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_PPC_ICP_NATIVE
  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         unsigned long flags;
 139
 140         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 141         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
 142             vc->preempt_tb != TB_NIL) {
 143                 vc->stolen_tb += mftb() - vc->preempt_tb;
 144                 vc->preempt_tb = TB_NIL;
 145         }
 146         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 147             vcpu->arch.busy_preempt != TB_NIL) {
 148                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 149                 vcpu->arch.busy_preempt = TB_NIL;
 150         }
 151         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 152 }
 153
 154 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 155 {
 156         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 157         unsigned long flags;
 158
 159         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 160         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
 161                 vc->preempt_tb = mftb();
 162         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 163                 vcpu->arch.busy_preempt = mftb();
 164         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 165 }
 166
 167 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
 168 {
 169         vcpu->arch.shregs.msr = msr;
 170         kvmppc_end_cede(vcpu);
 171 }
 172
 173 void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 174 {
 175         vcpu->arch.pvr = pvr;
 176 }
 177
 178 int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 179 {
 180         unsigned long pcr = 0;
 181         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 182
 183         if (arch_compat) {
 184                 if (!cpu_has_feature(CPU_FTR_ARCH_206))
 185                         return -EINVAL; /* 970 has no compat mode support */
 186
 187                 switch (arch_compat) {
 188                 case PVR_ARCH_205:
 189                         /*
 190                          * If an arch bit is set in PCR, all the defined
 191                          * higher-order arch bits also have to be set.
 192                          */
 193                         pcr = PCR_ARCH_206 | PCR_ARCH_205;
 194                         break;
 195                 case PVR_ARCH_206:
 196                 case PVR_ARCH_206p:
 197                         pcr = PCR_ARCH_206;
 198                         break;
 199                 case PVR_ARCH_207:
 200                         break;
 201                 default:
 202                         return -EINVAL;
 203                 }
 204
 205                 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
 206                         /* POWER7 can't emulate POWER8 */
 207                         if (!(pcr & PCR_ARCH_206))
 208                                 return -EINVAL;
 209                         pcr &= ~PCR_ARCH_206;
 210                 }
 211         }
 212
 213         spin_lock(&vc->lock);
 214         vc->arch_compat = arch_compat;
 215         vc->pcr = pcr;
 216         spin_unlock(&vc->lock);
 217
 218         return 0;
 219 }
 220
 221 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 222 {
 223         int r;
 224
 225         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 226         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 227                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
 228         for (r = 0; r < 16; ++r)
 229                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 230                        r, kvmppc_get_gpr(vcpu, r),
 231                        r+16, kvmppc_get_gpr(vcpu, r+16));
 232         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 233                vcpu->arch.ctr, vcpu->arch.lr);
 234         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 235                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 236         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 237                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 238         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 239                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 240         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 241                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 242         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 243         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 244                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 245         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 246         for (r = 0; r < vcpu->arch.slb_max; ++r)
 247                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 248                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 249         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 250                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 251                vcpu->arch.last_inst);
 252 }
 253
 254 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 255 {
 256         int r;
 257         struct kvm_vcpu *v, *ret = NULL;
 258
 259         mutex_lock(&kvm->lock);
 260         kvm_for_each_vcpu(r, v, kvm) {
 261                 if (v->vcpu_id == id) {
 262                         ret = v;
 263                         break;
 264                 }
 265         }
 266         mutex_unlock(&kvm->lock);
 267         return ret;
 268 }
 269
 270 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 271 {
 272         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 273         vpa->yield_count = 1;
 274 }
 275
 276 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 277                    unsigned long addr, unsigned long len)
 278 {
 279         /* check address is cacheline aligned */
 280         if (addr & (L1_CACHE_BYTES - 1))
 281                 return -EINVAL;
 282         spin_lock(&vcpu->arch.vpa_update_lock);
 283         if (v->next_gpa != addr || v->len != len) {
 284                 v->next_gpa = addr;
 285                 v->len = addr ? len : 0;
 286                 v->update_pending = 1;
 287         }
 288         spin_unlock(&vcpu->arch.vpa_update_lock);
 289         return 0;
 290 }
 291
 292 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 293 struct reg_vpa {
 294         u32 dummy;
 295         union {
 296                 u16 hword;
 297                 u32 word;
 298         } length;
 299 };
 300
 301 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 302 {
 303         if (vpap->update_pending)
 304                 return vpap->next_gpa != 0;
 305         return vpap->pinned_addr != NULL;
 306 }
 307
 308 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 309                                        unsigned long flags,
 310                                        unsigned long vcpuid, unsigned long vpa)
 311 {
 312         struct kvm *kvm = vcpu->kvm;
 313         unsigned long len, nb;
 314         void *va;
 315         struct kvm_vcpu *tvcpu;
 316         int err;
 317         int subfunc;
 318         struct kvmppc_vpa *vpap;
 319
 320         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 321         if (!tvcpu)
