Pileus Git - ~andy/linux/blob - 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 #include <linux/srcu.h>
  34
  35 #include <asm/reg.h>
  36 #include <asm/cputable.h>
  37 #include <asm/cacheflush.h>
  38 #include <asm/tlbflush.h>
  39 #include <asm/uaccess.h>
  40 #include <asm/io.h>
  41 #include <asm/kvm_ppc.h>
  42 #include <asm/kvm_book3s.h>
  43 #include <asm/mmu_context.h>
  44 #include <asm/lppaca.h>
  45 #include <asm/processor.h>
  46 #include <asm/cputhreads.h>
  47 #include <asm/page.h>
  48 #include <asm/hvcall.h>
  49 #include <asm/switch_to.h>
  50 #include <asm/smp.h>
  51 #include <linux/gfp.h>
  52 #include <linux/vmalloc.h>
  53 #include <linux/highmem.h>
  54 #include <linux/hugetlb.h>
  55
  56 /* #define EXIT_DEBUG */
  57 /* #define EXIT_DEBUG_SIMPLE */
  58 /* #define EXIT_DEBUG_INT */
  59
  60 /* Used to indicate that a guest page fault needs to be handled */
  61 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  62
  63 /* Used as a "null" value for timebase values */
  64 #define TB_NIL  (~(u64)0)
  65
  66 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
  67 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
  68
  69 /*
  70  * We use the vcpu_load/put functions to measure stolen time.
  71  * Stolen time is counted as time when either the vcpu is able to
  72  * run as part of a virtual core, but the task running the vcore
  73  * is preempted or sleeping, or when the vcpu needs something done
  74  * in the kernel by the task running the vcpu, but that task is
  75  * preempted or sleeping.  Those two things have to be counted
  76  * separately, since one of the vcpu tasks will take on the job
  77  * of running the core, and the other vcpu tasks in the vcore will
  78  * sleep waiting for it to do that, but that sleep shouldn't count
  79  * as stolen time.
  80  *
  81  * Hence we accumulate stolen time when the vcpu can run as part of
  82  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
  83  * needs its task to do other things in the kernel (for example,
  84  * service a page fault) in busy_stolen.  We don't accumulate
  85  * stolen time for a vcore when it is inactive, or for a vcpu
  86  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
  87  * a misnomer; it means that the vcpu task is not executing in
  88  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
  89  * the kernel.  We don't have any way of dividing up that time
  90  * between time that the vcpu is genuinely stopped, time that
  91  * the task is actively working on behalf of the vcpu, and time
  92  * that the task is preempted, so we don't count any of it as
  93  * stolen.
  94  *
  95  * Updates to busy_stolen are protected by arch.tbacct_lock;
  96  * updates to vc->stolen_tb are protected by the arch.tbacct_lock
  97  * of the vcpu that has taken responsibility for running the vcore
  98  * (i.e. vc->runner).  The stolen times are measured in units of
  99  * timebase ticks.  (Note that the != TB_NIL checks below are
 100  * purely defensive; they should never fail.)
 101  */
 102
 103 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 104 {
 105         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 106
 107         spin_lock(&vcpu->arch.tbacct_lock);
 108         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
 109             vc->preempt_tb != TB_NIL) {
 110                 vc->stolen_tb += mftb() - vc->preempt_tb;
 111                 vc->preempt_tb = TB_NIL;
 112         }
 113         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 114             vcpu->arch.busy_preempt != TB_NIL) {
 115                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 116                 vcpu->arch.busy_preempt = TB_NIL;
 117         }
 118         spin_unlock(&vcpu->arch.tbacct_lock);
 119 }
 120
 121 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
 122 {
 123         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 124
 125         spin_lock(&vcpu->arch.tbacct_lock);
 126         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
 127                 vc->preempt_tb = mftb();
 128         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 129                 vcpu->arch.busy_preempt = mftb();
 130         spin_unlock(&vcpu->arch.tbacct_lock);
 131 }
 132
 133 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
 134 {
 135         vcpu->arch.shregs.msr = msr;
 136         kvmppc_end_cede(vcpu);
 137 }
 138
 139 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
 140 {
 141         vcpu->arch.pvr = pvr;
 142 }
 143
 144 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 145 {
 146         int r;
 147
 148         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 149         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 150                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
 151         for (r = 0; r < 16; ++r)
 152                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 153                        r, kvmppc_get_gpr(vcpu, r),
 154                        r+16, kvmppc_get_gpr(vcpu, r+16));
 155         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 156                vcpu->arch.ctr, vcpu->arch.lr);
 157         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 158                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 159         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 160                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 161         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 162                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 163         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 164                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 165         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 166         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 167                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 168         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 169         for (r = 0; r < vcpu->arch.slb_max; ++r)
 170                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 171                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 172         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 173                vcpu->kvm->arch.lpcr, vcpu->kvm->arch.sdr1,
 174                vcpu->arch.last_inst);
 175 }
 176
 177 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 178 {
 179         int r;
 180         struct kvm_vcpu *v, *ret = NULL;
 181
 182         mutex_lock(&kvm->lock);
 183         kvm_for_each_vcpu(r, v, kvm) {
 184                 if (v->vcpu_id == id) {
 185                         ret = v;
 186                         break;
 187                 }
 188         }
 189         mutex_unlock(&kvm->lock);
 190         return ret;
 191 }
 192
 193 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 194 {
 195         vpa->shared_proc = 1;
 196         vpa->yield_count = 1;
 197 }
 198
 199 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 200                    unsigned long addr, unsigned long len)
 201 {
 202         /* check address is cacheline aligned */
 203         if (addr & (L1_CACHE_BYTES - 1))
 204                 return -EINVAL;
 205         spin_lock(&vcpu->arch.vpa_update_lock);
 206         if (v->next_gpa != addr || v->len != len) {
 207                 v->next_gpa = addr;
 208                 v->len = addr ? len : 0;
 209                 v->update_pending = 1;
 210         }
 211         spin_unlock(&vcpu->arch.vpa_update_lock);
 212         return 0;
 213 }
 214
 215 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 216 struct reg_vpa {
 217         u32 dummy;
 218         union {
 219                 u16 hword;
 220                 u32 word;
 221         } length;
 222 };
 223
 224 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 225 {
 226         if (vpap->update_pending)
 227                 return vpap->next_gpa != 0;
 228         return vpap->pinned_addr != NULL;
 229 }
 230
 231 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 232                                        unsigned long flags,
 233                                        unsigned long vcpuid, unsigned long vpa)
 234 {
 235         struct kvm *kvm = vcpu->kvm;
 236         unsigned long len, nb;
 237         void *va;
 238         struct kvm_vcpu *tvcpu;
 239         int err;
 240         int subfunc;
 241         struct kvmppc_vpa *vpap;
 242
 243         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 244         if (!tvcpu)
 245                 return H_PARAMETER;
 246
 247         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 248         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 249             subfunc == H_VPA_REG_SLB) {
 250                 /* Registering new area - address must be cache-line aligned */
 251                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 252                         return H_PARAMETER;
 253
 254                 /* convert logical addr to kernel addr and read length */
 255                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 256                 if (va == NULL)
 257                         return H_PARAMETER;
 258                 if (subfunc == H_VPA_REG_VPA)
 259                         len = ((struct reg_vpa *)va)->length.hword;
 260                 else
 261                         len = ((struct reg_vpa *)va)->length.word;
 262                 kvmppc_unpin_guest_page(kvm, va);
 263
 264                 /* Check length */
 265                 if (len > nb || len < sizeof(struct reg_vpa))
 266                         return H_PARAMETER;
 267         } else {
 268                 vpa = 0;
 269                 len = 0;
 270         }
 271
 272         err = H_PARAMETER;
 273         vpap = NULL;
 274         spin_lock(&tvcpu->arch.vpa_update_lock);
 275
 276         switch (subfunc) {
 277         case H_VPA_REG_VPA:             /* register VPA */
 278                 if (len < sizeof(struct lppaca))
 279                         break;
 280                 vpap = &tvcpu->arch.vpa;
 281                 err = 0;
 282                 break;
 283
 284         case H_VPA_REG_DTL:             /* register DTL */
 285                 if (len < sizeof(struct dtl_entry))
 286                         break;
 287                 len -= len % sizeof(struct dtl_entry);
 288
 289                 /* Check that they have previously registered a VPA */
 290                 err = H_RESOURCE;
 291                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 292                         break;
 293
 294                 vpap = &tvcpu->arch.dtl;
 295                 err = 0;
 296                 break;
 297
 298         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 299                 /* Check that they have previously registered a VPA */
 300                 err = H_RESOURCE;
 301                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 302                         break;
 303
 304                 vpap = &tvcpu->arch.slb_shadow;
 305                 err = 0;
 306                 break;
 307
 308         case H_VPA_DEREG_VPA:           /* deregister VPA */
 309                 /* Check they don't still have a DTL or SLB buf registered */
 310                 err = H_RESOURCE;
 311                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 312                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 313                         break;
 314
 315                 vpap = &tvcpu->arch.vpa;
 316                 err = 0;
 317                 break;
 318
 319         case H_VPA_DEREG_DTL:           /* deregister DTL */
 320                 vpap = &tvcpu->arch.dtl;
 321                 err = 0;
 322                 break;
 323
 324         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 325                 vpap = &tvcpu->arch.slb_shadow;
 326                 err = 0;
 327                 break;
 328         }
 329
 330         if (vpap) {
 331                 vpap->next_gpa = vpa;
 332                 vpap->len = len;
 333                 vpap->update_pending = 1;
 334         }
 335
 336         spin_unlock(&tvcpu->arch.vpa_update_lock);
 337
 338         return err;
 339 }
 340
 341 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 342 {
 343         struct kvm *kvm = vcpu->kvm;
 344         void *va;
 345         unsigned long nb;
 346         unsigned long gpa;
 347
 348         /*
 349          * We need to pin the page pointed to by vpap->next_gpa,
 350          * but we can't call kvmppc_pin_guest_page under the lock
 351          * as it does get_user_pages() and down_read().  So we
 352          * have to drop the lock, pin the page, then get the lock
 353          * again and check that a new area didn't get registered
 354          * in the meantime.
