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         case BOOK3S_INTERRUPT_H_DOORBELL:
 655                 vcpu->stat.ext_intr_exits++;
 656                 r = RESUME_GUEST;
 657                 break;
 658         case BOOK3S_INTERRUPT_PERFMON:
 659                 r = RESUME_GUEST;
 660                 break;
 661         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 662                 /*
 663                  * Deliver a machine check interrupt to the guest.
 664                  * We have to do this, even if the host has handled the
 665                  * machine check, because machine checks use SRR0/1 and
 666                  * the interrupt might have trashed guest state in them.
 667                  */
 668                 kvmppc_book3s_queue_irqprio(vcpu,
 669                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 670                 r = RESUME_GUEST;
 671                 break;
 672         case BOOK3S_INTERRUPT_PROGRAM:
 673         {
 674                 ulong flags;
 675                 /*
 676                  * Normally program interrupts are delivered directly
 677                  * to the guest by the hardware, but we can get here
 678                  * as a result of a hypervisor emulation interrupt
 679                  * (e40) getting turned into a 700 by BML RTAS.
 680                  */
 681                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 682                 kvmppc_core_queue_program(vcpu, flags);
 683                 r = RESUME_GUEST;
 684                 break;
 685         }
 686         case BOOK3S_INTERRUPT_SYSCALL:
 687         {
 688                 /* hcall - punt to userspace */
 689                 int i;
 690
 691                 /* hypercall with MSR_PR has already been handled in rmode,
 692                  * and never reaches here.
 693                  */
 694
 695                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 696                 for (i = 0; i < 9; ++i)
 697                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 698                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 699                 vcpu->arch.hcall_needed = 1;
 700                 r = RESUME_HOST;
 701                 break;
 702         }
 703         /*
 704          * We get these next two if the guest accesses a page which it thinks
 705          * it has mapped but which is not actually present, either because
 706          * it is for an emulated I/O device or because the corresonding
 707          * host page has been paged out.  Any other HDSI/HISI interrupts
 708          * have been handled already.
 709          */
 710         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 711                 r = RESUME_PAGE_FAULT;
 712                 break;
 713         case BOOK3S_INTERRUPT_H_INST_STORAGE:
 714                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
 715                 vcpu->arch.fault_dsisr = 0;
 716                 r = RESUME_PAGE_FAULT;
 717                 break;
 718         /*
 719          * This occurs if the guest executes an illegal instruction.
 720          * We just generate a program interrupt to the guest, since
 721          * we don't emulate any guest instructions at this stage.
 722          */
 723         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 724                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 725                 r = RESUME_GUEST;
 726                 break;
 727         /*
 728          * This occurs if the guest (kernel or userspace), does something that
 729          * is prohibited by HFSCR.  We just generate a program interrupt to
 730          * the guest.
 731          */
 732         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
 733                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 734                 r = RESUME_GUEST;
 735                 break;
 736         default:
 737                 kvmppc_dump_regs(vcpu);
 738                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 739                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 740                         vcpu->arch.shregs.msr);
 741                 run->hw.hardware_exit_reason = vcpu->arch.trap;
 742                 r = RESUME_HOST;
 743                 break;
 744         }
 745
 746         return r;
 747 }
 748
 749 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
 750                                             struct kvm_sregs *sregs)
 751 {
 752         int i;
 753
 754         memset(sregs, 0, sizeof(struct kvm_sregs));
 755         sregs->pvr = vcpu->arch.pvr;
 756         for (i = 0; i < vcpu->arch.slb_max; i++) {
 757                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
 758                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 759         }
 760
 761         return 0;
 762 }
 763
 764 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
 765                                             struct kvm_sregs *sregs)
 766 {
 767         int i, j;
 768
 769         kvmppc_set_pvr_hv(vcpu, sregs->pvr);
 770
 771         j = 0;
 772         for (i = 0; i < vcpu->arch.slb_nr; i++) {
 773                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
 774                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
 775                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
 776                         ++j;
 777                 }
 778         }
 779         vcpu->arch.slb_max = j;
 780
 781         return 0;
 782 }
 783
 784 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr)
 785 {
 786         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 787         u64 mask;
 788
 789         spin_lock(&vc->lock);
 790         /*
 791          * Userspace can only modify DPFD (default prefetch depth),
 792          * ILE (interrupt little-endian) and TC (translation control).
 793          * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
 794          */
 795         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
 796         if (cpu_has_feature(CPU_FTR_ARCH_207S))
 797                 mask |= LPCR_AIL;
 798         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
 799         spin_unlock(&vc->lock);
 800 }
 801
 802 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
 803                                  union kvmppc_one_reg *val)
 804 {
 805         int r = 0;
 806         long int i;
 807
 808         switch (id) {
 809         case KVM_REG_PPC_HIOR:
 810                 *val = get_reg_val(id, 0);
 811                 break;
 812         case KVM_REG_PPC_DABR:
 813                 *val = get_reg_val(id, vcpu->arch.dabr);
 814                 break;
 815         case KVM_REG_PPC_DSCR:
 816                 *val = get_reg_val(id, vcpu->arch.dscr);
 817                 break;
 818         case KVM_REG_PPC_PURR:
 819                 *val = get_reg_val(id, vcpu->arch.purr);
 820                 break;
 821         case KVM_REG_PPC_SPURR:
 822                 *val = get_reg_val(id, vcpu->arch.spurr);
 823                 break;
 824         case KVM_REG_PPC_AMR:
 825                 *val = get_reg_val(id, vcpu->arch.amr);
 826                 break;
 827         case KVM_REG_PPC_UAMOR:
 828                 *val = get_reg_val(id, vcpu->arch.uamor);
 829                 break;
 830         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
 831                 i = id - KVM_REG_PPC_MMCR0;
 832                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
 833                 break;
 834         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
 835                 i = id - KVM_REG_PPC_PMC1;
 836                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
 837                 break;
 838         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
 839                 i = id - KVM_REG_PPC_SPMC1;
 840                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
 841                 break;
 842         case KVM_REG_PPC_SIAR:
 843                 *val = get_reg_val(id, vcpu->arch.siar);
 844                 break;
 845         case KVM_REG_PPC_SDAR:
 846                 *val = get_reg_val(id, vcpu->arch.sdar);
 847                 break;
 848         case KVM_REG_PPC_SIER:
 849                 *val = get_reg_val(id, vcpu->arch.sier);
 850                 break;
 851         case KVM_REG_PPC_IAMR:
 852                 *val = get_reg_val(id, vcpu->arch.iamr);
 853                 break;
 854         case KVM_REG_PPC_TFHAR:
 855                 *val = get_reg_val(id, vcpu->arch.tfhar);
