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                         /*
 188                          * If an arch bit is set in PCR, all the defined
 189                          * higher-order arch bits also have to be set.
 190                          */
 191                         pcr = PCR_ARCH_206 | PCR_ARCH_205;
 192                         break;
 193                 case PVR_ARCH_206:
 194                 case PVR_ARCH_206p:
 195                         pcr = PCR_ARCH_206;
 196                         break;
 197                 case PVR_ARCH_207:
 198                         break;
 199                 default:
 200                         return -EINVAL;
 201                 }
 202
 203                 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
 204                         /* POWER7 can't emulate POWER8 */
 205                         if (!(pcr & PCR_ARCH_206))
 206                                 return -EINVAL;
 207                         pcr &= ~PCR_ARCH_206;
 208                 }
 209         }
 210
 211         spin_lock(&vc->lock);
 212         vc->arch_compat = arch_compat;
 213         vc->pcr = pcr;
 214         spin_unlock(&vc->lock);
 215
 216         return 0;
 217 }
 218
 219 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 220 {
 221         int r;
 222
 223         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 224         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 225                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
 226         for (r = 0; r < 16; ++r)
 227                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 228                        r, kvmppc_get_gpr(vcpu, r),
 229                        r+16, kvmppc_get_gpr(vcpu, r+16));
 230         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 231                vcpu->arch.ctr, vcpu->arch.lr);
 232         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 233                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 234         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 235                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 236         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 237                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 238         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 239                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 240         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 241         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 242                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 243         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 244         for (r = 0; r < vcpu->arch.slb_max; ++r)
 245                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 246                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 247         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 248                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 249                vcpu->arch.last_inst);
 250 }
 251
 252 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 253 {
 254         int r;
 255         struct kvm_vcpu *v, *ret = NULL;
 256
 257         mutex_lock(&kvm->lock);
 258         kvm_for_each_vcpu(r, v, kvm) {
 259                 if (v->vcpu_id == id) {
 260                         ret = v;
 261                         break;
 262                 }
 263         }
 264         mutex_unlock(&kvm->lock);
 265         return ret;
 266 }
 267
 268 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 269 {
 270         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 271         vpa->yield_count = 1;
 272 }
 273
 274 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 275                    unsigned long addr, unsigned long len)
 276 {
 277         /* check address is cacheline aligned */
 278         if (addr & (L1_CACHE_BYTES - 1))
 279                 return -EINVAL;
 280         spin_lock(&vcpu->arch.vpa_update_lock);
 281         if (v->next_gpa != addr || v->len != len) {
 282                 v->next_gpa = addr;
 283                 v->len = addr ? len : 0;
 284                 v->update_pending = 1;
 285         }
 286         spin_unlock(&vcpu->arch.vpa_update_lock);
 287         return 0;
 288 }
 289
 290 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 291 struct reg_vpa {
 292         u32 dummy;
 293         union {
 294                 u16 hword;
 295                 u32 word;
 296         } length;
 297 };
 298
 299 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 300 {
 301         if (vpap->update_pending)
 302                 return vpap->next_gpa != 0;
 303         return vpap->pinned_addr != NULL;
 304 }
 305
 306 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 307                                        unsigned long flags,
 308                                        unsigned long vcpuid, unsigned long vpa)
 309 {
 310         struct kvm *kvm = vcpu->kvm;
 311         unsigned long len, nb;
 312         void *va;
 313         struct kvm_vcpu *tvcpu;
 314         int err;
 315         int subfunc;
 316         struct kvmppc_vpa *vpap;
 317
 318         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 319         if (!tvcpu)
 320                 return H_PARAMETER;
 321
 322         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 323         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 324             subfunc == H_VPA_REG_SLB) {
 325                 /* Registering new area - address must be cache-line aligned */
 326                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 327                         return H_PARAMETER;
 328
 329                 /* convert logical addr to kernel addr and read length */
 330                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 331                 if (va == NULL)
 332                         return H_PARAMETER;
 333                 if (subfunc == H_VPA_REG_VPA)
 334                         len = ((struct reg_vpa *)va)->length.hword;
 335                 else
 336                         len = ((struct reg_vpa *)va)->length.word;
 337                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 338
 339                 /* Check length */
 340                 if (len > nb || len < sizeof(struct reg_vpa))
 341                         return H_PARAMETER;
 342         } else {
 343                 vpa = 0;
 344                 len = 0;
 345         }
 346
 347         err = H_PARAMETER;
 348         vpap = NULL;
 349         spin_lock(&tvcpu->arch.vpa_update_lock);
 350
 351         switch (subfunc) {
 352         case H_VPA_REG_VPA:             /* register VPA */
 353                 if (len < sizeof(struct lppaca))
 354                         break;
 355                 vpap = &tvcpu->arch.vpa;
 356                 err = 0;
 357                 break;
 358
 359         case H_VPA_REG_DTL:             /* register DTL */
 360                 if (len < sizeof(struct dtl_entry))
 361                         break;
 362                 len -= len % sizeof(struct dtl_entry);
 363
 364                 /* Check that they have previously registered a VPA */
 365                 err = H_RESOURCE;
 366                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 367                         break;
 368
 369                 vpap = &tvcpu->arch.dtl;
 370                 err = 0;
 371                 break;
 372
 373         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 374                 /* Check that they have previously registered a VPA */
 375                 err = H_RESOURCE;
 376                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 377                         break;
 378
 379                 vpap = &tvcpu->arch.slb_shadow;
 380                 err = 0;
 381                 break;
 382
 383         case H_VPA_DEREG_VPA:           /* deregister VPA */
 384                 /* Check they don't still have a DTL or SLB buf registered */
 385                 err = H_RESOURCE;
 386                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 387                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 388                         break;
 389
 390                 vpap = &tvcpu->arch.vpa;
 391                 err = 0;
 392                 break;
 393
 394         case H_VPA_DEREG_DTL:           /* deregister DTL */
 395                 vpap = &tvcpu->arch.dtl;
 396                 err = 0;
 397                 break;
 398
 399         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 400                 vpap = &tvcpu->arch.slb_shadow;
 401                 err = 0;
 402                 break;
 403         }
 404
 405         if (vpap) {
 406                 vpap->next_gpa = vpa;
 407                 vpap->len = len;
 408                 vpap->update_pending = 1;
 409         }
 410
 411         spin_unlock(&tvcpu->arch.vpa_update_lock);
 412
 413         return err;
 414 }
 415
 416 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 417 {
 418         struct kvm *kvm = vcpu->kvm;
 419         void *va;
 420         unsigned long nb;
 421         unsigned long gpa;
 422
 423         /*
 424          * We need to pin the page pointed to by vpap->next_gpa,
 425          * but we can't call kvmppc_pin_guest_page under the lock
 426          * as it does get_user_pages() and down_read().  So we
 427          * have to drop the lock, pin the page, then get the lock
 428          * again and check that a new area didn't get registered
 429          * in the meantime.
 430          */
 431         for (;;) {
 432                 gpa = vpap->next_gpa;
 433                 spin_unlock(&vcpu->arch.vpa_update_lock);
 434                 va = NULL;
 435                 nb = 0;
 436                 if (gpa)
 437                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 438                 spin_lock(&vcpu->arch.vpa_update_lock);
 439                 if (gpa == vpap->next_gpa)
 440                         break;
 441                 /* sigh... unpin that one and try again */
 442                 if (va)
 443                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 444         }
 445
 446         vpap->update_pending = 0;
 447         if (va && nb < vpap->len) {
 448                 /*
 449                  * If it's now too short, it must be that userspace
 450                  * has changed the mappings underlying guest memory,
 451                  * so unregister the region.
 452                  */
 453                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 454                 va = NULL;
 455         }
 456         if (vpap->pinned_addr)
 457                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 458                                         vpap->dirty);
 459         vpap->gpa = gpa;
 460         vpap->pinned_addr = va;
 461         vpap->dirty = false;
 462         if (va)
 463                 vpap->pinned_end = va + vpap->len;
 464 }
 465
 466 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 467 {
 468         if (!(vcpu->arch.vpa.update_pending ||
 469               vcpu->arch.slb_shadow.update_pending ||
 470               vcpu->arch.dtl.update_pending))
 471                 return;
 472
 473         spin_lock(&vcpu->arch.vpa_update_lock);
 474         if (vcpu->arch.vpa.update_pending) {
 475                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 476                 if (vcpu->arch.vpa.pinned_addr)
 477                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 478         }
 479         if (vcpu->arch.dtl.update_pending) {
 480                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 481                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 482                 vcpu->arch.dtl_index = 0;
 483         }
 484         if (vcpu->arch.slb_shadow.update_pending)
 485                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 486         spin_unlock(&vcpu->arch.vpa_update_lock);
 487 }
 488
 489 /*
 490  * Return the accumulated stolen time for the vcore up until `now'.
 491  * The caller should hold the vcore lock.
 492  */
 493 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 494 {
 495         u64 p;
 496
 497         /*
 498          * If we are the task running the vcore, then since we hold
 499          * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
 500          * can't be updated, so we don't need the tbacct_lock.
