 vcpu->arch.magic_page_pa = param1;
 154                 vcpu->arch.magic_page_ea = param2;
 155
 156                 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
 157
 158                 r = EV_SUCCESS;
 159                 break;
 160         }
 161         case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
 162                 r = EV_SUCCESS;
 163 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
 164                 /* XXX Missing magic page on 44x */
 165                 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
 166 #endif
 167
 168                 /* Second return value is in r4 */
 169                 break;
 170         case EV_HCALL_TOKEN(EV_IDLE):
 171                 r = EV_SUCCESS;
 172                 kvm_vcpu_block(vcpu);
 173                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 174                 break;
 175         default:
 176                 r = EV_UNIMPLEMENTED;
 177                 break;
 178         }
 179
 180         kvmppc_set_gpr(vcpu, 4, r2);
 181
 182         return r;
 183 }
 184
 185 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
 186 {
 187         int r = false;
 188
 189         /* We have to know what CPU to virtualize */
 190         if (!vcpu->arch.pvr)
 191                 goto out;
 192
 193         /* PAPR only works with book3s_64 */
 194         if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
 195                 goto out;
 196
 197 #ifdef CONFIG_KVM_BOOK3S_64_HV
 198         /* HV KVM can only do PAPR mode for now */
 199         if (!vcpu->arch.papr_enabled)
 200                 goto out;
 201 #endif
 202
 203 #ifdef CONFIG_KVM_BOOKE_HV
 204         if (!cpu_has_feature(CPU_FTR_EMB_HV))
 205                 goto out;
 206 #endif
 207
 208         r = true;
 209
 210 out:
 211         vcpu->arch.sane = r;
 212         return r ? 0 : -EINVAL;
 213 }
 214
 215 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
 216 {
 217         enum emulation_result er;
 218         int r;
 219
 220         er = kvmppc_emulate_instruction(run, vcpu);
 221         switch (er) {
 222         case EMULATE_DONE:
 223                 /* Future optimization: only reload non-volatiles if they were
 224                  * actually modified. */
 225                 r = RESUME_GUEST_NV;
 226                 break;
 227         case EMULATE_DO_MMIO:
 228                 run->exit_reason = KVM_EXIT_MMIO;
 229                 /* We must reload nonvolatiles because "update" load/store
 230                  * instructions modify register state. */
 231                 /* Future optimization: only reload non-volatiles if they were
 232                  * actually modified. */
 233                 r = RESUME_HOST_NV;
 234                 break;
 235         case EMULATE_FAIL:
 236                 /* XXX Deliver Program interrupt to guest. */
 237                 printk(KERN_EMERG "%s: emulation failed (%08x)\n", __func__,
 238                        kvmppc_get_last_inst(vcpu));
 239                 r = RESUME_HOST;
 240                 break;
 241         default:
 242                 WARN_ON(1);
 243                 r = RESUME_GUEST;
 244         }
 245
 246         return r;
 247 }
 248
 249 int kvm_arch_hardware_enable(void)
 250 {
 251         return 0;
 252 }
 253
 254 int kvm_arch_hardware_setup(void)
 255 {
 256         return 0;
 257 }
 258
 259 void kvm_arch_check_processor_compat(void *rtn)
 260 {
 261         *(int *)rtn = kvmppc_core_check_processor_compat();