 856                 break;
 857         case KVM_REG_PPC_TFIAR:
 858                 *val = get_reg_val(id, vcpu->arch.tfiar);
 859                 break;
 860         case KVM_REG_PPC_TEXASR:
 861                 *val = get_reg_val(id, vcpu->arch.texasr);
 862                 break;
 863         case KVM_REG_PPC_FSCR:
 864                 *val = get_reg_val(id, vcpu->arch.fscr);
 865                 break;
 866         case KVM_REG_PPC_PSPB:
 867                 *val = get_reg_val(id, vcpu->arch.pspb);
 868                 break;
 869         case KVM_REG_PPC_EBBHR:
 870                 *val = get_reg_val(id, vcpu->arch.ebbhr);
 871                 break;
 872         case KVM_REG_PPC_EBBRR:
 873                 *val = get_reg_val(id, vcpu->arch.ebbrr);
 874                 break;
 875         case KVM_REG_PPC_BESCR:
 876                 *val = get_reg_val(id, vcpu->arch.bescr);
 877                 break;
 878         case KVM_REG_PPC_TAR:
 879                 *val = get_reg_val(id, vcpu->arch.tar);
 880                 break;
 881         case KVM_REG_PPC_DPDES:
 882                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
 883                 break;
 884         case KVM_REG_PPC_DAWR:
 885                 *val = get_reg_val(id, vcpu->arch.dawr);
 886                 break;
 887         case KVM_REG_PPC_DAWRX:
 888                 *val = get_reg_val(id, vcpu->arch.dawrx);
 889                 break;
 890         case KVM_REG_PPC_CIABR:
 891                 *val = get_reg_val(id, vcpu->arch.ciabr);
 892                 break;
 893         case KVM_REG_PPC_IC:
 894                 *val = get_reg_val(id, vcpu->arch.ic);
 895                 break;
 896         case KVM_REG_PPC_VTB:
 897                 *val = get_reg_val(id, vcpu->arch.vtb);
 898                 break;
 899         case KVM_REG_PPC_CSIGR:
 900                 *val = get_reg_val(id, vcpu->arch.csigr);
 901                 break;
 902         case KVM_REG_PPC_TACR:
 903                 *val = get_reg_val(id, vcpu->arch.tacr);
 904                 break;
 905         case KVM_REG_PPC_TCSCR:
 906                 *val = get_reg_val(id, vcpu->arch.tcscr);
 907                 break;
 908         case KVM_REG_PPC_PID:
 909                 *val = get_reg_val(id, vcpu->arch.pid);
 910                 break;
 911         case KVM_REG_PPC_ACOP:
 912                 *val = get_reg_val(id, vcpu->arch.acop);
 913                 break;
 914         case KVM_REG_PPC_WORT:
 915                 *val = get_reg_val(id, vcpu->arch.wort);
 916                 break;
 917         case KVM_REG_PPC_VPA_ADDR:
 918                 spin_lock(&vcpu->arch.vpa_update_lock);
 919                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
 920                 spin_unlock(&vcpu->arch.vpa_update_lock);
 921                 break;
 922         case KVM_REG_PPC_VPA_SLB:
 923                 spin_lock(&vcpu->arch.vpa_update_lock);
 924                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
 925                 val->vpaval.length = vcpu->arch.slb_shadow.len;
 926                 spin_unlock(&vcpu->arch.vpa_update_lock);
 927                 break;
 928         case KVM_REG_PPC_VPA_DTL:
 929                 spin_lock(&vcpu->arch.vpa_update_lock);
 930                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
 931                 val->vpaval.length = vcpu->arch.dtl.len;
 932                 spin_unlock(&vcpu->arch.vpa_update_lock);
 933                 break;
 934         case KVM_REG_PPC_TB_OFFSET:
 935                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
 936                 break;
 937         case KVM_REG_PPC_LPCR:
 938                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
 939                 break;
 940         case KVM_REG_PPC_PPR:
 941                 *val = get_reg_val(id, vcpu->arch.ppr);
 942                 break;
 943         case KVM_REG_PPC_ARCH_COMPAT:
 944                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
 945                 break;
 946         default:
 947                 r = -EINVAL;
 948                 break;
 949         }
 950
 951         return r;
 952 }
 953
 954 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
 955                                  union kvmppc_one_reg *val)
 956 {
 957         int r = 0;
 958         long int i;
 959         unsigned long addr, len;
 960
 961         switch (id) {
 962         case KVM_REG_PPC_HIOR:
 963                 /* Only allow this to be set to zero */
 964                 if (set_reg_val(id, *val))
 965                         r = -EINVAL;
 966                 break;
 967         case KVM_REG_PPC_DABR:
 968                 vcpu->arch.dabr = set_reg_val(id, *val);
 969                 break;
 970         case KVM_REG_PPC_DSCR:
 971                 vcpu->arch.dscr = set_reg_val(id, *val);
 972                 break;
 973         case KVM_REG_PPC_PURR:
 974                 vcpu->arch.purr = set_reg_val(id, *val);
 975                 break;
 976         case KVM_REG_PPC_SPURR:
 977                 vcpu->arch.spurr = set_reg_val(id, *val);
 978                 break;
 979         case KVM_REG_PPC_AMR:
 980                 vcpu->arch.amr = set_reg_val(id, *val);
 981                 break;
 982         case KVM_REG_PPC_UAMOR:
 983                 vcpu->arch.uamor = set_reg_val(id, *val);
 984                 break;
 985         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
 986                 i = id - KVM_REG_PPC_MMCR0;
 987                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
 988                 break;
 989         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
 990                 i = id - KVM_REG_PPC_PMC1;
 991                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
 992                 break;
 993         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
 994                 i = id - KVM_REG_PPC_SPMC1;
 995                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
 996                 break;
 997         case KVM_REG_PPC_SIAR:
 998                 vcpu->arch.siar = set_reg_val(id, *val);
 999                 break;
1000         case KVM_REG_PPC_SDAR:
1001                 vcpu->arch.sdar = set_reg_val(id, *val);
1002                 break;
1003         case KVM_REG_PPC_SIER:
1004                 vcpu->arch.sier = set_reg_val(id, *val);
1005                 break;
1006         case KVM_REG_PPC_IAMR:
1007                 vcpu->arch.iamr = set_reg_val(id, *val);
1008                 break;
1009         case KVM_REG_PPC_TFHAR:
1010                 vcpu->arch.tfhar = set_reg_val(id, *val);
1011                 break;
1012         case KVM_REG_PPC_TFIAR:
1013                 vcpu->arch.tfiar = set_reg_val(id, *val);
1014                 break;
1015         case KVM_REG_PPC_TEXASR:
1016                 vcpu->arch.texasr = set_reg_val(id, *val);
1017                 break;
1018         case KVM_REG_PPC_FSCR:
1019                 vcpu->arch.fscr = set_reg_val(id, *val);
1020                 break;
1021         case KVM_REG_PPC_PSPB:
1022                 vcpu->arch.pspb = set_reg_val(id, *val);
1023                 break;
1024         case KVM_REG_PPC_EBBHR:
1025                 vcpu->arch.ebbhr = set_reg_val(id, *val);
1026                 break;
1027         case KVM_REG_PPC_EBBRR:
1028                 vcpu->arch.ebbrr = set_reg_val(id, *val);
1029                 break;
1030         case KVM_REG_PPC_BESCR:
1031                 vcpu->arch.bescr = set_reg_val(id, *val);
1032                 break;
1033         case KVM_REG_PPC_TAR:
1034                 vcpu->arch.tar = set_reg_val(id, *val);
1035                 break;
1036         case KVM_REG_PPC_DPDES:
1037                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1038                 break;
1039         case KVM_REG_PPC_DAWR:
1040                 vcpu->arch.dawr = set_reg_val(id, *val);
1041                 break;
1042         case KVM_REG_PPC_DAWRX:
1043                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1044                 break;
1045         case KVM_REG_PPC_CIABR:
1046                 vcpu->arch.ciabr = set_reg_val(id, *val);
1047                 /* Don't allow setting breakpoints in hypervisor code */
1048                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1049                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1050                 break;
1051         case KVM_REG_PPC_IC:
1052                 vcpu->arch.ic = set_reg_val(id, *val);
1053                 break;
1054         case KVM_REG_PPC_VTB:
1055                 vcpu->arch.vtb = set_reg_val(id, *val);
1056                 break;
1057         case KVM_REG_PPC_CSIGR:
1058                 vcpu->arch.csigr = set_reg_val(id, *val);
1059                 break;
1060         case KVM_REG_PPC_TACR:
1061                 vcpu->arch.tacr = set_reg_val(id, *val);
1062                 break;
1063         case KVM_REG_PPC_TCSCR:
1064                 vcpu->arch.tcscr = set_reg_val(id, *val);
1065                 break;
1066         case KVM_REG_PPC_PID:
1067                 vcpu->arch.pid = set_reg_val(id, *val);
1068                 break;
1069         case KVM_REG_PPC_ACOP:
1070                 vcpu->arch.acop = set_reg_val(id, *val);
1071                 break;
1072         case KVM_REG_PPC_WORT:
1073                 vcpu->arch.wort = set_reg_val(id, *val);
1074                 break;
1075         case KVM_REG_PPC_VPA_ADDR:
1076                 addr = set_reg_val(id, *val);
1077                 r = -EINVAL;
1078                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1079                               vcpu->arch.dtl.next_gpa))
1080                         break;
1081                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1082                 break;
1083         case KVM_REG_PPC_VPA_SLB:
1084                 addr = val->vpaval.addr;
1085                 len = val->vpaval.length;
1086                 r = -EINVAL;
1087                 if (addr && !vcpu->arch.vpa.next_gpa)
1088                         break;
1089                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1090                 break;
1091         case KVM_REG_PPC_VPA_DTL:
1092                 addr = val->vpaval.addr;
1093                 len = val->vpaval.length;
1094                 r = -EINVAL;
1095                 if (addr && (len < sizeof(struct dtl_entry) ||
1096                              !vcpu->arch.vpa.next_gpa))
1097                         break;
1098                 len -= len % sizeof(struct dtl_entry);
1099                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1100                 break;
1101         case KVM_REG_PPC_TB_OFFSET:
1102                 /* round up to multiple of 2^24 */
1103                 vcpu->arch.vcore->tb_offset =
1104                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1105                 break;
1106         case KVM_REG_PPC_LPCR:
1107                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val));
1108                 break;
1109         case KVM_REG_PPC_PPR:
1110                 vcpu->arch.ppr = set_reg_val(id, *val);
1111                 break;
1112         case KVM_REG_PPC_ARCH_COMPAT:
1113                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1114                 break;
1115         default:
1116                 r = -EINVAL;
1117                 break;
1118         }
1119
1120         return r;
1121 }
1122
1123 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1124                                                    unsigned int id)
1125 {
1126         struct kvm_vcpu *vcpu;
1127         int err = -EINVAL;
1128         int core;
1129         struct kvmppc_vcore *vcore;
1130
1131         core = id / threads_per_core;
1132         if (core >= KVM_MAX_VCORES)
1133                 goto out;
1134
1135         err = -ENOMEM;
1136         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1137         if (!vcpu)
1138                 goto out;
1139
1140         err = kvm_vcpu_init(vcpu, kvm, id);
1141         if (err)
1142                 goto free_vcpu;
1143
1144         vcpu->arch.shared = &vcpu->arch.shregs;
1145         vcpu->arch.mmcr[0] = MMCR0_FC;
1146         vcpu->arch.ctrl = CTRL_RUNLATCH;
1147         /* default to host PVR, since we can't spoof it */
1148         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1149         spin_lock_init(&vcpu->arch.vpa_update_lock);
1150         spin_lock_init(&vcpu->arch.tbacct_lock);
1151         vcpu->arch.busy_preempt = TB_NIL;
1152
1153         kvmppc_mmu_book3s_hv_init(vcpu);
1154
1155         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1156
1157         init_waitqueue_head(&vcpu->arch.cpu_run);
1158
1159         mutex_lock(&kvm->lock);
1160         vcore = kvm->arch.vcores[core];
1161         if (!vcore) {
1162                 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1163                 if (vcore) {
1164                         INIT_LIST_HEAD(&vcore->runnable_threads);
1165                         spin_lock_init(&vcore->lock);
1166                         init_waitqueue_head(&vcore->wq);
1167                         vcore->preempt_tb = TB_NIL;
1168                         vcore->lpcr = kvm->arch.lpcr;
1169                         vcore->first_vcpuid = core * threads_per_core;
1170                         vcore->kvm = kvm;
1171                 }
1172                 kvm->arch.vcores[core] = vcore;
1173                 kvm->arch.online_vcores++;
1174         }
1175         mutex_unlock(&kvm->lock);
1176
1177         if (!vcore)
1178                 goto free_vcpu;
1179
1180         spin_lock(&vcore->lock);
1181         ++vcore->num_threads;
1182         spin_unlock(&vcore->lock);
1183         vcpu->arch.vcore = vcore;
1184         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1185
1186         vcpu->arch.cpu_type = KVM_CPU_3S_64;
1187         kvmppc_sanity_check(vcpu);
1188
1189         return vcpu;
1190
1191 free_vcpu:
1192         kmem_cache_free(kvm_vcpu_cache, vcpu);
1193 out:
1194         return ERR_PTR(err);
1195 }
1196
1197 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1198 {
1199         if (vpa->pinned_addr)
1200                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1201                                         vpa->dirty);
1202 }
1203
1204 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1205 {
1206         spin_lock(&vcpu->arch.vpa_update_lock);
1207         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1208         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1209         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1210         spin_unlock(&vcpu->arch.vpa_update_lock);
1211         kvm_vcpu_uninit(vcpu);
1212         kmem_cache_free(kvm_vcpu_cache, vcpu);
1213 }
1214
1215 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1216 {
1217         /* Indicate we want to get back into the guest */
1218         return 1;
1219 }
1220
1221 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1222 {
1223         unsigned long dec_nsec, now;
1224
1225         now = get_tb();
1226         if (now > vcpu->arch.dec_expires) {
1227                 /* decrementer has already gone negative */
1228                 kvmppc_core_queue_dec(vcpu);
1229                 kvmppc_core_prepare_to_enter(vcpu);
1230                 return;
1231         }
1232         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1233                    / tb_ticks_per_sec;
1234         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1235                       HRTIMER_MODE_REL);
1236         vcpu->arch.timer_running = 1;
1237 }
1238
1239 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1240 {
1241         vcpu->arch.ceded = 0;
1242         if (vcpu->arch.timer_running) {
1243                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1244                 vcpu->arch.timer_running = 0;
1245         }
1246 }
1247
1248 extern void __kvmppc_vcore_entry(void);
1249
1250 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1251                                    struct kvm_vcpu *vcpu)
1252 {
1253         u64 now;
1254
1255         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1256                 return;
1257         spin_lock(&vcpu->arch.tbacct_lock);
1258         now = mftb();
1259         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1260                 vcpu->arch.stolen_logged;
1261         vcpu->arch.busy_preempt = now;
1262         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1263         spin_unlock(&vcpu->arch.tbacct_lock);
1264         --vc->n_runnable;
1265         list_del(&vcpu->arch.run_list);
1266 }
1267
1268 static int kvmppc_grab_hwthread(int cpu)
1269 {
1270         struct paca_struct *tpaca;
1271         long timeout = 1000;
1272
1273         tpaca = &paca[cpu];
1274
1275         /* Ensure the thread won't go into the kernel if it wakes */
1276         tpaca->kvm_hstate.hwthread_req = 1;
1277         tpaca->kvm_hstate.kvm_vcpu = NULL;
1278
1279         /*
1280          * If the thread is already executing in the kernel (e.g. handling
1281          * a stray interrupt), wait for it to get back to nap mode.
