arch/powerpc/kvm/book3s_hv.c

   1 /*
   2  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
   3  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
   4  *
   5  * Authors:
   6  *    Paul Mackerras <paulus@au1.ibm.com>
   7  *    Alexander Graf <agraf@suse.de>
   8  *    Kevin Wolf <mail@kevin-wolf.de>
   9  *
  10  * Description: KVM functions specific to running on Book 3S
  11  * processors in hypervisor mode (specifically POWER7 and later).
  12  *
  13  * This file is derived from arch/powerpc/kvm/book3s.c,
  14  * by Alexander Graf <agraf@suse.de>.
  15  *
  16  * This program is free software; you can redistribute it and/or modify
  17  * it under the terms of the GNU General Public License, version 2, as
  18  * published by the Free Software Foundation.
  19  */
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/err.h>
  23 #include <linux/slab.h>
  24 #include <linux/preempt.h>
  25 #include <linux/sched.h>
  26 #include <linux/delay.h>
  27 #include <linux/export.h>
  28 #include <linux/fs.h>
  29 #include <linux/anon_inodes.h>
  30 #include <linux/cpumask.h>
  31 #include <linux/spinlock.h>
  32 #include <linux/page-flags.h>
  33
  34 #include <asm/reg.h>
  35 #include <asm/cputable.h>
  36 #include <asm/cacheflush.h>
  37 #include <asm/tlbflush.h>
  38 #include <asm/uaccess.h>
  39 #include <asm/io.h>
  40 #include <asm/kvm_ppc.h>
  41 #include <asm/kvm_book3s.h>
  42 #include <asm/mmu_context.h>
  43 #include <asm/lppaca.h>
  44 #include <asm/processor.h>
  45 #include <asm/cputhreads.h>
  46 #include <asm/page.h>
  47 #include <asm/hvcall.h>
  48 #include <linux/gfp.h>
  49 #include <linux/sched.h>
  50 #include <linux/vmalloc.h>
  51 #include <linux/highmem.h>
  52 #include <linux/hugetlb.h>
  53
  54 /* #define EXIT_DEBUG */
  55 /* #define EXIT_DEBUG_SIMPLE */
  56 /* #define EXIT_DEBUG_INT */
  57
  58 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
  59 static int kvmppc_hv_setup_rma(struct kvm_vcpu *vcpu);
  60
  61 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
  62 {
  63         local_paca->kvm_hstate.kvm_vcpu = vcpu;
  64         local_paca->kvm_hstate.kvm_vcore = vcpu->arch.vcore;
  65 }
  66
  67 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
  68 {
  69 }
  70
  71 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
  72 {
  73         vcpu->arch.shregs.msr = msr;
  74         kvmppc_end_cede(vcpu);
  75 }
  76
  77 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
  78 {
  79         vcpu->arch.pvr = pvr;
  80 }
  81
  82 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
  83 {
  84         int r;
  85
  86         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
  87         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
  88                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
  89         for (r = 0; r < 16; ++r)
  90                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
  91                        r, kvmppc_get_gpr(vcpu, r),
  92                        r+16, kvmppc_get_gpr(vcpu, r+16));
  93         pr_err("ctr = %.16lx  lr  = %.16lx\n",
  94                vcpu->arch.ctr, vcpu->arch.lr);
  95         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
  96                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
  97         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
  98                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
  99         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 100                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 101         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 102                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 103         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 104         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 105                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 106         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 107         for (r = 0; r < vcpu->arch.slb_max; ++r)
 108                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 109                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 110         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 111                vcpu->kvm->arch.lpcr, vcpu->kvm->arch.sdr1,
 112                vcpu->arch.last_inst);
 113 }
 114
 115 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 116 {
 117         int r;
 118         struct kvm_vcpu *v, *ret = NULL;
 119
 120         mutex_lock(&kvm->lock);
 121         kvm_for_each_vcpu(r, v, kvm) {
 122                 if (v->vcpu_id == id) {
 123                         ret = v;
 124                         break;
 125                 }
 126         }
 127         mutex_unlock(&kvm->lock);
 128         return ret;
 129 }
 130
 131 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 132 {
 133         vpa->shared_proc = 1;
 134         vpa->yield_count = 1;
 135 }
 136
 137 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 138                                        unsigned long flags,
 139                                        unsigned long vcpuid, unsigned long vpa)
 140 {
 141         struct kvm *kvm = vcpu->kvm;
 142         unsigned long len, nb;
 143         void *va;
 144         struct kvm_vcpu *tvcpu;
 145         int err = H_PARAMETER;
 146
 147         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 148         if (!tvcpu)
 149                 return H_PARAMETER;
 150
 151         flags >>= 63 - 18;
 152         flags &= 7;
 153         if (flags == 0 || flags == 4)
 154                 return H_PARAMETER;
 155         if (flags < 4) {
 156                 if (vpa & 0x7f)
 157                         return H_PARAMETER;
 158                 if (flags >= 2 && !tvcpu->arch.vpa)
 159                         return H_RESOURCE;
 160                 /* registering new area; convert logical addr to real */
 161                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 162                 if (va == NULL)
 163                         return H_PARAMETER;
 164                 if (flags <= 1)
 165                         len = *(unsigned short *)(va + 4);
 166                 else
 167                         len = *(unsigned int *)(va + 4);
 168                 if (len > nb)
 169                         goto out_unpin;
 170                 switch (flags) {
 171                 case 1:         /* register VPA */
 172                         if (len < 640)
 173                                 goto out_unpin;
 174                         if (tvcpu->arch.vpa)
 175                                 kvmppc_unpin_guest_page(kvm, vcpu->arch.vpa);
 176                         tvcpu->arch.vpa = va;
 177                         init_vpa(vcpu, va);
 178                         break;
 179                 case 2:         /* register DTL */
