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 #include <linux/miscdevice.h>
  35
  36 #include <asm/reg.h>
  37 #include <asm/cputable.h>
  38 #include <asm/cacheflush.h>
  39 #include <asm/tlbflush.h>
  40 #include <asm/uaccess.h>
  41 #include <asm/io.h>
  42 #include <asm/kvm_ppc.h>
  43 #include <asm/kvm_book3s.h>
  44 #include <asm/mmu_context.h>
  45 #include <asm/lppaca.h>
  46 #include <asm/processor.h>
  47 #include <asm/cputhreads.h>
  48 #include <asm/page.h>
  49 #include <asm/hvcall.h>
  50 #include <asm/switch_to.h>
  51 #include <asm/smp.h>
  52 #include <linux/gfp.h>
  53 #include <linux/vmalloc.h>
  54 #include <linux/highmem.h>
  55 #include <linux/hugetlb.h>
  56 #include <linux/module.h>
  57
  58 #include "book3s.h"
  59
  60 /* #define EXIT_DEBUG */
  61 /* #define EXIT_DEBUG_SIMPLE */
  62 /* #define EXIT_DEBUG_INT */
  63
  64 /* Used to indicate that a guest page fault needs to be handled */
  65 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  66
  67 /* Used as a "null" value for timebase values */
  68 #define TB_NIL  (~(u64)0)
  69
  70 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
  71 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
  72
  73 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
  74 {
  75         int me;
  76         int cpu = vcpu->cpu;
  77         wait_queue_head_t *wqp;
  78
  79         wqp = kvm_arch_vcpu_wq(vcpu);
  80         if (waitqueue_active(wqp)) {
  81                 wake_up_interruptible(wqp);
  82                 ++vcpu->stat.halt_wakeup;
  83         }
  84
  85         me = get_cpu();
  86
  87         /* CPU points to the first thread of the core */
  88         if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
  89 #ifdef CONFIG_KVM_XICS
  90                 int real_cpu = cpu + vcpu->arch.ptid;
  91                 if (paca[real_cpu].kvm_hstate.xics_phys)
  92                         xics_wake_cpu(real_cpu);
  93                 else
  94 #endif
  95                 if (cpu_online(cpu))
  96                         smp_send_reschedule(cpu);
  97         }
  98         put_cpu();
  99 }
 100
 101 /*
 102  * We use the vcpu_load/put functions to measure stolen time.
 103  * Stolen time is counted as time when either the vcpu is able to
 104  * run as part of a virtual core, but the task running the vcore
 105  * is preempted or sleeping, or when the vcpu needs something done
 106  * in the kernel by the task running the vcpu, but that task is
 107  * preempted or sleeping.  Those two things have to be counted
 108  * separately, since one of the vcpu tasks will take on the job
 109  * of running the core, and the other vcpu tasks in the vcore will
 110  * sleep waiting for it to do that, but that sleep shouldn't count
 111  * as stolen time.
 112  *
 113  * Hence we accumulate stolen time when the vcpu can run as part of
 114  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 115  * needs its task to do other things in the kernel (for example,
 116  * service a page fault) in busy_stolen.  We don't accumulate
 117  * stolen time for a vcore when it is inactive, or for a vcpu
 118  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 119  * a misnomer; it means that the vcpu task is not executing in
 120  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 121  * the kernel.  We don't have any way of dividing up that time
 122  * between time that the vcpu is genuinely stopped, time that
 123  * the task is actively working on behalf of the vcpu, and time
 124  * that the task is preempted, so we don't count any of it as
 125  * stolen.
 126  *
 127  * Updates to busy_stolen are protected by arch.tbacct_lock;
 128  * updates to vc->stolen_tb are protected by the arch.tbacct_lock
 129  * of the vcpu that has taken responsibility for running the vcore
 130  * (i.e. vc->runner).  The stolen times are measured in units of
 131  * timebase ticks.  (Note that the != TB_NIL checks below are
 132  * purely defensive; they should never fail.)
 133  */
 134
 135 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 136 {
 137         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 138
 139         spin_lock(&vcpu->arch.tbacct_lock);
 140         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
 141             vc->preempt_tb != TB_NIL) {
 142                 vc->stolen_tb += mftb() - vc->preempt_tb;
 143                 vc->preempt_tb = TB_NIL;
 144         }
 145         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 146             vcpu->arch.busy_preempt != TB_NIL) {
 147                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 148                 vcpu->arch.busy_preempt = TB_NIL;
 149         }
 150         spin_unlock(&vcpu->arch.tbacct_lock);
 151 }
 152
 153 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 154 {
 155         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 156
 157         spin_lock(&vcpu->arch.tbacct_lock);
 158         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
 159                 vc->preempt_tb = mftb();
 160         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 161                 vcpu->arch.busy_preempt = mftb();
 162         spin_unlock(&vcpu->arch.tbacct_lock);
 163 }
 164
 165 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
 166 {
 167         vcpu->arch.shregs.msr = msr;
 168         kvmppc_end_cede(vcpu);
 169 }
 170
 171 void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 172 {
 173         vcpu->arch.pvr = pvr;
 174 }
 175
 176 int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 177 {
 178         unsigned long pcr = 0;
 179         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 180
 181         if (arch_compat) {
 182                 if (!cpu_has_feature(CPU_FTR_ARCH_206))
 183                         return -EINVAL; /* 970 has no compat mode support */
 184
 185                 switch (arch_compat) {
 186                 case PVR_ARCH_205:
 187                         pcr = PCR_ARCH_205;
 188                         break;
 189                 case PVR_ARCH_206:
 190                 case PVR_ARCH_206p:
 191                         break;
 192                 default:
 193                         return -EINVAL;
 194                 }
 195         }
 196
 197         spin_lock(&vc->lock);
 198         vc->arch_compat = arch_compat;
 199         vc->pcr = pcr;
 200         spin_unlock(&vc->lock);
 201
 202         return 0;
 203 }
 204
 205 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 206 {
 207         int r;
 208
 209         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 210         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 211                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
 212         for (r = 0; r < 16; ++r)
 213                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 214                        r, kvmppc_get_gpr(vcpu, r),
 215                        r+16, kvmppc_get_gpr(vcpu, r+16));
 216         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 217                vcpu->arch.ctr, vcpu->arch.lr);
 218         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 219                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 220         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 221                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 222         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 223                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 224         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 225                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 226         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 227         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 228                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 229         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 230         for (r = 0; r < vcpu->arch.slb_max; ++r)
 231                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 232                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 233         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 234                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 235                vcpu->arch.last_inst);
 236 }
 237
 238 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 239 {
 240         int r;
 241         struct kvm_vcpu *v, *ret = NULL;
 242
 243         mutex_lock(&kvm->lock);
 244         kvm_for_each_vcpu(r, v, kvm) {
 245                 if (v->vcpu_id == id) {
 246                         ret = v;
 247                         break;
 248                 }
 249         }
 250         mutex_unlock(&kvm->lock);
 251         return ret;
 252 }
 253
 254 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 255 {
 256         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 257         vpa->yield_count = 1;
 258 }
 259
 260 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 261                    unsigned long addr, unsigned long len)
 262 {
 263         /* check address is cacheline aligned */
 264         if (addr & (L1_CACHE_BYTES - 1))
 265                 return -EINVAL;
 266         spin_lock(&vcpu->arch.vpa_update_lock);
 267         if (v->next_gpa != addr || v->len != len) {
 268                 v->next_gpa = addr;
 269                 v->len = addr ? len : 0;
 270                 v->update_pending = 1;
 271         }
 272         spin_unlock(&vcpu->arch.vpa_update_lock);
 273         return 0;
 274 }
 275
 276 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 277 struct reg_vpa {
 278         u32 dummy;
 279         union {
 280                 u16 hword;
 281                 u32 word;
 282         } length;
 283 };
 284
 285 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 286 {
 287         if (vpap->update_pending)
 288                 return vpap->next_gpa != 0;
 289         return vpap->pinned_addr != NULL;
 290 }
 291
 292 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 293                                        unsigned long flags,
 294                                        unsigned long vcpuid, unsigned long vpa)
 295 {
 296         struct kvm *kvm = vcpu->kvm;
 297         unsigned long len, nb;
 298         void *va;
 299         struct kvm_vcpu *tvcpu;
 300         int err;
 301         int subfunc;
 302         struct kvmppc_vpa *vpap;
 303
 304         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 305         if (!tvcpu)
 306                 return H_PARAMETER;
 307
 308         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 309         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 310             subfunc == H_VPA_REG_SLB) {
 311                 /* Registering new area - address must be cache-line aligned */
 312                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 313                         return H_PARAMETER;
 314
 315                 /* convert logical addr to kernel addr and read length */
 316                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 317                 if (va == NULL)
 318                         return H_PARAMETER;
 319                 if (subfunc == H_VPA_REG_VPA)
 320                         len = ((struct reg_vpa *)va)->length.hword;
 321                 else
 322                         len = ((struct reg_vpa *)va)->length.word;
 323                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 324
 325                 /* Check length */
 326                 if (len > nb || len < sizeof(struct reg_vpa))
 327                         return H_PARAMETER;
 328         } else {
 329                 vpa = 0;
 330                 len = 0;
 331         }
 332
 333         err = H_PARAMETER;
 334         vpap = NULL;
 335         spin_lock(&tvcpu->arch.vpa_update_lock);
 336
 337         switch (subfunc) {
 338         case H_VPA_REG_VPA:             /* register VPA */
 339                 if (len < sizeof(struct lppaca))
 340                         break;
 341                 vpap = &tvcpu->arch.vpa;
 342                 err = 0;
 343                 break;
 344
 345         case H_VPA_REG_DTL:             /* register DTL */
 346                 if (len < sizeof(struct dtl_entry))
 347                         break;
 348                 len -= len % sizeof(struct dtl_entry);
 349
 350                 /* Check that they have previously registered a VPA */
 351                 err = H_RESOURCE;
 352                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 353                         break;
 354
 355                 vpap = &tvcpu->arch.dtl;
 356                 err = 0;
 357                 break;
 358
 359         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 360                 /* Check that they have previously registered a VPA */
 361                 err = H_RESOURCE;
 362                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 363                         break;
 364
 365                 vpap = &tvcpu->arch.slb_shadow;
 366                 err = 0;
 367                 break;
 368
 369         case H_VPA_DEREG_VPA:           /* deregister VPA */
 370                 /* Check they don't still have a DTL or SLB buf registered */
 371                 err = H_RESOURCE;
 372                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 373                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 374                         break;
 375
 376                 vpap = &tvcpu->arch.vpa;
 377                 err = 0;
 378                 break;
 379
 380         case H_VPA_DEREG_DTL:           /* deregister DTL */
 381                 vpap = &tvcpu->arch.dtl;
 382                 err = 0;
 383                 break;
 384
 385         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 386                 vpap = &tvcpu->arch.slb_shadow;
 387                 err = 0;
 388                 break;
 389         }
 390
 391         if (vpap) {
 392                 vpap->next_gpa = vpa;
 393                 vpap->len = len;
 394                 vpap->update_pending = 1;
 395         }
 396
 397         spin_unlock(&tvcpu->arch.vpa_update_lock);
 398
 399         return err;
 400 }
 401
 402 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 403 {
 404         struct kvm *kvm = vcpu->kvm;
 405         void *va;
 406         unsigned long nb;
 407         unsigned long gpa;
 408
 409         /*
 410          * We need to pin the page pointed to by vpap->next_gpa,
 411          * but we can't call kvmppc_pin_guest_page under the lock
 412          * as it does get_user_pages() and down_read().  So we
 413          * have to drop the lock, pin the page, then get the lock
 414          * again and check that a new area didn't get registered
 415          * in the meantime.