 322                 return H_PARAMETER;
 323
 324         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 325         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 326             subfunc == H_VPA_REG_SLB) {
 327                 /* Registering new area - address must be cache-line aligned */
 328                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 329                         return H_PARAMETER;
 330
 331                 /* convert logical addr to kernel addr and read length */
 332                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 333                 if (va == NULL)
 334                         return H_PARAMETER;
 335                 if (subfunc == H_VPA_REG_VPA)
 336                         len = ((struct reg_vpa *)va)->length.hword;
 337                 else
 338                         len = ((struct reg_vpa *)va)->length.word;
 339                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 340
 341                 /* Check length */
 342                 if (len > nb || len < sizeof(struct reg_vpa))
 343                         return H_PARAMETER;
 344         } else {
 345                 vpa = 0;
 346                 len = 0;
 347         }
 348
 349         err = H_PARAMETER;
 350         vpap = NULL;
 351         spin_lock(&tvcpu->arch.vpa_update_lock);
 352
 353         switch (subfunc) {
 354         case H_VPA_REG_VPA:             /* register VPA */
 355                 if (len < sizeof(struct lppaca))
 356                         break;
 357                 vpap = &tvcpu->arch.vpa;
 358                 err = 0;
 359                 break;
 360
 361         case H_VPA_REG_DTL:             /* register DTL */
 362                 if (len < sizeof(struct dtl_entry))
 363                         break;
 364                 len -= len % sizeof(struct dtl_entry);
 365
 366                 /* Check that they have previously registered a VPA */
 367                 err = H_RESOURCE;
 368                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 369                         break;
 370
 371                 vpap = &tvcpu->arch.dtl;
 372                 err = 0;
 373                 break;
 374
 375         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 376                 /* Check that they have previously registered a VPA */
 377                 err = H_RESOURCE;
 378                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 379                         break;
 380
 381                 vpap = &tvcpu->arch.slb_shadow;
 382                 err = 0;
 383                 break;
 384
 385         case H_VPA_DEREG_VPA:           /* deregister VPA */
 386                 /* Check they don't still have a DTL or SLB buf registered */
 387                 err = H_RESOURCE;
 388                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 389                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 390                         break;
 391
 392                 vpap = &tvcpu->arch.vpa;
 393                 err = 0;
 394                 break;
 395
 396         case H_VPA_DEREG_DTL:           /* deregister DTL */
 397                 vpap = &tvcpu->arch.dtl;
 398                 err = 0;
 399                 break;
 400
 401         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 402                 vpap = &tvcpu->arch.slb_shadow;
 403                 err = 0;
 404                 break;
 405         }
 406
 407         if (vpap) {
 408                 vpap->next_gpa = vpa;
 409                 vpap->len = len;
 410                 vpap->update_pending = 1;
 411         }
 412
 413         spin_unlock(&tvcpu->arch.vpa_update_lock);
 414
 415         return err;
 416 }
 417
 418 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 419 {
 420         struct kvm *kvm = vcpu->kvm;
 421         void *va;
 422         unsigned long nb;
 423         unsigned long gpa;
 424
 425         /*
 426          * We need to pin the page pointed to by vpap->next_gpa,
 427          * but we can't call kvmppc_pin_guest_page under the lock
 428          * as it does get_user_pages() and down_read().  So we
 429          * have to drop the lock, pin the page, then get the lock
 430          * again and check that a new area didn't get registered
 431          * in the meantime.
 432          */
 433         for (;;) {
 434                 gpa = vpap->next_gpa;
 435                 spin_unlock(&vcpu->arch.vpa_update_lock);
 436                 va = NULL;
 437                 nb = 0;
 438                 if (gpa)
 439                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 440                 spin_lock(&vcpu->arch.vpa_update_lock);
 441                 if (gpa == vpap->next_gpa)
 442                         break;
 443                 /* sigh... unpin that one and try again */
 444                 if (va)
 445                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 446         }
 447
 448         vpap->update_pending = 0;
 449         if (va && nb < vpap->len) {
 450                 /*
 451                  * If it's now too short, it must be that userspace
 452                  * has changed the mappings underlying guest memory,
 453                  * so unregister the region.
 454                  */
 455                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 456                 va = NULL;
 457         }
 458         if (vpap->pinned_addr)
 459                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 460                                         vpap->dirty);
 461         vpap->gpa = gpa;
 462         vpap->pinned_addr = va;
 463         vpap->dirty = false;
 464         if (va)
 465                 vpap->pinned_end = va + vpap->len;
 466 }
 467
 468 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 469 {
 470         if (!(vcpu->arch.vpa.update_pending ||
 471               vcpu->arch.slb_shadow.update_pending ||
 472               vcpu->arch.dtl.update_pending))
 473                 return;
 474
 475         spin_lock(&vcpu->arch.vpa_update_lock);
 476         if (vcpu->arch.vpa.update_pending) {
 477                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 478                 if (vcpu->arch.vpa.pinned_addr)
 479                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 480         }
 481         if (vcpu->arch.dtl.update_pending) {
 482                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 483                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 484                 vcpu->arch.dtl_index = 0;
 485         }
 486         if (vcpu->arch.slb_shadow.update_pending)
 487                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 488         spin_unlock(&vcpu->arch.vpa_update_lock);
 489 }
 490
 491 /*
 492  * Return the accumulated stolen time for the vcore up until `now'.
 493  * The caller should hold the vcore lock.
 494  */
 495 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 496 {
 497         u64 p;
 498
 499         /*
 500          * If we are the task running the vcore, then since we hold
 501          * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
 502          * can't be updated, so we don't need the tbacct_lock.