 180                         if (len < 48)
 181                                 goto out_unpin;
 182                         len -= len % 48;
 183                         if (tvcpu->arch.dtl)
 184                                 kvmppc_unpin_guest_page(kvm, vcpu->arch.dtl);
 185                         tvcpu->arch.dtl = va;
 186                         tvcpu->arch.dtl_end = va + len;
 187                         break;
 188                 case 3:         /* register SLB shadow buffer */
 189                         if (len < 16)
 190                                 goto out_unpin;
 191                         if (tvcpu->arch.slb_shadow)
 192                                 kvmppc_unpin_guest_page(kvm, vcpu->arch.slb_shadow);
 193                         tvcpu->arch.slb_shadow = va;
 194                         break;
 195                 }
 196         } else {
 197                 switch (flags) {
 198                 case 5:         /* unregister VPA */
 199                         if (tvcpu->arch.slb_shadow || tvcpu->arch.dtl)
 200                                 return H_RESOURCE;
 201                         if (!tvcpu->arch.vpa)
 202                                 break;
 203                         kvmppc_unpin_guest_page(kvm, tvcpu->arch.vpa);
 204                         tvcpu->arch.vpa = NULL;
 205                         break;
 206                 case 6:         /* unregister DTL */
 207                         if (!tvcpu->arch.dtl)
 208                                 break;
 209                         kvmppc_unpin_guest_page(kvm, tvcpu->arch.dtl);
 210                         tvcpu->arch.dtl = NULL;
 211                         break;
 212                 case 7:         /* unregister SLB shadow buffer */
 213                         if (!tvcpu->arch.slb_shadow)
 214                                 break;
 215                         kvmppc_unpin_guest_page(kvm, tvcpu->arch.slb_shadow);
 216                         tvcpu->arch.slb_shadow = NULL;
 217                         break;
 218                 }
 219         }
 220         return H_SUCCESS;
 221
 222  out_unpin:
 223         kvmppc_unpin_guest_page(kvm, va);
 224         return err;
 225 }
 226
 227 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 228 {
 229         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 230         unsigned long target, ret = H_SUCCESS;
 231         struct kvm_vcpu *tvcpu;
 232
 233         switch (req) {
 234         case H_ENTER:
 235                 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
 236                                               kvmppc_get_gpr(vcpu, 5),
 237                                               kvmppc_get_gpr(vcpu, 6),
 238                                               kvmppc_get_gpr(vcpu, 7));
 239                 break;
 240         case H_CEDE:
 241                 break;
 242         case H_PROD:
 243                 target = kvmppc_get_gpr(vcpu, 4);
 244                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 245                 if (!tvcpu) {
 246                         ret = H_PARAMETER;
 247                         break;
 248                 }
 249                 tvcpu->arch.prodded = 1;
 250                 smp_mb();
 251                 if (vcpu->arch.ceded) {
 252                         if (waitqueue_active(&vcpu->wq)) {
 253                                 wake_up_interruptible(&vcpu->wq);
 254                                 vcpu->stat.halt_wakeup++;
 255                         }
 256                 }
 257                 break;
 258         case H_CONFER:
 259                 break;
 260         case H_REGISTER_VPA:
 261                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 262                                         kvmppc_get_gpr(vcpu, 5),
 263                                         kvmppc_get_gpr(vcpu, 6));
 264                 break;
 265         default:
 266                 return RESUME_HOST;
 267         }
 268         kvmppc_set_gpr(vcpu, 3, ret);
 269         vcpu->arch.hcall_needed = 0;
 270         return RESUME_GUEST;
 271 }
 272
 273 static int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 274                               struct task_struct *tsk)
 275 {
 276         int r = RESUME_HOST;
 277
 278         vcpu->stat.sum_exits++;
 279
 280         run->exit_reason = KVM_EXIT_UNKNOWN;
 281         run->ready_for_interrupt_injection = 1;
 282         switch (vcpu->arch.trap) {
 283         /* We're good on these - the host merely wanted to get our attention */
 284         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 285                 vcpu->stat.dec_exits++;
 286                 r = RESUME_GUEST;
 287                 break;
 288         case BOOK3S_INTERRUPT_EXTERNAL:
 289                 vcpu->stat.ext_intr_exits++;
 290                 r = RESUME_GUEST;
 291                 break;
 292         case BOOK3S_INTERRUPT_PERFMON:
 293                 r = RESUME_GUEST;
 294                 break;
 295         case BOOK3S_INTERRUPT_PROGRAM:
 296         {
 297                 ulong flags;
 298                 /*
 299                  * Normally program interrupts are delivered directly
 300                  * to the guest by the hardware, but we can get here
 301                  * as a result of a hypervisor emulation interrupt
 302                  * (e40) getting turned into a 700 by BML RTAS.
 303                  */
 304                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 305                 kvmppc_core_queue_program(vcpu, flags);
 306                 r = RESUME_GUEST;
 307                 break;
 308         }
 309         case BOOK3S_INTERRUPT_SYSCALL:
 310         {
 311                 /* hcall - punt to userspace */
 312                 int i;
 313
 314                 if (vcpu->arch.shregs.msr & MSR_PR) {
 315                         /* sc 1 from userspace - reflect to guest syscall */
 316                         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_SYSCALL);
 317                         r = RESUME_GUEST;
 318                         break;
 319                 }
 320                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 321                 for (i = 0; i < 9; ++i)
 322                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 323                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 324                 vcpu->arch.hcall_needed = 1;
 325                 r = RESUME_HOST;
 326                 break;
 327         }
 328         /*
 329          * We get these next two if the guest accesses a page which it thinks
 330          * it has mapped but which is not actually present, either because
 331          * it is for an emulated I/O device or because the corresonding
 332          * host page has been paged out.  Any other HDSI/HISI interrupts
 333          * have been handled already.