 416          */
 417         for (;;) {
 418                 gpa = vpap->next_gpa;
 419                 spin_unlock(&vcpu->arch.vpa_update_lock);
 420                 va = NULL;
 421                 nb = 0;
 422                 if (gpa)
 423                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 424                 spin_lock(&vcpu->arch.vpa_update_lock);
 425                 if (gpa == vpap->next_gpa)
 426                         break;
 427                 /* sigh... unpin that one and try again */
 428                 if (va)
 429                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 430         }
 431
 432         vpap->update_pending = 0;
 433         if (va && nb < vpap->len) {
 434                 /*
 435                  * If it's now too short, it must be that userspace
 436                  * has changed the mappings underlying guest memory,
 437                  * so unregister the region.
 438                  */
 439                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 440                 va = NULL;
 441         }
 442         if (vpap->pinned_addr)
 443                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 444                                         vpap->dirty);
 445         vpap->gpa = gpa;
 446         vpap->pinned_addr = va;
 447         vpap->dirty = false;
 448         if (va)
 449                 vpap->pinned_end = va + vpap->len;
 450 }
 451
 452 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 453 {
 454         if (!(vcpu->arch.vpa.update_pending ||
 455               vcpu->arch.slb_shadow.update_pending ||
 456               vcpu->arch.dtl.update_pending))
 457                 return;
 458
 459         spin_lock(&vcpu->arch.vpa_update_lock);
 460         if (vcpu->arch.vpa.update_pending) {
 461                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 462                 if (vcpu->arch.vpa.pinned_addr)
 463                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 464         }
 465         if (vcpu->arch.dtl.update_pending) {
 466                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 467                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 468                 vcpu->arch.dtl_index = 0;
 469         }
 470         if (vcpu->arch.slb_shadow.update_pending)
 471                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 472         spin_unlock(&vcpu->arch.vpa_update_lock);
 473 }
 474
 475 /*
 476  * Return the accumulated stolen time for the vcore up until `now'.
 477  * The caller should hold the vcore lock.
 478  */
 479 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 480 {
 481         u64 p;
 482
 483         /*
 484          * If we are the task running the vcore, then since we hold
 485          * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
 486          * can't be updated, so we don't need the tbacct_lock.
 487          * If the vcore is inactive, it can't become active (since we
 488          * hold the vcore lock), so the vcpu load/put functions won't
 489          * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
 490          */
 491         if (vc->vcore_state != VCORE_INACTIVE &&
 492             vc->runner->arch.run_task != current) {
 493                 spin_lock(&vc->runner->arch.tbacct_lock);
 494                 p = vc->stolen_tb;
 495                 if (vc->preempt_tb != TB_NIL)
 496                         p += now - vc->preempt_tb;
 497                 spin_unlock(&vc->runner->arch.tbacct_lock);
 498         } else {
 499                 p = vc->stolen_tb;
 500         }
 501         return p;
 502 }
 503
 504 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 505                                     struct kvmppc_vcore *vc)
 506 {
 507         struct dtl_entry *dt;
 508         struct lppaca *vpa;
 509         unsigned long stolen;
 510         unsigned long core_stolen;
 511         u64 now;
 512
 513         dt = vcpu->arch.dtl_ptr;
 514         vpa = vcpu->arch.vpa.pinned_addr;
 515         now = mftb();
 516         core_stolen = vcore_stolen_time(vc, now);
 517         stolen = core_stolen - vcpu->arch.stolen_logged;
 518         vcpu->arch.stolen_logged = core_stolen;
 519         spin_lock(&vcpu->arch.tbacct_lock);
 520         stolen += vcpu->arch.busy_stolen;
 521         vcpu->arch.busy_stolen = 0;
 522         spin_unlock(&vcpu->arch.tbacct_lock);
 523         if (!dt || !vpa)
 524                 return;
 525         memset(dt, 0, sizeof(struct dtl_entry));
 526         dt->dispatch_reason = 7;
 527         dt->processor_id = vc->pcpu + vcpu->arch.ptid;
 528         dt->timebase = now + vc->tb_offset;
 529         dt->enqueue_to_dispatch_time = stolen;
 530         dt->srr0 = kvmppc_get_pc(vcpu);
 531         dt->srr1 = vcpu->arch.shregs.msr;
 532         ++dt;
 533         if (dt == vcpu->arch.dtl.pinned_end)
 534                 dt = vcpu->arch.dtl.pinned_addr;
 535         vcpu->arch.dtl_ptr = dt;
 536         /* order writing *dt vs. writing vpa->dtl_idx */
 537         smp_wmb();
 538         vpa->dtl_idx = ++vcpu->arch.dtl_index;
 539         vcpu->arch.dtl.dirty = true;
 540 }
 541
 542 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 543 {
 544         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 545         unsigned long target, ret = H_SUCCESS;
 546         struct kvm_vcpu *tvcpu;
 547         int idx, rc;
 548
 549         switch (req) {
 550         case H_ENTER:
 551                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 552                 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
 553                                               kvmppc_get_gpr(vcpu, 5),
 554                                               kvmppc_get_gpr(vcpu, 6),
 555                                               kvmppc_get_gpr(vcpu, 7));
 556                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 557                 break;
 558         case H_CEDE:
 559                 break;
 560         case H_PROD:
 561                 target = kvmppc_get_gpr(vcpu, 4);
 562                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 563                 if (!tvcpu) {
 564                         ret = H_PARAMETER;
 565                         break;
 566                 }
 567                 tvcpu->arch.prodded = 1;
 568                 smp_mb();
 569                 if (vcpu->arch.ceded) {
 570                         if (waitqueue_active(&vcpu->wq)) {
 571                                 wake_up_interruptible(&vcpu->wq);
 572                                 vcpu->stat.halt_wakeup++;
 573                         }
 574                 }
 575                 break;
 576         case H_CONFER:
 577                 target = kvmppc_get_gpr(vcpu, 4);
 578                 if (target == -1)
 579                         break;
 580                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 581                 if (!tvcpu) {
 582                         ret = H_PARAMETER;
 583                         break;
 584                 }
 585                 kvm_vcpu_yield_to(tvcpu);
 586                 break;
 587         case H_REGISTER_VPA:
 588                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 589                                         kvmppc_get_gpr(vcpu, 5),
 590                                         kvmppc_get_gpr(vcpu, 6));
 591                 break;
 592         case H_RTAS:
 593                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 594                         return RESUME_HOST;
 595
 596                 rc = kvmppc_rtas_hcall(vcpu);
 597
 598                 if (rc == -ENOENT)
 599                         return RESUME_HOST;
 600                 else if (rc == 0)
 601                         break;
 602
 603                 /* Send the error out to userspace via KVM_RUN */
 604                 return rc;
 605
 606         case H_XIRR:
 607         case H_CPPR:
 608         case H_EOI:
 609         case H_IPI:
 610         case H_IPOLL:
 611         case H_XIRR_X:
 612                 if (kvmppc_xics_enabled(vcpu)) {
 613                         ret = kvmppc_xics_hcall(vcpu, req);
 614                         break;
 615                 } /* fallthrough */
 616         default:
 617                 return RESUME_HOST;
 618         }
 619         kvmppc_set_gpr(vcpu, 3, ret);
 620         vcpu->arch.hcall_needed = 0;
 621         return RESUME_GUEST;
 622 }
 623
 624 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
 625                                  struct task_struct *tsk)
 626 {
 627         int r = RESUME_HOST;
 628
 629         vcpu->stat.sum_exits++;
 630
 631         run->exit_reason = KVM_EXIT_UNKNOWN;
 632         run->ready_for_interrupt_injection = 1;
 633         switch (vcpu->arch.trap) {
 634         /* We're good on these - the host merely wanted to get our attention */
 635         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 636                 vcpu->stat.dec_exits++;
 637                 r = RESUME_GUEST;
 638                 break;
 639         case BOOK3S_INTERRUPT_EXTERNAL:
 640                 vcpu->stat.ext_intr_exits++;
 641                 r = RESUME_GUEST;
 642                 break;
 643         case BOOK3S_INTERRUPT_PERFMON:
 644                 r = RESUME_GUEST;
 645                 break;
 646         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 647                 /*
 648                  * Deliver a machine check interrupt to the guest.