 355          */
 356         for (;;) {
 357                 gpa = vpap->next_gpa;
 358                 spin_unlock(&vcpu->arch.vpa_update_lock);
 359                 va = NULL;
 360                 nb = 0;
 361                 if (gpa)
 362                         va = kvmppc_pin_guest_page(kvm, vpap->next_gpa, &nb);
 363                 spin_lock(&vcpu->arch.vpa_update_lock);
 364                 if (gpa == vpap->next_gpa)
 365                         break;
 366                 /* sigh... unpin that one and try again */
 367                 if (va)
 368                         kvmppc_unpin_guest_page(kvm, va);
 369         }
 370
 371         vpap->update_pending = 0;
 372         if (va && nb < vpap->len) {
 373                 /*
 374                  * If it's now too short, it must be that userspace
 375                  * has changed the mappings underlying guest memory,
 376                  * so unregister the region.
 377                  */
 378                 kvmppc_unpin_guest_page(kvm, va);
 379                 va = NULL;
 380         }
 381         if (vpap->pinned_addr)
 382                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr);
 383         vpap->pinned_addr = va;
 384         if (va)
 385                 vpap->pinned_end = va + vpap->len;
 386 }
 387
 388 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 389 {
 390         if (!(vcpu->arch.vpa.update_pending ||
 391               vcpu->arch.slb_shadow.update_pending ||
 392               vcpu->arch.dtl.update_pending))
 393                 return;
 394
 395         spin_lock(&vcpu->arch.vpa_update_lock);
 396         if (vcpu->arch.vpa.update_pending) {
 397                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 398                 if (vcpu->arch.vpa.pinned_addr)
 399                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 400         }
 401         if (vcpu->arch.dtl.update_pending) {
 402                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 403                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 404                 vcpu->arch.dtl_index = 0;
 405         }
 406         if (vcpu->arch.slb_shadow.update_pending)
 407                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 408         spin_unlock(&vcpu->arch.vpa_update_lock);
 409 }
 410
 411 /*
 412  * Return the accumulated stolen time for the vcore up until `now'.
 413  * The caller should hold the vcore lock.
 414  */
 415 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 416 {
 417         u64 p;
 418
 419         /*
 420          * If we are the task running the vcore, then since we hold
 421          * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
 422          * can't be updated, so we don't need the tbacct_lock.
 423          * If the vcore is inactive, it can't become active (since we
 424          * hold the vcore lock), so the vcpu load/put functions won't
 425          * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
 426          */
 427         if (vc->vcore_state != VCORE_INACTIVE &&
 428             vc->runner->arch.run_task != current) {
 429                 spin_lock(&vc->runner->arch.tbacct_lock);
 430                 p = vc->stolen_tb;
 431                 if (vc->preempt_tb != TB_NIL)
 432                         p += now - vc->preempt_tb;
 433                 spin_unlock(&vc->runner->arch.tbacct_lock);
 434         } else {
 435                 p = vc->stolen_tb;
 436         }
 437         return p;
 438 }
 439
 440 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 441                                     struct kvmppc_vcore *vc)
 442 {
 443         struct dtl_entry *dt;
 444         struct lppaca *vpa;
 445         unsigned long stolen;
 446         unsigned long core_stolen;
 447         u64 now;
 448
 449         dt = vcpu->arch.dtl_ptr;
 450         vpa = vcpu->arch.vpa.pinned_addr;
 451         now = mftb();
 452         core_stolen = vcore_stolen_time(vc, now);
 453         stolen = core_stolen - vcpu->arch.stolen_logged;
 454         vcpu->arch.stolen_logged = core_stolen;
 455         spin_lock(&vcpu->arch.tbacct_lock);
 456         stolen += vcpu->arch.busy_stolen;
 457         vcpu->arch.busy_stolen = 0;
 458         spin_unlock(&vcpu->arch.tbacct_lock);
 459         if (!dt || !vpa)
 460                 return;
 461         memset(dt, 0, sizeof(struct dtl_entry));
 462         dt->dispatch_reason = 7;
 463         dt->processor_id = vc->pcpu + vcpu->arch.ptid;
 464         dt->timebase = now;
 465         dt->enqueue_to_dispatch_time = stolen;
 466         dt->srr0 = kvmppc_get_pc(vcpu);
 467         dt->srr1 = vcpu->arch.shregs.msr;
 468         ++dt;
 469         if (dt == vcpu->arch.dtl.pinned_end)
 470                 dt = vcpu->arch.dtl.pinned_addr;
 471         vcpu->arch.dtl_ptr = dt;
 472         /* order writing *dt vs. writing vpa->dtl_idx */
 473         smp_wmb();
 474         vpa->dtl_idx = ++vcpu->arch.dtl_index;
 475 }
 476
 477 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 478 {
 479         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 480         unsigned long target, ret = H_SUCCESS;
 481         struct kvm_vcpu *tvcpu;
 482         int idx;
 483
 484         switch (req) {
 485         case H_ENTER:
 486                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 487                 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
 488                                               kvmppc_get_gpr(vcpu, 5),
 489                                               kvmppc_get_gpr(vcpu, 6),
 490                                               kvmppc_get_gpr(vcpu, 7));
 491                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 492                 break;
 493         case H_CEDE:
 494                 break;
 495         case H_PROD:
 496                 target = kvmppc_get_gpr(vcpu, 4);
 497                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 498                 if (!tvcpu) {
 499                         ret = H_PARAMETER;
 500                         break;
 501                 }
 502                 tvcpu->arch.prodded = 1;
 503                 smp_mb();
 504                 if (vcpu->arch.ceded) {
 505                         if (waitqueue_active(&vcpu->wq)) {
 506                                 wake_up_interruptible(&vcpu->wq);
 507                                 vcpu->stat.halt_wakeup++;
 508                         }
 509                 }
 510                 break;
 511         case H_CONFER:
 512                 break;
 513         case H_REGISTER_VPA:
 514                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 515                                         kvmppc_get_gpr(vcpu, 5),
 516                                         kvmppc_get_gpr(vcpu, 6));
 517                 break;
 518         default:
 519                 return RESUME_HOST;
 520         }
 521         kvmppc_set_gpr(vcpu, 3, ret);
 522         vcpu->arch.hcall_needed = 0;
 523         return RESUME_GUEST;
 524 }
 525
 526 static int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 527                               struct task_struct *tsk)
 528 {
 529         int r = RESUME_HOST;
 530
 531         vcpu->stat.sum_exits++;
 532
 533         run->exit_reason = KVM_EXIT_UNKNOWN;
 534         run->ready_for_interrupt_injection = 1;
 535         switch (vcpu->arch.trap) {
 536         /* We're good on these - the host merely wanted to get our attention */
 537         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 538                 vcpu->stat.dec_exits++;
 539                 r = RESUME_GUEST;
 540                 break;
 541         case BOOK3S_INTERRUPT_EXTERNAL:
 542                 vcpu->stat.ext_intr_exits++;
 543                 r = RESUME_GUEST;
 544                 break;
 545         case BOOK3S_INTERRUPT_PERFMON:
 546                 r = RESUME_GUEST;
 547                 break;
 548         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 549                 /*
 550                  * Deliver a machine check interrupt to the guest.