 334          */
 335         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 336                 r = kvmppc_book3s_hv_page_fault(run, vcpu,
 337                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 338                 break;
 339         case BOOK3S_INTERRUPT_H_INST_STORAGE:
 340                 r = kvmppc_book3s_hv_page_fault(run, vcpu,
 341                                 kvmppc_get_pc(vcpu), 0);
 342                 break;
 343         /*
 344          * This occurs if the guest executes an illegal instruction.
 345          * We just generate a program interrupt to the guest, since
 346          * we don't emulate any guest instructions at this stage.
 347          */
 348         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 349                 kvmppc_core_queue_program(vcpu, 0x80000);
 350                 r = RESUME_GUEST;
 351                 break;
 352         default:
 353                 kvmppc_dump_regs(vcpu);
 354                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 355                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 356                         vcpu->arch.shregs.msr);
 357                 r = RESUME_HOST;
 358                 BUG();
 359                 break;
 360         }
 361
 362         return r;
 363 }
 364
 365 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 366                                   struct kvm_sregs *sregs)
 367 {
 368         int i;
 369
 370         sregs->pvr = vcpu->arch.pvr;
 371
 372         memset(sregs, 0, sizeof(struct kvm_sregs));
 373         for (i = 0; i < vcpu->arch.slb_max; i++) {
 374                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
 375                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 376         }
 377
 378         return 0;
 379 }
 380
 381 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 382                                   struct kvm_sregs *sregs)
 383 {
 384         int i, j;
 385
 386         kvmppc_set_pvr(vcpu, sregs->pvr);
 387
 388         j = 0;
 389         for (i = 0; i < vcpu->arch.slb_nr; i++) {
 390                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
 391                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
 392                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
 393                         ++j;
 394                 }
 395         }
 396         vcpu->arch.slb_max = j;
 397
 398         return 0;
 399 }
 400
 401 int kvmppc_core_check_processor_compat(void)
 402 {
 403         if (cpu_has_feature(CPU_FTR_HVMODE))
 404                 return 0;
 405         return -EIO;
 406 }
 407
 408 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
 409 {
 410         struct kvm_vcpu *vcpu;
 411         int err = -EINVAL;
 412         int core;
 413         struct kvmppc_vcore *vcore;
 414
 415         core = id / threads_per_core;
 416         if (core >= KVM_MAX_VCORES)
 417                 goto out;
 418
 419         err = -ENOMEM;
 420         vcpu = kzalloc(sizeof(struct kvm_vcpu), GFP_KERNEL);
 421         if (!vcpu)
 422                 goto out;
 423
 424         err = kvm_vcpu_init(vcpu, kvm, id);
 425         if (err)
 426                 goto free_vcpu;
 427
 428         vcpu->arch.shared = &vcpu->arch.shregs;
 429         vcpu->arch.last_cpu = -1;
 430         vcpu->arch.mmcr[0] = MMCR0_FC;
 431         vcpu->arch.ctrl = CTRL_RUNLATCH;
 432         /* default to host PVR, since we can't spoof it */
 433         vcpu->arch.pvr = mfspr(SPRN_PVR);
 434         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
 435
 436         kvmppc_mmu_book3s_hv_init(vcpu);
 437
 438         /*
 439          * We consider the vcpu stopped until we see the first run ioctl for it.
 440          */
 441         vcpu->arch.state = KVMPPC_VCPU_STOPPED;
 442
 443         init_waitqueue_head(&vcpu->arch.cpu_run);
 444
 445         mutex_lock(&kvm->lock);
 446         vcore = kvm->arch.vcores[core];
 447         if (!vcore) {
 448                 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
 449                 if (vcore) {
 450                         INIT_LIST_HEAD(&vcore->runnable_threads);
 451                         spin_lock_init(&vcore->lock);
 452                         init_waitqueue_head(&vcore->wq);
 453                 }
 454                 kvm->arch.vcores[core] = vcore;
 455         }
 456         mutex_unlock(&kvm->lock);
 457
 458         if (!vcore)
 459                 goto free_vcpu;
 460
 461         spin_lock(&vcore->lock);
 462         ++vcore->num_threads;
 463         spin_unlock(&vcore->lock);
 464         vcpu->arch.vcore = vcore;
 465
 466         vcpu->arch.cpu_type = KVM_CPU_3S_64;
 467         kvmppc_sanity_check(vcpu);
 468
 469         return vcpu;
 470
 471 free_vcpu:
 472         kfree(vcpu);
 473 out:
 474         return ERR_PTR(err);
 475 }
 476
 477 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
 478 {
 479         if (vcpu->arch.dtl)
 480                 kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.dtl);
 481         if (vcpu->arch.slb_shadow)
 482                 kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.slb_shadow);
 483         if (vcpu->arch.vpa)
 484                 kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.vpa);
 485         kvm_vcpu_uninit(vcpu);
 486         kfree(vcpu);
 487 }
 488
 489 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
 490 {
 491         unsigned long dec_nsec, now;
 492
 493         now = get_tb();
 494         if (now > vcpu->arch.dec_expires) {
 495                 /* decrementer has already gone negative */
 496                 kvmppc_core_queue_dec(vcpu);
 497                 kvmppc_core_prepare_to_enter(vcpu);
 498                 return;
 499         }
 500         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
 501                    / tb_ticks_per_sec;
 502         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
 503                       HRTIMER_MODE_REL);
 504         vcpu->arch.timer_running = 1;
 505 }
 506
 507 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
 508 {
 509         vcpu->arch.ceded = 0;
 510         if (vcpu->arch.timer_running) {
 511                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
 512                 vcpu->arch.timer_running = 0;
 513         }
 514 }
 515
 516 extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
 517 extern void xics_wake_cpu(int cpu);
 518
 519 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
 520                                    struct kvm_vcpu *vcpu)
 521 {
 522         struct kvm_vcpu *v;
 523
 524         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
 525                 return;
 526         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
 527         --vc->n_runnable;
 528         ++vc->n_busy;
 529         /* decrement the physical thread id of each following vcpu */
 530         v = vcpu;
 531         list_for_each_entry_continue(v, &vc->runnable_threads, arch.run_list)
 532                 --v->arch.ptid;
 533         list_del(&vcpu->arch.run_list);
 534 }
 535
 536 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
 537 {
 538         int cpu;
 539         struct paca_struct *tpaca;
 540         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 541
 542         if (vcpu->arch.timer_running) {
 543                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
 544                 vcpu->arch.timer_running = 0;
 545         }
 546         cpu = vc->pcpu + vcpu->arch.ptid;
 547         tpaca = &paca[cpu];
 548         tpaca->kvm_hstate.kvm_vcpu = vcpu;
 549         tpaca->kvm_hstate.kvm_vcore = vc;
 550         tpaca->kvm_hstate.napping = 0;
 551         vcpu->cpu = vc->pcpu;
 552         smp_wmb();
 553 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
 554         if (vcpu->arch.ptid) {
 555                 tpaca->cpu_start = 0x80;
 556                 wmb();
 557                 xics_wake_cpu(cpu);
 558                 ++vc->n_woken;
 559         }
 560 #endif
 561 }
 562
 563 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
 564 {
 565         int i;
 566
 567         HMT_low();
 568         i = 0;
 569         while (vc->nap_count < vc->n_woken) {
 570                 if (++i >= 1000000) {
 571                         pr_err("kvmppc_wait_for_nap timeout %d %d\n",
 572                                vc->nap_count, vc->n_woken);
 573                         break;
 574                 }
 575                 cpu_relax();
 576         }
 577         HMT_medium();
 578 }
 579
 580 /*
 581  * Check that we are on thread 0 and that any other threads in
 582  * this core are off-line.