 503          * If the vcore is inactive, it can't become active (since we
 504          * hold the vcore lock), so the vcpu load/put functions won't
 505          * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
 506          */
 507         if (vc->vcore_state != VCORE_INACTIVE &&
 508             vc->runner->arch.run_task != current) {
 509                 spin_lock_irq(&vc->runner->arch.tbacct_lock);
 510                 p = vc->stolen_tb;
 511                 if (vc->preempt_tb != TB_NIL)
 512                         p += now - vc->preempt_tb;
 513                 spin_unlock_irq(&vc->runner->arch.tbacct_lock);
 514         } else {
 515                 p = vc->stolen_tb;
 516         }
 517         return p;
 518 }
 519
 520 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 521                                     struct kvmppc_vcore *vc)
 522 {
 523         struct dtl_entry *dt;
 524         struct lppaca *vpa;
 525         unsigned long stolen;
 526         unsigned long core_stolen;
 527         u64 now;
 528
 529         dt = vcpu->arch.dtl_ptr;
 530         vpa = vcpu->arch.vpa.pinned_addr;
 531         now = mftb();
 532         core_stolen = vcore_stolen_time(vc, now);
 533         stolen = core_stolen - vcpu->arch.stolen_logged;
 534         vcpu->arch.stolen_logged = core_stolen;
 535         spin_lock_irq(&vcpu->arch.tbacct_lock);
 536         stolen += vcpu->arch.busy_stolen;
 537         vcpu->arch.busy_stolen = 0;
 538         spin_unlock_irq(&vcpu->arch.tbacct_lock);
 539         if (!dt || !vpa)
 540                 return;
 541         memset(dt, 0, sizeof(struct dtl_entry));
 542         dt->dispatch_reason = 7;
 543         dt->processor_id = vc->pcpu + vcpu->arch.ptid;
 544         dt->timebase = now + vc->tb_offset;
 545         dt->enqueue_to_dispatch_time = stolen;
 546         dt->srr0 = kvmppc_get_pc(vcpu);
 547         dt->srr1 = vcpu->arch.shregs.msr;
 548         ++dt;
 549         if (dt == vcpu->arch.dtl.pinned_end)
 550                 dt = vcpu->arch.dtl.pinned_addr;
 551         vcpu->arch.dtl_ptr = dt;
 552         /* order writing *dt vs. writing vpa->dtl_idx */
 553         smp_wmb();
 554         vpa->dtl_idx = ++vcpu->arch.dtl_index;
 555         vcpu->arch.dtl.dirty = true;
 556 }
 557
 558 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 559 {
 560         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 561         unsigned long target, ret = H_SUCCESS;
 562         struct kvm_vcpu *tvcpu;
 563         int idx, rc;
 564
 565         switch (req) {
 566         case H_ENTER:
 567                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 568                 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
 569                                               kvmppc_get_gpr(vcpu, 5),
 570                                               kvmppc_get_gpr(vcpu, 6),
 571                                               kvmppc_get_gpr(vcpu, 7));
 572                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 573                 break;
 574         case H_CEDE:
 575                 break;
 576         case H_PROD:
 577                 target = kvmppc_get_gpr(vcpu, 4);
 578                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 579                 if (!tvcpu) {
 580                         ret = H_PARAMETER;
 581                         break;
 582                 }
 583                 tvcpu->arch.prodded = 1;
 584                 smp_mb();
 585                 if (vcpu->arch.ceded) {
 586                         if (waitqueue_active(&vcpu->wq)) {
 587                                 wake_up_interruptible(&vcpu->wq);
 588                                 vcpu->stat.halt_wakeup++;
 589                         }
 590                 }
 591                 break;
 592         case H_CONFER:
 593                 target = kvmppc_get_gpr(vcpu, 4);
 594                 if (target == -1)
 595                         break;
 596                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 597                 if (!tvcpu) {
 598                         ret = H_PARAMETER;
 599                         break;
 600                 }
 601                 kvm_vcpu_yield_to(tvcpu);
 602                 break;
 603         case H_REGISTER_VPA:
 604                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 605                                         kvmppc_get_gpr(vcpu, 5),
 606                                         kvmppc_get_gpr(vcpu, 6));
 607                 break;
 608         case H_RTAS:
 609                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 610                         return RESUME_HOST;
 611
 612                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 613                 rc = kvmppc_rtas_hcall(vcpu);
 614                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 615
 616                 if (rc == -ENOENT)
 617                         return RESUME_HOST;
 618                 else if (rc == 0)
 619                         break;
 620
 621                 /* Send the error out to userspace via KVM_RUN */
 622                 return rc;
 623
 624         case H_XIRR:
 625         case H_CPPR:
 626         case H_EOI:
 627         case H_IPI:
 628         case H_IPOLL:
 629         case H_XIRR_X:
 630                 if (kvmppc_xics_enabled(vcpu)) {
 631                         ret = kvmppc_xics_hcall(vcpu, req);
 632                         break;
 633                 } /* fallthrough */
 634         default:
 635                 return RESUME_HOST;
 636         }
 637         kvmppc_set_gpr(vcpu, 3, ret);
 638         vcpu->arch.hcall_needed = 0;
 639         return RESUME_GUEST;
 640 }
 641
 642 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
 643                                  struct task_struct *tsk)
 644 {
 645         int r = RESUME_HOST;
 646
 647         vcpu->stat.sum_exits++;
 648
 649         run->exit_reason = KVM_EXIT_UNKNOWN;
 650         run->ready_for_interrupt_injection = 1;
 651         switch (vcpu->arch.trap) {
 652         /* We're good on these - the host merely wanted to get our attention */
 653         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 654                 vcpu->stat.dec_exits++;
 655                 r = RESUME_GUEST;
 656                 break;
 657         case BOOK3S_INTERRUPT_EXTERNAL:
 658         case BOOK3S_INTERRUPT_H_DOORBELL:
 659                 vcpu->stat.ext_intr_exits++;
 660                 r = RESUME_GUEST;
 661                 break;
 662         case BOOK3S_INTERRUPT_PERFMON:
 663                 r = RESUME_GUEST;
 664                 break;
 665         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 666                 /*
 667                  * Deliver a machine check interrupt to the guest.
 668                  * We have to do this, even if the host has handled the
 669                  * machine check, because machine checks use SRR0/1 and
 670                  * the interrupt might have trashed guest state in them.
 671                  */
 672                 kvmppc_book3s_queue_irqprio(vcpu,
 673                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 674                 r = RESUME_GUEST;
 675                 break;
 676         case BOOK3S_INTERRUPT_PROGRAM:
 677         {
 678                 ulong flags;
 679                 /*
 680                  * Normally program interrupts are delivered directly
 681                  * to the guest by the hardware, but we can get here
 682                  * as a result of a hypervisor emulation interrupt
 683                  * (e40) getting turned into a 700 by BML RTAS.