 501          * If the vcore is inactive, it can't become active (since we
 502          * hold the vcore lock), so the vcpu load/put functions won't
 503          * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
 504          */
 505         if (vc->vcore_state != VCORE_INACTIVE &&
 506             vc->runner->arch.run_task != current) {
 507                 spin_lock(&vc->runner->arch.tbacct_lock);
 508                 p = vc->stolen_tb;
 509                 if (vc->preempt_tb != TB_NIL)
 510                         p += now - vc->preempt_tb;
 511                 spin_unlock(&vc->runner->arch.tbacct_lock);
 512         } else {
 513                 p = vc->stolen_tb;
 514         }
 515         return p;
 516 }
 517
 518 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 519                                     struct kvmppc_vcore *vc)
 520 {
 521         struct dtl_entry *dt;
 522         struct lppaca *vpa;
 523         unsigned long stolen;
 524         unsigned long core_stolen;
 525         u64 now;
 526
 527         dt = vcpu->arch.dtl_ptr;
 528         vpa = vcpu->arch.vpa.pinned_addr;
 529         now = mftb();
 530         core_stolen = vcore_stolen_time(vc, now);
 531         stolen = core_stolen - vcpu->arch.stolen_logged;
 532         vcpu->arch.stolen_logged = core_stolen;
 533         spin_lock(&vcpu->arch.tbacct_lock);
 534         stolen += vcpu->arch.busy_stolen;
 535         vcpu->arch.busy_stolen = 0;
 536         spin_unlock(&vcpu->arch.tbacct_lock);
 537         if (!dt || !vpa)
 538                 return;
 539         memset(dt, 0, sizeof(struct dtl_entry));
 540         dt->dispatch_reason = 7;
 541         dt->processor_id = vc->pcpu + vcpu->arch.ptid;
 542         dt->timebase = now + vc->tb_offset;
 543         dt->enqueue_to_dispatch_time = stolen;
 544         dt->srr0 = kvmppc_get_pc(vcpu);
 545         dt->srr1 = vcpu->arch.shregs.msr;
 546         ++dt;
 547         if (dt == vcpu->arch.dtl.pinned_end)
 548                 dt = vcpu->arch.dtl.pinned_addr;
 549         vcpu->arch.dtl_ptr = dt;
 550         /* order writing *dt vs. writing vpa->dtl_idx */
 551         smp_wmb();
 552         vpa->dtl_idx = ++vcpu->arch.dtl_index;
 553         vcpu->arch.dtl.dirty = true;
 554 }
 555
 556 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 557 {
 558         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 559         unsigned long target, ret = H_SUCCESS;
 560         struct kvm_vcpu *tvcpu;
 561         int idx, rc;
 562
 563         switch (req) {
 564         case H_ENTER:
 565                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 566                 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
 567                                               kvmppc_get_gpr(vcpu, 5),
 568                                               kvmppc_get_gpr(vcpu, 6),
 569                                               kvmppc_get_gpr(vcpu, 7));
 570                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 571                 break;
 572         case H_CEDE:
 573                 break;
 574         case H_PROD:
 575                 target = kvmppc_get_gpr(vcpu, 4);
 576                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 577                 if (!tvcpu) {
 578                         ret = H_PARAMETER;
 579                         break;
 580                 }
 581                 tvcpu->arch.prodded = 1;
 582                 smp_mb();
 583                 if (vcpu->arch.ceded) {
 584                         if (waitqueue_active(&vcpu->wq)) {
 585                                 wake_up_interruptible(&vcpu->wq);
 586                                 vcpu->stat.halt_wakeup++;
 587                         }
 588                 }
 589                 break;
 590         case H_CONFER:
 591                 target = kvmppc_get_gpr(vcpu, 4);
 592                 if (target == -1)
 593                         break;
 594                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 595                 if (!tvcpu) {
 596                         ret = H_PARAMETER;
 597                         break;
 598                 }
 599                 kvm_vcpu_yield_to(tvcpu);
 600                 break;
 601         case H_REGISTER_VPA:
 602                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 603                                         kvmppc_get_gpr(vcpu, 5),
 604                                         kvmppc_get_gpr(vcpu, 6));
 605                 break;
 606         case H_RTAS:
 607                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 608                         return RESUME_HOST;
 609
 610                 rc = kvmppc_rtas_hcall(vcpu);
 611
 612                 if (rc == -ENOENT)
 613                         return RESUME_HOST;
 614                 else if (rc == 0)
 615                         break;
 616
 617                 /* Send the error out to userspace via KVM_RUN */
 618                 return rc;
 619
 620         case H_XIRR:
 621         case H_CPPR:
 622         case H_EOI:
 623         case H_IPI:
 624         case H_IPOLL:
 625         case H_XIRR_X:
 626                 if (kvmppc_xics_enabled(vcpu)) {
 627                         ret = kvmppc_xics_hcall(vcpu, req);
 628                         break;
 629                 } /* fallthrough */
 630         default:
 631                 return RESUME_HOST;
 632         }
 633         kvmppc_set_gpr(vcpu, 3, ret);
 634         vcpu->arch.hcall_needed = 0;
 635         return RESUME_GUEST;
 636 }
 637
 638 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
 639                                  struct task_struct *tsk)
 640 {
 641         int r = RESUME_HOST;
 642
 643         vcpu->stat.sum_exits++;
 644
 645         run->exit_reason = KVM_EXIT_UNKNOWN;
 646         run->ready_for_interrupt_injection = 1;
 647         switch (vcpu->arch.trap) {
 648         /* We're good on these - the host merely wanted to get our attention */
 649         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 650                 vcpu->stat.dec_exits++;
 651                 r = RESUME_GUEST;
 652                 break;
 653         case BOOK3S_INTERRUPT_EXTERNAL:
 654                 vcpu->stat.ext_intr_exits++;
 655                 r = RESUME_GUEST;
 656                 break;
 657         case BOOK3S_INTERRUPT_PERFMON:
 658                 r = RESUME_GUEST;
 659                 break;
 660         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 661                 /*
 662                  * Deliver a machine check interrupt to the guest.
 663                  * We have to do this, even if the host has handled the
 664                  * machine check, because machine checks use SRR0/1 and
 665                  * the interrupt might have trashed guest state in them.
 666                  */
 667                 kvmppc_book3s_queue_irqprio(vcpu,
 668                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 669                 r = RESUME_GUEST;
 670                 break;
 671         case BOOK3S_INTERRUPT_PROGRAM:
 672         {
 673                 ulong flags;
 674                 /*
 675                  * Normally program interrupts are delivered directly
 676                  * to the guest by the hardware, but we can get here
 677                  * as a result of a hypervisor emulation interrupt
 678                  * (e40) getting turned into a 700 by BML RTAS.
 679                  */
 680                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 681                 kvmppc_core_queue_program(vcpu, flags);
 682                 r = RESUME_GUEST;
 683                 break;
 684         }
 685         case BOOK3S_INTERRUPT_SYSCALL:
 686         {
 687                 /* hcall - punt to userspace */
 688                 int i;
 689
 690                 /* hypercall with MSR_PR has already been handled in rmode,
 691                  * and never reaches here.
 692                  */
 693
 694                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 695                 for (i = 0; i < 9; ++i)
 696                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 697                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 698                 vcpu->arch.hcall_needed = 1;
 699                 r = RESUME_HOST;
 700                 break;
 701         }
 702         /*
 703          * We get these next two if the guest accesses a page which it thinks
 704          * it has mapped but which is not actually present, either because
 705          * it is for an emulated I/O device or because the corresonding
 706          * host page has been paged out.  Any other HDSI/HISI interrupts
 707          * have been handled already.
 708          */
 709         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 710                 r = RESUME_PAGE_FAULT;
 711                 break;
 712         case BOOK3S_INTERRUPT_H_INST_STORAGE:
 713                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
 714                 vcpu->arch.fault_dsisr = 0;
 715                 r = RESUME_PAGE_FAULT;
 716                 break;
 717         /*
 718          * This occurs if the guest executes an illegal instruction.
 719          * We just generate a program interrupt to the guest, since
 720          * we don't emulate any guest instructions at this stage.
 721          */
 722         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 723                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 724                 r = RESUME_GUEST;
 725                 break;
 726         /*
 727          * This occurs if the guest (kernel or userspace), does something that
 728          * is prohibited by HFSCR.  We just generate a program interrupt to
 729          * the guest.