 551                  * We have to do this, even if the host has handled the
 552                  * machine check, because machine checks use SRR0/1 and
 553                  * the interrupt might have trashed guest state in them.
 554                  */
 555                 kvmppc_book3s_queue_irqprio(vcpu,
 556                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 557                 r = RESUME_GUEST;
 558                 break;
 559         case BOOK3S_INTERRUPT_PROGRAM:
 560         {
 561                 ulong flags;
 562                 /*
 563                  * Normally program interrupts are delivered directly
 564                  * to the guest by the hardware, but we can get here
 565                  * as a result of a hypervisor emulation interrupt
 566                  * (e40) getting turned into a 700 by BML RTAS.
 567                  */
 568                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 569                 kvmppc_core_queue_program(vcpu, flags);
 570                 r = RESUME_GUEST;
 571                 break;
 572         }
 573         case BOOK3S_INTERRUPT_SYSCALL:
 574         {
 575                 /* hcall - punt to userspace */
 576                 int i;
 577
 578                 if (vcpu->arch.shregs.msr & MSR_PR) {
 579                         /* sc 1 from userspace - reflect to guest syscall */
 580                         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_SYSCALL);
 581                         r = RESUME_GUEST;
 582                         break;
 583                 }
 584                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 585                 for (i = 0; i < 9; ++i)
 586                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 587                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 588                 vcpu->arch.hcall_needed = 1;
 589                 r = RESUME_HOST;
 590                 break;
 591         }
 592         /*
 593          * We get these next two if the guest accesses a page which it thinks
 594          * it has mapped but which is not actually present, either because
 595          * it is for an emulated I/O device or because the corresonding
 596          * host page has been paged out.  Any other HDSI/HISI interrupts
 597          * have been handled already.
 598          */
 599         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 600                 r = RESUME_PAGE_FAULT;
 601                 break;
 602         case BOOK3S_INTERRUPT_H_INST_STORAGE:
 603                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
 604                 vcpu->arch.fault_dsisr = 0;
 605                 r = RESUME_PAGE_FAULT;
 606                 break;
 607         /*
 608          * This occurs if the guest executes an illegal instruction.
 609          * We just generate a program interrupt to the guest, since
 610          * we don't emulate any guest instructions at this stage.
 611          */
 612         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 613                 kvmppc_core_queue_program(vcpu, 0x80000);
 614                 r = RESUME_GUEST;
 615                 break;
 616         default:
 617                 kvmppc_dump_regs(vcpu);
 618                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 619                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 620                         vcpu->arch.shregs.msr);
 621                 r = RESUME_HOST;
 622                 BUG();
 623                 break;
 624         }
 625
 626         return r;
 627 }
 628
 629 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 630                                   struct kvm_sregs *sregs)
 631 {
 632         int i;
 633
 634         sregs->pvr = vcpu->arch.pvr;
 635
 636         memset(sregs, 0, sizeof(struct kvm_sregs));
 637         for (i = 0; i < vcpu->arch.slb_max; i++) {
 638                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
 639                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 640         }
 641
 642         return 0;
 643 }
 644
 645 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 646                                   struct kvm_sregs *sregs)
 647 {
 648         int i, j;
 649
 650         kvmppc_set_pvr(vcpu, sregs->pvr);
 651
 652         j = 0;
 653         for (i = 0; i < vcpu->arch.slb_nr; i++) {
 654                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
 655                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
 656                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
 657                         ++j;
 658                 }
 659         }
 660         vcpu->arch.slb_max = j;
 661
 662         return 0;
 663 }
 664
 665 int kvmppc_get_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
 666 {
 667         int r = 0;
 668         long int i;
 669
 670         switch (id) {
 671         case KVM_REG_PPC_HIOR:
 672                 *val = get_reg_val(id, 0);
 673                 break;
 674         case KVM_REG_PPC_DABR:
 675                 *val = get_reg_val(id, vcpu->arch.dabr);
 676                 break;
 677         case KVM_REG_PPC_DSCR:
 678                 *val = get_reg_val(id, vcpu->arch.dscr);
 679                 break;
 680         case KVM_REG_PPC_PURR:
 681                 *val = get_reg_val(id, vcpu->arch.purr);
 682                 break;
 683         case KVM_REG_PPC_SPURR:
 684                 *val = get_reg_val(id, vcpu->arch.spurr);
 685                 break;
 686         case KVM_REG_PPC_AMR:
 687                 *val = get_reg_val(id, vcpu->arch.amr);
 688                 break;
 689         case KVM_REG_PPC_UAMOR:
 690                 *val = get_reg_val(id, vcpu->arch.uamor);
 691                 break;
 692         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
 693                 i = id - KVM_REG_PPC_MMCR0;
 694                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
 695                 break;
 696         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
 697                 i = id - KVM_REG_PPC_PMC1;
 698                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
 699                 break;
 700 #ifdef CONFIG_VSX
 701         case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
 702                 if (cpu_has_feature(CPU_FTR_VSX)) {
 703                         /* VSX => FP reg i is stored in arch.vsr[2*i] */
 704                         long int i = id - KVM_REG_PPC_FPR0;
 705                         *val = get_reg_val(id, vcpu->arch.vsr[2 * i]);
 706                 } else {
 707                         /* let generic code handle it */
 708                         r = -EINVAL;
 709                 }
 710                 break;
 711         case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
 712                 if (cpu_has_feature(CPU_FTR_VSX)) {
 713                         long int i = id - KVM_REG_PPC_VSR0;
 714                         val->vsxval[0] = vcpu->arch.vsr[2 * i];
 715                         val->vsxval[1] = vcpu->arch.vsr[2 * i + 1];
 716                 } else {
 717                         r = -ENXIO;
 718                 }
 719                 break;
 720 #endif /* CONFIG_VSX */
 721         case KVM_REG_PPC_VPA_ADDR:
 722                 spin_lock(&vcpu->arch.vpa_update_lock);
 723                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
 724                 spin_unlock(&vcpu->arch.vpa_update_lock);
 725                 break;
 726         case KVM_REG_PPC_VPA_SLB:
 727                 spin_lock(&vcpu->arch.vpa_update_lock);
 728                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
 729                 val->vpaval.length = vcpu->arch.slb_shadow.len;
 730                 spin_unlock(&vcpu->arch.vpa_update_lock);
 731                 break;
 732         case KVM_REG_PPC_VPA_DTL:
 733                 spin_lock(&vcpu->arch.vpa_update_lock);
 734                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
 735                 val->vpaval.length = vcpu->arch.dtl.len;
 736                 spin_unlock(&vcpu->arch.vpa_update_lock);
 737                 break;
 738         default:
 739                 r = -EINVAL;
 740                 break;
 741         }
 742
 743         return r;
 744 }
 745
 746 int kvmppc_set_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
 747 {
 748         int r = 0;
 749         long int i;
 750         unsigned long addr, len;
 751
 752         switch (id) {
 753         case KVM_REG_PPC_HIOR:
 754                 /* Only allow this to be set to zero */
 755                 if (set_reg_val(id, *val))
 756                         r = -EINVAL;
 757                 break;
 758         case KVM_REG_PPC_DABR:
 759                 vcpu->arch.dabr = set_reg_val(id, *val);
 760                 break;
 761         case KVM_REG_PPC_DSCR:
 762                 vcpu->arch.dscr = set_reg_val(id, *val);