 262 }
 263
 264 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 265 {
 266         if (type)
 267                 return -EINVAL;
 268
 269         return kvmppc_core_init_vm(kvm);
 270 }
 271
 272 void kvm_arch_destroy_vm(struct kvm *kvm)
 273 {
 274         unsigned int i;
 275         struct kvm_vcpu *vcpu;
 276
 277         kvm_for_each_vcpu(i, vcpu, kvm)
 278                 kvm_arch_vcpu_free(vcpu);
 279
 280         mutex_lock(&kvm->lock);
 281         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
 282                 kvm->vcpus[i] = NULL;
 283
 284         atomic_set(&kvm->online_vcpus, 0);
 285
 286         kvmppc_core_destroy_vm(kvm);
 287
 288         mutex_unlock(&kvm->lock);
 289 }
 290
 291 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 292 {
 293         int r;
 294
 295         switch (ext) {
 296 #ifdef CONFIG_BOOKE
 297         case KVM_CAP_PPC_BOOKE_SREGS:
 298         case KVM_CAP_PPC_BOOKE_WATCHDOG:
 299         case KVM_CAP_PPC_EPR:
 300 #else
 301         case KVM_CAP_PPC_SEGSTATE:
 302         case KVM_CAP_PPC_HIOR:
 303         case KVM_CAP_PPC_PAPR:
 304 #endif
 305         case KVM_CAP_PPC_UNSET_IRQ:
 306         case KVM_CAP_PPC_IRQ_LEVEL:
 307         case KVM_CAP_ENABLE_CAP:
 308         case KVM_CAP_ONE_REG:
 309         case KVM_CAP_IOEVENTFD:
 310         case KVM_CAP_DEVICE_CTRL:
 311                 r = 1;
 312                 break;
 313 #ifndef CONFIG_KVM_BOOK3S_64_HV
 314         case KVM_CAP_PPC_PAIRED_SINGLES:
 315         case KVM_CAP_PPC_OSI:
 316         case KVM_CAP_PPC_GET_PVINFO:
 317 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 318         case KVM_CAP_SW_TLB:
 319 #endif
 320 #ifdef CONFIG_KVM_MPIC
 321         case KVM_CAP_IRQ_MPIC:
 322 #endif
 323                 r = 1;
 324                 break;
 325         case KVM_CAP_COALESCED_MMIO:
 326                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 327                 break;
 328 #endif
 329 #ifdef CONFIG_PPC_BOOK3S_64
 330         case KVM_CAP_SPAPR_TCE:
 331         case KVM_CAP_PPC_ALLOC_HTAB:
 332         case KVM_CAP_PPC_RTAS:
 333 #ifdef CONFIG_KVM_XICS
 334         case KVM_CAP_IRQ_XICS:
 335 #endif
 336                 r = 1;
 337                 break;
 338 #endif /* CONFIG_PPC_BOOK3S_64 */
 339 #ifdef CONFIG_KVM_BOOK3S_64_HV
 340         case KVM_CAP_PPC_SMT:
 341                 r = threads_per_core;
 342                 break;
 343         case KVM_CAP_PPC_RMA:
 344                 r = 1;
 345                 /* PPC970 requires an RMA */
 346                 if (cpu_has_feature(CPU_FTR_ARCH_201))
 347                         r = 2;
 348                 break;
 349 #endif
 350         case KVM_CAP_SYNC_MMU:
 351 #ifdef CONFIG_KVM_BOOK3S_64_HV
 352                 r = cpu_has_feature(CPU_FTR_ARCH_206) ? 1 : 0;
 353 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 354                 r = 1;
 355 #else
 356                 r = 0;
 357                 break;
 358 #endif
 359 #ifdef CONFIG_KVM_BOOK3S_64_HV
 360         case KVM_CAP_PPC_HTAB_FD:
 361                 r = 1;
 362                 break;
 363 #endif
 364                 break;
 365         case KVM_CAP_NR_VCPUS:
 366                 /*
 367                  * Recommending a number of CPUs is somewhat arbitrary; we
 368                  * return the number of present CPUs for -HV (since a host
 369                  * will have secondary threads "offline"), and for other KVM
 370                  * implementations just count online CPUs.