 730          */
 731         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
 732                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 733                 r = RESUME_GUEST;
 734                 break;
 735         default:
 736                 kvmppc_dump_regs(vcpu);
 737                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 738                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 739                         vcpu->arch.shregs.msr);
 740                 run->hw.hardware_exit_reason = vcpu->arch.trap;
 741                 r = RESUME_HOST;
 742                 break;
 743         }
 744
 745         return r;
 746 }
 747
 748 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
 749                                             struct kvm_sregs *sregs)
 750 {
 751         int i;
 752
 753         memset(sregs, 0, sizeof(struct kvm_sregs));
 754         sregs->pvr = vcpu->arch.pvr;
 755         for (i = 0; i < vcpu->arch.slb_max; i++) {
 756                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
 757                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 758         }
 759
 760         return 0;
 761 }
 762
 763 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
 764                                             struct kvm_sregs *sregs)
 765 {
 766         int i, j;
 767
 768         kvmppc_set_pvr_hv(vcpu, sregs->pvr);
 769
 770         j = 0;
 771         for (i = 0; i < vcpu->arch.slb_nr; i++) {
 772                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
 773                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
 774                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
 775                         ++j;
 776                 }
 777         }
 778         vcpu->arch.slb_max = j;
 779
 780         return 0;
 781 }
 782
 783 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr)
 784 {
 785         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 786         u64 mask;
 787
 788         spin_lock(&vc->lock);
 789         /*
 790          * Userspace can only modify DPFD (default prefetch depth),
 791          * ILE (interrupt little-endian) and TC (translation control).
 792          * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
 793          */
 794         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
 795         if (cpu_has_feature(CPU_FTR_ARCH_207S))
 796                 mask |= LPCR_AIL;
 797         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
 798         spin_unlock(&vc->lock);
 799 }
 800
 801 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
 802                                  union kvmppc_one_reg *val)
 803 {
 804         int r = 0;
 805         long int i;
 806
 807         switch (id) {
 808         case KVM_REG_PPC_HIOR:
 809                 *val = get_reg_val(id, 0);
 810                 break;
 811         case KVM_REG_PPC_DABR:
 812                 *val = get_reg_val(id, vcpu->arch.dabr);
 813                 break;
 814         case KVM_REG_PPC_DSCR:
 815                 *val = get_reg_val(id, vcpu->arch.dscr);
 816                 break;
 817         case KVM_REG_PPC_PURR:
 818                 *val = get_reg_val(id, vcpu->arch.purr);
 819                 break;
 820         case KVM_REG_PPC_SPURR:
 821                 *val = get_reg_val(id, vcpu->arch.spurr);
 822                 break;
 823         case KVM_REG_PPC_AMR:
 824                 *val = get_reg_val(id, vcpu->arch.amr);
 825                 break;
 826         case KVM_REG_PPC_UAMOR:
 827                 *val = get_reg_val(id, vcpu->arch.uamor);
 828                 break;
 829         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
 830                 i = id - KVM_REG_PPC_MMCR0;
 831                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
 832                 break;
 833         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
 834                 i = id - KVM_REG_PPC_PMC1;
 835                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
 836                 break;
 837         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
 838                 i = id - KVM_REG_PPC_SPMC1;
 839                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
 840                 break;
 841         case KVM_REG_PPC_SIAR:
 842                 *val = get_reg_val(id, vcpu->arch.siar);
 843                 break;
 844         case KVM_REG_PPC_SDAR:
 845                 *val = get_reg_val(id, vcpu->arch.sdar);
 846                 break;
 847         case KVM_REG_PPC_SIER:
 848                 *val = get_reg_val(id, vcpu->arch.sier);
 849                 break;
 850         case KVM_REG_PPC_IAMR:
 851                 *val = get_reg_val(id, vcpu->arch.iamr);
 852                 break;
 853         case KVM_REG_PPC_TFHAR:
 854                 *val = get_reg_val(id, vcpu->arch.tfhar);
 855                 break;
 856         case KVM_REG_PPC_TFIAR:
 857                 *val = get_reg_val(id, vcpu->arch.tfiar);
 858                 break;
 859         case KVM_REG_PPC_TEXASR:
 860                 *val = get_reg_val(id, vcpu->arch.texasr);
 861                 break;
 862         case KVM_REG_PPC_FSCR:
 863                 *val = get_reg_val(id, vcpu->arch.fscr);
 864                 break;
 865         case KVM_REG_PPC_PSPB:
 866                 *val = get_reg_val(id, vcpu->arch.pspb);
 867                 break;
 868         case KVM_REG_PPC_EBBHR:
 869                 *val = get_reg_val(id, vcpu->arch.ebbhr);
 870                 break;
 871         case KVM_REG_PPC_EBBRR:
 872                 *val = get_reg_val(id, vcpu->arch.ebbrr);
 873                 break;
 874         case KVM_REG_PPC_BESCR:
 875                 *val = get_reg_val(id, vcpu->arch.bescr);
 876                 break;
 877         case KVM_REG_PPC_TAR:
 878                 *val = get_reg_val(id, vcpu->arch.tar);
 879                 break;
 880         case KVM_REG_PPC_DPDES:
 881                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
 882                 break;
 883         case KVM_REG_PPC_DAWR:
 884                 *val = get_reg_val(id, vcpu->arch.dawr);
 885                 break;
 886         case KVM_REG_PPC_DAWRX:
 887                 *val = get_reg_val(id, vcpu->arch.dawrx);
 888                 break;
 889         case KVM_REG_PPC_CIABR:
 890                 *val = get_reg_val(id, vcpu->arch.ciabr);
 891                 break;
 892         case KVM_REG_PPC_IC:
 893                 *val = get_reg_val(id, vcpu->arch.ic);
 894                 break;
 895         case KVM_REG_PPC_VTB:
 896                 *val = get_reg_val(id, vcpu->arch.vtb);
 897                 break;
 898         case KVM_REG_PPC_CSIGR:
 899                 *val = get_reg_val(id, vcpu->arch.csigr);
 900                 break;
 901         case KVM_REG_PPC_TACR:
 902                 *val = get_reg_val(id, vcpu->arch.tacr);
 903                 break;
 904         case KVM_REG_PPC_TCSCR:
 905                 *val = get_reg_val(id, vcpu->arch.tcscr);
 906                 break;
 907         case KVM_REG_PPC_PID:
 908                 *val = get_reg_val(id, vcpu->arch.pid);
 909                 break;
 910         case KVM_REG_PPC_ACOP:
 911                 *val = get_reg_val(id, vcpu->arch.acop);
 912                 break;
 913         case KVM_REG_PPC_WORT:
 914                 *val = get_reg_val(id, vcpu->arch.wort);
 915                 break;
 916         case KVM_REG_PPC_VPA_ADDR:
 917                 spin_lock(&vcpu->arch.vpa_update_lock);
 918                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
 919                 spin_unlock(&vcpu->arch.vpa_update_lock);
 920                 break;
 921         case KVM_REG_PPC_VPA_SLB:
 922                 spin_lock(&vcpu->arch.vpa_update_lock);
 923                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
 924                 val->vpaval.length = vcpu->arch.slb_shadow.len;
 925                 spin_unlock(&vcpu->arch.vpa_update_lock);
 926                 break;
 927         case KVM_REG_PPC_VPA_DTL:
 928                 spin_lock(&vcpu->arch.vpa_update_lock);
 929                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
 930                 val->vpaval.length = vcpu->arch.dtl.len;
 931                 spin_unlock(&vcpu->arch.vpa_update_lock);
 932                 break;
 933         case KVM_REG_PPC_TB_OFFSET:
 934                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
 935                 break;
 936         case KVM_REG_PPC_LPCR:
 937                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
 938                 break;
 939         case KVM_REG_PPC_PPR:
 940                 *val = get_reg_val(id, vcpu->arch.ppr);
 941                 break;
 942         case KVM_REG_PPC_ARCH_COMPAT:
 943                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
 944                 break;
 945         default:
 946                 r = -EINVAL;
 947                 break;
 948         }
 949
 950         return r;
 951 }
 952
 953 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
 954                                  union kvmppc_one_reg *val)
 955 {
 956         int r = 0;
 957         long int i;
 958         unsigned long addr, len;
 959
 960         switch (id) {
 961         case KVM_REG_PPC_HIOR:
 962                 /* Only allow this to be set to zero */
 963                 if (set_reg_val(id, *val))
 964                         r = -EINVAL;
 965                 break;
 966         case KVM_REG_PPC_DABR:
 967                 vcpu->arch.dabr = set_reg_val(id, *val);
 968                 break;
 969         case KVM_REG_PPC_DSCR:
 970                 vcpu->arch.dscr = set_reg_val(id, *val);
 971                 break;
 972         case KVM_REG_PPC_PURR:
 973                 vcpu->arch.purr = set_reg_val(id, *val);
 974                 break;
 975         case KVM_REG_PPC_SPURR:
 976                 vcpu->arch.spurr = set_reg_val(id, *val);
 977                 break;
 978         case KVM_REG_PPC_AMR:
 979                 vcpu->arch.amr = set_reg_val(id, *val);
 980                 break;
 981         case KVM_REG_PPC_UAMOR:
 982                 vcpu->arch.uamor = set_reg_val(id, *val);
 983                 break;
 984         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
 985                 i = id - KVM_REG_PPC_MMCR0;
 986                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
 987                 break;
 988         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
 989                 i = id - KVM_REG_PPC_PMC1;
 990                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
 991                 break;
 992         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
 993                 i = id - KVM_REG_PPC_SPMC1;
 994                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
 995                 break;
 996         case KVM_REG_PPC_SIAR:
 997                 vcpu->arch.siar = set_reg_val(id, *val);