 763                 break;
 764         case KVM_REG_PPC_PURR:
 765                 vcpu->arch.purr = set_reg_val(id, *val);
 766                 break;
 767         case KVM_REG_PPC_SPURR:
 768                 vcpu->arch.spurr = set_reg_val(id, *val);
 769                 break;
 770         case KVM_REG_PPC_AMR:
 771                 vcpu->arch.amr = set_reg_val(id, *val);
 772                 break;
 773         case KVM_REG_PPC_UAMOR:
 774                 vcpu->arch.uamor = set_reg_val(id, *val);
 775                 break;
 776         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
 777                 i = id - KVM_REG_PPC_MMCR0;
 778                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
 779                 break;
 780         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
 781                 i = id - KVM_REG_PPC_PMC1;
 782                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
 783                 break;
 784 #ifdef CONFIG_VSX
 785         case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
 786                 if (cpu_has_feature(CPU_FTR_VSX)) {
 787                         /* VSX => FP reg i is stored in arch.vsr[2*i] */
 788                         long int i = id - KVM_REG_PPC_FPR0;
 789                         vcpu->arch.vsr[2 * i] = set_reg_val(id, *val);
 790                 } else {
 791                         /* let generic code handle it */
 792                         r = -EINVAL;
 793                 }
 794                 break;
 795         case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
 796                 if (cpu_has_feature(CPU_FTR_VSX)) {
 797                         long int i = id - KVM_REG_PPC_VSR0;
 798                         vcpu->arch.vsr[2 * i] = val->vsxval[0];
 799                         vcpu->arch.vsr[2 * i + 1] = val->vsxval[1];
 800                 } else {
 801                         r = -ENXIO;
 802                 }
 803                 break;
 804 #endif /* CONFIG_VSX */
 805         case KVM_REG_PPC_VPA_ADDR:
 806                 addr = set_reg_val(id, *val);
 807                 r = -EINVAL;
 808                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
 809                               vcpu->arch.dtl.next_gpa))
 810                         break;
 811                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
 812                 break;
 813         case KVM_REG_PPC_VPA_SLB:
 814                 addr = val->vpaval.addr;
 815                 len = val->vpaval.length;
 816                 r = -EINVAL;
 817                 if (addr && !vcpu->arch.vpa.next_gpa)
 818                         break;
 819                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
 820                 break;
 821         case KVM_REG_PPC_VPA_DTL:
 822                 addr = val->vpaval.addr;
 823                 len = val->vpaval.length;
 824                 r = -EINVAL;
 825                 if (addr && (len < sizeof(struct dtl_entry) ||
 826                              !vcpu->arch.vpa.next_gpa))
 827                         break;
 828                 len -= len % sizeof(struct dtl_entry);
 829                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
 830                 break;
 831         default:
 832                 r = -EINVAL;
 833                 break;
 834         }
 835
 836         return r;
 837 }
 838
 839 int kvmppc_core_check_processor_compat(void)
 840 {
 841         if (cpu_has_feature(CPU_FTR_HVMODE))
 842                 return 0;
 843         return -EIO;
 844 }
 845
 846 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
 847 {
 848         struct kvm_vcpu *vcpu;
 849         int err = -EINVAL;
 850         int core;
 851         struct kvmppc_vcore *vcore;
 852
 853         core = id / threads_per_core;
 854         if (core >= KVM_MAX_VCORES)
 855                 goto out;
 856
 857         err = -ENOMEM;
 858         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 859         if (!vcpu)
 860                 goto out;
 861
 862         err = kvm_vcpu_init(vcpu, kvm, id);
 863         if (err)
 864                 goto free_vcpu;
 865
 866         vcpu->arch.shared = &vcpu->arch.shregs;
 867         vcpu->arch.mmcr[0] = MMCR0_FC;
 868         vcpu->arch.ctrl = CTRL_RUNLATCH;
 869         /* default to host PVR, since we can't spoof it */
 870         vcpu->arch.pvr = mfspr(SPRN_PVR);
 871         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
 872         spin_lock_init(&vcpu->arch.vpa_update_lock);
 873         spin_lock_init(&vcpu->arch.tbacct_lock);
 874         vcpu->arch.busy_preempt = TB_NIL;
 875
 876         kvmppc_mmu_book3s_hv_init(vcpu);
 877
 878         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
 879
 880         init_waitqueue_head(&vcpu->arch.cpu_run);
 881
 882         mutex_lock(&kvm->lock);
 883         vcore = kvm->arch.vcores[core];
 884         if (!vcore) {
 885                 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
 886                 if (vcore) {
 887                         INIT_LIST_HEAD(&vcore->runnable_threads);
 888                         spin_lock_init(&vcore->lock);
 889                         init_waitqueue_head(&vcore->wq);
 890                         vcore->preempt_tb = TB_NIL;
 891                 }
 892                 kvm->arch.vcores[core] = vcore;
 893                 kvm->arch.online_vcores++;
 894         }
 895         mutex_unlock(&kvm->lock);
 896
 897         if (!vcore)
 898                 goto free_vcpu;
 899
 900         spin_lock(&vcore->lock);
 901         ++vcore->num_threads;
 902         spin_unlock(&vcore->lock);
 903         vcpu->arch.vcore = vcore;
 904
 905         vcpu->arch.cpu_type = KVM_CPU_3S_64;
 906         kvmppc_sanity_check(vcpu);
 907
 908         return vcpu;
 909
 910 free_vcpu:
 911         kmem_cache_free(kvm_vcpu_cache, vcpu);
 912 out:
 913         return ERR_PTR(err);
 914 }
 915
 916 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
 917 {
 918         spin_lock(&vcpu->arch.vpa_update_lock);
 919         if (vcpu->arch.dtl.pinned_addr)
 920                 kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.dtl.pinned_addr);
 921         if (vcpu->arch.slb_shadow.pinned_addr)
 922                 kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.slb_shadow.pinned_addr);
 923         if (vcpu->arch.vpa.pinned_addr)
 924                 kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.vpa.pinned_addr);
 925         spin_unlock(&vcpu->arch.vpa_update_lock);
 926         kvm_vcpu_uninit(vcpu);
 927         kmem_cache_free(kvm_vcpu_cache, vcpu);
 928 }
 929
 930 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
 931 {
 932         unsigned long dec_nsec, now;
 933
 934         now = get_tb();
 935         if (now > vcpu->arch.dec_expires) {
 936                 /* decrementer has already gone negative */
 937                 kvmppc_core_queue_dec(vcpu);
 938                 kvmppc_core_prepare_to_enter(vcpu);
 939                 return;
 940         }
 941         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
 942                    / tb_ticks_per_sec;
 943         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
 944                       HRTIMER_MODE_REL);
 945         vcpu->arch.timer_running = 1;
 946 }
 947
 948 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
 949 {
 950         vcpu->arch.ceded = 0;
 951         if (vcpu->arch.timer_running) {
 952                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
 953                 vcpu->arch.timer_running = 0;
 954         }
 955 }
 956
 957 extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
 958 extern void xics_wake_cpu(int cpu);
 959
 960 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
 961                                    struct kvm_vcpu *vcpu)
 962 {
 963         u64 now;
 964
 965         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
 966                 return;
 967         spin_lock(&vcpu->arch.tbacct_lock);
 968         now = mftb();
 969         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
 970                 vcpu->arch.stolen_logged;
 971         vcpu->arch.busy_preempt = now;
 972         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
 973         spin_unlock(&vcpu->arch.tbacct_lock);
 974         --vc->n_runnable;
 975         list_del(&vcpu->arch.run_list);
 976 }
 977
 978 static int kvmppc_grab_hwthread(int cpu)
 979 {
 980         struct paca_struct *tpaca;
 981         long timeout = 1000;
 982
 983         tpaca = &paca[cpu];
 984
 985         /* Ensure the thread won't go into the kernel if it wakes */
 986         tpaca->kvm_hstate.hwthread_req = 1;
 987         tpaca->kvm_hstate.kvm_vcpu = NULL;
 988
 989         /*
 990          * If the thread is already executing in the kernel (e.g. handling
 991          * a stray interrupt), wait for it to get back to nap mode.