 649                  * We have to do this, even if the host has handled the
 650                  * machine check, because machine checks use SRR0/1 and
 651                  * the interrupt might have trashed guest state in them.
 652                  */
 653                 kvmppc_book3s_queue_irqprio(vcpu,
 654                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 655                 r = RESUME_GUEST;
 656                 break;
 657         case BOOK3S_INTERRUPT_PROGRAM:
 658         {
 659                 ulong flags;
 660                 /*
 661                  * Normally program interrupts are delivered directly
 662                  * to the guest by the hardware, but we can get here
 663                  * as a result of a hypervisor emulation interrupt
 664                  * (e40) getting turned into a 700 by BML RTAS.
 665                  */
 666                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 667                 kvmppc_core_queue_program(vcpu, flags);
 668                 r = RESUME_GUEST;
 669                 break;
 670         }
 671         case BOOK3S_INTERRUPT_SYSCALL:
 672         {
 673                 /* hcall - punt to userspace */
 674                 int i;
 675
 676                 /* hypercall with MSR_PR has already been handled in rmode,
 677                  * and never reaches here.
 678                  */
 679
 680                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 681                 for (i = 0; i < 9; ++i)
 682                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 683                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 684                 vcpu->arch.hcall_needed = 1;
 685                 r = RESUME_HOST;
 686                 break;
 687         }
 688         /*
 689          * We get these next two if the guest accesses a page which it thinks
 690          * it has mapped but which is not actually present, either because
 691          * it is for an emulated I/O device or because the corresonding
 692          * host page has been paged out.  Any other HDSI/HISI interrupts
 693          * have been handled already.
 694          */
 695         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 696                 r = RESUME_PAGE_FAULT;
 697                 break;
 698         case BOOK3S_INTERRUPT_H_INST_STORAGE:
 699                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
 700                 vcpu->arch.fault_dsisr = 0;
 701                 r = RESUME_PAGE_FAULT;
 702                 break;
 703         /*
 704          * This occurs if the guest executes an illegal instruction.
 705          * We just generate a program interrupt to the guest, since
 706          * we don't emulate any guest instructions at this stage.
 707          */
 708         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 709                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 710                 r = RESUME_GUEST;
 711                 break;
 712         /*
 713          * This occurs if the guest (kernel or userspace), does something that
 714          * is prohibited by HFSCR.  We just generate a program interrupt to
 715          * the guest.
 716          */
 717         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
 718                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 719                 r = RESUME_GUEST;
 720                 break;
 721         default:
 722                 kvmppc_dump_regs(vcpu);
 723                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 724                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 725                         vcpu->arch.shregs.msr);
 726                 run->hw.hardware_exit_reason = vcpu->arch.trap;
 727                 r = RESUME_HOST;
 728                 break;
 729         }
 730
 731         return r;
 732 }
 733
 734 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
 735                                             struct kvm_sregs *sregs)
 736 {
 737         int i;
 738
 739         memset(sregs, 0, sizeof(struct kvm_sregs));
 740         sregs->pvr = vcpu->arch.pvr;
 741         for (i = 0; i < vcpu->arch.slb_max; i++) {
 742                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
 743                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 744         }
 745
 746         return 0;
 747 }
 748
 749 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
 750                                             struct kvm_sregs *sregs)
 751 {
 752         int i, j;
 753
 754         kvmppc_set_pvr_hv(vcpu, sregs->pvr);
 755
 756         j = 0;
 757         for (i = 0; i < vcpu->arch.slb_nr; i++) {
 758                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
 759                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
 760                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
 761                         ++j;
 762                 }
 763         }
 764         vcpu->arch.slb_max = j;
 765
 766         return 0;
 767 }
 768
 769 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr)
 770 {
 771         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 772         u64 mask;
 773
 774         spin_lock(&vc->lock);
 775         /*
 776          * Userspace can only modify DPFD (default prefetch depth),
 777          * ILE (interrupt little-endian) and TC (translation control).
 778          */
 779         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
 780         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
 781         spin_unlock(&vc->lock);
 782 }
 783
 784 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
 785                                  union kvmppc_one_reg *val)
 786 {
 787         int r = 0;
 788         long int i;
 789
 790         switch (id) {
 791         case KVM_REG_PPC_HIOR:
 792                 *val = get_reg_val(id, 0);
 793                 break;
 794         case KVM_REG_PPC_DABR:
 795                 *val = get_reg_val(id, vcpu->arch.dabr);
 796                 break;
 797         case KVM_REG_PPC_DSCR:
 798                 *val = get_reg_val(id, vcpu->arch.dscr);
 799                 break;
 800         case KVM_REG_PPC_PURR:
 801                 *val = get_reg_val(id, vcpu->arch.purr);
 802                 break;
 803         case KVM_REG_PPC_SPURR:
 804                 *val = get_reg_val(id, vcpu->arch.spurr);
 805                 break;
 806         case KVM_REG_PPC_AMR:
 807                 *val = get_reg_val(id, vcpu->arch.amr);
 808                 break;
 809         case KVM_REG_PPC_UAMOR:
 810                 *val = get_reg_val(id, vcpu->arch.uamor);
 811                 break;
 812         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
 813                 i = id - KVM_REG_PPC_MMCR0;
 814                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
 815                 break;
 816         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
 817                 i = id - KVM_REG_PPC_PMC1;
 818                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
 819                 break;
 820         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
 821                 i = id - KVM_REG_PPC_SPMC1;
 822                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
 823                 break;
 824         case KVM_REG_PPC_SIAR:
 825                 *val = get_reg_val(id, vcpu->arch.siar);
 826                 break;
 827         case KVM_REG_PPC_SDAR:
 828                 *val = get_reg_val(id, vcpu->arch.sdar);
 829                 break;
 830         case KVM_REG_PPC_SIER:
 831                 *val = get_reg_val(id, vcpu->arch.sier);
 832                 break;
 833         case KVM_REG_PPC_IAMR:
 834                 *val = get_reg_val(id, vcpu->arch.iamr);
 835                 break;
 836         case KVM_REG_PPC_TFHAR:
 837                 *val = get_reg_val(id, vcpu->arch.tfhar);
 838                 break;
 839         case KVM_REG_PPC_TFIAR:
 840                 *val = get_reg_val(id, vcpu->arch.tfiar);
 841                 break;
 842         case KVM_REG_PPC_TEXASR:
 843                 *val = get_reg_val(id, vcpu->arch.texasr);
 844                 break;
 845         case KVM_REG_PPC_FSCR:
 846                 *val = get_reg_val(id, vcpu->arch.fscr);
 847                 break;
 848         case KVM_REG_PPC_PSPB:
 849                 *val = get_reg_val(id, vcpu->arch.pspb);
 850                 break;
 851         case KVM_REG_PPC_EBBHR:
 852                 *val = get_reg_val(id, vcpu->arch.ebbhr);
 853                 break;
 854         case KVM_REG_PPC_EBBRR:
 855                 *val = get_reg_val(id, vcpu->arch.ebbrr);
 856                 break;
 857         case KVM_REG_PPC_BESCR:
 858                 *val = get_reg_val(id, vcpu->arch.bescr);
 859                 break;
 860         case KVM_REG_PPC_TAR:
 861                 *val = get_reg_val(id, vcpu->arch.tar);
 862                 break;
 863         case KVM_REG_PPC_DPDES:
 864                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
 865                 break;
 866         case KVM_REG_PPC_DAWR:
 867                 *val = get_reg_val(id, vcpu->arch.dawr);
 868                 break;
 869         case KVM_REG_PPC_DAWRX:
 870                 *val = get_reg_val(id, vcpu->arch.dawrx);
 871                 break;
 872         case KVM_REG_PPC_CIABR:
 873                 *val = get_reg_val(id, vcpu->arch.ciabr);
 874                 break;
 875         case KVM_REG_PPC_IC:
 876                 *val = get_reg_val(id, vcpu->arch.ic);
 877                 break;
 878         case KVM_REG_PPC_VTB:
 879                 *val = get_reg_val(id, vcpu->arch.vtb);
 880                 break;
 881         case KVM_REG_PPC_CSIGR:
 882                 *val = get_reg_val(id, vcpu->arch.csigr);
 883                 break;
 884         case KVM_REG_PPC_TACR:
 885                 *val = get_reg_val(id, vcpu->arch.tacr);
 886                 break;
 887         case KVM_REG_PPC_TCSCR:
 888                 *val = get_reg_val(id, vcpu->arch.tcscr);
 889                 break;
 890         case KVM_REG_PPC_PID:
 891                 *val = get_reg_val(id, vcpu->arch.pid);
 892                 break;
 893         case KVM_REG_PPC_ACOP:
 894                 *val = get_reg_val(id, vcpu->arch.acop);
 895                 break;
 896         case KVM_REG_PPC_WORT:
 897                 *val = get_reg_val(id, vcpu->arch.wort);
 898                 break;
 899         case KVM_REG_PPC_VPA_ADDR:
 900                 spin_lock(&vcpu->arch.vpa_update_lock);
 901                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
 902                 spin_unlock(&vcpu->arch.vpa_update_lock);
 903                 break;
 904         case KVM_REG_PPC_VPA_SLB:
 905                 spin_lock(&vcpu->arch.vpa_update_lock);
 906                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
 907                 val->vpaval.length = vcpu->arch.slb_shadow.len;
 908                 spin_unlock(&vcpu->arch.vpa_update_lock);
 909                 break;
 910         case KVM_REG_PPC_VPA_DTL:
 911                 spin_lock(&vcpu->arch.vpa_update_lock);
 912                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
 913                 val->vpaval.length = vcpu->arch.dtl.len;
 914                 spin_unlock(&vcpu->arch.vpa_update_lock);
 915                 break;
 916         case KVM_REG_PPC_TB_OFFSET:
 917                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
 918                 break;
 919         case KVM_REG_PPC_LPCR:
 920                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
 921                 break;
 922         case KVM_REG_PPC_PPR:
 923                 *val = get_reg_val(id, vcpu->arch.ppr);
 924                 break;
 925         case KVM_REG_PPC_ARCH_COMPAT:
 926                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
 927                 break;
 928         default:
 929                 r = -EINVAL;
 930                 break;
 931         }
 932
 933         return r;
 934 }
 935
 936 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
 937                                  union kvmppc_one_reg *val)
 938 {
 939         int r = 0;
 940         long int i;
 941         unsigned long addr, len;
 942
 943         switch (id) {
 944         case KVM_REG_PPC_HIOR:
 945                 /* Only allow this to be set to zero */
 946                 if (set_reg_val(id, *val))
 947                         r = -EINVAL;
 948                 break;
 949         case KVM_REG_PPC_DABR:
 950                 vcpu->arch.dabr = set_reg_val(id, *val);
 951                 break;
 952         case KVM_REG_PPC_DSCR:
 953                 vcpu->arch.dscr = set_reg_val(id, *val);
 954                 break;
 955         case KVM_REG_PPC_PURR:
 956                 vcpu->arch.purr = set_reg_val(id, *val);
 957                 break;
 958         case KVM_REG_PPC_SPURR:
 959                 vcpu->arch.spurr = set_reg_val(id, *val);
 960                 break;
 961         case KVM_REG_PPC_AMR:
 962                 vcpu->arch.amr = set_reg_val(id, *val);
 963                 break;
 964         case KVM_REG_PPC_UAMOR:
 965                 vcpu->arch.uamor = set_reg_val(id, *val);
 966                 break;
 967         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
 968                 i = id - KVM_REG_PPC_MMCR0;
 969                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
 970                 break;
 971         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
 972                 i = id - KVM_REG_PPC_PMC1;
 973                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
 974                 break;
 975         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
 976                 i = id - KVM_REG_PPC_SPMC1;
 977                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
 978                 break;
 979         case KVM_REG_PPC_SIAR:
 980                 vcpu->arch.siar = set_reg_val(id, *val);
 981                 break;
 982         case KVM_REG_PPC_SDAR:
 983                 vcpu->arch.sdar = set_reg_val(id, *val);
 984                 break;
 985         case KVM_REG_PPC_SIER:
 986                 vcpu->arch.sier = set_reg_val(id, *val);
 987                 break;
 988         case KVM_REG_PPC_IAMR:
 989                 vcpu->arch.iamr = set_reg_val(id, *val);
 990                 break;
 991         case KVM_REG_PPC_TFHAR:
 992                 vcpu->arch.tfhar = set_reg_val(id, *val);
 993                 break;
 994         case KVM_REG_PPC_TFIAR:
 995                 vcpu->arch.tfiar = set_reg_val(id, *val);
 996                 break;
 997         case KVM_REG_PPC_TEXASR:
 998                 vcpu->arch.texasr = set_reg_val(id, *val);
 999                 break;
1000         case KVM_REG_PPC_FSCR:
1001                 vcpu->arch.fscr = set_reg_val(id, *val);
1002                 break;
1003         case KVM_REG_PPC_PSPB:
1004                 vcpu->arch.pspb = set_reg_val(id, *val);
1005                 break;
1006         case KVM_REG_PPC_EBBHR:
1007                 vcpu->arch.ebbhr = set_reg_val(id, *val);
1008                 break;
1009         case KVM_REG_PPC_EBBRR:
1010                 vcpu->arch.ebbrr = set_reg_val(id, *val);
1011                 break;
1012         case KVM_REG_PPC_BESCR:
1013                 vcpu->arch.bescr = set_reg_val(id, *val);
1014                 break;
1015         case KVM_REG_PPC_TAR:
1016                 vcpu->arch.tar = set_reg_val(id, *val);
1017                 break;
1018         case KVM_REG_PPC_DPDES:
1019                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1020                 break;
1021         case KVM_REG_PPC_DAWR:
1022                 vcpu->arch.dawr = set_reg_val(id, *val);
1023                 break;
1024         case KVM_REG_PPC_DAWRX:
1025                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1026                 break;
1027         case KVM_REG_PPC_CIABR:
1028                 vcpu->arch.ciabr = set_reg_val(id, *val);
1029                 /* Don't allow setting breakpoints in hypervisor code */
1030                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1031                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1032                 break;
1033         case KVM_REG_PPC_IC:
1034                 vcpu->arch.ic = set_reg_val(id, *val);
1035                 break;
1036         case KVM_REG_PPC_VTB:
1037                 vcpu->arch.vtb = set_reg_val(id, *val);
1038                 break;
1039         case KVM_REG_PPC_CSIGR:
1040                 vcpu->arch.csigr = set_reg_val(id, *val);
1041                 break;
1042         case KVM_REG_PPC_TACR:
1043                 vcpu->arch.tacr = set_reg_val(id, *val);
1044                 break;
1045         case KVM_REG_PPC_TCSCR:
1046                 vcpu->arch.tcscr = set_reg_val(id, *val);
1047                 break;
1048         case KVM_REG_PPC_PID:
1049                 vcpu->arch.pid = set_reg_val(id, *val);
1050                 break;
1051         case KVM_REG_PPC_ACOP:
1052                 vcpu->arch.acop = set_reg_val(id, *val);
1053                 break;
1054         case KVM_REG_PPC_WORT:
1055                 vcpu->arch.wort = set_reg_val(id, *val);
1056                 break;
1057         case KVM_REG_PPC_VPA_ADDR:
1058                 addr = set_reg_val(id, *val);
1059                 r = -EINVAL;
1060                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1061                               vcpu->arch.dtl.next_gpa))
1062                         break;
1063                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1064                 break;
1065         case KVM_REG_PPC_VPA_SLB:
1066                 addr = val->vpaval.addr;
1067                 len = val->vpaval.length;
1068                 r = -EINVAL;
1069                 if (addr && !vcpu->arch.vpa.next_gpa)
1070                         break;
1071                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1072                 break;
1073         case KVM_REG_PPC_VPA_DTL:
1074                 addr = val->vpaval.addr;
1075                 len = val->vpaval.length;
1076                 r = -EINVAL;
1077                 if (addr && (len < sizeof(struct dtl_entry) ||
1078                              !vcpu->arch.vpa.next_gpa))
1079                         break;
1080                 len -= len % sizeof(struct dtl_entry);
1081                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1082                 break;
1083         case KVM_REG_PPC_TB_OFFSET:
1084                 /* round up to multiple of 2^24 */
1085                 vcpu->arch.vcore->tb_offset =
1086                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1087                 break;
1088         case KVM_REG_PPC_LPCR:
1089                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val));
1090                 break;
1091         case KVM_REG_PPC_PPR:
1092                 vcpu->arch.ppr = set_reg_val(id, *val);
1093                 break;
1094         case KVM_REG_PPC_ARCH_COMPAT:
1095                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1096                 break;
1097         default:
1098                 r = -EINVAL;
1099                 break;
1100         }
1101
1102         return r;
1103 }
1104
1105 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1106                                                    unsigned int id)
1107 {
1108         struct kvm_vcpu *vcpu;
1109         int err = -EINVAL;
1110         int core;
1111         struct kvmppc_vcore *vcore;
1112
1113         core = id / threads_per_core;
1114         if (core >= KVM_MAX_VCORES)
1115                 goto out;
1116
1117         err = -ENOMEM;
1118         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1119         if (!vcpu)
1120                 goto out;
1121
1122         err = kvm_vcpu_init(vcpu, kvm, id);
1123         if (err)
1124                 goto free_vcpu;
1125
1126         vcpu->arch.shared = &vcpu->arch.shregs;
1127         vcpu->arch.mmcr[0] = MMCR0_FC;
1128         vcpu->arch.ctrl = CTRL_RUNLATCH;
1129         /* default to host PVR, since we can't spoof it */