 998                 break;
 999         case KVM_REG_PPC_SDAR:
1000                 vcpu->arch.sdar = set_reg_val(id, *val);
1001                 break;
1002         case KVM_REG_PPC_SIER:
1003                 vcpu->arch.sier = set_reg_val(id, *val);
1004                 break;
1005         case KVM_REG_PPC_IAMR:
1006                 vcpu->arch.iamr = set_reg_val(id, *val);
1007                 break;
1008         case KVM_REG_PPC_TFHAR:
1009                 vcpu->arch.tfhar = set_reg_val(id, *val);
1010                 break;
1011         case KVM_REG_PPC_TFIAR:
1012                 vcpu->arch.tfiar = set_reg_val(id, *val);
1013                 break;
1014         case KVM_REG_PPC_TEXASR:
1015                 vcpu->arch.texasr = set_reg_val(id, *val);
1016                 break;
1017         case KVM_REG_PPC_FSCR:
1018                 vcpu->arch.fscr = set_reg_val(id, *val);
1019                 break;
1020         case KVM_REG_PPC_PSPB:
1021                 vcpu->arch.pspb = set_reg_val(id, *val);
1022                 break;
1023         case KVM_REG_PPC_EBBHR:
1024                 vcpu->arch.ebbhr = set_reg_val(id, *val);
1025                 break;
1026         case KVM_REG_PPC_EBBRR:
1027                 vcpu->arch.ebbrr = set_reg_val(id, *val);
1028                 break;
1029         case KVM_REG_PPC_BESCR:
1030                 vcpu->arch.bescr = set_reg_val(id, *val);
1031                 break;
1032         case KVM_REG_PPC_TAR:
1033                 vcpu->arch.tar = set_reg_val(id, *val);
1034                 break;
1035         case KVM_REG_PPC_DPDES:
1036                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1037                 break;
1038         case KVM_REG_PPC_DAWR:
1039                 vcpu->arch.dawr = set_reg_val(id, *val);
1040                 break;
1041         case KVM_REG_PPC_DAWRX:
1042                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1043                 break;
1044         case KVM_REG_PPC_CIABR:
1045                 vcpu->arch.ciabr = set_reg_val(id, *val);
1046                 /* Don't allow setting breakpoints in hypervisor code */
1047                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1048                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1049                 break;
1050         case KVM_REG_PPC_IC:
1051                 vcpu->arch.ic = set_reg_val(id, *val);
1052                 break;
1053         case KVM_REG_PPC_VTB:
1054                 vcpu->arch.vtb = set_reg_val(id, *val);
1055                 break;
1056         case KVM_REG_PPC_CSIGR:
1057                 vcpu->arch.csigr = set_reg_val(id, *val);
1058                 break;
1059         case KVM_REG_PPC_TACR:
1060                 vcpu->arch.tacr = set_reg_val(id, *val);
1061                 break;
1062         case KVM_REG_PPC_TCSCR:
1063                 vcpu->arch.tcscr = set_reg_val(id, *val);
1064                 break;
1065         case KVM_REG_PPC_PID:
1066                 vcpu->arch.pid = set_reg_val(id, *val);
1067                 break;
1068         case KVM_REG_PPC_ACOP:
1069                 vcpu->arch.acop = set_reg_val(id, *val);
1070                 break;
1071         case KVM_REG_PPC_WORT:
1072                 vcpu->arch.wort = set_reg_val(id, *val);
1073                 break;
1074         case KVM_REG_PPC_VPA_ADDR:
1075                 addr = set_reg_val(id, *val);
1076                 r = -EINVAL;
1077                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1078                               vcpu->arch.dtl.next_gpa))
1079                         break;
1080                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1081                 break;
1082         case KVM_REG_PPC_VPA_SLB:
1083                 addr = val->vpaval.addr;
1084                 len = val->vpaval.length;
1085                 r = -EINVAL;
1086                 if (addr && !vcpu->arch.vpa.next_gpa)
1087                         break;
1088                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1089                 break;
1090         case KVM_REG_PPC_VPA_DTL:
1091                 addr = val->vpaval.addr;
1092                 len = val->vpaval.length;
1093                 r = -EINVAL;
1094                 if (addr && (len < sizeof(struct dtl_entry) ||
1095                              !vcpu->arch.vpa.next_gpa))
1096                         break;
1097                 len -= len % sizeof(struct dtl_entry);
1098                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1099                 break;
1100         case KVM_REG_PPC_TB_OFFSET:
1101                 /* round up to multiple of 2^24 */
1102                 vcpu->arch.vcore->tb_offset =
1103                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1104                 break;
1105         case KVM_REG_PPC_LPCR:
1106                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val));
1107                 break;
1108         case KVM_REG_PPC_PPR:
1109                 vcpu->arch.ppr = set_reg_val(id, *val);
1110                 break;
1111         case KVM_REG_PPC_ARCH_COMPAT:
1112                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1113                 break;
1114         default:
1115                 r = -EINVAL;
1116                 break;
1117         }
1118
1119         return r;
1120 }
1121
1122 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1123                                                    unsigned int id)
1124 {
1125         struct kvm_vcpu *vcpu;
1126         int err = -EINVAL;
1127         int core;
1128         struct kvmppc_vcore *vcore;
1129
1130         core = id / threads_per_core;
1131         if (core >= KVM_MAX_VCORES)
1132                 goto out;
1133
1134         err = -ENOMEM;
1135         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1136         if (!vcpu)
1137                 goto out;
1138
1139         err = kvm_vcpu_init(vcpu, kvm, id);
1140         if (err)
1141                 goto free_vcpu;
1142
1143         vcpu->arch.shared = &vcpu->arch.shregs;
1144         vcpu->arch.mmcr[0] = MMCR0_FC;
1145         vcpu->arch.ctrl = CTRL_RUNLATCH;
1146         /* default to host PVR, since we can't spoof it */
1147         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1148         spin_lock_init(&vcpu->arch.vpa_update_lock);
1149         spin_lock_init(&vcpu->arch.tbacct_lock);
1150         vcpu->arch.busy_preempt = TB_NIL;
1151
1152         kvmppc_mmu_book3s_hv_init(vcpu);
1153
1154         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1155
1156         init_waitqueue_head(&vcpu->arch.cpu_run);
1157
1158         mutex_lock(&kvm->lock);
1159         vcore = kvm->arch.vcores[core];
1160         if (!vcore) {
1161                 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1162                 if (vcore) {
1163                         INIT_LIST_HEAD(&vcore->runnable_threads);
1164                         spin_lock_init(&vcore->lock);
1165                         init_waitqueue_head(&vcore->wq);
1166                         vcore->preempt_tb = TB_NIL;
1167                         vcore->lpcr = kvm->arch.lpcr;
1168                         vcore->first_vcpuid = core * threads_per_core;
1169                         vcore->kvm = kvm;
1170                 }
1171                 kvm->arch.vcores[core] = vcore;
1172                 kvm->arch.online_vcores++;
1173         }
1174         mutex_unlock(&kvm->lock);
1175
1176         if (!vcore)
1177                 goto free_vcpu;
1178
1179         spin_lock(&vcore->lock);
1180         ++vcore->num_threads;
1181         spin_unlock(&vcore->lock);
1182         vcpu->arch.vcore = vcore;
1183         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1184
1185         vcpu->arch.cpu_type = KVM_CPU_3S_64;
1186         kvmppc_sanity_check(vcpu);
1187
1188         return vcpu;
1189
1190 free_vcpu:
1191         kmem_cache_free(kvm_vcpu_cache, vcpu);
1192 out:
1193         return ERR_PTR(err);
1194 }
1195
1196 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1197 {
1198         if (vpa->pinned_addr)
1199                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1200                                         vpa->dirty);
1201 }
1202
1203 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1204 {
1205         spin_lock(&vcpu->arch.vpa_update_lock);
1206         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1207         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1208         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1209         spin_unlock(&vcpu->arch.vpa_update_lock);
1210         kvm_vcpu_uninit(vcpu);
1211         kmem_cache_free(kvm_vcpu_cache, vcpu);
1212 }
1213
1214 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1215 {
1216         /* Indicate we want to get back into the guest */
1217         return 1;
1218 }
1219
1220 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1221 {
1222         unsigned long dec_nsec, now;
1223
1224         now = get_tb();
1225         if (now > vcpu->arch.dec_expires) {
1226                 /* decrementer has already gone negative */
1227                 kvmppc_core_queue_dec(vcpu);
1228                 kvmppc_core_prepare_to_enter(vcpu);
1229                 return;
1230         }
1231         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1232                    / tb_ticks_per_sec;
1233         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1234                       HRTIMER_MODE_REL);
1235         vcpu->arch.timer_running = 1;
1236 }
1237
1238 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1239 {
1240         vcpu->arch.ceded = 0;
1241         if (vcpu->arch.timer_running) {
1242                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1243                 vcpu->arch.timer_running = 0;
1244         }
1245 }
1246
1247 extern void __kvmppc_vcore_entry(void);
1248
1249 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1250                                    struct kvm_vcpu *vcpu)
1251 {
1252         u64 now;
1253
1254         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1255                 return;
1256         spin_lock(&vcpu->arch.tbacct_lock);
1257         now = mftb();
1258         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1259                 vcpu->arch.stolen_logged;
1260         vcpu->arch.busy_preempt = now;
1261         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1262         spin_unlock(&vcpu->arch.tbacct_lock);
1263         --vc->n_runnable;
1264         list_del(&vcpu->arch.run_list);
1265 }
1266
1267 static int kvmppc_grab_hwthread(int cpu)
1268 {
1269         struct paca_struct *tpaca;
1270         long timeout = 1000;
1271
1272         tpaca = &paca[cpu];
1273
1274         /* Ensure the thread won't go into the kernel if it wakes */
1275         tpaca->kvm_hstate.hwthread_req = 1;
1276         tpaca->kvm_hstate.kvm_vcpu = NULL;
1277
1278         /*
1279          * If the thread is already executing in the kernel (e.g. handling
1280          * a stray interrupt), wait for it to get back to nap mode.