 371                  */
 372 #ifdef CONFIG_KVM_BOOK3S_64_HV
 373                 r = num_present_cpus();
 374 #else
 375                 r = num_online_cpus();
 376 #endif
 377                 break;
 378         case KVM_CAP_MAX_VCPUS:
 379                 r = KVM_MAX_VCPUS;
 380                 break;
 381 #ifdef CONFIG_PPC_BOOK3S_64
 382         case KVM_CAP_PPC_GET_SMMU_INFO:
 383                 r = 1;
 384                 break;
 385 #endif
 386         default:
 387                 r = 0;
 388                 break;
 389         }
 390         return r;
 391
 392 }
 393
 394 long kvm_arch_dev_ioctl(struct file *filp,
 395                         unsigned int ioctl, unsigned long arg)
 396 {
 397         return -EINVAL;
 398 }
 399
 400 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 401                            struct kvm_memory_slot *dont)
 402 {
 403         kvmppc_core_free_memslot(kvm, free, dont);
 404 }
 405
 406 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
 407                             unsigned long npages)
 408 {
 409         return kvmppc_core_create_memslot(kvm, slot, npages);
 410 }
 411
 412 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 413                                    struct kvm_memory_slot *memslot,
 414                                    struct kvm_userspace_memory_region *mem,
 415                                    enum kvm_mr_change change)
 416 {
 417         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
 418 }
 419
 420 void kvm_arch_commit_memory_region(struct kvm *kvm,
 421                                    struct kvm_userspace_memory_region *mem,
 422                                    const struct kvm_memory_slot *old,
 423                                    enum kvm_mr_change change)
 424 {
 425         kvmppc_core_commit_memory_region(kvm, mem, old);
 426 }
 427
 428 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 429                                    struct kvm_memory_slot *slot)
 430 {
 431         kvmppc_core_flush_memslot(kvm, slot);
 432 }
 433
 434 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 435 {
 436         struct kvm_vcpu *vcpu;
 437         vcpu = kvmppc_core_vcpu_create(kvm, id);
 438         if (!IS_ERR(vcpu)) {
 439                 vcpu->arch.wqp = &vcpu->wq;
 440                 kvmppc_create_vcpu_debugfs(vcpu, id);
 441         }
 442         return vcpu;
 443 }
 444
 445 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 446 {
 447         return 0;
 448 }
 449
 450 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 451 {
 452         /* Make sure we're not using the vcpu anymore */
 453         hrtimer_cancel(&vcpu->arch.dec_timer);
 454         tasklet_kill(&vcpu->arch.tasklet);
 455
 456         kvmppc_remove_vcpu_debugfs(vcpu);
 457
 458         switch (vcpu->arch.irq_type) {
 459         case KVMPPC_IRQ_MPIC:
 460                 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
 461                 break;
 462         case KVMPPC_IRQ_XICS:
 463                 kvmppc_xics_free_icp(vcpu);
 464                 break;
 465         }
 466
 467         kvmppc_core_vcpu_free(vcpu);
 468 }
 469
 470 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 471 {
 472         kvm_arch_vcpu_free(vcpu);
 473 }
 474
 475 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 476 {
 477         return kvmppc_core_pending_dec(vcpu);
 478 }
 479
 480 /*
 481  * low level hrtimer wake routine. Because this runs in hardirq context
 482  * we schedule a tasklet to do the real work.
 483  */
 484 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
 485 {
 486         struct kvm_vcpu *vcpu;
 487
 488         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
 489         tasklet_schedule(&vcpu->arch.tasklet);
 490
 491         return HRTIMER_NORESTART;
 492 }
 493
 494 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 495 {
 496         int ret;
 497
 498         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
 499         tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
 500         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
 501         vcpu->arch.dec_expires = ~(u64)0;
 502
 503 #ifdef CONFIG_KVM_EXIT_TIMING
 504         mutex_init(&vcpu->arch.exit_timing_lock);
 505 #endif
 506         ret = kvmppc_subarch_vcpu_init(vcpu);
 507         return ret;
 508 }
 509
 510 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 511 {
 512         kvmppc_mmu_destroy(vcpu);
 513         kvmppc_subarch_vcpu_uninit(vcpu);
 514 }
 515
 516 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 517 {
 518 #ifdef CONFIG_BOOKE
 519         /*
 520          * vrsave (formerly usprg0) isn't used by Linux, but may
 521          * be used by the guest.
 522          *
 523          * On non-booke this is associated with Altivec and
 524          * is handled by code in book3s.c.