1130         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1131         spin_lock_init(&vcpu->arch.vpa_update_lock);
1132         spin_lock_init(&vcpu->arch.tbacct_lock);
1133         vcpu->arch.busy_preempt = TB_NIL;
1134
1135         kvmppc_mmu_book3s_hv_init(vcpu);
1136
1137         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1138
1139         init_waitqueue_head(&vcpu->arch.cpu_run);
1140
1141         mutex_lock(&kvm->lock);
1142         vcore = kvm->arch.vcores[core];
1143         if (!vcore) {
1144                 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1145                 if (vcore) {
1146                         INIT_LIST_HEAD(&vcore->runnable_threads);
1147                         spin_lock_init(&vcore->lock);
1148                         init_waitqueue_head(&vcore->wq);
1149                         vcore->preempt_tb = TB_NIL;
1150                         vcore->lpcr = kvm->arch.lpcr;
1151                         vcore->first_vcpuid = core * threads_per_core;
1152                         vcore->kvm = kvm;
1153                 }
1154                 kvm->arch.vcores[core] = vcore;
1155                 kvm->arch.online_vcores++;
1156         }
1157         mutex_unlock(&kvm->lock);
1158
1159         if (!vcore)
1160                 goto free_vcpu;
1161
1162         spin_lock(&vcore->lock);
1163         ++vcore->num_threads;
1164         spin_unlock(&vcore->lock);
1165         vcpu->arch.vcore = vcore;
1166         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1167
1168         vcpu->arch.cpu_type = KVM_CPU_3S_64;
1169         kvmppc_sanity_check(vcpu);
1170
1171         return vcpu;
1172
1173 free_vcpu:
1174         kmem_cache_free(kvm_vcpu_cache, vcpu);
1175 out:
1176         return ERR_PTR(err);
1177 }
1178
1179 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1180 {
1181         if (vpa->pinned_addr)
1182                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1183                                         vpa->dirty);
1184 }
1185
1186 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1187 {
1188         spin_lock(&vcpu->arch.vpa_update_lock);
1189         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1190         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1191         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1192         spin_unlock(&vcpu->arch.vpa_update_lock);
1193         kvm_vcpu_uninit(vcpu);
1194         kmem_cache_free(kvm_vcpu_cache, vcpu);
1195 }
1196
1197 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1198 {
1199         /* Indicate we want to get back into the guest */
1200         return 1;
1201 }
1202
1203 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1204 {
1205         unsigned long dec_nsec, now;
1206
1207         now = get_tb();
1208         if (now > vcpu->arch.dec_expires) {
1209                 /* decrementer has already gone negative */
1210                 kvmppc_core_queue_dec(vcpu);
1211                 kvmppc_core_prepare_to_enter(vcpu);
1212                 return;
1213         }
1214         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1215                    / tb_ticks_per_sec;
1216         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1217                       HRTIMER_MODE_REL);
1218         vcpu->arch.timer_running = 1;
1219 }
1220
1221 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1222 {
1223         vcpu->arch.ceded = 0;
1224         if (vcpu->arch.timer_running) {
1225                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1226                 vcpu->arch.timer_running = 0;
1227         }
1228 }
1229
1230 extern void __kvmppc_vcore_entry(void);
1231
1232 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1233                                    struct kvm_vcpu *vcpu)
1234 {
1235         u64 now;
1236
1237         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1238                 return;
1239         spin_lock(&vcpu->arch.tbacct_lock);
1240         now = mftb();
1241         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1242                 vcpu->arch.stolen_logged;
1243         vcpu->arch.busy_preempt = now;
1244         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1245         spin_unlock(&vcpu->arch.tbacct_lock);
1246         --vc->n_runnable;
1247         list_del(&vcpu->arch.run_list);
1248 }
1249
1250 static int kvmppc_grab_hwthread(int cpu)
1251 {
1252         struct paca_struct *tpaca;
1253         long timeout = 1000;
1254
1255         tpaca = &paca[cpu];
1256
1257         /* Ensure the thread won't go into the kernel if it wakes */
1258         tpaca->kvm_hstate.hwthread_req = 1;
1259         tpaca->kvm_hstate.kvm_vcpu = NULL;
1260
1261         /*
1262          * If the thread is already executing in the kernel (e.g. handling
1263          * a stray interrupt), wait for it to get back to nap mode.
1264          * The smp_mb() is to ensure that our setting of hwthread_req
1265          * is visible before we look at hwthread_state, so if this
1266          * races with the code at system_reset_pSeries and the thread
1267          * misses our setting of hwthread_req, we are sure to see its
1268          * setting of hwthread_state, and vice versa.
1269          */
1270         smp_mb();
1271         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1272                 if (--timeout <= 0) {
1273                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1274                         return -EBUSY;
1275                 }
1276                 udelay(1);
1277         }
1278         return 0;
1279 }
1280
1281 static void kvmppc_release_hwthread(int cpu)
1282 {
1283         struct paca_struct *tpaca;
1284
1285         tpaca = &paca[cpu];
1286         tpaca->kvm_hstate.hwthread_req = 0;
1287         tpaca->kvm_hstate.kvm_vcpu = NULL;
1288 }
1289
1290 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1291 {
1292         int cpu;
1293         struct paca_struct *tpaca;
1294         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1295
1296         if (vcpu->arch.timer_running) {
1297                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1298                 vcpu->arch.timer_running = 0;
1299         }
1300         cpu = vc->pcpu + vcpu->arch.ptid;
1301         tpaca = &paca[cpu];
1302         tpaca->kvm_hstate.kvm_vcpu = vcpu;
1303         tpaca->kvm_hstate.kvm_vcore = vc;
1304         tpaca->kvm_hstate.ptid = vcpu->arch.ptid;
1305         vcpu->cpu = vc->pcpu;
1306         smp_wmb();
1307 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1308         if (cpu != smp_processor_id()) {
1309 #ifdef CONFIG_KVM_XICS
1310                 xics_wake_cpu(cpu);
1311 #endif
1312                 if (vcpu->arch.ptid)
1313                         ++vc->n_woken;
1314         }
1315 #endif
1316 }
1317
1318 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1319 {
1320         int i;
1321
1322         HMT_low();
1323         i = 0;
1324         while (vc->nap_count < vc->n_woken) {
1325                 if (++i >= 1000000) {
1326                         pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1327                                vc->nap_count, vc->n_woken);
1328                         break;
1329                 }
1330                 cpu_relax();
1331         }
1332         HMT_medium();
1333 }
1334
1335 /*
1336  * Check that we are on thread 0 and that any other threads in
1337  * this core are off-line.  Then grab the threads so they can't
1338  * enter the kernel.
1339  */
1340 static int on_primary_thread(void)
1341 {
1342         int cpu = smp_processor_id();
1343         int thr = cpu_thread_in_core(cpu);
1344
1345         if (thr)
1346                 return 0;
1347         while (++thr < threads_per_core)
1348                 if (cpu_online(cpu + thr))
1349                         return 0;
1350
1351         /* Grab all hw threads so they can't go into the kernel */
1352         for (thr = 1; thr < threads_per_core; ++thr) {
1353                 if (kvmppc_grab_hwthread(cpu + thr)) {
1354                         /* Couldn't grab one; let the others go */
1355                         do {
1356                                 kvmppc_release_hwthread(cpu + thr);
1357                         } while (--thr > 0);
1358                         return 0;
1359                 }
1360         }
1361         return 1;
1362 }
1363
1364 /*
1365  * Run a set of guest threads on a physical core.
1366  * Called with vc->lock held.
1367  */
1368 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1369 {
1370         struct kvm_vcpu *vcpu, *vnext;
1371         long ret;
1372         u64 now;
1373         int i, need_vpa_update;
1374         int srcu_idx;
1375         struct kvm_vcpu *vcpus_to_update[threads_per_core];
1376
1377         /* don't start if any threads have a signal pending */
1378         need_vpa_update = 0;
1379         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1380                 if (signal_pending(vcpu->arch.run_task))
1381                         return;
1382                 if (vcpu->arch.vpa.update_pending ||
1383                     vcpu->arch.slb_shadow.update_pending ||
1384                     vcpu->arch.dtl.update_pending)
1385                         vcpus_to_update[need_vpa_update++] = vcpu;
1386         }
1387
1388         /*
1389          * Initialize *vc, in particular vc->vcore_state, so we can
1390          * drop the vcore lock if necessary.
1391          */
1392         vc->n_woken = 0;
1393         vc->nap_count = 0;
1394         vc->entry_exit_count = 0;
1395         vc->vcore_state = VCORE_STARTING;
1396         vc->in_guest = 0;
1397         vc->napping_threads = 0;
1398
1399         /*
1400          * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1401          * which can't be called with any spinlocks held.
1402          */
1403         if (need_vpa_update) {
1404                 spin_unlock(&vc->lock);
1405                 for (i = 0; i < need_vpa_update; ++i)
1406                         kvmppc_update_vpas(vcpus_to_update[i]);
1407                 spin_lock(&vc->lock);
1408         }
1409
1410         /*
1411          * Make sure we are running on thread 0, and that
1412          * secondary threads are offline.