1281          * The smp_mb() is to ensure that our setting of hwthread_req
1282          * is visible before we look at hwthread_state, so if this
1283          * races with the code at system_reset_pSeries and the thread
1284          * misses our setting of hwthread_req, we are sure to see its
1285          * setting of hwthread_state, and vice versa.
1286          */
1287         smp_mb();
1288         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1289                 if (--timeout <= 0) {
1290                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1291                         return -EBUSY;
1292                 }
1293                 udelay(1);
1294         }
1295         return 0;
1296 }
1297
1298 static void kvmppc_release_hwthread(int cpu)
1299 {
1300         struct paca_struct *tpaca;
1301
1302         tpaca = &paca[cpu];
1303         tpaca->kvm_hstate.hwthread_req = 0;
1304         tpaca->kvm_hstate.kvm_vcpu = NULL;
1305 }
1306
1307 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1308 {
1309         int cpu;
1310         struct paca_struct *tpaca;
1311         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1312
1313         if (vcpu->arch.timer_running) {
1314                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1315                 vcpu->arch.timer_running = 0;
1316         }
1317         cpu = vc->pcpu + vcpu->arch.ptid;
1318         tpaca = &paca[cpu];
1319         tpaca->kvm_hstate.kvm_vcpu = vcpu;
1320         tpaca->kvm_hstate.kvm_vcore = vc;
1321         tpaca->kvm_hstate.ptid = vcpu->arch.ptid;
1322         vcpu->cpu = vc->pcpu;
1323         smp_wmb();
1324 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1325         if (cpu != smp_processor_id()) {
1326 #ifdef CONFIG_KVM_XICS
1327                 xics_wake_cpu(cpu);
1328 #endif
1329                 if (vcpu->arch.ptid)
1330                         ++vc->n_woken;
1331         }
1332 #endif
1333 }
1334
1335 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1336 {
1337         int i;
1338
1339         HMT_low();
1340         i = 0;
1341         while (vc->nap_count < vc->n_woken) {
1342                 if (++i >= 1000000) {
1343                         pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1344                                vc->nap_count, vc->n_woken);
1345                         break;
1346                 }
1347                 cpu_relax();
1348         }
1349         HMT_medium();
1350 }
1351
1352 /*
1353  * Check that we are on thread 0 and that any other threads in
1354  * this core are off-line.  Then grab the threads so they can't
1355  * enter the kernel.
1356  */
1357 static int on_primary_thread(void)
1358 {
1359         int cpu = smp_processor_id();
1360         int thr = cpu_thread_in_core(cpu);
1361
1362         if (thr)
1363                 return 0;
1364         while (++thr < threads_per_core)
1365                 if (cpu_online(cpu + thr))
1366                         return 0;
1367
1368         /* Grab all hw threads so they can't go into the kernel */
1369         for (thr = 1; thr < threads_per_core; ++thr) {
1370                 if (kvmppc_grab_hwthread(cpu + thr)) {
1371                         /* Couldn't grab one; let the others go */
1372                         do {
1373                                 kvmppc_release_hwthread(cpu + thr);
1374                         } while (--thr > 0);
1375                         return 0;
1376                 }
1377         }
1378         return 1;
1379 }
1380
1381 /*
1382  * Run a set of guest threads on a physical core.
1383  * Called with vc->lock held.
1384  */
1385 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1386 {
1387         struct kvm_vcpu *vcpu, *vnext;
1388         long ret;
1389         u64 now;
1390         int i, need_vpa_update;
1391         int srcu_idx;
1392         struct kvm_vcpu *vcpus_to_update[threads_per_core];
1393
1394         /* don't start if any threads have a signal pending */
1395         need_vpa_update = 0;
1396         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1397                 if (signal_pending(vcpu->arch.run_task))
1398                         return;
1399                 if (vcpu->arch.vpa.update_pending ||
1400                     vcpu->arch.slb_shadow.update_pending ||
1401                     vcpu->arch.dtl.update_pending)
1402                         vcpus_to_update[need_vpa_update++] = vcpu;
1403         }
1404
1405         /*
1406          * Initialize *vc, in particular vc->vcore_state, so we can
1407          * drop the vcore lock if necessary.
1408          */
1409         vc->n_woken = 0;
1410         vc->nap_count = 0;
1411         vc->entry_exit_count = 0;
1412         vc->vcore_state = VCORE_STARTING;
1413         vc->in_guest = 0;
1414         vc->napping_threads = 0;
1415
1416         /*
1417          * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1418          * which can't be called with any spinlocks held.
1419          */
1420         if (need_vpa_update) {
1421                 spin_unlock(&vc->lock);
1422                 for (i = 0; i < need_vpa_update; ++i)
1423                         kvmppc_update_vpas(vcpus_to_update[i]);
1424                 spin_lock(&vc->lock);
1425         }
1426
1427         /*
1428          * Make sure we are running on thread 0, and that
1429          * secondary threads are offline.
1430          */
1431         if (threads_per_core > 1 && !on_primary_thread()) {
1432                 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1433                         vcpu->arch.ret = -EBUSY;
1434                 goto out;
1435         }
1436
1437         vc->pcpu = smp_processor_id();
1438         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1439                 kvmppc_start_thread(vcpu);
1440                 kvmppc_create_dtl_entry(vcpu, vc);
1441         }
1442
1443         /* Set this explicitly in case thread 0 doesn't have a vcpu */
1444         get_paca()->kvm_hstate.kvm_vcore = vc;
1445         get_paca()->kvm_hstate.ptid = 0;
1446
1447         vc->vcore_state = VCORE_RUNNING;
1448         preempt_disable();
1449         spin_unlock(&vc->lock);
1450
1451         kvm_guest_enter();
1452
1453         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
1454
1455         __kvmppc_vcore_entry();
1456
1457         spin_lock(&vc->lock);
1458         /* disable sending of IPIs on virtual external irqs */
1459         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1460                 vcpu->cpu = -1;
1461         /* wait for secondary threads to finish writing their state to memory */
1462         if (vc->nap_count < vc->n_woken)
1463                 kvmppc_wait_for_nap(vc);
1464         for (i = 0; i < threads_per_core; ++i)
1465                 kvmppc_release_hwthread(vc->pcpu + i);
1466         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
1467         vc->vcore_state = VCORE_EXITING;
1468         spin_unlock(&vc->lock);
1469
1470         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
1471
1472         /* make sure updates to secondary vcpu structs are visible now */
1473         smp_mb();
1474         kvm_guest_exit();
1475
1476         preempt_enable();
1477         kvm_resched(vcpu);
1478
1479         spin_lock(&vc->lock);
1480         now = get_tb();
1481         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1482                 /* cancel pending dec exception if dec is positive */
1483                 if (now < vcpu->arch.dec_expires &&
1484                     kvmppc_core_pending_dec(vcpu))
1485                         kvmppc_core_dequeue_dec(vcpu);
1486
1487                 ret = RESUME_GUEST;
1488                 if (vcpu->arch.trap)
1489                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1490                                                     vcpu->arch.run_task);
1491
1492                 vcpu->arch.ret = ret;
1493                 vcpu->arch.trap = 0;
1494
1495                 if (vcpu->arch.ceded) {
1496                         if (ret != RESUME_GUEST)
1497                                 kvmppc_end_cede(vcpu);
1498                         else
1499                                 kvmppc_set_timer(vcpu);
1500                 }
1501         }
1502
1503  out:
1504         vc->vcore_state = VCORE_INACTIVE;
1505         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1506                                  arch.run_list) {
1507                 if (vcpu->arch.ret != RESUME_GUEST) {
1508                         kvmppc_remove_runnable(vc, vcpu);
1509                         wake_up(&vcpu->arch.cpu_run);
1510                 }
1511         }
1512 }
1513
1514 /*
1515  * Wait for some other vcpu thread to execute us, and
1516  * wake us up when we need to handle something in the host.