 583  */
 584 static int on_primary_thread(void)
 585 {
 586         int cpu = smp_processor_id();
 587         int thr = cpu_thread_in_core(cpu);
 588
 589         if (thr)
 590                 return 0;
 591         while (++thr < threads_per_core)
 592                 if (cpu_online(cpu + thr))
 593                         return 0;
 594         return 1;
 595 }
 596
 597 /*
 598  * Run a set of guest threads on a physical core.
 599  * Called with vc->lock held.
 600  */
 601 static int kvmppc_run_core(struct kvmppc_vcore *vc)
 602 {
 603         struct kvm_vcpu *vcpu, *vcpu0, *vnext;
 604         long ret;
 605         u64 now;
 606         int ptid;
 607
 608         /* don't start if any threads have a signal pending */
 609         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
 610                 if (signal_pending(vcpu->arch.run_task))
 611                         return 0;
 612
 613         /*
 614          * Make sure we are running on thread 0, and that
 615          * secondary threads are offline.
 616          * XXX we should also block attempts to bring any
 617          * secondary threads online.
 618          */
 619         if (threads_per_core > 1 && !on_primary_thread()) {
 620                 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
 621                         vcpu->arch.ret = -EBUSY;
 622                 goto out;
 623         }
 624
 625         /*
 626          * Assign physical thread IDs, first to non-ceded vcpus
 627          * and then to ceded ones.
 628          */
 629         ptid = 0;
 630         vcpu0 = NULL;
 631         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
 632                 if (!vcpu->arch.ceded) {
 633                         if (!ptid)
 634                                 vcpu0 = vcpu;
 635                         vcpu->arch.ptid = ptid++;
 636                 }
 637         }
 638         if (!vcpu0)
 639                 return 0;               /* nothing to run */
 640         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
 641                 if (vcpu->arch.ceded)
 642                         vcpu->arch.ptid = ptid++;
 643
 644         vc->n_woken = 0;
 645         vc->nap_count = 0;
 646         vc->entry_exit_count = 0;
 647         vc->vcore_state = VCORE_RUNNING;
 648         vc->in_guest = 0;
 649         vc->pcpu = smp_processor_id();
 650         vc->napping_threads = 0;
 651         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
 652                 kvmppc_start_thread(vcpu);
 653
 654         preempt_disable();
 655         spin_unlock(&vc->lock);
 656
 657         kvm_guest_enter();
 658         __kvmppc_vcore_entry(NULL, vcpu0);
 659
 660         spin_lock(&vc->lock);
 661         /* disable sending of IPIs on virtual external irqs */
 662         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
 663                 vcpu->cpu = -1;
 664         /* wait for secondary threads to finish writing their state to memory */
 665         if (vc->nap_count < vc->n_woken)
 666                 kvmppc_wait_for_nap(vc);
 667         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
 668         vc->vcore_state = VCORE_EXITING;
 669         spin_unlock(&vc->lock);
 670
 671         /* make sure updates to secondary vcpu structs are visible now */
 672         smp_mb();
 673         kvm_guest_exit();
 674
 675         preempt_enable();
 676         kvm_resched(vcpu);
 677
 678         now = get_tb();
 679         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
 680                 /* cancel pending dec exception if dec is positive */
 681                 if (now < vcpu->arch.dec_expires &&
 682                     kvmppc_core_pending_dec(vcpu))
 683                         kvmppc_core_dequeue_dec(vcpu);
 684
 685                 ret = RESUME_GUEST;
 686                 if (vcpu->arch.trap)
 687                         ret = kvmppc_handle_exit(vcpu->arch.kvm_run, vcpu,
 688                                                  vcpu->arch.run_task);
 689
 690                 vcpu->arch.ret = ret;
 691                 vcpu->arch.trap = 0;
 692
 693                 if (vcpu->arch.ceded) {
 694                         if (ret != RESUME_GUEST)
 695                                 kvmppc_end_cede(vcpu);
 696                         else
 697                                 kvmppc_set_timer(vcpu);
 698                 }
 699         }
 700
 701         spin_lock(&vc->lock);
 702  out:
 703         vc->vcore_state = VCORE_INACTIVE;
 704         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
 705                                  arch.run_list) {
 706                 if (vcpu->arch.ret != RESUME_GUEST) {
 707                         kvmppc_remove_runnable(vc, vcpu);
 708                         wake_up(&vcpu->arch.cpu_run);
 709                 }
 710         }
 711
 712         return 1;
 713 }
 714
 715 /*
 716  * Wait for some other vcpu thread to execute us, and
 717  * wake us up when we need to handle something in the host.