1413          */
1414         if (threads_per_core > 1 && !on_primary_thread()) {
1415                 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1416                         vcpu->arch.ret = -EBUSY;
1417                 goto out;
1418         }
1419
1420         vc->pcpu = smp_processor_id();
1421         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1422                 kvmppc_start_thread(vcpu);
1423                 kvmppc_create_dtl_entry(vcpu, vc);
1424         }
1425
1426         /* Set this explicitly in case thread 0 doesn't have a vcpu */
1427         get_paca()->kvm_hstate.kvm_vcore = vc;
1428         get_paca()->kvm_hstate.ptid = 0;
1429
1430         vc->vcore_state = VCORE_RUNNING;
1431         preempt_disable();
1432         spin_unlock(&vc->lock);
1433
1434         kvm_guest_enter();
1435
1436         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
1437
1438         __kvmppc_vcore_entry();
1439
1440         spin_lock(&vc->lock);
1441         /* disable sending of IPIs on virtual external irqs */
1442         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1443                 vcpu->cpu = -1;
1444         /* wait for secondary threads to finish writing their state to memory */
1445         if (vc->nap_count < vc->n_woken)
1446                 kvmppc_wait_for_nap(vc);
1447         for (i = 0; i < threads_per_core; ++i)
1448                 kvmppc_release_hwthread(vc->pcpu + i);
1449         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
1450         vc->vcore_state = VCORE_EXITING;
1451         spin_unlock(&vc->lock);
1452
1453         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
1454
1455         /* make sure updates to secondary vcpu structs are visible now */
1456         smp_mb();
1457         kvm_guest_exit();
1458
1459         preempt_enable();
1460         kvm_resched(vcpu);
1461
1462         spin_lock(&vc->lock);
1463         now = get_tb();
1464         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1465                 /* cancel pending dec exception if dec is positive */
1466                 if (now < vcpu->arch.dec_expires &&
1467                     kvmppc_core_pending_dec(vcpu))
1468                         kvmppc_core_dequeue_dec(vcpu);
1469
1470                 ret = RESUME_GUEST;
1471                 if (vcpu->arch.trap)
1472                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1473                                                     vcpu->arch.run_task);
1474
1475                 vcpu->arch.ret = ret;
1476                 vcpu->arch.trap = 0;
1477
1478                 if (vcpu->arch.ceded) {
1479                         if (ret != RESUME_GUEST)
1480                                 kvmppc_end_cede(vcpu);
1481                         else
1482                                 kvmppc_set_timer(vcpu);
1483                 }
1484         }
1485
1486  out:
1487         vc->vcore_state = VCORE_INACTIVE;
1488         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1489                                  arch.run_list) {
1490                 if (vcpu->arch.ret != RESUME_GUEST) {
1491                         kvmppc_remove_runnable(vc, vcpu);
1492                         wake_up(&vcpu->arch.cpu_run);
1493                 }
1494         }
1495 }
1496
1497 /*
1498  * Wait for some other vcpu thread to execute us, and
1499  * wake us up when we need to handle something in the host.
1500  */
1501 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1502 {
1503         DEFINE_WAIT(wait);
1504
1505         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1506         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1507                 schedule();
1508         finish_wait(&vcpu->arch.cpu_run, &wait);
1509 }
1510
1511 /*
1512  * All the vcpus in this vcore are idle, so wait for a decrementer
1513  * or external interrupt to one of the vcpus.  vc->lock is held.
1514  */
1515 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1516 {
1517         DEFINE_WAIT(wait);
1518
1519         prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1520         vc->vcore_state = VCORE_SLEEPING;
1521         spin_unlock(&vc->lock);
1522         schedule();
1523         finish_wait(&vc->wq, &wait);
1524         spin_lock(&vc->lock);
1525         vc->vcore_state = VCORE_INACTIVE;
1526 }
1527
1528 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1529 {
1530         int n_ceded;
1531         struct kvmppc_vcore *vc;
1532         struct kvm_vcpu *v, *vn;
1533
1534         kvm_run->exit_reason = 0;
1535         vcpu->arch.ret = RESUME_GUEST;
1536         vcpu->arch.trap = 0;
1537         kvmppc_update_vpas(vcpu);
1538
1539         /*
1540          * Synchronize with other threads in this virtual core
1541          */
1542         vc = vcpu->arch.vcore;
1543         spin_lock(&vc->lock);
1544         vcpu->arch.ceded = 0;
1545         vcpu->arch.run_task = current;
1546         vcpu->arch.kvm_run = kvm_run;
1547         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
1548         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1549         vcpu->arch.busy_preempt = TB_NIL;
1550         list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1551         ++vc->n_runnable;
1552
1553         /*
1554          * This happens the first time this is called for a vcpu.
1555          * If the vcore is already running, we may be able to start
1556          * this thread straight away and have it join in.
1557          */
1558         if (!signal_pending(current)) {
1559                 if (vc->vcore_state == VCORE_RUNNING &&
1560                     VCORE_EXIT_COUNT(vc) == 0) {
1561                         kvmppc_create_dtl_entry(vcpu, vc);
1562                         kvmppc_start_thread(vcpu);
1563                 } else if (vc->vcore_state == VCORE_SLEEPING) {
1564                         wake_up(&vc->wq);
1565                 }
1566
1567         }
1568
1569         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1570                !signal_pending(current)) {
1571                 if (vc->vcore_state != VCORE_INACTIVE) {
1572                         spin_unlock(&vc->lock);
1573                         kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1574                         spin_lock(&vc->lock);
1575                         continue;
1576                 }
1577                 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1578                                          arch.run_list) {
1579                         kvmppc_core_prepare_to_enter(v);
1580                         if (signal_pending(v->arch.run_task)) {
1581                                 kvmppc_remove_runnable(vc, v);
1582                                 v->stat.signal_exits++;
1583                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1584                                 v->arch.ret = -EINTR;
1585                                 wake_up(&v->arch.cpu_run);
1586                         }
1587                 }
1588                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1589                         break;
1590                 vc->runner = vcpu;
1591                 n_ceded = 0;
1592                 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
1593                         if (!v->arch.pending_exceptions)
1594                                 n_ceded += v->arch.ceded;
1595                         else
1596                                 v->arch.ceded = 0;
1597                 }
1598                 if (n_ceded == vc->n_runnable)
1599                         kvmppc_vcore_blocked(vc);
1600                 else
1601                         kvmppc_run_core(vc);
1602                 vc->runner = NULL;
1603         }
1604
1605         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1606                (vc->vcore_state == VCORE_RUNNING ||
1607                 vc->vcore_state == VCORE_EXITING)) {
1608                 spin_unlock(&vc->lock);
1609                 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1610                 spin_lock(&vc->lock);
1611         }
1612
1613         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1614                 kvmppc_remove_runnable(vc, vcpu);
1615                 vcpu->stat.signal_exits++;
1616                 kvm_run->exit_reason = KVM_EXIT_INTR;
1617                 vcpu->arch.ret = -EINTR;
1618         }
1619
1620         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1621                 /* Wake up some vcpu to run the core */
1622                 v = list_first_entry(&vc->runnable_threads,
1623                                      struct kvm_vcpu, arch.run_list);
1624                 wake_up(&v->arch.cpu_run);
1625         }
1626
1627         spin_unlock(&vc->lock);
1628         return vcpu->arch.ret;
1629 }
1630
1631 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
1632 {
1633         int r;
1634         int srcu_idx;
1635
1636         if (!vcpu->arch.sane) {
1637                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1638                 return -EINVAL;
1639         }
1640
1641         kvmppc_core_prepare_to_enter(vcpu);
1642
1643         /* No need to go into the guest when all we'll do is come back out */
1644         if (signal_pending(current)) {
1645                 run->exit_reason = KVM_EXIT_INTR;
1646                 return -EINTR;
1647         }
1648
1649         atomic_inc(&vcpu->kvm->arch.vcpus_running);
1650         /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1651         smp_mb();
1652
1653         /* On the first time here, set up HTAB and VRMA or RMA */
1654         if (!vcpu->kvm->arch.rma_setup_done) {
1655                 r = kvmppc_hv_setup_htab_rma(vcpu);
1656                 if (r)
1657                         goto out;
1658         }
1659
1660         flush_fp_to_thread(current);
1661         flush_altivec_to_thread(current);
1662         flush_vsx_to_thread(current);
1663         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1664         vcpu->arch.pgdir = current->mm->pgd;
1665         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1666
1667         do {
1668                 r = kvmppc_run_vcpu(run, vcpu);
1669
1670                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1671                     !(vcpu->arch.shregs.msr & MSR_PR)) {
1672                         r = kvmppc_pseries_do_hcall(vcpu);
1673                         kvmppc_core_prepare_to_enter(vcpu);
1674                 } else if (r == RESUME_PAGE_FAULT) {
1675                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1676                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
1677                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1678                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1679                 }
1680         } while (r == RESUME_GUEST);
1681
1682  out:
1683         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1684         atomic_dec(&vcpu->kvm->arch.vcpus_running);
1685         return r;
1686 }
1687
1688
1689 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1690    Assumes POWER7 or PPC970. */
1691 static inline int lpcr_rmls(unsigned long rma_size)
1692 {
1693         switch (rma_size) {
1694         case 32ul << 20:        /* 32 MB */
1695                 if (cpu_has_feature(CPU_FTR_ARCH_206))
1696                         return 8;       /* only supported on POWER7 */
1697                 return -1;
1698         case 64ul << 20:        /* 64 MB */
1699                 return 3;
1700         case 128ul << 20:       /* 128 MB */
1701                 return 7;
1702         case 256ul << 20:       /* 256 MB */
1703                 return 4;
1704         case 1ul << 30:         /* 1 GB */
1705                 return 2;
1706         case 16ul << 30:        /* 16 GB */
1707                 return 1;
1708         case 256ul << 30:       /* 256 GB */
1709                 return 0;
1710         default:
1711                 return -1;
1712         }
1713 }
1714
1715 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1716 {
1717         struct page *page;
1718         struct kvm_rma_info *ri = vma->vm_file->private_data;
1719
1720         if (vmf->pgoff >= kvm_rma_pages)
1721                 return VM_FAULT_SIGBUS;
1722
1723         page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1724         get_page(page);
1725         vmf->page = page;
1726         return 0;
1727 }
1728
1729 static const struct vm_operations_struct kvm_rma_vm_ops = {
1730         .fault = kvm_rma_fault,
1731 };
1732
1733 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1734 {
1735         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1736         vma->vm_ops = &kvm_rma_vm_ops;
1737         return 0;
1738 }
1739
1740 static int kvm_rma_release(struct inode *inode, struct file *filp)
1741 {
1742         struct kvm_rma_info *ri = filp->private_data;
1743
1744         kvm_release_rma(ri);
1745         return 0;
1746 }
1747
1748 static const struct file_operations kvm_rma_fops = {
1749         .mmap           = kvm_rma_mmap,
1750         .release        = kvm_rma_release,
1751 };
1752
1753 static long kvm_vm_ioctl_allocate_rma(struct kvm *kvm,
1754                                       struct kvm_allocate_rma *ret)
1755 {
1756         long fd;
1757         struct kvm_rma_info *ri;
1758         /*
1759          * Only do this on PPC970 in HV mode
1760          */
1761         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
1762             !cpu_has_feature(CPU_FTR_ARCH_201))
1763                 return -EINVAL;
1764
1765         if (!kvm_rma_pages)
1766                 return -EINVAL;
1767
1768         ri = kvm_alloc_rma();
1769         if (!ri)
1770                 return -ENOMEM;
1771
1772         fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC);
1773         if (fd < 0)
1774                 kvm_release_rma(ri);
1775
1776         ret->rma_size = kvm_rma_pages << PAGE_SHIFT;
1777         return fd;
1778 }
1779
1780 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1781                                      int linux_psize)
1782 {
1783         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1784
1785         if (!def->shift)
1786                 return;
1787         (*sps)->page_shift = def->shift;
1788         (*sps)->slb_enc = def->sllp;
1789         (*sps)->enc[0].page_shift = def->shift;
1790         /*
1791          * Only return base page encoding. We don't want to return
1792          * all the supporting pte_enc, because our H_ENTER doesn't
1793          * support MPSS yet. Once they do, we can start passing all
1794          * support pte_enc here
1795          */
1796         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
1797         (*sps)++;
1798 }
1799
1800 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
1801                                          struct kvm_ppc_smmu_info *info)
1802 {
1803         struct kvm_ppc_one_seg_page_size *sps;
1804
1805         info->flags = KVM_PPC_PAGE_SIZES_REAL;
1806         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1807                 info->flags |= KVM_PPC_1T_SEGMENTS;
1808         info->slb_size = mmu_slb_size;
1809
1810         /* We only support these sizes for now, and no muti-size segments */
1811         sps = &info->sps[0];
1812         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1813         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1814         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1815
1816         return 0;
1817 }
1818
1819 /*
1820  * Get (and clear) the dirty memory log for a memory slot.