1517  */
1518 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1519 {
1520         DEFINE_WAIT(wait);
1521
1522         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1523         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1524                 schedule();
1525         finish_wait(&vcpu->arch.cpu_run, &wait);
1526 }
1527
1528 /*
1529  * All the vcpus in this vcore are idle, so wait for a decrementer
1530  * or external interrupt to one of the vcpus.  vc->lock is held.
1531  */
1532 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1533 {
1534         DEFINE_WAIT(wait);
1535
1536         prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1537         vc->vcore_state = VCORE_SLEEPING;
1538         spin_unlock(&vc->lock);
1539         schedule();
1540         finish_wait(&vc->wq, &wait);
1541         spin_lock(&vc->lock);
1542         vc->vcore_state = VCORE_INACTIVE;
1543 }
1544
1545 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1546 {
1547         int n_ceded;
1548         struct kvmppc_vcore *vc;
1549         struct kvm_vcpu *v, *vn;
1550
1551         kvm_run->exit_reason = 0;
1552         vcpu->arch.ret = RESUME_GUEST;
1553         vcpu->arch.trap = 0;
1554         kvmppc_update_vpas(vcpu);
1555
1556         /*
1557          * Synchronize with other threads in this virtual core
1558          */
1559         vc = vcpu->arch.vcore;
1560         spin_lock(&vc->lock);
1561         vcpu->arch.ceded = 0;
1562         vcpu->arch.run_task = current;
1563         vcpu->arch.kvm_run = kvm_run;
1564         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
1565         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1566         vcpu->arch.busy_preempt = TB_NIL;
1567         list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1568         ++vc->n_runnable;
1569
1570         /*
1571          * This happens the first time this is called for a vcpu.
1572          * If the vcore is already running, we may be able to start
1573          * this thread straight away and have it join in.
1574          */
1575         if (!signal_pending(current)) {
1576                 if (vc->vcore_state == VCORE_RUNNING &&
1577                     VCORE_EXIT_COUNT(vc) == 0) {
1578                         kvmppc_create_dtl_entry(vcpu, vc);
1579                         kvmppc_start_thread(vcpu);
1580                 } else if (vc->vcore_state == VCORE_SLEEPING) {
1581                         wake_up(&vc->wq);
1582                 }
1583
1584         }
1585
1586         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1587                !signal_pending(current)) {
1588                 if (vc->vcore_state != VCORE_INACTIVE) {
1589                         spin_unlock(&vc->lock);
1590                         kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1591                         spin_lock(&vc->lock);
1592                         continue;
1593                 }
1594                 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1595                                          arch.run_list) {
1596                         kvmppc_core_prepare_to_enter(v);
1597                         if (signal_pending(v->arch.run_task)) {
1598                                 kvmppc_remove_runnable(vc, v);
1599                                 v->stat.signal_exits++;
1600                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1601                                 v->arch.ret = -EINTR;
1602                                 wake_up(&v->arch.cpu_run);
1603                         }
1604                 }
1605                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1606                         break;
1607                 vc->runner = vcpu;
1608                 n_ceded = 0;
1609                 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
1610                         if (!v->arch.pending_exceptions)
1611                                 n_ceded += v->arch.ceded;
1612                         else
1613                                 v->arch.ceded = 0;
1614                 }
1615                 if (n_ceded == vc->n_runnable)
1616                         kvmppc_vcore_blocked(vc);
1617                 else
1618                         kvmppc_run_core(vc);
1619                 vc->runner = NULL;
1620         }
1621
1622         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1623                (vc->vcore_state == VCORE_RUNNING ||
1624                 vc->vcore_state == VCORE_EXITING)) {
1625                 spin_unlock(&vc->lock);
1626                 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1627                 spin_lock(&vc->lock);
1628         }
1629
1630         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1631                 kvmppc_remove_runnable(vc, vcpu);
1632                 vcpu->stat.signal_exits++;
1633                 kvm_run->exit_reason = KVM_EXIT_INTR;
1634                 vcpu->arch.ret = -EINTR;
1635         }
1636
1637         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1638                 /* Wake up some vcpu to run the core */
1639                 v = list_first_entry(&vc->runnable_threads,
1640                                      struct kvm_vcpu, arch.run_list);
1641                 wake_up(&v->arch.cpu_run);
1642         }
1643
1644         spin_unlock(&vc->lock);
1645         return vcpu->arch.ret;
1646 }
1647
1648 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
1649 {
1650         int r;
1651         int srcu_idx;
1652
1653         if (!vcpu->arch.sane) {
1654                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1655                 return -EINVAL;
1656         }
1657
1658         kvmppc_core_prepare_to_enter(vcpu);
1659
1660         /* No need to go into the guest when all we'll do is come back out */
1661         if (signal_pending(current)) {
1662                 run->exit_reason = KVM_EXIT_INTR;
1663                 return -EINTR;
1664         }
1665
1666         atomic_inc(&vcpu->kvm->arch.vcpus_running);
1667         /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1668         smp_mb();
1669
1670         /* On the first time here, set up HTAB and VRMA or RMA */
1671         if (!vcpu->kvm->arch.rma_setup_done) {
1672                 r = kvmppc_hv_setup_htab_rma(vcpu);
1673                 if (r)
1674                         goto out;
1675         }
1676
1677         flush_fp_to_thread(current);
1678         flush_altivec_to_thread(current);
1679         flush_vsx_to_thread(current);
1680         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1681         vcpu->arch.pgdir = current->mm->pgd;
1682         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1683
1684         do {
1685                 r = kvmppc_run_vcpu(run, vcpu);
1686
1687                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1688                     !(vcpu->arch.shregs.msr & MSR_PR)) {
1689                         r = kvmppc_pseries_do_hcall(vcpu);
1690                         kvmppc_core_prepare_to_enter(vcpu);
1691                 } else if (r == RESUME_PAGE_FAULT) {
1692                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1693                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
1694                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1695                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1696                 }
1697         } while (r == RESUME_GUEST);
1698
1699  out:
1700         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1701         atomic_dec(&vcpu->kvm->arch.vcpus_running);
1702         return r;
1703 }
1704
1705
1706 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1707    Assumes POWER7 or PPC970. */
1708 static inline int lpcr_rmls(unsigned long rma_size)
1709 {
1710         switch (rma_size) {
1711         case 32ul << 20:        /* 32 MB */
1712                 if (cpu_has_feature(CPU_FTR_ARCH_206))
1713                         return 8;       /* only supported on POWER7 */
1714                 return -1;
1715         case 64ul << 20:        /* 64 MB */
1716                 return 3;
1717         case 128ul << 20:       /* 128 MB */
1718                 return 7;
1719         case 256ul << 20:       /* 256 MB */
1720                 return 4;
1721         case 1ul << 30:         /* 1 GB */
1722                 return 2;
1723         case 16ul << 30:        /* 16 GB */
1724                 return 1;
1725         case 256ul << 30:       /* 256 GB */
1726                 return 0;
1727         default:
1728                 return -1;
1729         }
1730 }
1731
1732 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1733 {
1734         struct page *page;
1735         struct kvm_rma_info *ri = vma->vm_file->private_data;
1736
1737         if (vmf->pgoff >= kvm_rma_pages)
1738                 return VM_FAULT_SIGBUS;
1739
1740         page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1741         get_page(page);
1742         vmf->page = page;
1743         return 0;
1744 }
1745
1746 static const struct vm_operations_struct kvm_rma_vm_ops = {
1747         .fault = kvm_rma_fault,
1748 };
1749
1750 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1751 {
1752         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1753         vma->vm_ops = &kvm_rma_vm_ops;
1754         return 0;
1755 }
1756
1757 static int kvm_rma_release(struct inode *inode, struct file *filp)
1758 {
1759         struct kvm_rma_info *ri = filp->private_data;
1760
1761         kvm_release_rma(ri);
1762         return 0;
1763 }
1764
1765 static const struct file_operations kvm_rma_fops = {
1766         .mmap           = kvm_rma_mmap,
1767         .release        = kvm_rma_release,
1768 };
1769
1770 static long kvm_vm_ioctl_allocate_rma(struct kvm *kvm,
1771                                       struct kvm_allocate_rma *ret)
1772 {
1773         long fd;
1774         struct kvm_rma_info *ri;
1775         /*
1776          * Only do this on PPC970 in HV mode
1777          */
1778         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
1779             !cpu_has_feature(CPU_FTR_ARCH_201))
1780                 return -EINVAL;
1781
1782         if (!kvm_rma_pages)
1783                 return -EINVAL;
1784
1785         ri = kvm_alloc_rma();
1786         if (!ri)
1787                 return -ENOMEM;
1788
1789         fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC);
1790         if (fd < 0)
1791                 kvm_release_rma(ri);
1792
1793         ret->rma_size = kvm_rma_pages << PAGE_SHIFT;
1794         return fd;
1795 }
1796
1797 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1798                                      int linux_psize)
1799 {
1800         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1801
1802         if (!def->shift)
1803                 return;
1804         (*sps)->page_shift = def->shift;
1805         (*sps)->slb_enc = def->sllp;
1806         (*sps)->enc[0].page_shift = def->shift;
1807         /*
1808          * Only return base page encoding. We don't want to return
1809          * all the supporting pte_enc, because our H_ENTER doesn't
1810          * support MPSS yet. Once they do, we can start passing all
1811          * support pte_enc here
1812          */
1813         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
1814         (*sps)++;
1815 }
1816
1817 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
1818                                          struct kvm_ppc_smmu_info *info)
1819 {
1820         struct kvm_ppc_one_seg_page_size *sps;
1821
1822         info->flags = KVM_PPC_PAGE_SIZES_REAL;
1823         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1824                 info->flags |= KVM_PPC_1T_SEGMENTS;
1825         info->slb_size = mmu_slb_size;
1826
1827         /* We only support these sizes for now, and no muti-size segments */
1828         sps = &info->sps[0];
1829         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1830         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1831         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1832
1833         return 0;
1834 }
1835
1836 /*
1837  * Get (and clear) the dirty memory log for a memory slot.