 718  */
 719 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
 720 {
 721         DEFINE_WAIT(wait);
 722
 723         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
 724         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
 725                 schedule();
 726         finish_wait(&vcpu->arch.cpu_run, &wait);
 727 }
 728
 729 /*
 730  * All the vcpus in this vcore are idle, so wait for a decrementer
 731  * or external interrupt to one of the vcpus.  vc->lock is held.
 732  */
 733 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
 734 {
 735         DEFINE_WAIT(wait);
 736         struct kvm_vcpu *v;
 737         int all_idle = 1;
 738
 739         prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
 740         vc->vcore_state = VCORE_SLEEPING;
 741         spin_unlock(&vc->lock);
 742         list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
 743                 if (!v->arch.ceded || v->arch.pending_exceptions) {
 744                         all_idle = 0;
 745                         break;
 746                 }
 747         }
 748         if (all_idle)
 749                 schedule();
 750         finish_wait(&vc->wq, &wait);
 751         spin_lock(&vc->lock);
 752         vc->vcore_state = VCORE_INACTIVE;
 753 }
 754
 755 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
 756 {
 757         int n_ceded;
 758         int prev_state;
 759         struct kvmppc_vcore *vc;
 760         struct kvm_vcpu *v, *vn;
 761
 762         kvm_run->exit_reason = 0;
 763         vcpu->arch.ret = RESUME_GUEST;
 764         vcpu->arch.trap = 0;
 765
 766         /*
 767          * Synchronize with other threads in this virtual core
 768          */
 769         vc = vcpu->arch.vcore;
 770         spin_lock(&vc->lock);
 771         vcpu->arch.ceded = 0;
 772         vcpu->arch.run_task = current;
 773         vcpu->arch.kvm_run = kvm_run;
 774         prev_state = vcpu->arch.state;
 775         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
 776         list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
 777         ++vc->n_runnable;
 778
 779         /*
 780          * This happens the first time this is called for a vcpu.
 781          * If the vcore is already running, we may be able to start
 782          * this thread straight away and have it join in.
 783          */
 784         if (prev_state == KVMPPC_VCPU_STOPPED) {
 785                 if (vc->vcore_state == VCORE_RUNNING &&
 786                     VCORE_EXIT_COUNT(vc) == 0) {
 787                         vcpu->arch.ptid = vc->n_runnable - 1;
 788                         kvmppc_start_thread(vcpu);
 789                 }
 790
 791         } else if (prev_state == KVMPPC_VCPU_BUSY_IN_HOST)
 792                 --vc->n_busy;
 793
 794         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
 795                !signal_pending(current)) {
 796                 if (vc->n_busy || vc->vcore_state != VCORE_INACTIVE) {
 797                         spin_unlock(&vc->lock);
 798                         kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
 799                         spin_lock(&vc->lock);
 800                         continue;
 801                 }
 802                 n_ceded = 0;
 803                 list_for_each_entry(v, &vc->runnable_threads, arch.run_list)
 804                         n_ceded += v->arch.ceded;
 805                 if (n_ceded == vc->n_runnable)
 806                         kvmppc_vcore_blocked(vc);
 807                 else
 808                         kvmppc_run_core(vc);
 809
 810                 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
 811                                          arch.run_list) {
 812                         kvmppc_core_prepare_to_enter(v);
 813                         if (signal_pending(v->arch.run_task)) {
 814                                 kvmppc_remove_runnable(vc, v);
 815                                 v->stat.signal_exits++;
 816                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
 817                                 v->arch.ret = -EINTR;
 818                                 wake_up(&v->arch.cpu_run);
 819                         }
 820                 }
 821         }
 822
 823         if (signal_pending(current)) {
 824                 if (vc->vcore_state == VCORE_RUNNING ||
 825                     vc->vcore_state == VCORE_EXITING) {
 826                         spin_unlock(&vc->lock);
 827                         kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
 828                         spin_lock(&vc->lock);
 829                 }
 830                 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
 831                         kvmppc_remove_runnable(vc, vcpu);
 832                         vcpu->stat.signal_exits++;
 833                         kvm_run->exit_reason = KVM_EXIT_INTR;
 834                         vcpu->arch.ret = -EINTR;
 835                 }
 836         }
 837
 838         spin_unlock(&vc->lock);
 839         return vcpu->arch.ret;
 840 }
 841
 842 int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
 843 {
 844         int r;
 845
 846         if (!vcpu->arch.sane) {
 847                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
 848                 return -EINVAL;
 849         }
 850
 851         kvmppc_core_prepare_to_enter(vcpu);
 852
 853         /* No need to go into the guest when all we'll do is come back out */
 854         if (signal_pending(current)) {
 855                 run->exit_reason = KVM_EXIT_INTR;
 856                 return -EINTR;
 857         }
 858
 859         /* On the first time here, set up VRMA or RMA */
 860         if (!vcpu->kvm->arch.rma_setup_done) {
 861                 r = kvmppc_hv_setup_rma(vcpu);
 862                 if (r)
 863                         return r;
 864         }
 865
 866         flush_fp_to_thread(current);
 867         flush_altivec_to_thread(current);
 868         flush_vsx_to_thread(current);
 869         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
 870         vcpu->arch.pgdir = current->mm->pgd;
 871
 872         do {
 873                 r = kvmppc_run_vcpu(run, vcpu);
 874
 875                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
 876                     !(vcpu->arch.shregs.msr & MSR_PR)) {
 877                         r = kvmppc_pseries_do_hcall(vcpu);