 684                  */
 685                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 686                 kvmppc_core_queue_program(vcpu, flags);
 687                 r = RESUME_GUEST;
 688                 break;
 689         }
 690         case BOOK3S_INTERRUPT_SYSCALL:
 691         {
 692                 /* hcall - punt to userspace */
 693                 int i;
 694
 695                 /* hypercall with MSR_PR has already been handled in rmode,
 696                  * and never reaches here.
 697                  */
 698
 699                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 700                 for (i = 0; i < 9; ++i)
 701                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 702                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 703                 vcpu->arch.hcall_needed = 1;
 704                 r = RESUME_HOST;
 705                 break;
 706         }
 707         /*
 708          * We get these next two if the guest accesses a page which it thinks
 709          * it has mapped but which is not actually present, either because
 710          * it is for an emulated I/O device or because the corresonding
 711          * host page has been paged out.  Any other HDSI/HISI interrupts
 712          * have been handled already.
 713          */
 714         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 715                 r = RESUME_PAGE_FAULT;
 716                 break;
 717         case BOOK3S_INTERRUPT_H_INST_STORAGE:
 718                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
 719                 vcpu->arch.fault_dsisr = 0;
 720                 r = RESUME_PAGE_FAULT;
 721                 break;
 722         /*
 723          * This occurs if the guest executes an illegal instruction.
 724          * We just generate a program interrupt to the guest, since
 725          * we don't emulate any guest instructions at this stage.
 726          */
 727         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 728                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 729                 r = RESUME_GUEST;
 730                 break;
 731         /*
 732          * This occurs if the guest (kernel or userspace), does something that
 733          * is prohibited by HFSCR.  We just generate a program interrupt to
 734          * the guest.
 735          */
 736         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
 737                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 738                 r = RESUME_GUEST;
 739                 break;
 740         default:
 741                 kvmppc_dump_regs(vcpu);
 742                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 743                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 744                         vcpu->arch.shregs.msr);
 745                 run->hw.hardware_exit_reason = vcpu->arch.trap;
 746                 r = RESUME_HOST;
 747                 break;
 748         }
 749
 750         return r;
 751 }
 752
 753 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
 754                                             struct kvm_sregs *sregs)
 755 {
 756         int i;
 757
 758         memset(sregs, 0, sizeof(struct kvm_sregs));
 759         sregs->pvr = vcpu->arch.pvr;
 760         for (i = 0; i < vcpu->arch.slb_max; i++) {
 761                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
 762                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 763         }
 764
 765         return 0;
 766 }
 767
 768 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
 769                                             struct kvm_sregs *sregs)
 770 {
 771         int i, j;
 772
 773         kvmppc_set_pvr_hv(vcpu, sregs->pvr);
 774
 775         j = 0;
 776         for (i = 0; i < vcpu->arch.slb_nr; i++) {
 777                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
 778                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
 779                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
 780                         ++j;
 781                 }
 782         }
 783         vcpu->arch.slb_max = j;
 784
 785         return 0;
 786 }
 787
 788 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr)
 789 {
 790         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 791         u64 mask;
 792
 793         spin_lock(&vc->lock);
 794         /*
 795          * If ILE (interrupt little-endian) has changed, update the
 796          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
 797          */
 798         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
 799                 struct kvm *kvm = vcpu->kvm;
 800                 struct kvm_vcpu *vcpu;
 801                 int i;
 802
 803                 mutex_lock(&kvm->lock);
 804                 kvm_for_each_vcpu(i, vcpu, kvm) {
 805                         if (vcpu->arch.vcore != vc)
 806                                 continue;
 807                         if (new_lpcr & LPCR_ILE)
 808                                 vcpu->arch.intr_msr |= MSR_LE;
 809                         else
 810                                 vcpu->arch.intr_msr &= ~MSR_LE;
 811                 }
 812                 mutex_unlock(&kvm->lock);
 813         }
 814
 815         /*
 816          * Userspace can only modify DPFD (default prefetch depth),
 817          * ILE (interrupt little-endian) and TC (translation control).