 525          */
 526         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
 527 #endif
 528         kvmppc_core_vcpu_load(vcpu, cpu);
 529 }
 530
 531 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 532 {
 533         kvmppc_core_vcpu_put(vcpu);
 534 #ifdef CONFIG_BOOKE
 535         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
 536 #endif
 537 }
 538
 539 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
 540                                      struct kvm_run *run)
 541 {
 542         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
 543 }
 544
 545 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
 546                                       struct kvm_run *run)
 547 {
 548         u64 uninitialized_var(gpr);
 549
 550         if (run->mmio.len > sizeof(gpr)) {
 551                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
 552                 return;
 553         }
 554
 555         if (vcpu->arch.mmio_is_bigendian) {
 556                 switch (run->mmio.len) {
 557                 case 8: gpr = *(u64 *)run->mmio.data; break;
 558                 case 4: gpr = *(u32 *)run->mmio.data; break;
 559                 case 2: gpr = *(u16 *)run->mmio.data; break;
 560                 case 1: gpr = *(u8 *)run->mmio.data; break;
 561                 }
 562         } else {
 563                 /* Convert BE data from userland back to LE. */
 564                 switch (run->mmio.len) {
 565                 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
 566                 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
 567                 case 1: gpr = *(u8 *)run->mmio.data; break;
 568                 }
 569         }
 570
 571         if (vcpu->arch.mmio_sign_extend) {
 572                 switch (run->mmio.len) {
 573 #ifdef CONFIG_PPC64
 574                 case 4:
 575                         gpr = (s64)(s32)gpr;
 576                         break;
 577 #endif
 578                 case 2:
 579                         gpr = (s64)(s16)gpr;
 580                         break;
 581                 case 1:
 582                         gpr = (s64)(s8)gpr;
 583                         break;
 584                 }
 585         }
 586
 587         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 588
 589         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
 590         case KVM_MMIO_REG_GPR:
 591                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 592                 break;
 593         case KVM_MMIO_REG_FPR:
 594                 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 595                 break;
 596 #ifdef CONFIG_PPC_BOOK3S
 597         case KVM_MMIO_REG_QPR:
 598                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 599                 break;
 600         case KVM_MMIO_REG_FQPR:
 601                 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 602                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 603                 break;
 604 #endif
 605         default:
 606                 BUG();
 607         }
 608 }
 609
 610 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
 611                        unsigned int rt, unsigned int bytes, int is_bigendian)
 612 {
 613         int idx, ret;
 614
 615         if (bytes > sizeof(run->mmio.data)) {
 616                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 617                        run->mmio.len);
 618         }
 619
 620         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 621         run->mmio.len = bytes;
 622         run->mmio.is_write = 0;
 623
 624         vcpu->arch.io_gpr = rt;
 625         vcpu->arch.mmio_is_bigendian = is_bigendian;
 626         vcpu->mmio_needed = 1;
 627         vcpu->mmio_is_write = 0;
 628         vcpu->arch.mmio_sign_extend = 0;
 629
 630         idx = srcu_read_lock(&vcpu->kvm->srcu);
 631
 632         ret = kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
 633                               bytes, &run->mmio.data);
 634
 635         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 636
 637         if (!ret) {
 638                 kvmppc_complete_mmio_load(vcpu, run);
 639                 vcpu->mmio_needed = 0;
 640                 return EMULATE_DONE;
 641         }
 642
 643         return EMULATE_DO_MMIO;
 644 }
 645
 646 /* Same as above, but sign extends */
 647 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
 648                         unsigned int rt, unsigned int bytes, int is_bigendian)
 649 {
 650         int r;
 651
 652         vcpu->arch.mmio_sign_extend = 1;
 653         r = kvmppc_handle_load(run, vcpu, rt, bytes, is_bigendian);
 654
 655         return r;
 656 }
 657
 658 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
 659                         u64 val, unsigned int bytes, int is_bigendian)
 660 {
 661         void *data = run->mmio.data;
 662         int idx, ret;
 663
 664         if (bytes > sizeof(run->mmio.data)) {