 878                         kvmppc_core_prepare_to_enter(vcpu);
 879                 }
 880         } while (r == RESUME_GUEST);
 881         return r;
 882 }
 883
 884 static long kvmppc_stt_npages(unsigned long window_size)
 885 {
 886         return ALIGN((window_size >> SPAPR_TCE_SHIFT)
 887                      * sizeof(u64), PAGE_SIZE) / PAGE_SIZE;
 888 }
 889
 890 static void release_spapr_tce_table(struct kvmppc_spapr_tce_table *stt)
 891 {
 892         struct kvm *kvm = stt->kvm;
 893         int i;
 894
 895         mutex_lock(&kvm->lock);
 896         list_del(&stt->list);
 897         for (i = 0; i < kvmppc_stt_npages(stt->window_size); i++)
 898                 __free_page(stt->pages[i]);
 899         kfree(stt);
 900         mutex_unlock(&kvm->lock);
 901
 902         kvm_put_kvm(kvm);
 903 }
 904
 905 static int kvm_spapr_tce_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 906 {
 907         struct kvmppc_spapr_tce_table *stt = vma->vm_file->private_data;
 908         struct page *page;
 909
 910         if (vmf->pgoff >= kvmppc_stt_npages(stt->window_size))
 911                 return VM_FAULT_SIGBUS;
 912
 913         page = stt->pages[vmf->pgoff];
 914         get_page(page);
 915         vmf->page = page;
 916         return 0;
 917 }
 918
 919 static const struct vm_operations_struct kvm_spapr_tce_vm_ops = {
 920         .fault = kvm_spapr_tce_fault,
 921 };
 922
 923 static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma)
 924 {
 925         vma->vm_ops = &kvm_spapr_tce_vm_ops;
 926         return 0;
 927 }
 928
 929 static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
 930 {
 931         struct kvmppc_spapr_tce_table *stt = filp->private_data;
 932
 933         release_spapr_tce_table(stt);
 934         return 0;
 935 }
 936
 937 static struct file_operations kvm_spapr_tce_fops = {
 938         .mmap           = kvm_spapr_tce_mmap,
 939         .release        = kvm_spapr_tce_release,
 940 };
 941
 942 long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
 943                                    struct kvm_create_spapr_tce *args)
 944 {
 945         struct kvmppc_spapr_tce_table *stt = NULL;
 946         long npages;
 947         int ret = -ENOMEM;
 948         int i;
 949
 950         /* Check this LIOBN hasn't been previously allocated */
 951         list_for_each_entry(stt, &kvm->arch.spapr_tce_tables, list) {
 952                 if (stt->liobn == args->liobn)
 953                         return -EBUSY;
 954         }
 955
 956         npages = kvmppc_stt_npages(args->window_size);
 957
 958         stt = kzalloc(sizeof(*stt) + npages* sizeof(struct page *),
 959                       GFP_KERNEL);
 960         if (!stt)
 961                 goto fail;
 962
 963         stt->liobn = args->liobn;
 964         stt->window_size = args->window_size;
 965         stt->kvm = kvm;
 966
 967         for (i = 0; i < npages; i++) {
 968                 stt->pages[i] = alloc_page(GFP_KERNEL | __GFP_ZERO);
 969                 if (!stt->pages[i])
 970                         goto fail;
 971         }
 972
 973         kvm_get_kvm(kvm);
 974
 975         mutex_lock(&kvm->lock);
 976         list_add(&stt->list, &kvm->arch.spapr_tce_tables);
 977
 978         mutex_unlock(&kvm->lock);
 979
 980         return anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
 981                                 stt, O_RDWR);
 982
 983 fail:
 984         if (stt) {
 985                 for (i = 0; i < npages; i++)
 986                         if (stt->pages[i])
 987                                 __free_page(stt->pages[i]);
 988
 989                 kfree(stt);
 990         }
 991         return ret;
 992 }
 993
 994 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
 995    Assumes POWER7 or PPC970. */
 996 static inline int lpcr_rmls(unsigned long rma_size)
 997 {
 998         switch (rma_size) {
 999         case 32ul << 20:        /* 32 MB */
1000                 if (cpu_has_feature(CPU_FTR_ARCH_206))
1001                         return 8;       /* only supported on POWER7 */
1002                 return -1;
1003         case 64ul << 20:        /* 64 MB */
1004                 return 3;
1005         case 128ul << 20:       /* 128 MB */
1006                 return 7;
1007         case 256ul << 20:       /* 256 MB */
1008                 return 4;
1009         case 1ul << 30:         /* 1 GB */
1010                 return 2;
1011         case 16ul << 30:        /* 16 GB */
1012                 return 1;
1013         case 256ul << 30:       /* 256 GB */
1014                 return 0;
1015         default:
1016                 return -1;
1017         }
1018 }
1019
1020 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1021 {
1022         struct kvmppc_rma_info *ri = vma->vm_file->private_data;
1023         struct page *page;
1024
1025         if (vmf->pgoff >= ri->npages)
1026                 return VM_FAULT_SIGBUS;
1027
1028         page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1029         get_page(page);
1030         vmf->page = page;
1031         return 0;
1032 }
1033
1034 static const struct vm_operations_struct kvm_rma_vm_ops = {
1035         .fault = kvm_rma_fault,
1036 };
1037
1038 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1039 {
1040         vma->vm_flags |= VM_RESERVED;
1041         vma->vm_ops = &kvm_rma_vm_ops;
1042         return 0;
1043 }
1044
1045 static int kvm_rma_release(struct inode *inode, struct file *filp)
1046 {
1047         struct kvmppc_rma_info *ri = filp->private_data;
1048
1049         kvm_release_rma(ri);
1050         return 0;
1051 }
1052
1053 static struct file_operations kvm_rma_fops = {
1054         .mmap           = kvm_rma_mmap,
1055         .release        = kvm_rma_release,
1056 };
1057
1058 long kvm_vm_ioctl_allocate_rma(struct kvm *kvm, struct kvm_allocate_rma *ret)
1059 {
1060         struct kvmppc_rma_info *ri;
1061         long fd;
1062
1063         ri = kvm_alloc_rma();
1064         if (!ri)
1065                 return -ENOMEM;
1066
1067         fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR);
1068         if (fd < 0)
1069                 kvm_release_rma(ri);
1070
1071         ret->rma_size = ri->npages << PAGE_SHIFT;
1072         return fd;
1073 }
1074
1075 /*
1076  * Get (and clear) the dirty memory log for a memory slot.