1821  */
1822 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
1823                                          struct kvm_dirty_log *log)
1824 {
1825         struct kvm_memory_slot *memslot;
1826         int r;
1827         unsigned long n;
1828
1829         mutex_lock(&kvm->slots_lock);
1830
1831         r = -EINVAL;
1832         if (log->slot >= KVM_USER_MEM_SLOTS)
1833                 goto out;
1834
1835         memslot = id_to_memslot(kvm->memslots, log->slot);
1836         r = -ENOENT;
1837         if (!memslot->dirty_bitmap)
1838                 goto out;
1839
1840         n = kvm_dirty_bitmap_bytes(memslot);
1841         memset(memslot->dirty_bitmap, 0, n);
1842
1843         r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
1844         if (r)
1845                 goto out;
1846
1847         r = -EFAULT;
1848         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1849                 goto out;
1850
1851         r = 0;
1852 out:
1853         mutex_unlock(&kvm->slots_lock);
1854         return r;
1855 }
1856
1857 static void unpin_slot(struct kvm_memory_slot *memslot)
1858 {
1859         unsigned long *physp;
1860         unsigned long j, npages, pfn;
1861         struct page *page;
1862
1863         physp = memslot->arch.slot_phys;
1864         npages = memslot->npages;
1865         if (!physp)
1866                 return;
1867         for (j = 0; j < npages; j++) {
1868                 if (!(physp[j] & KVMPPC_GOT_PAGE))
1869                         continue;
1870                 pfn = physp[j] >> PAGE_SHIFT;
1871                 page = pfn_to_page(pfn);
1872                 SetPageDirty(page);
1873                 put_page(page);
1874         }
1875 }
1876
1877 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
1878                                         struct kvm_memory_slot *dont)
1879 {
1880         if (!dont || free->arch.rmap != dont->arch.rmap) {
1881                 vfree(free->arch.rmap);
1882                 free->arch.rmap = NULL;
1883         }
1884         if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
1885                 unpin_slot(free);
1886                 vfree(free->arch.slot_phys);
1887                 free->arch.slot_phys = NULL;
1888         }
1889 }
1890
1891 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
1892                                          unsigned long npages)
1893 {
1894         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
1895         if (!slot->arch.rmap)
1896                 return -ENOMEM;
1897         slot->arch.slot_phys = NULL;
1898
1899         return 0;
1900 }
1901
1902 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
1903                                         struct kvm_memory_slot *memslot,
1904                                         struct kvm_userspace_memory_region *mem)
1905 {
1906         unsigned long *phys;
1907
1908         /* Allocate a slot_phys array if needed */
1909         phys = memslot->arch.slot_phys;
1910         if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
1911                 phys = vzalloc(memslot->npages * sizeof(unsigned long));
1912                 if (!phys)
1913                         return -ENOMEM;
1914                 memslot->arch.slot_phys = phys;
1915         }
1916
1917         return 0;
1918 }
1919
1920 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
1921                                 struct kvm_userspace_memory_region *mem,
1922                                 const struct kvm_memory_slot *old)
1923 {
1924         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
1925         struct kvm_memory_slot *memslot;
1926
1927         if (npages && old->npages) {
1928                 /*
1929                  * If modifying a memslot, reset all the rmap dirty bits.
1930                  * If this is a new memslot, we don't need to do anything
1931                  * since the rmap array starts out as all zeroes,
1932                  * i.e. no pages are dirty.
1933                  */
1934                 memslot = id_to_memslot(kvm->memslots, mem->slot);
1935                 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
1936         }
1937 }
1938
1939 /*
1940  * Update LPCR values in kvm->arch and in vcores.
1941  * Caller must hold kvm->lock.
1942  */
1943 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
1944 {
1945         long int i;
1946         u32 cores_done = 0;
1947
1948         if ((kvm->arch.lpcr & mask) == lpcr)
1949                 return;
1950
1951         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
1952
1953         for (i = 0; i < KVM_MAX_VCORES; ++i) {
1954                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
1955                 if (!vc)
1956                         continue;
1957                 spin_lock(&vc->lock);
1958                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
1959                 spin_unlock(&vc->lock);
1960                 if (++cores_done >= kvm->arch.online_vcores)
1961                         break;
1962         }
1963 }
1964
1965 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
1966 {
1967         return;
1968 }
1969
1970 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
1971 {
1972         int err = 0;
1973         struct kvm *kvm = vcpu->kvm;
1974         struct kvm_rma_info *ri = NULL;
1975         unsigned long hva;
1976         struct kvm_memory_slot *memslot;
1977         struct vm_area_struct *vma;
1978         unsigned long lpcr = 0, senc;
1979         unsigned long lpcr_mask = 0;
1980         unsigned long psize, porder;
1981         unsigned long rma_size;
1982         unsigned long rmls;
1983         unsigned long *physp;
1984         unsigned long i, npages;
1985         int srcu_idx;
1986
1987         mutex_lock(&kvm->lock);
1988         if (kvm->arch.rma_setup_done)
1989                 goto out;       /* another vcpu beat us to it */
1990
1991         /* Allocate hashed page table (if not done already) and reset it */
1992         if (!kvm->arch.hpt_virt) {
1993                 err = kvmppc_alloc_hpt(kvm, NULL);
1994                 if (err) {
1995                         pr_err("KVM: Couldn't alloc HPT\n");
1996                         goto out;
1997                 }
1998         }
1999
2000         /* Look up the memslot for guest physical address 0 */
2001         srcu_idx = srcu_read_lock(&kvm->srcu);
2002         memslot = gfn_to_memslot(kvm, 0);
2003
2004         /* We must have some memory at 0 by now */
2005         err = -EINVAL;
2006         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2007                 goto out_srcu;
2008
2009         /* Look up the VMA for the start of this memory slot */
2010         hva = memslot->userspace_addr;
2011         down_read(&current->mm->mmap_sem);
2012         vma = find_vma(current->mm, hva);
2013         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2014                 goto up_out;
2015
2016         psize = vma_kernel_pagesize(vma);
2017         porder = __ilog2(psize);
2018
2019         /* Is this one of our preallocated RMAs? */
2020         if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
2021             hva == vma->vm_start)
2022                 ri = vma->vm_file->private_data;
2023
2024         up_read(&current->mm->mmap_sem);
2025
2026         if (!ri) {
2027                 /* On POWER7, use VRMA; on PPC970, give up */
2028                 err = -EPERM;
2029                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2030                         pr_err("KVM: CPU requires an RMO\n");
2031                         goto out_srcu;
2032                 }
2033
2034                 /* We can handle 4k, 64k or 16M pages in the VRMA */
2035                 err = -EINVAL;
2036                 if (!(psize == 0x1000 || psize == 0x10000 ||
2037                       psize == 0x1000000))
2038                         goto out_srcu;
2039
2040                 /* Update VRMASD field in the LPCR */
2041                 senc = slb_pgsize_encoding(psize);
2042                 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2043                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
2044                 lpcr_mask = LPCR_VRMASD;
2045                 /* the -4 is to account for senc values starting at 0x10 */
2046                 lpcr = senc << (LPCR_VRMASD_SH - 4);
2047
2048                 /* Create HPTEs in the hash page table for the VRMA */
2049                 kvmppc_map_vrma(vcpu, memslot, porder);
2050
2051         } else {
2052                 /* Set up to use an RMO region */
2053                 rma_size = kvm_rma_pages;
2054                 if (rma_size > memslot->npages)
2055                         rma_size = memslot->npages;
2056                 rma_size <<= PAGE_SHIFT;
2057                 rmls = lpcr_rmls(rma_size);
2058                 err = -EINVAL;
2059                 if ((long)rmls < 0) {
2060                         pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
2061                         goto out_srcu;
2062                 }
2063                 atomic_inc(&ri->use_count);
2064                 kvm->arch.rma = ri;
2065
2066                 /* Update LPCR and RMOR */
2067                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2068                         /* PPC970; insert RMLS value (split field) in HID4 */
2069                         lpcr_mask = (1ul << HID4_RMLS0_SH) |
2070                                 (3ul << HID4_RMLS2_SH) | HID4_RMOR;
2071                         lpcr = ((rmls >> 2) << HID4_RMLS0_SH) |
2072                                 ((rmls & 3) << HID4_RMLS2_SH);
2073                         /* RMOR is also in HID4 */