 992          * The smp_mb() is to ensure that our setting of hwthread_req
 993          * is visible before we look at hwthread_state, so if this
 994          * races with the code at system_reset_pSeries and the thread
 995          * misses our setting of hwthread_req, we are sure to see its
 996          * setting of hwthread_state, and vice versa.
 997          */
 998         smp_mb();
 999         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1000                 if (--timeout <= 0) {
1001                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1002                         return -EBUSY;
1003                 }
1004                 udelay(1);
1005         }
1006         return 0;
1007 }
1008
1009 static void kvmppc_release_hwthread(int cpu)
1010 {
1011         struct paca_struct *tpaca;
1012
1013         tpaca = &paca[cpu];
1014         tpaca->kvm_hstate.hwthread_req = 0;
1015         tpaca->kvm_hstate.kvm_vcpu = NULL;
1016 }
1017
1018 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1019 {
1020         int cpu;
1021         struct paca_struct *tpaca;
1022         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1023
1024         if (vcpu->arch.timer_running) {
1025                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1026                 vcpu->arch.timer_running = 0;
1027         }
1028         cpu = vc->pcpu + vcpu->arch.ptid;
1029         tpaca = &paca[cpu];
1030         tpaca->kvm_hstate.kvm_vcpu = vcpu;
1031         tpaca->kvm_hstate.kvm_vcore = vc;
1032         tpaca->kvm_hstate.napping = 0;
1033         vcpu->cpu = vc->pcpu;
1034         smp_wmb();
1035 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1036         if (vcpu->arch.ptid) {
1037                 xics_wake_cpu(cpu);
1038                 ++vc->n_woken;
1039         }
1040 #endif
1041 }
1042
1043 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1044 {
1045         int i;
1046
1047         HMT_low();
1048         i = 0;
1049         while (vc->nap_count < vc->n_woken) {
1050                 if (++i >= 1000000) {
1051                         pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1052                                vc->nap_count, vc->n_woken);
1053                         break;
1054                 }
1055                 cpu_relax();
1056         }
1057         HMT_medium();
1058 }
1059
1060 /*
1061  * Check that we are on thread 0 and that any other threads in
1062  * this core are off-line.  Then grab the threads so they can't
1063  * enter the kernel.
1064  */
1065 static int on_primary_thread(void)
1066 {
1067         int cpu = smp_processor_id();
1068         int thr = cpu_thread_in_core(cpu);
1069
1070         if (thr)
1071                 return 0;
1072         while (++thr < threads_per_core)
1073                 if (cpu_online(cpu + thr))
1074                         return 0;
1075
1076         /* Grab all hw threads so they can't go into the kernel */
1077         for (thr = 1; thr < threads_per_core; ++thr) {
1078                 if (kvmppc_grab_hwthread(cpu + thr)) {
1079                         /* Couldn't grab one; let the others go */
1080                         do {
1081                                 kvmppc_release_hwthread(cpu + thr);
1082                         } while (--thr > 0);
1083                         return 0;
1084                 }
1085         }
1086         return 1;
1087 }
1088
1089 /*
1090  * Run a set of guest threads on a physical core.
1091  * Called with vc->lock held.
1092  */
1093 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1094 {
1095         struct kvm_vcpu *vcpu, *vcpu0, *vnext;
1096         long ret;
1097         u64 now;
1098         int ptid, i, need_vpa_update;
1099         int srcu_idx;
1100         struct kvm_vcpu *vcpus_to_update[threads_per_core];
1101
1102         /* don't start if any threads have a signal pending */
1103         need_vpa_update = 0;
1104         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1105                 if (signal_pending(vcpu->arch.run_task))
1106                         return;
1107                 if (vcpu->arch.vpa.update_pending ||
1108                     vcpu->arch.slb_shadow.update_pending ||
1109                     vcpu->arch.dtl.update_pending)
1110                         vcpus_to_update[need_vpa_update++] = vcpu;
1111         }
1112
1113         /*
1114          * Initialize *vc, in particular vc->vcore_state, so we can
1115          * drop the vcore lock if necessary.
1116          */
1117         vc->n_woken = 0;
1118         vc->nap_count = 0;
1119         vc->entry_exit_count = 0;
1120         vc->vcore_state = VCORE_STARTING;
1121         vc->in_guest = 0;
1122         vc->napping_threads = 0;
1123
1124         /*
1125          * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1126          * which can't be called with any spinlocks held.
1127          */
1128         if (need_vpa_update) {
1129                 spin_unlock(&vc->lock);
1130                 for (i = 0; i < need_vpa_update; ++i)
1131                         kvmppc_update_vpas(vcpus_to_update[i]);
1132                 spin_lock(&vc->lock);
1133         }
1134
1135         /*
1136          * Assign physical thread IDs, first to non-ceded vcpus
1137          * and then to ceded ones.
1138          */
1139         ptid = 0;
1140         vcpu0 = NULL;
1141         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1142                 if (!vcpu->arch.ceded) {
1143                         if (!ptid)
1144                                 vcpu0 = vcpu;
1145                         vcpu->arch.ptid = ptid++;
1146                 }
1147         }
1148         if (!vcpu0)
1149                 goto out;       /* nothing to run; should never happen */
1150         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1151                 if (vcpu->arch.ceded)
1152                         vcpu->arch.ptid = ptid++;
1153
1154         /*
1155          * Make sure we are running on thread 0, and that
1156          * secondary threads are offline.