1077  */
1078 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1079 {
1080         struct kvm_memory_slot *memslot;
1081         int r;
1082         unsigned long n;
1083
1084         mutex_lock(&kvm->slots_lock);
1085
1086         r = -EINVAL;
1087         if (log->slot >= KVM_MEMORY_SLOTS)
1088                 goto out;
1089
1090         memslot = id_to_memslot(kvm->memslots, log->slot);
1091         r = -ENOENT;
1092         if (!memslot->dirty_bitmap)
1093                 goto out;
1094
1095         n = kvm_dirty_bitmap_bytes(memslot);
1096         memset(memslot->dirty_bitmap, 0, n);
1097
1098         r = kvmppc_hv_get_dirty_log(kvm, memslot);
1099         if (r)
1100                 goto out;
1101
1102         r = -EFAULT;
1103         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1104                 goto out;
1105
1106         r = 0;
1107 out:
1108         mutex_unlock(&kvm->slots_lock);
1109         return r;
1110 }
1111
1112 static unsigned long slb_pgsize_encoding(unsigned long psize)
1113 {
1114         unsigned long senc = 0;
1115
1116         if (psize > 0x1000) {
1117                 senc = SLB_VSID_L;
1118                 if (psize == 0x10000)
1119                         senc |= SLB_VSID_LP_01;
1120         }
1121         return senc;
1122 }
1123
1124 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1125                                 struct kvm_userspace_memory_region *mem)
1126 {
1127         unsigned long npages;
1128         unsigned long *phys;
1129
1130         /* Allocate a slot_phys array */
1131         phys = kvm->arch.slot_phys[mem->slot];
1132         if (!kvm->arch.using_mmu_notifiers && !phys) {
1133                 npages = mem->memory_size >> PAGE_SHIFT;
1134                 phys = vzalloc(npages * sizeof(unsigned long));
1135                 if (!phys)
1136                         return -ENOMEM;
1137                 kvm->arch.slot_phys[mem->slot] = phys;
1138                 kvm->arch.slot_npages[mem->slot] = npages;
1139         }
1140
1141         return 0;
1142 }
1143
1144 static void unpin_slot(struct kvm *kvm, int slot_id)
1145 {
1146         unsigned long *physp;
1147         unsigned long j, npages, pfn;
1148         struct page *page;
1149
1150         physp = kvm->arch.slot_phys[slot_id];
1151         npages = kvm->arch.slot_npages[slot_id];
1152         if (physp) {
1153                 spin_lock(&kvm->arch.slot_phys_lock);
1154                 for (j = 0; j < npages; j++) {
1155                         if (!(physp[j] & KVMPPC_GOT_PAGE))
1156                                 continue;
1157                         pfn = physp[j] >> PAGE_SHIFT;
1158                         page = pfn_to_page(pfn);
1159                         if (PageHuge(page))
1160                                 page = compound_head(page);
1161                         SetPageDirty(page);
1162                         put_page(page);
1163                 }
1164                 kvm->arch.slot_phys[slot_id] = NULL;
1165                 spin_unlock(&kvm->arch.slot_phys_lock);
1166                 vfree(physp);
1167         }
1168 }
1169
1170 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1171                                 struct kvm_userspace_memory_region *mem)
1172 {
1173 }
1174
1175 static int kvmppc_hv_setup_rma(struct kvm_vcpu *vcpu)
1176 {
1177         int err = 0;
1178         struct kvm *kvm = vcpu->kvm;
1179         struct kvmppc_rma_info *ri = NULL;
1180         unsigned long hva;
1181         struct kvm_memory_slot *memslot;
1182         struct vm_area_struct *vma;
1183         unsigned long lpcr, senc;
1184         unsigned long psize, porder;
1185         unsigned long rma_size;
1186         unsigned long rmls;
1187         unsigned long *physp;
1188         unsigned long i, npages;
1189
1190         mutex_lock(&kvm->lock);
1191         if (kvm->arch.rma_setup_done)
1192                 goto out;       /* another vcpu beat us to it */
1193
1194         /* Look up the memslot for guest physical address 0 */
1195         memslot = gfn_to_memslot(kvm, 0);
1196
1197         /* We must have some memory at 0 by now */
1198         err = -EINVAL;
1199         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1200                 goto out;
1201
1202         /* Look up the VMA for the start of this memory slot */
1203         hva = memslot->userspace_addr;
1204         down_read(&current->mm->mmap_sem);
1205         vma = find_vma(current->mm, hva);
1206         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
1207                 goto up_out;
1208
1209         psize = vma_kernel_pagesize(vma);
1210         porder = __ilog2(psize);
1211
1212         /* Is this one of our preallocated RMAs? */
1213         if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
1214             hva == vma->vm_start)
1215                 ri = vma->vm_file->private_data;
1216
1217         up_read(&current->mm->mmap_sem);
1218
1219         if (!ri) {
1220                 /* On POWER7, use VRMA; on PPC970, give up */
1221                 err = -EPERM;
1222                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1223                         pr_err("KVM: CPU requires an RMO\n");
1224                         goto out;
1225                 }
1226
1227                 /* We can handle 4k, 64k or 16M pages in the VRMA */
1228                 err = -EINVAL;
1229                 if (!(psize == 0x1000 || psize == 0x10000 ||
1230                       psize == 0x1000000))
1231                         goto out;
1232
1233                 /* Update VRMASD field in the LPCR */
1234                 senc = slb_pgsize_encoding(psize);
1235                 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1236                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
1237                 lpcr = kvm->arch.lpcr & ~LPCR_VRMASD;
1238                 lpcr |= senc << (LPCR_VRMASD_SH - 4);
1239                 kvm->arch.lpcr = lpcr;
1240
1241                 /* Create HPTEs in the hash page table for the VRMA */
1242                 kvmppc_map_vrma(vcpu, memslot, porder);