2074                         lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
2075                                 << HID4_RMOR_SH;
2076                 } else {
2077                         /* POWER7 */
2078                         lpcr_mask = LPCR_VPM0 | LPCR_VRMA_L | LPCR_RMLS;
2079                         lpcr = rmls << LPCR_RMLS_SH;
2080                         kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT;
2081                 }
2082                 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
2083                         ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
2084
2085                 /* Initialize phys addrs of pages in RMO */
2086                 npages = kvm_rma_pages;
2087                 porder = __ilog2(npages);
2088                 physp = memslot->arch.slot_phys;
2089                 if (physp) {
2090                         if (npages > memslot->npages)
2091                                 npages = memslot->npages;
2092                         spin_lock(&kvm->arch.slot_phys_lock);
2093                         for (i = 0; i < npages; ++i)
2094                                 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
2095                                         porder;
2096                         spin_unlock(&kvm->arch.slot_phys_lock);
2097                 }
2098         }
2099
2100         kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
2101
2102         /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
2103         smp_wmb();
2104         kvm->arch.rma_setup_done = 1;
2105         err = 0;
2106  out_srcu:
2107         srcu_read_unlock(&kvm->srcu, srcu_idx);
2108  out:
2109         mutex_unlock(&kvm->lock);
2110         return err;
2111
2112  up_out:
2113         up_read(&current->mm->mmap_sem);
2114         goto out_srcu;
2115 }
2116
2117 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2118 {
2119         unsigned long lpcr, lpid;
2120
2121         /* Allocate the guest's logical partition ID */
2122
2123         lpid = kvmppc_alloc_lpid();
2124         if ((long)lpid < 0)
2125                 return -ENOMEM;
2126         kvm->arch.lpid = lpid;
2127
2128         /*
2129          * Since we don't flush the TLB when tearing down a VM,
2130          * and this lpid might have previously been used,
2131          * make sure we flush on each core before running the new VM.
2132          */
2133         cpumask_setall(&kvm->arch.need_tlb_flush);
2134
2135         kvm->arch.rma = NULL;
2136
2137         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
2138
2139         if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2140                 /* PPC970; HID4 is effectively the LPCR */
2141                 kvm->arch.host_lpid = 0;
2142                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
2143                 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
2144                 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
2145                         ((lpid & 0xf) << HID4_LPID5_SH);
2146         } else {
2147                 /* POWER7; init LPCR for virtual RMA mode */
2148                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
2149                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2150                 lpcr &= LPCR_PECE | LPCR_LPES;
2151                 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
2152                         LPCR_VPM0 | LPCR_VPM1;
2153                 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2154                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
2155         }
2156         kvm->arch.lpcr = lpcr;
2157
2158         kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
2159         spin_lock_init(&kvm->arch.slot_phys_lock);
2160
2161         /*
2162          * Don't allow secondary CPU threads to come online
2163          * while any KVM VMs exist.
2164          */
2165         inhibit_secondary_onlining();
2166
2167         return 0;
2168 }
2169
2170 static void kvmppc_free_vcores(struct kvm *kvm)
2171 {
2172         long int i;
2173
2174         for (i = 0; i < KVM_MAX_VCORES; ++i)
2175                 kfree(kvm->arch.vcores[i]);
2176         kvm->arch.online_vcores = 0;
2177 }
2178
2179 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
2180 {
2181         uninhibit_secondary_onlining();
2182
2183         kvmppc_free_vcores(kvm);
2184         if (kvm->arch.rma) {
2185                 kvm_release_rma(kvm->arch.rma);
2186                 kvm->arch.rma = NULL;
2187         }
2188
2189         kvmppc_free_hpt(kvm);
2190 }
2191
2192 /* We don't need to emulate any privileged instructions or dcbz */
2193 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2194                                      unsigned int inst, int *advance)
2195 {
2196         return EMULATE_FAIL;
2197 }
2198
2199 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2200                                         ulong spr_val)
2201 {
2202         return EMULATE_FAIL;
2203 }
2204
2205 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2206                                         ulong *spr_val)
2207 {
2208         return EMULATE_FAIL;
2209 }
2210
2211 static int kvmppc_core_check_processor_compat_hv(void)
2212 {
2213         if (!cpu_has_feature(CPU_FTR_HVMODE))
2214                 return -EIO;
2215         return 0;
2216 }
2217
2218 static long kvm_arch_vm_ioctl_hv(struct file *filp,
2219                                  unsigned int ioctl, unsigned long arg)
2220 {
2221         struct kvm *kvm __maybe_unused = filp->private_data;
2222         void __user *argp = (void __user *)arg;
2223         long r;
2224
2225         switch (ioctl) {
2226
2227         case KVM_ALLOCATE_RMA: {
2228                 struct kvm_allocate_rma rma;
2229                 struct kvm *kvm = filp->private_data;
2230
2231                 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
2232                 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
2233                         r = -EFAULT;
2234                 break;
2235         }
2236
2237         case KVM_PPC_ALLOCATE_HTAB: {
2238                 u32 htab_order;
2239
2240                 r = -EFAULT;
2241                 if (get_user(htab_order, (u32 __user *)argp))
2242                         break;
2243                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2244                 if (r)
2245                         break;
2246                 r = -EFAULT;
2247                 if (put_user(htab_order, (u32 __user *)argp))
2248                         break;
2249                 r = 0;
2250                 break;
2251         }
2252
2253         case KVM_PPC_GET_HTAB_FD: {
2254                 struct kvm_get_htab_fd ghf;
2255
2256                 r = -EFAULT;
2257                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
2258                         break;
2259                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2260                 break;
2261         }
2262
2263         default:
2264                 r = -ENOTTY;
2265         }
2266
2267         return r;
2268 }
2269
2270 static struct kvmppc_ops kvm_ops_hv = {
2271         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2272         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2273         .get_one_reg = kvmppc_get_one_reg_hv,
2274         .set_one_reg = kvmppc_set_one_reg_hv,
2275         .vcpu_load   = kvmppc_core_vcpu_load_hv,
2276         .vcpu_put    = kvmppc_core_vcpu_put_hv,
2277         .set_msr     = kvmppc_set_msr_hv,
2278         .vcpu_run    = kvmppc_vcpu_run_hv,
2279         .vcpu_create = kvmppc_core_vcpu_create_hv,
2280         .vcpu_free   = kvmppc_core_vcpu_free_hv,
2281         .check_requests = kvmppc_core_check_requests_hv,
2282         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
2283         .flush_memslot  = kvmppc_core_flush_memslot_hv,
2284         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2285         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
2286         .unmap_hva = kvm_unmap_hva_hv,
2287         .unmap_hva_range = kvm_unmap_hva_range_hv,
2288         .age_hva  = kvm_age_hva_hv,
2289         .test_age_hva = kvm_test_age_hva_hv,
2290         .set_spte_hva = kvm_set_spte_hva_hv,
2291         .mmu_destroy  = kvmppc_mmu_destroy_hv,
2292         .free_memslot = kvmppc_core_free_memslot_hv,
2293         .create_memslot = kvmppc_core_create_memslot_hv,
2294         .init_vm =  kvmppc_core_init_vm_hv,
2295         .destroy_vm = kvmppc_core_destroy_vm_hv,
2296         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2297         .emulate_op = kvmppc_core_emulate_op_hv,
2298         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2299         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2300         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2301         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
2302 };
2303
2304 static int kvmppc_book3s_init_hv(void)
2305 {
2306         int r;
2307         /*
2308          * FIXME!! Do we need to check on all cpus ?
2309          */
2310         r = kvmppc_core_check_processor_compat_hv();
2311         if (r < 0)
2312                 return r;
2313
2314         kvm_ops_hv.owner = THIS_MODULE;
2315         kvmppc_hv_ops = &kvm_ops_hv;
2316
2317         r = kvmppc_mmu_hv_init();
2318         return r;
2319 }
2320
2321 static void kvmppc_book3s_exit_hv(void)
2322 {
2323         kvmppc_hv_ops = NULL;
2324 }
2325
2326 module_init(kvmppc_book3s_init_hv);
2327 module_exit(kvmppc_book3s_exit_hv);
2328 MODULE_LICENSE("GPL");
2329 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2330 MODULE_ALIAS("devname:kvm");