 665                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 666                        run->mmio.len);
 667         }
 668
 669         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 670         run->mmio.len = bytes;
 671         run->mmio.is_write = 1;
 672         vcpu->mmio_needed = 1;
 673         vcpu->mmio_is_write = 1;
 674
 675         /* Store the value at the lowest bytes in 'data'. */
 676         if (is_bigendian) {
 677                 switch (bytes) {
 678                 case 8: *(u64 *)data = val; break;
 679                 case 4: *(u32 *)data = val; break;
 680                 case 2: *(u16 *)data = val; break;
 681                 case 1: *(u8  *)data = val; break;
 682                 }
 683         } else {
 684                 /* Store LE value into 'data'. */
 685                 switch (bytes) {
 686                 case 4: st_le32(data, val); break;
 687                 case 2: st_le16(data, val); break;
 688                 case 1: *(u8 *)data = val; break;
 689                 }
 690         }
 691
 692         idx = srcu_read_lock(&vcpu->kvm->srcu);
 693
 694         ret = kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
 695                                bytes, &run->mmio.data);
 696
 697         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 698
 699         if (!ret) {
 700                 vcpu->mmio_needed = 0;
 701                 return EMULATE_DONE;
 702         }
 703
 704         return EMULATE_DO_MMIO;
 705 }
 706
 707 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 708 {
 709         int r;
 710         sigset_t sigsaved;
 711
 712         if (vcpu->sigset_active)
 713                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 714
 715         if (vcpu->mmio_needed) {
 716                 if (!vcpu->mmio_is_write)
 717                         kvmppc_complete_mmio_load(vcpu, run);
 718                 vcpu->mmio_needed = 0;
 719         } else if (vcpu->arch.dcr_needed) {
 720                 if (!vcpu->arch.dcr_is_write)
 721                         kvmppc_complete_dcr_load(vcpu, run);
 722                 vcpu->arch.dcr_needed = 0;
 723         } else if (vcpu->arch.osi_needed) {
 724                 u64 *gprs = run->osi.gprs;
 725                 int i;
 726
 727                 for (i = 0; i < 32; i++)
 728                         kvmppc_set_gpr(vcpu, i, gprs[i]);
 729                 vcpu->arch.osi_needed = 0;
 730         } else if (vcpu->arch.hcall_needed) {
 731                 int i;
 732
 733                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
 734                 for (i = 0; i < 9; ++i)
 735                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
 736                 vcpu->arch.hcall_needed = 0;
 737 #ifdef CONFIG_BOOKE
 738         } else if (vcpu->arch.epr_needed) {
 739                 kvmppc_set_epr(vcpu, run->epr.epr);
 740                 vcpu->arch.epr_needed = 0;
 741 #endif
 742         }
 743
 744         r = kvmppc_vcpu_run(run, vcpu);
 745
 746         if (vcpu->sigset_active)
 747                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 748
 749         return r;
 750 }
 751
 752 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
 753 {
 754         if (irq->irq == KVM_INTERRUPT_UNSET) {
 755                 kvmppc_core_dequeue_external(vcpu);
 756                 return 0;
 757         }
 758
 759         kvmppc_core_queue_external(vcpu, irq);
 760
 761         kvm_vcpu_kick(vcpu);
 762
 763         return 0;
 764 }
 765
 766 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
 767                                      struct kvm_enable_cap *cap)
 768 {
 769         int r;
 770
 771         if (cap->flags)
 772                 return -EINVAL;
 773
 774         switch (cap->cap) {
 775         case KVM_CAP_PPC_OSI:
 776                 r = 0;
 777                 vcpu->arch.osi_enabled = true;
 778                 break;
 779         case KVM_CAP_PPC_PAPR:
 780                 r = 0;
 781                 vcpu->arch.papr_enabled = true;
 782                 break;
 783         case KVM_CAP_PPC_EPR:
 784                 r = 0;
 785                 if (cap->args[0])
 786                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
 787                 else
 788                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
 789                 break;
 790 #ifdef CONFIG_BOOKE
 791         case KVM_CAP_PPC_BOOKE_WATCHDOG:
 792                 r = 0;
 793                 vcpu->arch.watchdog_enabled = true;
 794                 break;
 795 #endif
 796 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 797         case KVM_CAP_SW_TLB: {
 798                 struct kvm_config_tlb cfg;
 799                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
 800
 801                 r = -EFAULT;
 802                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
 803                         break;
 804
 805                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
 806                 break;