 818          * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
 819          */
 820         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
 821         if (cpu_has_feature(CPU_FTR_ARCH_207S))
 822                 mask |= LPCR_AIL;
 823         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
 824         spin_unlock(&vc->lock);
 825 }
 826
 827 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
 828                                  union kvmppc_one_reg *val)
 829 {
 830         int r = 0;
 831         long int i;
 832
 833         switch (id) {
 834         case KVM_REG_PPC_HIOR:
 835                 *val = get_reg_val(id, 0);
 836                 break;
 837         case KVM_REG_PPC_DABR:
 838                 *val = get_reg_val(id, vcpu->arch.dabr);
 839                 break;
 840         case KVM_REG_PPC_DABRX:
 841                 *val = get_reg_val(id, vcpu->arch.dabrx);
 842                 break;
 843         case KVM_REG_PPC_DSCR:
 844                 *val = get_reg_val(id, vcpu->arch.dscr);
 845                 break;
 846         case KVM_REG_PPC_PURR:
 847                 *val = get_reg_val(id, vcpu->arch.purr);
 848                 break;
 849         case KVM_REG_PPC_SPURR:
 850                 *val = get_reg_val(id, vcpu->arch.spurr);
 851                 break;
 852         case KVM_REG_PPC_AMR:
 853                 *val = get_reg_val(id, vcpu->arch.amr);
 854                 break;
 855         case KVM_REG_PPC_UAMOR:
 856                 *val = get_reg_val(id, vcpu->arch.uamor);
 857                 break;
 858         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
 859                 i = id - KVM_REG_PPC_MMCR0;
 860                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
 861                 break;
 862         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
 863                 i = id - KVM_REG_PPC_PMC1;
 864                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
 865                 break;
 866         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
 867                 i = id - KVM_REG_PPC_SPMC1;
 868                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
 869                 break;
 870         case KVM_REG_PPC_SIAR:
 871                 *val = get_reg_val(id, vcpu->arch.siar);
 872                 break;
 873         case KVM_REG_PPC_SDAR:
 874                 *val = get_reg_val(id, vcpu->arch.sdar);
 875                 break;
 876         case KVM_REG_PPC_SIER:
 877                 *val = get_reg_val(id, vcpu->arch.sier);
 878                 break;
 879         case KVM_REG_PPC_IAMR:
 880                 *val = get_reg_val(id, vcpu->arch.iamr);
 881                 break;
 882         case KVM_REG_PPC_PSPB:
 883                 *val = get_reg_val(id, vcpu->arch.pspb);
 884                 break;
 885         case KVM_REG_PPC_DPDES:
 886                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
 887                 break;
 888         case KVM_REG_PPC_DAWR:
 889                 *val = get_reg_val(id, vcpu->arch.dawr);
 890                 break;
 891         case KVM_REG_PPC_DAWRX:
 892                 *val = get_reg_val(id, vcpu->arch.dawrx);
 893                 break;
 894         case KVM_REG_PPC_CIABR:
 895                 *val = get_reg_val(id, vcpu->arch.ciabr);
 896                 break;
 897         case KVM_REG_PPC_IC:
 898                 *val = get_reg_val(id, vcpu->arch.ic);
 899                 break;
 900         case KVM_REG_PPC_VTB:
 901                 *val = get_reg_val(id, vcpu->arch.vtb);
 902                 break;
 903         case KVM_REG_PPC_CSIGR:
 904                 *val = get_reg_val(id, vcpu->arch.csigr);
 905                 break;
 906         case KVM_REG_PPC_TACR:
 907                 *val = get_reg_val(id, vcpu->arch.tacr);
 908                 break;
 909         case KVM_REG_PPC_TCSCR:
 910                 *val = get_reg_val(id, vcpu->arch.tcscr);
 911                 break;
 912         case KVM_REG_PPC_PID:
 913                 *val = get_reg_val(id, vcpu->arch.pid);
 914                 break;
 915         case KVM_REG_PPC_ACOP:
 916                 *val = get_reg_val(id, vcpu->arch.acop);
 917                 break;
 918         case KVM_REG_PPC_WORT:
 919                 *val = get_reg_val(id, vcpu->arch.wort);
 920                 break;
 921         case KVM_REG_PPC_VPA_ADDR:
 922                 spin_lock(&vcpu->arch.vpa_update_lock);
 923                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
 924                 spin_unlock(&vcpu->arch.vpa_update_lock);
 925                 break;
 926         case KVM_REG_PPC_VPA_SLB:
 927                 spin_lock(&vcpu->arch.vpa_update_lock);
 928                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
 929                 val->vpaval.length = vcpu->arch.slb_shadow.len;
 930                 spin_unlock(&vcpu->arch.vpa_update_lock);
 931                 break;
 932         case KVM_REG_PPC_VPA_DTL:
 933                 spin_lock(&vcpu->arch.vpa_update_lock);
 934                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
 935                 val->vpaval.length = vcpu->arch.dtl.len;
 936                 spin_unlock(&vcpu->arch.vpa_update_lock);
 937                 break;
 938         case KVM_REG_PPC_TB_OFFSET:
 939                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
 940                 break;
 941         case KVM_REG_PPC_LPCR:
 942                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
 943                 break;
 944         case KVM_REG_PPC_PPR:
 945                 *val = get_reg_val(id, vcpu->arch.ppr);
 946                 break;
 947 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
 948         case KVM_REG_PPC_TFHAR:
 949                 *val = get_reg_val(id, vcpu->arch.tfhar);
 950                 break;
 951         case KVM_REG_PPC_TFIAR:
 952                 *val = get_reg_val(id, vcpu->arch.tfiar);
 953                 break;
 954         case KVM_REG_PPC_TEXASR:
 955                 *val = get_reg_val(id, vcpu->arch.texasr);
 956                 break;
 957         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
 958                 i = id - KVM_REG_PPC_TM_GPR0;
 959                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
 960                 break;
 961         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
 962         {
 963                 int j;
 964                 i = id - KVM_REG_PPC_TM_VSR0;
 965                 if (i < 32)
 966                         for (j = 0; j < TS_FPRWIDTH; j++)
 967                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
 968                 else {
 969                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
 970                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
 971                         else
 972                                 r = -ENXIO;
 973                 }
 974                 break;
 975         }