1282          * The smp_mb() is to ensure that our setting of hwthread_req
1283          * is visible before we look at hwthread_state, so if this
1284          * races with the code at system_reset_pSeries and the thread
1285          * misses our setting of hwthread_req, we are sure to see its
1286          * setting of hwthread_state, and vice versa.
1287          */
1288         smp_mb();
1289         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1290                 if (--timeout <= 0) {
1291                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1292                         return -EBUSY;
1293                 }
1294                 udelay(1);
1295         }
1296         return 0;
1297 }
1298
1299 static void kvmppc_release_hwthread(int cpu)
1300 {
1301         struct paca_struct *tpaca;
1302
1303         tpaca = &paca[cpu];
1304         tpaca->kvm_hstate.hwthread_req = 0;
1305         tpaca->kvm_hstate.kvm_vcpu = NULL;
1306 }
1307
1308 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1309 {
1310         int cpu;
1311         struct paca_struct *tpaca;
1312         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1313
1314         if (vcpu->arch.timer_running) {
1315                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1316                 vcpu->arch.timer_running = 0;
1317         }
1318         cpu = vc->pcpu + vcpu->arch.ptid;
1319         tpaca = &paca[cpu];
1320         tpaca->kvm_hstate.kvm_vcpu = vcpu;
1321         tpaca->kvm_hstate.kvm_vcore = vc;
1322         tpaca->kvm_hstate.ptid = vcpu->arch.ptid;
1323         vcpu->cpu = vc->pcpu;
1324         smp_wmb();
1325 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1326         if (cpu != smp_processor_id()) {
1327 #ifdef CONFIG_KVM_XICS
1328                 xics_wake_cpu(cpu);
1329 #endif
1330                 if (vcpu->arch.ptid)
1331                         ++vc->n_woken;
1332         }
1333 #endif
1334 }
1335
1336 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1337 {
1338         int i;
1339
1340         HMT_low();
1341         i = 0;
1342         while (vc->nap_count < vc->n_woken) {
1343                 if (++i >= 1000000) {
1344                         pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1345                                vc->nap_count, vc->n_woken);
1346                         break;
1347                 }
1348                 cpu_relax();
1349         }
1350         HMT_medium();
1351 }
1352
1353 /*
1354  * Check that we are on thread 0 and that any other threads in
1355  * this core are off-line.  Then grab the threads so they can't
1356  * enter the kernel.
1357  */
1358 static int on_primary_thread(void)
1359 {
1360         int cpu = smp_processor_id();
1361         int thr = cpu_thread_in_core(cpu);
1362
1363         if (thr)
1364                 return 0;
1365         while (++thr < threads_per_core)
1366                 if (cpu_online(cpu + thr))
1367                         return 0;
1368
1369         /* Grab all hw threads so they can't go into the kernel */
1370         for (thr = 1; thr < threads_per_core; ++thr) {
1371                 if (kvmppc_grab_hwthread(cpu + thr)) {
1372                         /* Couldn't grab one; let the others go */
1373                         do {
1374                                 kvmppc_release_hwthread(cpu + thr);
1375                         } while (--thr > 0);
1376                         return 0;
1377                 }
1378         }
1379         return 1;
1380 }
1381
1382 /*
1383  * Run a set of guest threads on a physical core.
1384  * Called with vc->lock held.
1385  */
1386 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1387 {
1388         struct kvm_vcpu *vcpu, *vnext;
1389         long ret;
1390         u64 now;
1391         int i, need_vpa_update;
1392         int srcu_idx;
1393         struct kvm_vcpu *vcpus_to_update[threads_per_core];
1394
1395         /* don't start if any threads have a signal pending */
1396         need_vpa_update = 0;
1397         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1398                 if (signal_pending(vcpu->arch.run_task))
1399                         return;
1400                 if (vcpu->arch.vpa.update_pending ||
1401                     vcpu->arch.slb_shadow.update_pending ||
1402                     vcpu->arch.dtl.update_pending)
1403                         vcpus_to_update[need_vpa_update++] = vcpu;
1404         }
1405
1406         /*
1407          * Initialize *vc, in particular vc->vcore_state, so we can
1408          * drop the vcore lock if necessary.
1409          */
1410         vc->n_woken = 0;
1411         vc->nap_count = 0;
1412         vc->entry_exit_count = 0;
1413         vc->vcore_state = VCORE_STARTING;
1414         vc->in_guest = 0;
1415         vc->napping_threads = 0;
1416
1417         /*
1418          * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1419          * which can't be called with any spinlocks held.
1420          */
1421         if (need_vpa_update) {
1422                 spin_unlock(&vc->lock);
1423                 for (i = 0; i < need_vpa_update; ++i)
1424                         kvmppc_update_vpas(vcpus_to_update[i]);
1425                 spin_lock(&vc->lock);
1426         }
1427
1428         /*
1429          * Make sure we are running on thread 0, and that
1430          * secondary threads are offline.