 807         }
 808 #endif
 809 #ifdef CONFIG_KVM_MPIC
 810         case KVM_CAP_IRQ_MPIC: {
 811                 struct file *filp;
 812                 struct kvm_device *dev;
 813
 814                 r = -EBADF;
 815                 filp = fget(cap->args[0]);
 816                 if (!filp)
 817                         break;
 818
 819                 r = -EPERM;
 820                 dev = kvm_device_from_filp(filp);
 821                 if (dev)
 822                         r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
 823
 824                 fput(filp);
 825                 break;
 826         }
 827 #endif
 828 #ifdef CONFIG_KVM_XICS
 829         case KVM_CAP_IRQ_XICS: {
 830                 struct file *filp;
 831                 struct kvm_device *dev;
 832
 833                 r = -EBADF;
 834                 filp = fget(cap->args[0]);
 835                 if (!filp)
 836                         break;
 837
 838                 r = -EPERM;
 839                 dev = kvm_device_from_filp(filp);
 840                 if (dev)
 841                         r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
 842
 843                 fput(filp);
 844                 break;
 845         }
 846 #endif /* CONFIG_KVM_XICS */
 847         default:
 848                 r = -EINVAL;
 849                 break;
 850         }
 851
 852         if (!r)
 853                 r = kvmppc_sanity_check(vcpu);
 854
 855         return r;
 856 }
 857
 858 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 859                                     struct kvm_mp_state *mp_state)
 860 {
 861         return -EINVAL;
 862 }
 863
 864 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 865                                     struct kvm_mp_state *mp_state)
 866 {
 867         return -EINVAL;
 868 }
 869
 870 long kvm_arch_vcpu_ioctl(struct file *filp,
 871                          unsigned int ioctl, unsigned long arg)
 872 {
 873         struct kvm_vcpu *vcpu = filp->private_data;
 874         void __user *argp = (void __user *)arg;
 875         long r;
 876
 877         switch (ioctl) {
 878         case KVM_INTERRUPT: {
 879                 struct kvm_interrupt irq;
 880                 r = -EFAULT;
 881                 if (copy_from_user(&irq, argp, sizeof(irq)))
 882                         goto out;
 883                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
 884                 goto out;
 885         }
 886
 887         case KVM_ENABLE_CAP:
 888         {
 889                 struct kvm_enable_cap cap;
 890                 r = -EFAULT;
 891                 if (copy_from_user(&cap, argp, sizeof(cap)))
 892                         goto out;
 893                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
 894                 break;
 895         }
 896
 897         case KVM_SET_ONE_REG:
 898         case KVM_GET_ONE_REG:
 899         {
 900                 struct kvm_one_reg reg;
 901                 r = -EFAULT;
 902                 if (copy_from_user(&reg, argp, sizeof(reg)))
 903                         goto out;
 904                 if (ioctl == KVM_SET_ONE_REG)
 905                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
 906                 else
 907                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
 908                 break;
 909         }
 910
 911 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 912         case KVM_DIRTY_TLB: {
 913                 struct kvm_dirty_tlb dirty;
 914                 r = -EFAULT;
 915                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
 916                         goto out;
 917                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
 918                 break;
 919         }
 920 #endif
 921         default:
 922                 r = -EINVAL;
 923         }
 924
 925 out:
 926         return r;
 927 }
 928
 929 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 930 {
 931         return VM_FAULT_SIGBUS;
 932 }
 933
 934 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
 935 {
 936         u32 inst_nop = 0x60000000;
 937 #ifdef CONFIG_KVM_BOOKE_HV
 938         u32 inst_sc1 = 0x44000022;
 939         pvinfo->hcall[0] = inst_sc1;
 940         pvinfo->hcall[1] = inst_nop;
 941         pvinfo->hcall[2] = inst_nop;
 942         pvinfo->hcall[3] = inst_nop;
 943 #else
 944         u32 inst_lis = 0x3c000000;
 945         u32 inst_ori = 0x60000000;
 946         u32 inst_sc = 0x44000002;
 947         u32 inst_imm_mask = 0xffff;
 948
 949         /*
 950          * The hypercall to get into KVM from within guest context is as
 951          * follows:
 952          *
 953          *    lis r0, r0, KVM_SC_MAGIC_R0@h
 954          *    ori r0, KVM_SC_MAGIC_R0@l
 955          *    sc
 956          *    nop
 957          */
 958         pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask);
 959         pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask);
 960         pvinfo->hcall[2] = inst_sc;