1157          */
1158         if (threads_per_core > 1 && !on_primary_thread()) {
1159                 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1160                         vcpu->arch.ret = -EBUSY;
1161                 goto out;
1162         }
1163
1164         vc->pcpu = smp_processor_id();
1165         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1166                 kvmppc_start_thread(vcpu);
1167                 kvmppc_create_dtl_entry(vcpu, vc);
1168         }
1169
1170         vc->vcore_state = VCORE_RUNNING;
1171         preempt_disable();
1172         spin_unlock(&vc->lock);
1173
1174         kvm_guest_enter();
1175
1176         srcu_idx = srcu_read_lock(&vcpu0->kvm->srcu);
1177
1178         __kvmppc_vcore_entry(NULL, vcpu0);
1179
1180         spin_lock(&vc->lock);
1181         /* disable sending of IPIs on virtual external irqs */
1182         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1183                 vcpu->cpu = -1;
1184         /* wait for secondary threads to finish writing their state to memory */
1185         if (vc->nap_count < vc->n_woken)
1186                 kvmppc_wait_for_nap(vc);
1187         for (i = 0; i < threads_per_core; ++i)
1188                 kvmppc_release_hwthread(vc->pcpu + i);
1189         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
1190         vc->vcore_state = VCORE_EXITING;
1191         spin_unlock(&vc->lock);
1192
1193         srcu_read_unlock(&vcpu0->kvm->srcu, srcu_idx);
1194
1195         /* make sure updates to secondary vcpu structs are visible now */
1196         smp_mb();
1197         kvm_guest_exit();
1198
1199         preempt_enable();
1200         kvm_resched(vcpu);
1201
1202         spin_lock(&vc->lock);
1203         now = get_tb();
1204         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1205                 /* cancel pending dec exception if dec is positive */
1206                 if (now < vcpu->arch.dec_expires &&
1207                     kvmppc_core_pending_dec(vcpu))
1208                         kvmppc_core_dequeue_dec(vcpu);
1209
1210                 ret = RESUME_GUEST;
1211                 if (vcpu->arch.trap)
1212                         ret = kvmppc_handle_exit(vcpu->arch.kvm_run, vcpu,
1213                                                  vcpu->arch.run_task);
1214
1215                 vcpu->arch.ret = ret;
1216                 vcpu->arch.trap = 0;
1217
1218                 if (vcpu->arch.ceded) {
1219                         if (ret != RESUME_GUEST)
1220                                 kvmppc_end_cede(vcpu);
1221                         else
1222                                 kvmppc_set_timer(vcpu);
1223                 }
1224         }
1225
1226  out:
1227         vc->vcore_state = VCORE_INACTIVE;
1228         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1229                                  arch.run_list) {
1230                 if (vcpu->arch.ret != RESUME_GUEST) {
1231                         kvmppc_remove_runnable(vc, vcpu);
1232                         wake_up(&vcpu->arch.cpu_run);
1233                 }
1234         }
1235 }
1236
1237 /*
1238  * Wait for some other vcpu thread to execute us, and
1239  * wake us up when we need to handle something in the host.
1240  */
1241 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1242 {
1243         DEFINE_WAIT(wait);
1244
1245         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1246         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1247                 schedule();
1248         finish_wait(&vcpu->arch.cpu_run, &wait);
1249 }
1250
1251 /*
1252  * All the vcpus in this vcore are idle, so wait for a decrementer
1253  * or external interrupt to one of the vcpus.  vc->lock is held.
1254  */
1255 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1256 {
1257         DEFINE_WAIT(wait);
1258
1259         prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1260         vc->vcore_state = VCORE_SLEEPING;
1261         spin_unlock(&vc->lock);
1262         schedule();
1263         finish_wait(&vc->wq, &wait);
1264         spin_lock(&vc->lock);
1265         vc->vcore_state = VCORE_INACTIVE;
1266 }
1267
1268 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1269 {
1270         int n_ceded;
1271         struct kvmppc_vcore *vc;
1272         struct kvm_vcpu *v, *vn;
1273
1274         kvm_run->exit_reason = 0;
1275         vcpu->arch.ret = RESUME_GUEST;
1276         vcpu->arch.trap = 0;
1277         kvmppc_update_vpas(vcpu);
1278
1279         /*
1280          * Synchronize with other threads in this virtual core
1281          */
1282         vc = vcpu->arch.vcore;
1283         spin_lock(&vc->lock);
1284         vcpu->arch.ceded = 0;
1285         vcpu->arch.run_task = current;
1286         vcpu->arch.kvm_run = kvm_run;
1287         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
1288         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1289         vcpu->arch.busy_preempt = TB_NIL;
1290         list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1291         ++vc->n_runnable;
1292
1293         /*
1294          * This happens the first time this is called for a vcpu.
1295          * If the vcore is already running, we may be able to start
1296          * this thread straight away and have it join in.
1297          */
1298         if (!signal_pending(current)) {
1299                 if (vc->vcore_state == VCORE_RUNNING &&
1300                     VCORE_EXIT_COUNT(vc) == 0) {
1301                         vcpu->arch.ptid = vc->n_runnable - 1;
1302                         kvmppc_create_dtl_entry(vcpu, vc);
1303                         kvmppc_start_thread(vcpu);
1304                 } else if (vc->vcore_state == VCORE_SLEEPING) {
1305                         wake_up(&vc->wq);
1306                 }
1307
1308         }
1309
1310         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1311                !signal_pending(current)) {
1312                 if (vc->vcore_state != VCORE_INACTIVE) {
1313                         spin_unlock(&vc->lock);
1314                         kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1315                         spin_lock(&vc->lock);
1316                         continue;
1317                 }
1318                 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1319                                          arch.run_list) {
1320                         kvmppc_core_prepare_to_enter(v);
1321                         if (signal_pending(v->arch.run_task)) {
1322                                 kvmppc_remove_runnable(vc, v);
1323                                 v->stat.signal_exits++;
1324                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1325                                 v->arch.ret = -EINTR;
1326                                 wake_up(&v->arch.cpu_run);
1327                         }
1328                 }
1329                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1330                         break;
1331                 vc->runner = vcpu;
1332                 n_ceded = 0;
1333                 list_for_each_entry(v, &vc->runnable_threads, arch.run_list)
1334                         if (!v->arch.pending_exceptions)
1335                                 n_ceded += v->arch.ceded;
1336                 if (n_ceded == vc->n_runnable)
1337                         kvmppc_vcore_blocked(vc);
1338                 else
1339                         kvmppc_run_core(vc);
1340                 vc->runner = NULL;
1341         }
1342
1343         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1344                (vc->vcore_state == VCORE_RUNNING ||
1345                 vc->vcore_state == VCORE_EXITING)) {
1346                 spin_unlock(&vc->lock);
1347                 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1348                 spin_lock(&vc->lock);
1349         }
1350
1351         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1352                 kvmppc_remove_runnable(vc, vcpu);
1353                 vcpu->stat.signal_exits++;
1354                 kvm_run->exit_reason = KVM_EXIT_INTR;
1355                 vcpu->arch.ret = -EINTR;
1356         }
1357
1358         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1359                 /* Wake up some vcpu to run the core */
1360                 v = list_first_entry(&vc->runnable_threads,
1361                                      struct kvm_vcpu, arch.run_list);
1362                 wake_up(&v->arch.cpu_run);
1363         }
1364
1365         spin_unlock(&vc->lock);
1366         return vcpu->arch.ret;
1367 }
1368
1369 int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
1370 {
1371         int r;
1372         int srcu_idx;
1373
1374         if (!vcpu->arch.sane) {
1375                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1376                 return -EINVAL;
1377         }
1378
1379         kvmppc_core_prepare_to_enter(vcpu);
1380
1381         /* No need to go into the guest when all we'll do is come back out */
1382         if (signal_pending(current)) {
1383                 run->exit_reason = KVM_EXIT_INTR;
1384                 return -EINTR;
1385         }
1386
1387         atomic_inc(&vcpu->kvm->arch.vcpus_running);
1388         /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1389         smp_mb();