1243
1244         } else {
1245                 /* Set up to use an RMO region */
1246                 rma_size = ri->npages;
1247                 if (rma_size > memslot->npages)
1248                         rma_size = memslot->npages;
1249                 rma_size <<= PAGE_SHIFT;
1250                 rmls = lpcr_rmls(rma_size);
1251                 err = -EINVAL;
1252                 if (rmls < 0) {
1253                         pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
1254                         goto out;
1255                 }
1256                 atomic_inc(&ri->use_count);
1257                 kvm->arch.rma = ri;
1258
1259                 /* Update LPCR and RMOR */
1260                 lpcr = kvm->arch.lpcr;
1261                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1262                         /* PPC970; insert RMLS value (split field) in HID4 */
1263                         lpcr &= ~((1ul << HID4_RMLS0_SH) |
1264                                   (3ul << HID4_RMLS2_SH));
1265                         lpcr |= ((rmls >> 2) << HID4_RMLS0_SH) |
1266                                 ((rmls & 3) << HID4_RMLS2_SH);
1267                         /* RMOR is also in HID4 */
1268                         lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
1269                                 << HID4_RMOR_SH;
1270                 } else {
1271                         /* POWER7 */
1272                         lpcr &= ~(LPCR_VPM0 | LPCR_VRMA_L);
1273                         lpcr |= rmls << LPCR_RMLS_SH;
1274                         kvm->arch.rmor = kvm->arch.rma->base_pfn << PAGE_SHIFT;
1275                 }
1276                 kvm->arch.lpcr = lpcr;
1277                 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
1278                         ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
1279
1280                 /* Initialize phys addrs of pages in RMO */
1281                 npages = ri->npages;
1282                 porder = __ilog2(npages);
1283                 physp = kvm->arch.slot_phys[memslot->id];
1284                 spin_lock(&kvm->arch.slot_phys_lock);
1285                 for (i = 0; i < npages; ++i)
1286                         physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) + porder;
1287                 spin_unlock(&kvm->arch.slot_phys_lock);
1288         }
1289
1290         /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
1291         smp_wmb();
1292         kvm->arch.rma_setup_done = 1;
1293         err = 0;
1294  out:
1295         mutex_unlock(&kvm->lock);
1296         return err;
1297
1298  up_out:
1299         up_read(&current->mm->mmap_sem);
1300         goto out;
1301 }
1302
1303 int kvmppc_core_init_vm(struct kvm *kvm)
1304 {
1305         long r;
1306         unsigned long lpcr;
1307
1308         /* Allocate hashed page table */
1309         r = kvmppc_alloc_hpt(kvm);
1310         if (r)
1311                 return r;
1312
1313         INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
1314
1315         kvm->arch.rma = NULL;
1316
1317         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
1318
1319         if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1320                 /* PPC970; HID4 is effectively the LPCR */
1321                 unsigned long lpid = kvm->arch.lpid;
1322                 kvm->arch.host_lpid = 0;
1323                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
1324                 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
1325                 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
1326                         ((lpid & 0xf) << HID4_LPID5_SH);
1327         } else {
1328                 /* POWER7; init LPCR for virtual RMA mode */
1329                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
1330                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
1331                 lpcr &= LPCR_PECE | LPCR_LPES;
1332                 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
1333                         LPCR_VPM0 | LPCR_VPM1;
1334                 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
1335                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
1336         }
1337         kvm->arch.lpcr = lpcr;
1338
1339         kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
1340         spin_lock_init(&kvm->arch.slot_phys_lock);
1341         return 0;
1342 }
1343
1344 void kvmppc_core_destroy_vm(struct kvm *kvm)
1345 {
1346         unsigned long i;
1347
1348         if (!kvm->arch.using_mmu_notifiers)
1349                 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
1350                         unpin_slot(kvm, i);
1351
1352         if (kvm->arch.rma) {
1353                 kvm_release_rma(kvm->arch.rma);
1354                 kvm->arch.rma = NULL;
1355         }
1356
1357         kvmppc_free_hpt(kvm);
1358         WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1359 }
1360
1361 /* These are stubs for now */
1362 void kvmppc_mmu_pte_pflush(struct kvm_vcpu *vcpu, ulong pa_start, ulong pa_end)
1363 {
1364 }
1365
1366 /* We don't need to emulate any privileged instructions or dcbz */
1367 int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu,
1368                            unsigned int inst, int *advance)
1369 {
1370         return EMULATE_FAIL;
1371 }
1372
1373 int kvmppc_core_emulate_mtspr(struct kvm_vcpu *vcpu, int sprn, int rs)
1374 {
1375         return EMULATE_FAIL;
1376 }
1377
1378 int kvmppc_core_emulate_mfspr(struct kvm_vcpu *vcpu, int sprn, int rt)
1379 {
1380         return EMULATE_FAIL;
1381 }
1382
1383 static int kvmppc_book3s_hv_init(void)
1384 {
1385         int r;
1386
1387         r = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1388
1389         if (r)
1390                 return r;
1391
1392         r = kvmppc_mmu_hv_init();
1393
1394         return r;
1395 }
1396
1397 static void kvmppc_book3s_hv_exit(void)
1398 {
1399         kvm_exit();
1400 }
1401
1402 module_init(kvmppc_book3s_hv_init);
1403 module_exit(kvmppc_book3s_hv_exit);