 961         pvinfo->hcall[3] = inst_nop;
 962 #endif
 963
 964         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
 965
 966         return 0;
 967 }
 968
 969 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
 970                           bool line_status)
 971 {
 972         if (!irqchip_in_kernel(kvm))
 973                 return -ENXIO;
 974
 975         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
 976                                         irq_event->irq, irq_event->level,
 977                                         line_status);
 978         return 0;
 979 }
 980
 981 long kvm_arch_vm_ioctl(struct file *filp,
 982                        unsigned int ioctl, unsigned long arg)
 983 {
 984         struct kvm *kvm __maybe_unused = filp->private_data;
 985         void __user *argp = (void __user *)arg;
 986         long r;
 987
 988         switch (ioctl) {
 989         case KVM_PPC_GET_PVINFO: {
 990                 struct kvm_ppc_pvinfo pvinfo;
 991                 memset(&pvinfo, 0, sizeof(pvinfo));
 992                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
 993                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
 994                         r = -EFAULT;
 995                         goto out;
 996                 }
 997
 998                 break;
 999         }
1000 #ifdef CONFIG_PPC_BOOK3S_64
1001         case KVM_CREATE_SPAPR_TCE: {
1002                 struct kvm_create_spapr_tce create_tce;
1003
1004                 r = -EFAULT;
1005                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1006                         goto out;
1007                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
1008                 goto out;
1009         }
1010 #endif /* CONFIG_PPC_BOOK3S_64 */
1011
1012 #ifdef CONFIG_KVM_BOOK3S_64_HV
1013         case KVM_ALLOCATE_RMA: {
1014                 struct kvm_allocate_rma rma;
1015                 struct kvm *kvm = filp->private_data;
1016
1017                 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
1018                 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
1019                         r = -EFAULT;
1020                 break;
1021         }
1022
1023         case KVM_PPC_ALLOCATE_HTAB: {
1024                 u32 htab_order;
1025
1026                 r = -EFAULT;
1027                 if (get_user(htab_order, (u32 __user *)argp))
1028                         break;
1029                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
1030                 if (r)
1031                         break;
1032                 r = -EFAULT;
1033                 if (put_user(htab_order, (u32 __user *)argp))
1034                         break;
1035                 r = 0;
1036                 break;
1037         }
1038
1039         case KVM_PPC_GET_HTAB_FD: {
1040                 struct kvm_get_htab_fd ghf;
1041
1042                 r = -EFAULT;
1043                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
1044                         break;
1045                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
1046                 break;
1047         }
1048 #endif /* CONFIG_KVM_BOOK3S_64_HV */
1049
1050 #ifdef CONFIG_PPC_BOOK3S_64
1051         case KVM_PPC_GET_SMMU_INFO: {
1052                 struct kvm_ppc_smmu_info info;
1053
1054                 memset(&info, 0, sizeof(info));
1055                 r = kvm_vm_ioctl_get_smmu_info(kvm, &info);
1056                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1057                         r = -EFAULT;
1058                 break;
1059         }
1060         case KVM_PPC_RTAS_DEFINE_TOKEN: {
1061                 struct kvm *kvm = filp->private_data;
1062
1063                 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1064                 break;
1065         }
1066 #endif /* CONFIG_PPC_BOOK3S_64 */
1067         default:
1068                 r = -ENOTTY;
1069         }
1070
1071 out:
1072         return r;
1073 }
1074
1075 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1076 static unsigned long nr_lpids;
1077
1078 long kvmppc_alloc_lpid(void)
1079 {
1080         long lpid;
1081
1082         do {
1083                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1084                 if (lpid >= nr_lpids) {
1085                         pr_err("%s: No LPIDs free\n", __func__);
1086                         return -ENOMEM;
1087                 }
1088         } while (test_and_set_bit(lpid, lpid_inuse));
1089
1090         return lpid;
1091 }
1092
1093 void kvmppc_claim_lpid(long lpid)
1094 {
1095         set_bit(lpid, lpid_inuse);
1096 }
1097
1098 void kvmppc_free_lpid(long lpid)
1099 {
1100         clear_bit(lpid, lpid_inuse);
1101 }
1102
1103 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1104 {
1105         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1106         memset(lpid_inuse, 0, sizeof(lpid_inuse));
1107 }
1108
1109 int kvm_arch_init(void *opaque)
1110 {
1111         return 0;
1112 }