1390
1391         /* On the first time here, set up HTAB and VRMA or RMA */
1392         if (!vcpu->kvm->arch.rma_setup_done) {
1393                 r = kvmppc_hv_setup_htab_rma(vcpu);
1394                 if (r)
1395                         goto out;
1396         }
1397
1398         flush_fp_to_thread(current);
1399         flush_altivec_to_thread(current);
1400         flush_vsx_to_thread(current);
1401         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1402         vcpu->arch.pgdir = current->mm->pgd;
1403         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1404
1405         do {
1406                 r = kvmppc_run_vcpu(run, vcpu);
1407
1408                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1409                     !(vcpu->arch.shregs.msr & MSR_PR)) {
1410                         r = kvmppc_pseries_do_hcall(vcpu);
1411                         kvmppc_core_prepare_to_enter(vcpu);
1412                 } else if (r == RESUME_PAGE_FAULT) {
1413                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1414                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
1415                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1416                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1417                 }
1418         } while (r == RESUME_GUEST);
1419
1420  out:
1421         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1422         atomic_dec(&vcpu->kvm->arch.vcpus_running);
1423         return r;
1424 }
1425
1426
1427 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1428    Assumes POWER7 or PPC970. */
1429 static inline int lpcr_rmls(unsigned long rma_size)
1430 {
1431         switch (rma_size) {
1432         case 32ul << 20:        /* 32 MB */
1433                 if (cpu_has_feature(CPU_FTR_ARCH_206))
1434                         return 8;       /* only supported on POWER7 */
1435                 return -1;
1436         case 64ul << 20:        /* 64 MB */
1437                 return 3;
1438         case 128ul << 20:       /* 128 MB */
1439                 return 7;
1440         case 256ul << 20:       /* 256 MB */
1441                 return 4;
1442         case 1ul << 30:         /* 1 GB */
1443                 return 2;
1444         case 16ul << 30:        /* 16 GB */
1445                 return 1;
1446         case 256ul << 30:       /* 256 GB */
1447                 return 0;
1448         default:
1449                 return -1;
1450         }
1451 }
1452
1453 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1454 {
1455         struct kvmppc_linear_info *ri = vma->vm_file->private_data;
1456         struct page *page;
1457
1458         if (vmf->pgoff >= ri->npages)
1459                 return VM_FAULT_SIGBUS;
1460
1461         page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1462         get_page(page);
1463         vmf->page = page;
1464         return 0;
1465 }
1466
1467 static const struct vm_operations_struct kvm_rma_vm_ops = {
1468         .fault = kvm_rma_fault,
1469 };
1470
1471 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1472 {
1473         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1474         vma->vm_ops = &kvm_rma_vm_ops;
1475         return 0;
1476 }
1477
1478 static int kvm_rma_release(struct inode *inode, struct file *filp)
1479 {
1480         struct kvmppc_linear_info *ri = filp->private_data;
1481
1482         kvm_release_rma(ri);
1483         return 0;
1484 }
1485
1486 static struct file_operations kvm_rma_fops = {
1487         .mmap           = kvm_rma_mmap,
1488         .release        = kvm_rma_release,
1489 };
1490
1491 long kvm_vm_ioctl_allocate_rma(struct kvm *kvm, struct kvm_allocate_rma *ret)
1492 {
1493         struct kvmppc_linear_info *ri;
1494         long fd;
1495
1496         ri = kvm_alloc_rma();
1497         if (!ri)
1498                 return -ENOMEM;
1499
1500         fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR);
1501         if (fd < 0)
1502                 kvm_release_rma(ri);
1503
1504         ret->rma_size = ri->npages << PAGE_SHIFT;
1505         return fd;
1506 }
1507
1508 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1509                                      int linux_psize)
1510 {
1511         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1512
1513         if (!def->shift)
1514                 return;
1515         (*sps)->page_shift = def->shift;
1516         (*sps)->slb_enc = def->sllp;
1517         (*sps)->enc[0].page_shift = def->shift;
1518         (*sps)->enc[0].pte_enc = def->penc;
1519         (*sps)++;
1520 }
1521
1522 int kvm_vm_ioctl_get_smmu_info(struct kvm *kvm, struct kvm_ppc_smmu_info *info)
1523 {
1524         struct kvm_ppc_one_seg_page_size *sps;
1525
1526         info->flags = KVM_PPC_PAGE_SIZES_REAL;
1527         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1528                 info->flags |= KVM_PPC_1T_SEGMENTS;
1529         info->slb_size = mmu_slb_size;
1530
1531         /* We only support these sizes for now, and no muti-size segments */
1532         sps = &info->sps[0];
1533         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1534         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1535         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1536
1537         return 0;
1538 }
1539
1540 /*
1541  * Get (and clear) the dirty memory log for a memory slot.
1542  */
1543 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1544 {
1545         struct kvm_memory_slot *memslot;
1546         int r;
1547         unsigned long n;
1548
1549         mutex_lock(&kvm->slots_lock);
1550
1551         r = -EINVAL;
1552         if (log->slot >= KVM_USER_MEM_SLOTS)
1553                 goto out;
1554
1555         memslot = id_to_memslot(kvm->memslots, log->slot);
1556         r = -ENOENT;
1557         if (!memslot->dirty_bitmap)
1558                 goto out;
1559
1560         n = kvm_dirty_bitmap_bytes(memslot);
1561         memset(memslot->dirty_bitmap, 0, n);
1562
1563         r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
1564         if (r)
1565                 goto out;
1566
1567         r = -EFAULT;
1568         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1569                 goto out;
1570
1571         r = 0;
1572 out:
1573         mutex_unlock(&kvm->slots_lock);
1574         return r;
1575 }
1576
1577 static void unpin_slot(struct kvm_memory_slot *memslot)
1578 {
1579         unsigned long *physp;
1580         unsigned long j, npages, pfn;
1581         struct page *page;
1582
1583         physp = memslot->arch.slot_phys;
1584         npages = memslot->npages;
1585         if (!physp)
1586                 return;
1587         for (j = 0; j < npages; j++) {
1588                 if (!(physp[j] & KVMPPC_GOT_PAGE))
1589                         continue;
1590                 pfn = physp[j] >> PAGE_SHIFT;
1591                 page = pfn_to_page(pfn);
1592                 SetPageDirty(page);
1593                 put_page(page);
1594         }
1595 }
1596
1597 void kvmppc_core_free_memslot(struct kvm_memory_slot *free,
1598                               struct kvm_memory_slot *dont)
1599 {
1600         if (!dont || free->arch.rmap != dont->arch.rmap) {
1601                 vfree(free->arch.rmap);
1602                 free->arch.rmap = NULL;
1603         }
1604         if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
1605                 unpin_slot(free);
1606                 vfree(free->arch.slot_phys);
1607                 free->arch.slot_phys = NULL;
1608         }
1609 }
1610
1611 int kvmppc_core_create_memslot(struct kvm_memory_slot *slot,
1612                                unsigned long npages)
1613 {
1614         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
1615         if (!slot->arch.rmap)
1616                 return -ENOMEM;
1617         slot->arch.slot_phys = NULL;
1618
1619         return 0;
1620 }
1621
1622 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1623                                       struct kvm_memory_slot *memslot,
1624                                       struct kvm_userspace_memory_region *mem)
1625 {
1626         unsigned long *phys;
1627
1628         /* Allocate a slot_phys array if needed */
1629         phys = memslot->arch.slot_phys;
1630         if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
1631                 phys = vzalloc(memslot->npages * sizeof(unsigned long));
1632                 if (!phys)
1633                         return -ENOMEM;
1634                 memslot->arch.slot_phys = phys;
1635         }
1636
1637         return 0;
1638 }
1639
1640 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1641                                       struct kvm_userspace_memory_region *mem,
1642                                       struct kvm_memory_slot old)
1643 {
1644         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
1645         struct kvm_memory_slot *memslot;
1646
1647         if (npages && old.npages) {
1648                 /*
1649                  * If modifying a memslot, reset all the rmap dirty bits.
1650                  * If this is a new memslot, we don't need to do anything
1651                  * since the rmap array starts out as all zeroes,
1652                  * i.e. no pages are dirty.