1431          */
1432         if (threads_per_core > 1 && !on_primary_thread()) {
1433                 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1434                         vcpu->arch.ret = -EBUSY;
1435                 goto out;
1436         }
1437
1438         vc->pcpu = smp_processor_id();
1439         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1440                 kvmppc_start_thread(vcpu);
1441                 kvmppc_create_dtl_entry(vcpu, vc);
1442         }
1443
1444         /* Set this explicitly in case thread 0 doesn't have a vcpu */
1445         get_paca()->kvm_hstate.kvm_vcore = vc;
1446         get_paca()->kvm_hstate.ptid = 0;
1447
1448         vc->vcore_state = VCORE_RUNNING;
1449         preempt_disable();
1450         spin_unlock(&vc->lock);
1451
1452         kvm_guest_enter();
1453
1454         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
1455
1456         __kvmppc_vcore_entry();
1457
1458         spin_lock(&vc->lock);
1459         /* disable sending of IPIs on virtual external irqs */
1460         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1461                 vcpu->cpu = -1;
1462         /* wait for secondary threads to finish writing their state to memory */
1463         if (vc->nap_count < vc->n_woken)
1464                 kvmppc_wait_for_nap(vc);
1465         for (i = 0; i < threads_per_core; ++i)
1466                 kvmppc_release_hwthread(vc->pcpu + i);
1467         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
1468         vc->vcore_state = VCORE_EXITING;
1469         spin_unlock(&vc->lock);
1470
1471         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
1472
1473         /* make sure updates to secondary vcpu structs are visible now */
1474         smp_mb();
1475         kvm_guest_exit();
1476
1477         preempt_enable();
1478         kvm_resched(vcpu);
1479
1480         spin_lock(&vc->lock);
1481         now = get_tb();
1482         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1483                 /* cancel pending dec exception if dec is positive */
1484                 if (now < vcpu->arch.dec_expires &&
1485                     kvmppc_core_pending_dec(vcpu))
1486                         kvmppc_core_dequeue_dec(vcpu);
1487
1488                 ret = RESUME_GUEST;
1489                 if (vcpu->arch.trap)
1490                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1491                                                     vcpu->arch.run_task);
1492
1493                 vcpu->arch.ret = ret;
1494                 vcpu->arch.trap = 0;
1495
1496                 if (vcpu->arch.ceded) {
1497                         if (ret != RESUME_GUEST)
1498                                 kvmppc_end_cede(vcpu);
1499                         else
1500                                 kvmppc_set_timer(vcpu);
1501                 }
1502         }
1503
1504  out:
1505         vc->vcore_state = VCORE_INACTIVE;
1506         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1507                                  arch.run_list) {
1508                 if (vcpu->arch.ret != RESUME_GUEST) {
1509                         kvmppc_remove_runnable(vc, vcpu);
1510                         wake_up(&vcpu->arch.cpu_run);
1511                 }
1512         }
1513 }
1514
1515 /*
1516  * Wait for some other vcpu thread to execute us, and
1517  * wake us up when we need to handle something in the host.
1518  */
1519 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1520 {
1521         DEFINE_WAIT(wait);
1522
1523         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1524         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1525                 schedule();
1526         finish_wait(&vcpu->arch.cpu_run, &wait);
1527 }
1528
1529 /*
1530  * All the vcpus in this vcore are idle, so wait for a decrementer
1531  * or external interrupt to one of the vcpus.  vc->lock is held.
1532  */
1533 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1534 {
1535         DEFINE_WAIT(wait);
1536
1537         prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1538         vc->vcore_state = VCORE_SLEEPING;
1539         spin_unlock(&vc->lock);
1540         schedule();
1541         finish_wait(&vc->wq, &wait);
1542         spin_lock(&vc->lock);
1543         vc->vcore_state = VCORE_INACTIVE;
1544 }
1545
1546 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1547 {
1548         int n_ceded;
1549         struct kvmppc_vcore *vc;
1550         struct kvm_vcpu *v, *vn;
1551
1552         kvm_run->exit_reason = 0;
1553         vcpu->arch.ret = RESUME_GUEST;
1554         vcpu->arch.trap = 0;
1555         kvmppc_update_vpas(vcpu);
1556
1557         /*
1558          * Synchronize with other threads in this virtual core
1559          */
1560         vc = vcpu->arch.vcore;
1561         spin_lock(&vc->lock);
1562         vcpu->arch.ceded = 0;
1563         vcpu->arch.run_task = current;
1564         vcpu->arch.kvm_run = kvm_run;
1565         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
1566         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1567         vcpu->arch.busy_preempt = TB_NIL;
1568         list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1569         ++vc->n_runnable;
1570
1571         /*
1572          * This happens the first time this is called for a vcpu.
1573          * If the vcore is already running, we may be able to start
1574          * this thread straight away and have it join in.
1575          */
1576         if (!signal_pending(current)) {
1577                 if (vc->vcore_state == VCORE_RUNNING &&
1578                     VCORE_EXIT_COUNT(vc) == 0) {
1579                         kvmppc_create_dtl_entry(vcpu, vc);
1580                         kvmppc_start_thread(vcpu);
1581                 } else if (vc->vcore_state == VCORE_SLEEPING) {
1582                         wake_up(&vc->wq);
1583                 }
1584
1585         }
1586
1587         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1588                !signal_pending(current)) {
1589                 if (vc->vcore_state != VCORE_INACTIVE) {
1590                         spin_unlock(&vc->lock);
1591                         kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1592                         spin_lock(&vc->lock);
1593                         continue;
1594                 }
1595                 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1596                                          arch.run_list) {
1597                         kvmppc_core_prepare_to_enter(v);
1598                         if (signal_pending(v->arch.run_task)) {
1599                                 kvmppc_remove_runnable(vc, v);
1600                                 v->stat.signal_exits++;
1601                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1602                                 v->arch.ret = -EINTR;
1603                                 wake_up(&v->arch.cpu_run);
1604                         }
1605                 }
1606                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1607                         break;
1608                 vc->runner = vcpu;
1609                 n_ceded = 0;
1610                 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
1611                         if (!v->arch.pending_exceptions)
1612                                 n_ceded += v->arch.ceded;
1613                         else
1614                                 v->arch.ceded = 0;
1615                 }
1616                 if (n_ceded == vc->n_runnable)
1617                         kvmppc_vcore_blocked(vc);
1618                 else
1619                         kvmppc_run_core(vc);
1620                 vc->runner = NULL;
1621         }
1622
1623         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1624                (vc->vcore_state == VCORE_RUNNING ||
1625                 vc->vcore_state == VCORE_EXITING)) {
1626                 spin_unlock(&vc->lock);
1627                 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1628                 spin_lock(&vc->lock);
1629         }
1630
1631         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1632                 kvmppc_remove_runnable(vc, vcpu);
1633                 vcpu->stat.signal_exits++;
1634                 kvm_run->exit_reason = KVM_EXIT_INTR;
1635                 vcpu->arch.ret = -EINTR;
1636         }
1637
1638         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1639                 /* Wake up some vcpu to run the core */
1640                 v = list_first_entry(&vc->runnable_threads,
1641                                      struct kvm_vcpu, arch.run_list);
1642                 wake_up(&v->arch.cpu_run);
1643         }
1644
1645         spin_unlock(&vc->lock);
1646         return vcpu->arch.ret;
1647 }
1648
1649 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
1650 {
1651         int r;
1652         int srcu_idx;
1653
1654         if (!vcpu->arch.sane) {
1655                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1656                 return -EINVAL;
1657         }
1658
1659         kvmppc_core_prepare_to_enter(vcpu);
1660
1661         /* No need to go into the guest when all we'll do is come back out */
1662         if (signal_pending(current)) {
1663                 run->exit_reason = KVM_EXIT_INTR;
1664                 return -EINTR;
1665         }
1666
1667         atomic_inc(&vcpu->kvm->arch.vcpus_running);
1668         /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1669         smp_mb();
1670
1671         /* On the first time here, set up HTAB and VRMA or RMA */
1672         if (!vcpu->kvm->arch.rma_setup_done) {
1673                 r = kvmppc_hv_setup_htab_rma(vcpu);