1838  */
1839 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
1840                                          struct kvm_dirty_log *log)
1841 {
1842         struct kvm_memory_slot *memslot;
1843         int r;
1844         unsigned long n;
1845
1846         mutex_lock(&kvm->slots_lock);
1847
1848         r = -EINVAL;
1849         if (log->slot >= KVM_USER_MEM_SLOTS)
1850                 goto out;
1851
1852         memslot = id_to_memslot(kvm->memslots, log->slot);
1853         r = -ENOENT;
1854         if (!memslot->dirty_bitmap)
1855                 goto out;
1856
1857         n = kvm_dirty_bitmap_bytes(memslot);
1858         memset(memslot->dirty_bitmap, 0, n);
1859
1860         r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
1861         if (r)
1862                 goto out;
1863
1864         r = -EFAULT;
1865         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1866                 goto out;
1867
1868         r = 0;
1869 out:
1870         mutex_unlock(&kvm->slots_lock);
1871         return r;
1872 }
1873
1874 static void unpin_slot(struct kvm_memory_slot *memslot)
1875 {
1876         unsigned long *physp;
1877         unsigned long j, npages, pfn;
1878         struct page *page;
1879
1880         physp = memslot->arch.slot_phys;
1881         npages = memslot->npages;
1882         if (!physp)
1883                 return;
1884         for (j = 0; j < npages; j++) {
1885                 if (!(physp[j] & KVMPPC_GOT_PAGE))
1886                         continue;
1887                 pfn = physp[j] >> PAGE_SHIFT;
1888                 page = pfn_to_page(pfn);
1889                 SetPageDirty(page);
1890                 put_page(page);
1891         }
1892 }
1893
1894 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
1895                                         struct kvm_memory_slot *dont)
1896 {
1897         if (!dont || free->arch.rmap != dont->arch.rmap) {
1898                 vfree(free->arch.rmap);
1899                 free->arch.rmap = NULL;
1900         }
1901         if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
1902                 unpin_slot(free);
1903                 vfree(free->arch.slot_phys);
1904                 free->arch.slot_phys = NULL;
1905         }
1906 }
1907
1908 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
1909                                          unsigned long npages)
1910 {
1911         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
1912         if (!slot->arch.rmap)
1913                 return -ENOMEM;
1914         slot->arch.slot_phys = NULL;
1915
1916         return 0;
1917 }
1918
1919 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
1920                                         struct kvm_memory_slot *memslot,
1921                                         struct kvm_userspace_memory_region *mem)
1922 {
1923         unsigned long *phys;
1924
1925         /* Allocate a slot_phys array if needed */
1926         phys = memslot->arch.slot_phys;
1927         if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
1928                 phys = vzalloc(memslot->npages * sizeof(unsigned long));
1929                 if (!phys)
1930                         return -ENOMEM;
1931                 memslot->arch.slot_phys = phys;
1932         }
1933
1934         return 0;
1935 }
1936
1937 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
1938                                 struct kvm_userspace_memory_region *mem,
1939                                 const struct kvm_memory_slot *old)
1940 {
1941         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
1942         struct kvm_memory_slot *memslot;
1943
1944         if (npages && old->npages) {
1945                 /*
1946                  * If modifying a memslot, reset all the rmap dirty bits.
1947                  * If this is a new memslot, we don't need to do anything
1948                  * since the rmap array starts out as all zeroes,
1949                  * i.e. no pages are dirty.
1950                  */
1951                 memslot = id_to_memslot(kvm->memslots, mem->slot);
1952                 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
1953         }
1954 }
1955
1956 /*
1957  * Update LPCR values in kvm->arch and in vcores.
1958  * Caller must hold kvm->lock.
1959  */
1960 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
1961 {
1962         long int i;
1963         u32 cores_done = 0;
1964
1965         if ((kvm->arch.lpcr & mask) == lpcr)
1966                 return;
1967
1968         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
1969
1970         for (i = 0; i < KVM_MAX_VCORES; ++i) {
1971                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
1972                 if (!vc)
1973                         continue;
1974                 spin_lock(&vc->lock);
1975                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
1976                 spin_unlock(&vc->lock);
1977                 if (++cores_done >= kvm->arch.online_vcores)
1978                         break;
1979         }
1980 }
1981
1982 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
1983 {
1984         return;
1985 }
1986
1987 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
1988 {
1989         int err = 0;
1990         struct kvm *kvm = vcpu->kvm;
1991         struct kvm_rma_info *ri = NULL;
1992         unsigned long hva;
1993         struct kvm_memory_slot *memslot;
1994         struct vm_area_struct *vma;
1995         unsigned long lpcr = 0, senc;
1996         unsigned long lpcr_mask = 0;
1997         unsigned long psize, porder;
1998         unsigned long rma_size;
1999         unsigned long rmls;
2000         unsigned long *physp;
2001         unsigned long i, npages;
2002         int srcu_idx;
2003
2004         mutex_lock(&kvm->lock);
2005         if (kvm->arch.rma_setup_done)
2006                 goto out;       /* another vcpu beat us to it */
2007
2008         /* Allocate hashed page table (if not done already) and reset it */
2009         if (!kvm->arch.hpt_virt) {
2010                 err = kvmppc_alloc_hpt(kvm, NULL);
2011                 if (err) {
2012                         pr_err("KVM: Couldn't alloc HPT\n");
2013                         goto out;
2014                 }
2015         }
2016
2017         /* Look up the memslot for guest physical address 0 */
2018         srcu_idx = srcu_read_lock(&kvm->srcu);
2019         memslot = gfn_to_memslot(kvm, 0);
2020
2021         /* We must have some memory at 0 by now */
2022         err = -EINVAL;
2023         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2024                 goto out_srcu;
2025
2026         /* Look up the VMA for the start of this memory slot */
2027         hva = memslot->userspace_addr;
2028         down_read(&current->mm->mmap_sem);
2029         vma = find_vma(current->mm, hva);
2030         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2031                 goto up_out;
2032
2033         psize = vma_kernel_pagesize(vma);
2034         porder = __ilog2(psize);
2035
2036         /* Is this one of our preallocated RMAs? */
2037         if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
2038             hva == vma->vm_start)
2039                 ri = vma->vm_file->private_data;
2040
2041         up_read(&current->mm->mmap_sem);
2042
2043         if (!ri) {
2044                 /* On POWER7, use VRMA; on PPC970, give up */
2045                 err = -EPERM;
2046                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2047                         pr_err("KVM: CPU requires an RMO\n");
2048                         goto out_srcu;
2049                 }
2050
2051                 /* We can handle 4k, 64k or 16M pages in the VRMA */
2052                 err = -EINVAL;
2053                 if (!(psize == 0x1000 || psize == 0x10000 ||
2054                       psize == 0x1000000))
2055                         goto out_srcu;
2056
2057                 /* Update VRMASD field in the LPCR */
2058                 senc = slb_pgsize_encoding(psize);
2059                 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2060                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
2061                 lpcr_mask = LPCR_VRMASD;
2062                 /* the -4 is to account for senc values starting at 0x10 */
2063                 lpcr = senc << (LPCR_VRMASD_SH - 4);
2064
2065                 /* Create HPTEs in the hash page table for the VRMA */
2066                 kvmppc_map_vrma(vcpu, memslot, porder);
2067
2068         } else {
2069                 /* Set up to use an RMO region */
2070                 rma_size = kvm_rma_pages;
2071                 if (rma_size > memslot->npages)
2072                         rma_size = memslot->npages;
2073                 rma_size <<= PAGE_SHIFT;
2074                 rmls = lpcr_rmls(rma_size);
2075                 err = -EINVAL;
2076                 if ((long)rmls < 0) {
2077                         pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
2078                         goto out_srcu;
2079                 }
2080                 atomic_inc(&ri->use_count);
2081                 kvm->arch.rma = ri;
2082
2083                 /* Update LPCR and RMOR */
2084                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2085                         /* PPC970; insert RMLS value (split field) in HID4 */
2086                         lpcr_mask = (1ul << HID4_RMLS0_SH) |
2087                                 (3ul << HID4_RMLS2_SH) | HID4_RMOR;
2088                         lpcr = ((rmls >> 2) << HID4_RMLS0_SH) |
2089                                 ((rmls & 3) << HID4_RMLS2_SH);
2090                         /* RMOR is also in HID4 */
2091                         lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
2092                                 << HID4_RMOR_SH;
2093                 } else {
2094                         /* POWER7 */
2095                         lpcr_mask = LPCR_VPM0 | LPCR_VRMA_L | LPCR_RMLS;
2096                         lpcr = rmls << LPCR_RMLS_SH;
2097                         kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT;
2098                 }
2099                 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
2100                         ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
2101
2102                 /* Initialize phys addrs of pages in RMO */
2103                 npages = kvm_rma_pages;
2104                 porder = __ilog2(npages);
2105                 physp = memslot->arch.slot_phys;
2106                 if (physp) {
2107                         if (npages > memslot->npages)
2108                                 npages = memslot->npages;
2109                         spin_lock(&kvm->arch.slot_phys_lock);
2110                         for (i = 0; i < npages; ++i)
2111                                 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
2112                                         porder;
2113                         spin_unlock(&kvm->arch.slot_phys_lock);
2114                 }
2115         }
2116
2117         kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
2118
2119         /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
2120         smp_wmb();
2121         kvm->arch.rma_setup_done = 1;
2122         err = 0;
2123  out_srcu:
2124         srcu_read_unlock(&kvm->srcu, srcu_idx);
2125  out:
2126         mutex_unlock(&kvm->lock);
2127         return err;
2128
2129  up_out:
2130         up_read(&current->mm->mmap_sem);
2131         goto out_srcu;
2132 }
2133
2134 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2135 {
2136         unsigned long lpcr, lpid;
2137
2138         /* Allocate the guest's logical partition ID */
2139
2140         lpid = kvmppc_alloc_lpid();
2141         if ((long)lpid < 0)
2142                 return -ENOMEM;
2143         kvm->arch.lpid = lpid;
2144
2145         /*
2146          * Since we don't flush the TLB when tearing down a VM,
2147          * and this lpid might have previously been used,
2148          * make sure we flush on each core before running the new VM.