1653                  */
1654                 memslot = id_to_memslot(kvm->memslots, mem->slot);
1655                 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
1656         }
1657 }
1658
1659 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
1660 {
1661         int err = 0;
1662         struct kvm *kvm = vcpu->kvm;
1663         struct kvmppc_linear_info *ri = NULL;
1664         unsigned long hva;
1665         struct kvm_memory_slot *memslot;
1666         struct vm_area_struct *vma;
1667         unsigned long lpcr, senc;
1668         unsigned long psize, porder;
1669         unsigned long rma_size;
1670         unsigned long rmls;
1671         unsigned long *physp;
1672         unsigned long i, npages;
1673         int srcu_idx;
1674
1675         mutex_lock(&kvm->lock);
1676         if (kvm->arch.rma_setup_done)
1677                 goto out;       /* another vcpu beat us to it */
1678
1679         /* Allocate hashed page table (if not done already) and reset it */
1680         if (!kvm->arch.hpt_virt) {
1681                 err = kvmppc_alloc_hpt(kvm, NULL);
1682                 if (err) {
1683                         pr_err("KVM: Couldn't alloc HPT\n");
1684                         goto out;
1685                 }
1686         }
1687
1688         /* Look up the memslot for guest physical address 0 */
1689         srcu_idx = srcu_read_lock(&kvm->srcu);
1690         memslot = gfn_to_memslot(kvm, 0);
1691
1692         /* We must have some memory at 0 by now */
1693         err = -EINVAL;
1694         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1695                 goto out_srcu;
1696
1697         /* Look up the VMA for the start of this memory slot */
1698         hva = memslot->userspace_addr;
1699         down_read(&current->mm->mmap_sem);
1700         vma = find_vma(current->mm, hva);
1701         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
1702                 goto up_out;
1703
1704         psize = vma_kernel_pagesize(vma);
1705         porder = __ilog2(psize);
1706
1707         /* Is this one of our preallocated RMAs? */
1708         if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
1709             hva == vma->vm_start)
1710                 ri = vma->vm_file->private_data;
1711
1712         up_read(&current->mm->mmap_sem);
1713
1714         if (!ri) {
1715                 /* On POWER7, use VRMA; on PPC970, give up */
1716                 err = -EPERM;
1717                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1718                         pr_err("KVM: CPU requires an RMO\n");
1719                         goto out_srcu;
1720                 }
1721
1722                 /* We can handle 4k, 64k or 16M pages in the VRMA */
1723                 err = -EINVAL;
1724                 if (!(psize == 0x1000 || psize == 0x10000 ||
1725                       psize == 0x1000000))
1726                         goto out_srcu;
1727
1728                 /* Update VRMASD field in the LPCR */
1729                 senc = slb_pgsize_encoding(psize);
1730                 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1731                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
1732                 lpcr = kvm->arch.lpcr & ~LPCR_VRMASD;
1733                 lpcr |= senc << (LPCR_VRMASD_SH - 4);
1734                 kvm->arch.lpcr = lpcr;
1735
1736                 /* Create HPTEs in the hash page table for the VRMA */
1737                 kvmppc_map_vrma(vcpu, memslot, porder);
1738
1739         } else {
1740                 /* Set up to use an RMO region */
1741                 rma_size = ri->npages;
1742                 if (rma_size > memslot->npages)
1743                         rma_size = memslot->npages;
1744                 rma_size <<= PAGE_SHIFT;
1745                 rmls = lpcr_rmls(rma_size);
1746                 err = -EINVAL;
1747                 if (rmls < 0) {
1748                         pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
1749                         goto out_srcu;
1750                 }
1751                 atomic_inc(&ri->use_count);
1752                 kvm->arch.rma = ri;
1753
1754                 /* Update LPCR and RMOR */
1755                 lpcr = kvm->arch.lpcr;
1756                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1757                         /* PPC970; insert RMLS value (split field) in HID4 */
1758                         lpcr &= ~((1ul << HID4_RMLS0_SH) |
1759                                   (3ul << HID4_RMLS2_SH));
1760                         lpcr |= ((rmls >> 2) << HID4_RMLS0_SH) |
1761                                 ((rmls & 3) << HID4_RMLS2_SH);
1762                         /* RMOR is also in HID4 */
1763                         lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
1764                                 << HID4_RMOR_SH;
1765                 } else {
1766                         /* POWER7 */
1767                         lpcr &= ~(LPCR_VPM0 | LPCR_VRMA_L);
1768                         lpcr |= rmls << LPCR_RMLS_SH;
1769                         kvm->arch.rmor = kvm->arch.rma->base_pfn << PAGE_SHIFT;
1770                 }
1771                 kvm->arch.lpcr = lpcr;
1772                 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
1773                         ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
1774
1775                 /* Initialize phys addrs of pages in RMO */
1776                 npages = ri->npages;
1777                 porder = __ilog2(npages);
1778                 physp = memslot->arch.slot_phys;
1779                 if (physp) {
1780                         if (npages > memslot->npages)
1781                                 npages = memslot->npages;
1782                         spin_lock(&kvm->arch.slot_phys_lock);
1783                         for (i = 0; i < npages; ++i)
1784                                 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
1785                                         porder;
1786                         spin_unlock(&kvm->arch.slot_phys_lock);
1787                 }
1788         }
1789
1790         /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
1791         smp_wmb();
1792         kvm->arch.rma_setup_done = 1;
1793         err = 0;
1794  out_srcu:
1795         srcu_read_unlock(&kvm->srcu, srcu_idx);
1796  out:
1797         mutex_unlock(&kvm->lock);
1798         return err;
1799
1800  up_out:
1801         up_read(&current->mm->mmap_sem);
1802         goto out;
1803 }
1804
1805 int kvmppc_core_init_vm(struct kvm *kvm)
1806 {
1807         unsigned long lpcr, lpid;
1808
1809         /* Allocate the guest's logical partition ID */
1810
1811         lpid = kvmppc_alloc_lpid();
1812         if (lpid < 0)
1813                 return -ENOMEM;
1814         kvm->arch.lpid = lpid;
1815
1816         /*
1817          * Since we don't flush the TLB when tearing down a VM,
1818          * and this lpid might have previously been used,
1819          * make sure we flush on each core before running the new VM.
1820          */
1821         cpumask_setall(&kvm->arch.need_tlb_flush);
1822
1823         INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
1824
1825         kvm->arch.rma = NULL;
1826
1827         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
1828
1829         if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1830                 /* PPC970; HID4 is effectively the LPCR */
1831                 kvm->arch.host_lpid = 0;
1832                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
1833                 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
1834                 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
1835                         ((lpid & 0xf) << HID4_LPID5_SH);
1836         } else {
1837                 /* POWER7; init LPCR for virtual RMA mode */
1838                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
1839                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
1840                 lpcr &= LPCR_PECE | LPCR_LPES;
1841                 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
1842                         LPCR_VPM0 | LPCR_VPM1;
1843                 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
1844                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
1845         }
1846         kvm->arch.lpcr = lpcr;
1847
1848         kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
1849         spin_lock_init(&kvm->arch.slot_phys_lock);
1850
1851         /*
1852          * Don't allow secondary CPU threads to come online
1853          * while any KVM VMs exist.
1854          */
1855         inhibit_secondary_onlining();
1856
1857         return 0;
1858 }
1859
1860 void kvmppc_core_destroy_vm(struct kvm *kvm)
1861 {
1862         uninhibit_secondary_onlining();
1863
1864         if (kvm->arch.rma) {
1865                 kvm_release_rma(kvm->arch.rma);
1866                 kvm->arch.rma = NULL;
1867         }
1868
1869         kvmppc_free_hpt(kvm);
1870         WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1871 }
1872
1873 /* These are stubs for now */
1874 void kvmppc_mmu_pte_pflush(struct kvm_vcpu *vcpu, ulong pa_start, ulong pa_end)
1875 {
1876 }
1877
1878 /* We don't need to emulate any privileged instructions or dcbz */
1879 int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu,
1880                            unsigned int inst, int *advance)
1881 {
1882         return EMULATE_FAIL;
1883 }
1884
1885 int kvmppc_core_emulate_mtspr(struct kvm_vcpu *vcpu, int sprn, ulong spr_val)
1886 {
1887         return EMULATE_FAIL;
1888 }
1889
1890 int kvmppc_core_emulate_mfspr(struct kvm_vcpu *vcpu, int sprn, ulong *spr_val)
1891 {
1892         return EMULATE_FAIL;
1893 }
1894
1895 static int kvmppc_book3s_hv_init(void)
1896 {
1897         int r;
1898
1899         r = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1900
1901         if (r)
1902                 return r;
1903
1904         r = kvmppc_mmu_hv_init();
1905
1906         return r;
1907 }
1908
1909 static void kvmppc_book3s_hv_exit(void)
1910 {
1911         kvm_exit();
1912 }
1913
1914 module_init(kvmppc_book3s_hv_init);
1915 module_exit(kvmppc_book3s_hv_exit);