1674                 if (r)
1675                         goto out;
1676         }
1677
1678         flush_fp_to_thread(current);
1679         flush_altivec_to_thread(current);
1680         flush_vsx_to_thread(current);
1681         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1682         vcpu->arch.pgdir = current->mm->pgd;
1683         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1684
1685         do {
1686                 r = kvmppc_run_vcpu(run, vcpu);
1687
1688                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1689                     !(vcpu->arch.shregs.msr & MSR_PR)) {
1690                         r = kvmppc_pseries_do_hcall(vcpu);
1691                         kvmppc_core_prepare_to_enter(vcpu);
1692                 } else if (r == RESUME_PAGE_FAULT) {
1693                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1694                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
1695                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1696                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1697                 }
1698         } while (r == RESUME_GUEST);
1699
1700  out:
1701         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1702         atomic_dec(&vcpu->kvm->arch.vcpus_running);
1703         return r;
1704 }
1705
1706
1707 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1708    Assumes POWER7 or PPC970. */
1709 static inline int lpcr_rmls(unsigned long rma_size)
1710 {
1711         switch (rma_size) {
1712         case 32ul << 20:        /* 32 MB */
1713                 if (cpu_has_feature(CPU_FTR_ARCH_206))
1714                         return 8;       /* only supported on POWER7 */
1715                 return -1;
1716         case 64ul << 20:        /* 64 MB */
1717                 return 3;
1718         case 128ul << 20:       /* 128 MB */
1719                 return 7;
1720         case 256ul << 20:       /* 256 MB */
1721                 return 4;
1722         case 1ul << 30:         /* 1 GB */
1723                 return 2;
1724         case 16ul << 30:        /* 16 GB */
1725                 return 1;
1726         case 256ul << 30:       /* 256 GB */
1727                 return 0;
1728         default:
1729                 return -1;
1730         }
1731 }
1732
1733 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1734 {
1735         struct page *page;
1736         struct kvm_rma_info *ri = vma->vm_file->private_data;
1737
1738         if (vmf->pgoff >= kvm_rma_pages)
1739                 return VM_FAULT_SIGBUS;
1740
1741         page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1742         get_page(page);
1743         vmf->page = page;
1744         return 0;
1745 }
1746
1747 static const struct vm_operations_struct kvm_rma_vm_ops = {
1748         .fault = kvm_rma_fault,
1749 };
1750
1751 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1752 {
1753         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1754         vma->vm_ops = &kvm_rma_vm_ops;
1755         return 0;
1756 }
1757
1758 static int kvm_rma_release(struct inode *inode, struct file *filp)
1759 {
1760         struct kvm_rma_info *ri = filp->private_data;
1761
1762         kvm_release_rma(ri);
1763         return 0;
1764 }
1765
1766 static const struct file_operations kvm_rma_fops = {
1767         .mmap           = kvm_rma_mmap,
1768         .release        = kvm_rma_release,
1769 };
1770
1771 static long kvm_vm_ioctl_allocate_rma(struct kvm *kvm,
1772                                       struct kvm_allocate_rma *ret)
1773 {
1774         long fd;
1775         struct kvm_rma_info *ri;
1776         /*
1777          * Only do this on PPC970 in HV mode
1778          */
1779         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
1780             !cpu_has_feature(CPU_FTR_ARCH_201))
1781                 return -EINVAL;
1782
1783         if (!kvm_rma_pages)
1784                 return -EINVAL;
1785
1786         ri = kvm_alloc_rma();
1787         if (!ri)
1788                 return -ENOMEM;
1789
1790         fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC);
1791         if (fd < 0)
1792                 kvm_release_rma(ri);
1793
1794         ret->rma_size = kvm_rma_pages << PAGE_SHIFT;
1795         return fd;
1796 }
1797
1798 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1799                                      int linux_psize)
1800 {
1801         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1802
1803         if (!def->shift)
1804                 return;
1805         (*sps)->page_shift = def->shift;
1806         (*sps)->slb_enc = def->sllp;
1807         (*sps)->enc[0].page_shift = def->shift;
1808         /*
1809          * Only return base page encoding. We don't want to return
1810          * all the supporting pte_enc, because our H_ENTER doesn't
1811          * support MPSS yet. Once they do, we can start passing all
1812          * support pte_enc here
1813          */
1814         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
1815         (*sps)++;
1816 }
1817
1818 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
1819                                          struct kvm_ppc_smmu_info *info)
1820 {
1821         struct kvm_ppc_one_seg_page_size *sps;
1822
1823         info->flags = KVM_PPC_PAGE_SIZES_REAL;
1824         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1825                 info->flags |= KVM_PPC_1T_SEGMENTS;
1826         info->slb_size = mmu_slb_size;
1827
1828         /* We only support these sizes for now, and no muti-size segments */
1829         sps = &info->sps[0];
1830         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1831         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1832         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1833
1834         return 0;
1835 }
1836
1837 /*
1838  * Get (and clear) the dirty memory log for a memory slot.
1839  */
1840 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
1841                                          struct kvm_dirty_log *log)
1842 {
1843         struct kvm_memory_slot *memslot;
1844         int r;
1845         unsigned long n;
1846
1847         mutex_lock(&kvm->slots_lock);
1848
1849         r = -EINVAL;
1850         if (log->slot >= KVM_USER_MEM_SLOTS)
1851                 goto out;
1852
1853         memslot = id_to_memslot(kvm->memslots, log->slot);
1854         r = -ENOENT;
1855         if (!memslot->dirty_bitmap)
1856                 goto out;
1857
1858         n = kvm_dirty_bitmap_bytes(memslot);
1859         memset(memslot->dirty_bitmap, 0, n);
1860
1861         r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
1862         if (r)
1863                 goto out;
1864
1865         r = -EFAULT;
1866         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1867                 goto out;
1868
1869         r = 0;
1870 out:
1871         mutex_unlock(&kvm->slots_lock);
1872         return r;
1873 }
1874
1875 static void unpin_slot(struct kvm_memory_slot *memslot)
1876 {
1877         unsigned long *physp;
1878         unsigned long j, npages, pfn;
1879         struct page *page;
1880
1881         physp = memslot->arch.slot_phys;
1882         npages = memslot->npages;
1883         if (!physp)
1884                 return;
1885         for (j = 0; j < npages; j++) {
1886                 if (!(physp[j] & KVMPPC_GOT_PAGE))
1887                         continue;
1888                 pfn = physp[j] >> PAGE_SHIFT;
1889                 page = pfn_to_page(pfn);
1890                 SetPageDirty(page);
1891                 put_page(page);
1892         }
1893 }
1894
1895 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
1896                                         struct kvm_memory_slot *dont)
1897 {
1898         if (!dont || free->arch.rmap != dont->arch.rmap) {
1899                 vfree(free->arch.rmap);
1900                 free->arch.rmap = NULL;
1901         }
1902         if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
1903                 unpin_slot(free);
1904                 vfree(free->arch.slot_phys);
1905                 free->arch.slot_phys = NULL;
1906         }
1907 }
1908
1909 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
1910                                          unsigned long npages)
1911 {
1912         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
1913         if (!slot->arch.rmap)
1914                 return -ENOMEM;
1915         slot->arch.slot_phys = NULL;
1916
1917         return 0;
1918 }
1919
1920 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
1921                                         struct kvm_memory_slot *memslot,
1922                                         struct kvm_userspace_memory_region *mem)
1923 {
1924         unsigned long *phys;
1925
1926         /* Allocate a slot_phys array if needed */
1927         phys = memslot->arch.slot_phys;
1928         if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
1929                 phys = vzalloc(memslot->npages * sizeof(unsigned long));
1930                 if (!phys)
1931                         return -ENOMEM;
1932                 memslot->arch.slot_phys = phys;
1933         }
1934
1935         return 0;
1936 }
1937
1938 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
1939                                 struct kvm_userspace_memory_region *mem,
1940                                 const struct kvm_memory_slot *old)
1941 {
1942         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
1943         struct kvm_memory_slot *memslot;
1944
1945         if (npages && old->npages) {
1946                 /*
1947                  * If modifying a memslot, reset all the rmap dirty bits.