2149          */
2150         cpumask_setall(&kvm->arch.need_tlb_flush);
2151
2152         kvm->arch.rma = NULL;
2153
2154         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
2155
2156         if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2157                 /* PPC970; HID4 is effectively the LPCR */
2158                 kvm->arch.host_lpid = 0;
2159                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
2160                 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
2161                 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
2162                         ((lpid & 0xf) << HID4_LPID5_SH);
2163         } else {
2164                 /* POWER7; init LPCR for virtual RMA mode */
2165                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
2166                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2167                 lpcr &= LPCR_PECE | LPCR_LPES;
2168                 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
2169                         LPCR_VPM0 | LPCR_VPM1;
2170                 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2171                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
2172                 /* On POWER8 turn on online bit to enable PURR/SPURR */
2173                 if (cpu_has_feature(CPU_FTR_ARCH_207S))
2174                         lpcr |= LPCR_ONL;
2175         }
2176         kvm->arch.lpcr = lpcr;
2177
2178         kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
2179         spin_lock_init(&kvm->arch.slot_phys_lock);
2180
2181         /*
2182          * Don't allow secondary CPU threads to come online
2183          * while any KVM VMs exist.
2184          */
2185         inhibit_secondary_onlining();
2186
2187         return 0;
2188 }
2189
2190 static void kvmppc_free_vcores(struct kvm *kvm)
2191 {
2192         long int i;
2193
2194         for (i = 0; i < KVM_MAX_VCORES; ++i)
2195                 kfree(kvm->arch.vcores[i]);
2196         kvm->arch.online_vcores = 0;
2197 }
2198
2199 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
2200 {
2201         uninhibit_secondary_onlining();
2202
2203         kvmppc_free_vcores(kvm);
2204         if (kvm->arch.rma) {
2205                 kvm_release_rma(kvm->arch.rma);
2206                 kvm->arch.rma = NULL;
2207         }
2208
2209         kvmppc_free_hpt(kvm);
2210 }
2211
2212 /* We don't need to emulate any privileged instructions or dcbz */
2213 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2214                                      unsigned int inst, int *advance)
2215 {
2216         return EMULATE_FAIL;
2217 }
2218
2219 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2220                                         ulong spr_val)
2221 {
2222         return EMULATE_FAIL;
2223 }
2224
2225 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2226                                         ulong *spr_val)
2227 {
2228         return EMULATE_FAIL;
2229 }
2230
2231 static int kvmppc_core_check_processor_compat_hv(void)
2232 {
2233         if (!cpu_has_feature(CPU_FTR_HVMODE))
2234                 return -EIO;
2235         return 0;
2236 }
2237
2238 static long kvm_arch_vm_ioctl_hv(struct file *filp,
2239                                  unsigned int ioctl, unsigned long arg)
2240 {
2241         struct kvm *kvm __maybe_unused = filp->private_data;
2242         void __user *argp = (void __user *)arg;
2243         long r;
2244
2245         switch (ioctl) {
2246
2247         case KVM_ALLOCATE_RMA: {
2248                 struct kvm_allocate_rma rma;
2249                 struct kvm *kvm = filp->private_data;
2250
2251                 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
2252                 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
2253                         r = -EFAULT;
2254                 break;
2255         }
2256
2257         case KVM_PPC_ALLOCATE_HTAB: {
2258                 u32 htab_order;
2259
2260                 r = -EFAULT;
2261                 if (get_user(htab_order, (u32 __user *)argp))
2262                         break;
2263                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2264                 if (r)
2265                         break;
2266                 r = -EFAULT;
2267                 if (put_user(htab_order, (u32 __user *)argp))
2268                         break;
2269                 r = 0;
2270                 break;
2271         }
2272
2273         case KVM_PPC_GET_HTAB_FD: {
2274                 struct kvm_get_htab_fd ghf;
2275
2276                 r = -EFAULT;
2277                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
2278                         break;
2279                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2280                 break;
2281         }
2282
2283         default:
2284                 r = -ENOTTY;
2285         }
2286
2287         return r;
2288 }
2289
2290 static struct kvmppc_ops kvm_ops_hv = {
2291         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2292         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2293         .get_one_reg = kvmppc_get_one_reg_hv,
2294         .set_one_reg = kvmppc_set_one_reg_hv,
2295         .vcpu_load   = kvmppc_core_vcpu_load_hv,
2296         .vcpu_put    = kvmppc_core_vcpu_put_hv,
2297         .set_msr     = kvmppc_set_msr_hv,
2298         .vcpu_run    = kvmppc_vcpu_run_hv,
2299         .vcpu_create = kvmppc_core_vcpu_create_hv,
2300         .vcpu_free   = kvmppc_core_vcpu_free_hv,
2301         .check_requests = kvmppc_core_check_requests_hv,
2302         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
2303         .flush_memslot  = kvmppc_core_flush_memslot_hv,
2304         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2305         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
2306         .unmap_hva = kvm_unmap_hva_hv,
2307         .unmap_hva_range = kvm_unmap_hva_range_hv,
2308         .age_hva  = kvm_age_hva_hv,
2309         .test_age_hva = kvm_test_age_hva_hv,
2310         .set_spte_hva = kvm_set_spte_hva_hv,
2311         .mmu_destroy  = kvmppc_mmu_destroy_hv,
2312         .free_memslot = kvmppc_core_free_memslot_hv,
2313         .create_memslot = kvmppc_core_create_memslot_hv,
2314         .init_vm =  kvmppc_core_init_vm_hv,
2315         .destroy_vm = kvmppc_core_destroy_vm_hv,
2316         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2317         .emulate_op = kvmppc_core_emulate_op_hv,
2318         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2319         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2320         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2321         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
2322 };
2323
2324 static int kvmppc_book3s_init_hv(void)
2325 {
2326         int r;
2327         /*
2328          * FIXME!! Do we need to check on all cpus ?
2329          */
2330         r = kvmppc_core_check_processor_compat_hv();
2331         if (r < 0)
2332                 return r;
2333
2334         kvm_ops_hv.owner = THIS_MODULE;
2335         kvmppc_hv_ops = &kvm_ops_hv;
2336
2337         r = kvmppc_mmu_hv_init();
2338         return r;
2339 }
2340
2341 static void kvmppc_book3s_exit_hv(void)
2342 {
2343         kvmppc_hv_ops = NULL;
2344 }
2345
2346 module_init(kvmppc_book3s_init_hv);
2347 module_exit(kvmppc_book3s_exit_hv);
2348 MODULE_LICENSE("GPL");
2349 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2350 MODULE_ALIAS("devname:kvm");