1948                  * If this is a new memslot, we don't need to do anything
1949                  * since the rmap array starts out as all zeroes,
1950                  * i.e. no pages are dirty.
1951                  */
1952                 memslot = id_to_memslot(kvm->memslots, mem->slot);
1953                 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
1954         }
1955 }
1956
1957 /*
1958  * Update LPCR values in kvm->arch and in vcores.
1959  * Caller must hold kvm->lock.
1960  */
1961 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
1962 {
1963         long int i;
1964         u32 cores_done = 0;
1965
1966         if ((kvm->arch.lpcr & mask) == lpcr)
1967                 return;
1968
1969         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
1970
1971         for (i = 0; i < KVM_MAX_VCORES; ++i) {
1972                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
1973                 if (!vc)
1974                         continue;
1975                 spin_lock(&vc->lock);
1976                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
1977                 spin_unlock(&vc->lock);
1978                 if (++cores_done >= kvm->arch.online_vcores)
1979                         break;
1980         }
1981 }
1982
1983 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
1984 {
1985         return;
1986 }
1987
1988 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
1989 {
1990         int err = 0;
1991         struct kvm *kvm = vcpu->kvm;
1992         struct kvm_rma_info *ri = NULL;
1993         unsigned long hva;
1994         struct kvm_memory_slot *memslot;
1995         struct vm_area_struct *vma;
1996         unsigned long lpcr = 0, senc;
1997         unsigned long lpcr_mask = 0;
1998         unsigned long psize, porder;
1999         unsigned long rma_size;
2000         unsigned long rmls;
2001         unsigned long *physp;
2002         unsigned long i, npages;
2003         int srcu_idx;
2004
2005         mutex_lock(&kvm->lock);
2006         if (kvm->arch.rma_setup_done)
2007                 goto out;       /* another vcpu beat us to it */
2008
2009         /* Allocate hashed page table (if not done already) and reset it */
2010         if (!kvm->arch.hpt_virt) {
2011                 err = kvmppc_alloc_hpt(kvm, NULL);
2012                 if (err) {
2013                         pr_err("KVM: Couldn't alloc HPT\n");
2014                         goto out;
2015                 }
2016         }
2017
2018         /* Look up the memslot for guest physical address 0 */
2019         srcu_idx = srcu_read_lock(&kvm->srcu);
2020         memslot = gfn_to_memslot(kvm, 0);
2021
2022         /* We must have some memory at 0 by now */
2023         err = -EINVAL;
2024         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2025                 goto out_srcu;
2026
2027         /* Look up the VMA for the start of this memory slot */
2028         hva = memslot->userspace_addr;
2029         down_read(&current->mm->mmap_sem);
2030         vma = find_vma(current->mm, hva);
2031         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2032                 goto up_out;
2033
2034         psize = vma_kernel_pagesize(vma);
2035         porder = __ilog2(psize);
2036
2037         /* Is this one of our preallocated RMAs? */
2038         if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
2039             hva == vma->vm_start)
2040                 ri = vma->vm_file->private_data;
2041
2042         up_read(&current->mm->mmap_sem);
2043
2044         if (!ri) {
2045                 /* On POWER7, use VRMA; on PPC970, give up */
2046                 err = -EPERM;
2047                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2048                         pr_err("KVM: CPU requires an RMO\n");
2049                         goto out_srcu;
2050                 }
2051
2052                 /* We can handle 4k, 64k or 16M pages in the VRMA */
2053                 err = -EINVAL;
2054                 if (!(psize == 0x1000 || psize == 0x10000 ||
2055                       psize == 0x1000000))
2056                         goto out_srcu;
2057
2058                 /* Update VRMASD field in the LPCR */
2059                 senc = slb_pgsize_encoding(psize);
2060                 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2061                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
2062                 lpcr_mask = LPCR_VRMASD;
2063                 /* the -4 is to account for senc values starting at 0x10 */
2064                 lpcr = senc << (LPCR_VRMASD_SH - 4);
2065
2066                 /* Create HPTEs in the hash page table for the VRMA */
2067                 kvmppc_map_vrma(vcpu, memslot, porder);
2068
2069         } else {
2070                 /* Set up to use an RMO region */
2071                 rma_size = kvm_rma_pages;
2072                 if (rma_size > memslot->npages)
2073                         rma_size = memslot->npages;
2074                 rma_size <<= PAGE_SHIFT;
2075                 rmls = lpcr_rmls(rma_size);
2076                 err = -EINVAL;
2077                 if ((long)rmls < 0) {
2078                         pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
2079                         goto out_srcu;
2080                 }
2081                 atomic_inc(&ri->use_count);
2082                 kvm->arch.rma = ri;
2083
2084                 /* Update LPCR and RMOR */
2085                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2086                         /* PPC970; insert RMLS value (split field) in HID4 */
2087                         lpcr_mask = (1ul << HID4_RMLS0_SH) |
2088                                 (3ul << HID4_RMLS2_SH) | HID4_RMOR;
2089                         lpcr = ((rmls >> 2) << HID4_RMLS0_SH) |
2090                                 ((rmls & 3) << HID4_RMLS2_SH);
2091                         /* RMOR is also in HID4 */
2092                         lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
2093                                 << HID4_RMOR_SH;
2094                 } else {
2095                         /* POWER7 */
2096                         lpcr_mask = LPCR_VPM0 | LPCR_VRMA_L | LPCR_RMLS;
2097                         lpcr = rmls << LPCR_RMLS_SH;
2098                         kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT;
2099                 }
2100                 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
2101                         ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
2102
2103                 /* Initialize phys addrs of pages in RMO */
2104                 npages = kvm_rma_pages;
2105                 porder = __ilog2(npages);
2106                 physp = memslot->arch.slot_phys;
2107                 if (physp) {
2108                         if (npages > memslot->npages)
2109                                 npages = memslot->npages;
2110                         spin_lock(&kvm->arch.slot_phys_lock);
2111                         for (i = 0; i < npages; ++i)
2112                                 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
2113                                         porder;
2114                         spin_unlock(&kvm->arch.slot_phys_lock);
2115                 }
2116         }
2117
2118         kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
2119
2120         /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
2121         smp_wmb();
2122         kvm->arch.rma_setup_done = 1;
2123         err = 0;
2124  out_srcu:
2125         srcu_read_unlock(&kvm->srcu, srcu_idx);
2126  out:
2127         mutex_unlock(&kvm->lock);
2128         return err;
2129
2130  up_out:
2131         up_read(&current->mm->mmap_sem);
2132         goto out_srcu;
2133 }
2134
2135 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2136 {
2137         unsigned long lpcr, lpid;
2138
2139         /* Allocate the guest's logical partition ID */
2140
2141         lpid = kvmppc_alloc_lpid();
2142         if ((long)lpid < 0)
2143                 return -ENOMEM;
2144         kvm->arch.lpid = lpid;
2145
2146         /*
2147          * Since we don't flush the TLB when tearing down a VM,
2148          * and this lpid might have previously been used,
2149          * make sure we flush on each core before running the new VM.
2150          */
2151         cpumask_setall(&kvm->arch.need_tlb_flush);
2152
2153         kvm->arch.rma = NULL;
2154
2155         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
2156
2157         if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2158                 /* PPC970; HID4 is effectively the LPCR */
2159                 kvm->arch.host_lpid = 0;
2160                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
2161                 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
2162                 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
2163                         ((lpid & 0xf) << HID4_LPID5_SH);
2164         } else {
2165                 /* POWER7; init LPCR for virtual RMA mode */
2166                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
2167                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2168                 lpcr &= LPCR_PECE | LPCR_LPES;
2169                 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
2170                         LPCR_VPM0 | LPCR_VPM1;
2171                 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2172                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
2173                 /* On POWER8 turn on online bit to enable PURR/SPURR */
2174                 if (cpu_has_feature(CPU_FTR_ARCH_207S))
2175                         lpcr |= LPCR_ONL;
2176         }
2177         kvm->arch.lpcr = lpcr;
2178
2179         kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
2180         spin_lock_init(&kvm->arch.slot_phys_lock);
2181
2182         /*
2183          * Don't allow secondary CPU threads to come online
2184          * while any KVM VMs exist.
2185          */
2186         inhibit_secondary_onlining();
2187
2188         return 0;
2189 }
2190
2191 static void kvmppc_free_vcores(struct kvm *kvm)
2192 {
2193         long int i;
2194
2195         for (i = 0; i < KVM_MAX_VCORES; ++i)
2196                 kfree(kvm->arch.vcores[i]);
2197         kvm->arch.online_vcores = 0;
2198 }
2199
2200 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
2201 {
2202         uninhibit_secondary_onlining();
2203
2204         kvmppc_free_vcores(kvm);
2205         if (kvm->arch.rma) {
2206                 kvm_release_rma(kvm->arch.rma);
2207                 kvm->arch.rma = NULL;
2208         }
2209
2210         kvmppc_free_hpt(kvm);
2211 }
2212
2213 /* We don't need to emulate any privileged instructions or dcbz */
2214 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2215                                      unsigned int inst, int *advance)
2216 {
2217         return EMULATE_FAIL;
2218 }
2219
2220 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2221                                         ulong spr_val)
2222 {
2223         return EMULATE_FAIL;
2224 }
2225
2226 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2227                                         ulong *spr_val)
2228 {
2229         return EMULATE_FAIL;
2230 }
2231
2232 static int kvmppc_core_check_processor_compat_hv(void)
2233 {
2234         if (!cpu_has_feature(CPU_FTR_HVMODE))
2235                 return -EIO;
2236         return 0;
2237 }
2238
2239 static long kvm_arch_vm_ioctl_hv(struct file *filp,
2240                                  unsigned int ioctl, unsigned long arg)
2241 {
2242         struct kvm *kvm __maybe_unused = filp->private_data;
2243         void __user *argp = (void __user *)arg;
2244         long r;
2245
2246         switch (ioctl) {
2247
2248         case KVM_ALLOCATE_RMA: {
2249                 struct kvm_allocate_rma rma;
2250                 struct kvm *kvm = filp->private_data;
2251
2252                 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
2253                 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
2254                         r = -EFAULT;
2255                 break;
2256         }
2257
2258         case KVM_PPC_ALLOCATE_HTAB: {
2259                 u32 htab_order;
2260
2261                 r = -EFAULT;
2262                 if (get_user(htab_order, (u32 __user *)argp))
2263                         break;
2264                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2265                 if (r)
2266                         break;
2267                 r = -EFAULT;
2268                 if (put_user(htab_order, (u32 __user *)argp))
2269                         break;
2270                 r = 0;
2271                 break;
2272         }
2273
2274         case KVM_PPC_GET_HTAB_FD: {
2275                 struct kvm_get_htab_fd ghf;
2276
2277                 r = -EFAULT;
2278                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
2279                         break;
2280                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2281                 break;
2282         }
2283
2284         default:
2285                 r = -ENOTTY;
2286         }
2287
2288         return r;
2289 }
2290
2291 static struct kvmppc_ops kvm_ops_hv = {
2292         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2293         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2294         .get_one_reg = kvmppc_get_one_reg_hv,
2295         .set_one_reg = kvmppc_set_one_reg_hv,
2296         .vcpu_load   = kvmppc_core_vcpu_load_hv,
2297         .vcpu_put    = kvmppc_core_vcpu_put_hv,
2298         .set_msr     = kvmppc_set_msr_hv,
2299         .vcpu_run    = kvmppc_vcpu_run_hv,
2300         .vcpu_create = kvmppc_core_vcpu_create_hv,
2301         .vcpu_free   = kvmppc_core_vcpu_free_hv,
2302         .check_requests = kvmppc_core_check_requests_hv,
2303         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
2304         .flush_memslot  = kvmppc_core_flush_memslot_hv,
2305         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2306         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
2307         .unmap_hva = kvm_unmap_hva_hv,
2308         .unmap_hva_range = kvm_unmap_hva_range_hv,
2309         .age_hva  = kvm_age_hva_hv,
2310         .test_age_hva = kvm_test_age_hva_hv,
2311         .set_spte_hva = kvm_set_spte_hva_hv,
2312         .mmu_destroy  = kvmppc_mmu_destroy_hv,
2313         .free_memslot = kvmppc_core_free_memslot_hv,
2314         .create_memslot = kvmppc_core_create_memslot_hv,
2315         .init_vm =  kvmppc_core_init_vm_hv,
2316         .destroy_vm = kvmppc_core_destroy_vm_hv,
2317         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2318         .emulate_op = kvmppc_core_emulate_op_hv,
2319         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2320         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2321         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2322         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
2323 };
2324
2325 static int kvmppc_book3s_init_hv(void)
2326 {
2327         int r;
2328         /*
2329          * FIXME!! Do we need to check on all cpus ?
2330          */
2331         r = kvmppc_core_check_processor_compat_hv();
2332         if (r < 0)
2333                 return r;
2334
2335         kvm_ops_hv.owner = THIS_MODULE;
2336         kvmppc_hv_ops = &kvm_ops_hv;
2337
2338         r = kvmppc_mmu_hv_init();
2339         return r;
2340 }
2341
2342 static void kvmppc_book3s_exit_hv(void)
2343 {
2344         kvmppc_hv_ops = NULL;
2345 }
2346
2347 module_init(kvmppc_book3s_init_hv);
2348 module_exit(kvmppc_book3s_exit_hv);
2349 MODULE_LICENSE("GPL");
2350 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2351 MODULE_ALIAS("devname:kvm");