 976         case KVM_REG_PPC_TM_CR:
 977                 *val = get_reg_val(id, vcpu->arch.cr_tm);
 978                 break;
 979         case KVM_REG_PPC_TM_LR:
 980                 *val = get_reg_val(id, vcpu->arch.lr_tm);
 981                 break;
 982         case KVM_REG_PPC_TM_CTR:
 983                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
 984                 break;
 985         case KVM_REG_PPC_TM_FPSCR:
 986                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
 987                 break;
 988         case KVM_REG_PPC_TM_AMR:
 989                 *val = get_reg_val(id, vcpu->arch.amr_tm);
 990                 break;
 991         case KVM_REG_PPC_TM_PPR:
 992                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
 993                 break;
 994         case KVM_REG_PPC_TM_VRSAVE:
 995                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
 996                 break;
 997         case KVM_REG_PPC_TM_VSCR:
 998                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
 999                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1000                 else
1001                         r = -ENXIO;
1002                 break;
1003         case KVM_REG_PPC_TM_DSCR:
1004                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1005                 break;
1006         case KVM_REG_PPC_TM_TAR:
1007                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1008                 break;
1009 #endif
1010         case KVM_REG_PPC_ARCH_COMPAT:
1011                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1012                 break;
1013         default:
1014                 r = -EINVAL;
1015                 break;
1016         }
1017
1018         return r;
1019 }
1020
1021 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1022                                  union kvmppc_one_reg *val)
1023 {
1024         int r = 0;
1025         long int i;
1026         unsigned long addr, len;
1027
1028         switch (id) {
1029         case KVM_REG_PPC_HIOR:
1030                 /* Only allow this to be set to zero */
1031                 if (set_reg_val(id, *val))
1032                         r = -EINVAL;
1033                 break;
1034         case KVM_REG_PPC_DABR:
1035                 vcpu->arch.dabr = set_reg_val(id, *val);
1036                 break;
1037         case KVM_REG_PPC_DABRX:
1038                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1039                 break;
1040         case KVM_REG_PPC_DSCR:
1041                 vcpu->arch.dscr = set_reg_val(id, *val);
1042                 break;
1043         case KVM_REG_PPC_PURR:
1044                 vcpu->arch.purr = set_reg_val(id, *val);
1045                 break;
1046         case KVM_REG_PPC_SPURR:
1047                 vcpu->arch.spurr = set_reg_val(id, *val);
1048                 break;
1049         case KVM_REG_PPC_AMR:
1050                 vcpu->arch.amr = set_reg_val(id, *val);
1051                 break;
1052         case KVM_REG_PPC_UAMOR:
1053                 vcpu->arch.uamor = set_reg_val(id, *val);
1054                 break;
1055         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1056                 i = id - KVM_REG_PPC_MMCR0;
1057                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1058                 break;
1059         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1060                 i = id - KVM_REG_PPC_PMC1;
1061                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1062                 break;
1063         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1064                 i = id - KVM_REG_PPC_SPMC1;
1065                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1066                 break;
1067         case KVM_REG_PPC_SIAR:
1068                 vcpu->arch.siar = set_reg_val(id, *val);
1069                 break;
1070         case KVM_REG_PPC_SDAR:
1071                 vcpu->arch.sdar = set_reg_val(id, *val);
1072                 break;
1073         case KVM_REG_PPC_SIER:
1074                 vcpu->arch.sier = set_reg_val(id, *val);
1075                 break;
1076         case KVM_REG_PPC_IAMR:
1077                 vcpu->arch.iamr = set_reg_val(id, *val);
1078                 break;
1079         case KVM_REG_PPC_PSPB:
1080                 vcpu->arch.pspb = set_reg_val(id, *val);
1081                 break;
1082         case KVM_REG_PPC_DPDES:
1083                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1084                 break;
1085         case KVM_REG_PPC_DAWR:
1086                 vcpu->arch.dawr = set_reg_val(id, *val);
1087                 break;
1088         case KVM_REG_PPC_DAWRX:
1089                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1090                 break;
1091         case KVM_REG_PPC_CIABR:
1092                 vcpu->arch.ciabr = set_reg_val(id, *val);
1093                 /* Don't allow setting breakpoints in hypervisor code */
1094                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1095                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1096                 break;
1097         case KVM_REG_PPC_IC:
1098                 vcpu->arch.ic = set_reg_val(id, *val);
1099                 break;
1100         case KVM_REG_PPC_VTB:
1101                 vcpu->arch.vtb = set_reg_val(id, *val);
1102                 break;
1103         case KVM_REG_PPC_CSIGR:
1104                 vcpu->arch.csigr = set_reg_val(id, *val);
1105                 break;
1106         case KVM_REG_PPC_TACR:
1107                 vcpu->arch.tacr = set_reg_val(id, *val);
1108                 break;
1109         case KVM_REG_PPC_TCSCR:
1110                 vcpu->arch.tcscr = set_reg_val(id, *val);
1111                 break;
1112         case KVM_REG_PPC_PID:
1113                 vcpu->arch.pid = set_reg_val(id, *val);
1114                 break;
1115         case KVM_REG_PPC_ACOP:
1116                 vcpu->arch.acop = set_reg_val(id, *val);
1117                 break;
1118         case KVM_REG_PPC_WORT:
1119                 vcpu->arch.wort = set_reg_val(id, *val);
1120                 break;
1121         case KVM_REG_PPC_VPA_ADDR:
1122                 addr = set_reg_val(id, *val);
1123                 r = -EINVAL;
1124                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1125                               vcpu->arch.dtl.next_gpa))
1126                         break;
1127                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1128                 break;
1129         case KVM_REG_PPC_VPA_SLB:
1130                 addr = val->vpaval.addr;
1131                 len = val->vpaval.length;
1132                 r = -EINVAL;
1133                 if (addr && !vcpu->arch.vpa.next_gpa)
1134                         break;
1135                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1136                 break;
1137         case KVM_REG_PPC_VPA_DTL:
1138                 addr = val->vpaval.addr;
1139                 len = val->vpaval.length;
1140                 r = -EINVAL;
1141                 if (addr && (len < sizeof(struct dtl_entry) ||
1142                              !vcpu->arch.vpa.next_gpa))
1143                         break;
1144                 len -= len % sizeof(struct dtl_entry);
1145                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1146                 break;
1147         case KVM_REG_PPC_TB_OFFSET:
1148                 /* round up to multiple of 2^24 */
1149                 vcpu->arch.vcore->tb_offset =
1150                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1151                 break;
1152         case KVM_REG_PPC_LPCR:
1153                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val));
1154                 break;
1155         case KVM_REG_PPC_PPR:
1156                 vcpu->arch.ppr = set_reg_val(id, *val);
1157                 break;
1158 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1159         case KVM_REG_PPC_TFHAR:
1160                 vcpu->arch.tfhar = set_reg_val(id, *val);
1161                 break;
1162         case KVM_REG_PPC_TFIAR:
1163                 vcpu->arch.tfiar = set_reg_val(id, *val);
1164                 break;
1165         case KVM_REG_PPC_TEXASR:
1166                 vcpu->arch.texasr = set_reg_val(id, *val);
1167                 break;
1168         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1169                 i = id - KVM_REG_PPC_TM_GPR0;
1170                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1171                 break;
1172         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1173         {
1174                 int j;
1175                 i = id - KVM_REG_PPC_TM_VSR0;
1176                 if (i < 32)
1177                         for (j = 0; j < TS_FPRWIDTH; j++)
1178                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1179                 else
1180                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1181                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1182                         else
1183                                 r = -ENXIO;
1184                 break;
1185         }
1186         case KVM_REG_PPC_TM_CR:
1187                 vcpu->arch.cr_tm = set_reg_val(id, *val);
1188                 break;
1189         case KVM_REG_PPC_TM_LR:
1190                 vcpu->arch.lr_tm = set_reg_val(id, *val);
1191                 break;
1192         case KVM_REG_PPC_TM_CTR:
1193                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1194                 break;
1195         case KVM_REG_PPC_TM_FPSCR:
1196                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1197                 break;
1198         case KVM_REG_PPC_TM_AMR:
1199                 vcpu->arch.amr_tm = set_reg_val(id, *val);
1200                 break;
1201         case KVM_REG_PPC_TM_PPR:
1202                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1203                 break;
1204         case KVM_REG_PPC_TM_VRSAVE:
1205                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1206                 break;
1207         case KVM_REG_PPC_TM_VSCR:
1208                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1209                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1210                 else
1211                         r = - ENXIO;
1212                 break;
1213         case KVM_REG_PPC_TM_DSCR:
1214                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1215                 break;
1216         case KVM_REG_PPC_TM_TAR:
1217                 vcpu->arch.tar_tm = set_reg_val(id, *val);
1218                 break;
1219 #endif
1220         case KVM_REG_PPC_ARCH_COMPAT:
1221                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1222                 break;
1223         default:
1224                 r = -EINVAL;
1225                 break;
1226         }
1227
1228         return r;
1229 }
1230
1231 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1232                                                    unsigned int id)
1233 {
1234         struct kvm_vcpu *vcpu;
1235         int err = -EINVAL;
1236         int core;
1237         struct kvmppc_vcore *vcore;
1238
1239         core = id / threads_per_subcore;
1240         if (core >= KVM_MAX_VCORES)
1241                 goto out;
1242
1243         err = -ENOMEM;
1244         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1245         if (!vcpu)
1246                 goto out;
1247
1248         err = kvm_vcpu_init(vcpu, kvm, id);
1249         if (err)
1250                 goto free_vcpu;
1251
1252         vcpu->arch.shared = &vcpu->arch.shregs;
1253 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1254         /*
1255          * The shared struct is never shared on HV,
1256          * so we can always use host endianness
1257          */
1258 #ifdef __BIG_ENDIAN__
1259         vcpu->arch.shared_big_endian = true;
1260 #else
1261         vcpu->arch.shared_big_endian = false;
1262 #endif
1263 #endif
1264         vcpu->arch.mmcr[0] = MMCR0_FC;
1265         vcpu->arch.ctrl = CTRL_RUNLATCH;
1266         /* default to host PVR, since we can't spoof it */
1267         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1268         spin_lock_init(&vcpu->arch.vpa_update_lock);
1269         spin_lock_init(&vcpu->arch.tbacct_lock);
1270         vcpu->arch.busy_preempt = TB_NIL;
1271         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1272
1273         kvmppc_mmu_book3s_hv_init(vcpu);
1274
1275         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1276
1277         init_waitqueue_head(&vcpu->arch.cpu_run);
1278
1279         mutex_lock(&kvm->lock);
1280         vcore = kvm->arch.vcores[core];
1281         if (!vcore) {
1282                 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1283                 if (vcore) {
1284                         INIT_LIST_HEAD(&vcore->runnable_threads);
1285                         spin_lock_init(&vcore->lock);
1286                         init_waitqueue_head(&vcore->wq);
1287                         vcore->preempt_tb = TB_NIL;
1288                         vcore->lpcr = kvm->arch.lpcr;
1289                         vcore->first_vcpuid = core * threads_per_subcore;
1290                         vcore->kvm = kvm;
1291                 }
1292                 kvm->arch.vcores[core] = vcore;
1293                 kvm->arch.online_vcores++;
1294         }
1295         mutex_unlock(&kvm->lock);
1296
1297         if (!vcore)
1298                 goto free_vcpu;
1299
1300         spin_lock(&vcore->lock);
1301         ++vcore->num_threads;
1302         spin_unlock(&vcore->lock);
1303         vcpu->arch.vcore = vcore;
1304         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1305
1306         vcpu->arch.cpu_type = KVM_CPU_3S_64;
1307         kvmppc_sanity_check(vcpu);
1308
1309         return vcpu;
1310
1311 free_vcpu:
1312         kmem_cache_free(kvm_vcpu_cache, vcpu);
1313 out:
1314         return ERR_PTR(err);
1315 }
1316
1317 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1318 {
1319         if (vpa->pinned_addr)
1320                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1321                                         vpa->dirty);
1322 }
1323
1324 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1325 {
1326         spin_lock(&vcpu->arch.vpa_update_lock);
1327         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1328         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1329         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1330         spin_unlock(&vcpu->arch.vpa_update_lock);
1331         kvm_vcpu_uninit(vcpu);
1332         kmem_cache_free(kvm_vcpu_cache, vcpu);
1333 }
1334
1335 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1336 {
1337         /* Indicate we want to get back into the guest */
1338         return 1;
1339 }
1340
1341 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1342 {
1343         unsigned long dec_nsec, now;
1344
1345         now = get_tb();
1346         if (now > vcpu->arch.dec_expires) {
1347                 /* decrementer has already gone negative */
1348                 kvmppc_core_queue_dec(vcpu);
1349                 kvmppc_core_prepare_to_enter(vcpu);
1350                 return;
1351         }
1352         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1353                    / tb_ticks_per_sec;
1354         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1355                       HRTIMER_MODE_REL);
1356         vcpu->arch.timer_running = 1;
1357 }
1358
1359 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1360 {
1361         vcpu->arch.ceded = 0;
1362         if (vcpu->arch.timer_running) {
1363                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1364                 vcpu->arch.timer_running = 0;
1365         }
1366 }
1367
1368 extern void __kvmppc_vcore_entry(void);
1369
1370 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1371                                    struct kvm_vcpu *vcpu)
1372 {
1373         u64 now;
1374
1375         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1376                 return;
1377         spin_lock_irq(&vcpu->arch.tbacct_lock);
1378         now = mftb();
1379         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1380                 vcpu->arch.stolen_logged;
1381         vcpu->arch.busy_preempt = now;
1382         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1383         spin_unlock_irq(&vcpu->arch.tbacct_lock);
1384         --vc->n_runnable;
1385         list_del(&vcpu->arch.run_list);
1386 }
1387
1388 static int kvmppc_grab_hwthread(int cpu)
1389 {
1390         struct paca_struct *tpaca;
1391         long timeout = 1000;
1392
1393         tpaca = &paca[cpu];
1394
1395         /* Ensure the thread won't go into the kernel if it wakes */
1396         tpaca->kvm_hstate.hwthread_req = 1;
1397         tpaca->kvm_hstate.kvm_vcpu = NULL;
1398
1399         /*
1400          * If the thread is already executing in the kernel (e.g. handling
1401          * a stray interrupt), wait for it to get back to nap mode.
1402          * The smp_mb() is to ensure that our setting of hwthread_req
1403          * is visible before we look at hwthread_state, so if this
1404          * races with the code at system_reset_pSeries and the thread
1405          * misses our setting of hwthread_req, we are sure to see its
1406          * setting of hwthread_state, and vice versa.
1407          */
1408         smp_mb();
1409         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1410                 if (--timeout <= 0) {
1411                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1412                         return -EBUSY;
1413                 }
1414                 udelay(1);
1415         }
1416         return 0;
1417 }
1418
1419 static void kvmppc_release_hwthread(int cpu)
1420 {
1421         struct paca_struct *tpaca;
1422
1423         tpaca = &paca[cpu];
1424         tpaca->kvm_hstate.hwthread_req = 0;
1425         tpaca->kvm_hstate.kvm_vcpu = NULL;
1426 }
1427
1428 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1429 {
1430         int cpu;
1431         struct paca_struct *tpaca;
1432         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1433
1434         if (vcpu->arch.timer_running) {
1435                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1436                 vcpu->arch.timer_running = 0;
1437         }
1438         cpu = vc->pcpu + vcpu->arch.ptid;
1439         tpaca = &paca[cpu];
1440         tpaca->kvm_hstate.kvm_vcpu = vcpu;
1441         tpaca->kvm_hstate.kvm_vcore = vc;
1442         tpaca->kvm_hstate.ptid = vcpu->arch.ptid;
1443         vcpu->cpu = vc->pcpu;
1444         smp_wmb();
1445 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1446         if (cpu != smp_processor_id()) {
1447                 xics_wake_cpu(cpu);
1448                 if (vcpu->arch.ptid)
1449                         ++vc->n_woken;
1450         }
1451 #endif
1452 }
1453
1454 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1455 {
1456         int i;
1457
1458         HMT_low();
1459         i = 0;
1460         while (vc->nap_count < vc->n_woken) {
1461                 if (++i >= 1000000) {
1462                         pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1463                                vc->nap_count, vc->n_woken);
1464                         break;
1465                 }
1466                 cpu_relax();
1467         }
1468         HMT_medium();
1469 }
1470
1471 /*
1472  * Check that we are on thread 0 and that any other threads in
1473  * this core are off-line.  Then grab the threads so they can't
1474  * enter the kernel.
1475  */
1476 static int on_primary_thread(void)
1477 {
1478         int cpu = smp_processor_id();
1479         int thr;
1480
1481         /* Are we on a primary subcore? */
1482         if (cpu_thread_in_subcore(cpu))
1483                 return 0;
1484
1485         thr = 0;
1486         while (++thr < threads_per_subcore)
1487                 if (cpu_online(cpu + thr))
1488                         return 0;
1489
1490         /* Grab all hw threads so they can't go into the kernel */
1491         for (thr = 1; thr < threads_per_subcore; ++thr) {
1492                 if (kvmppc_grab_hwthread(cpu + thr)) {
1493                         /* Couldn't grab one; let the others go */
1494                         do {
1495                                 kvmppc_release_hwthread(cpu + thr);
1496                         } while (--thr > 0);
1497                         return 0;
1498                 }
1499         }
1500         return 1;
1501 }
1502
1503 /*
1504  * Run a set of guest threads on a physical core.
1505  * Called with vc->lock held.
1506  */
1507 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1508 {
1509         struct kvm_vcpu *vcpu, *vnext;
1510         long ret;
1511         u64 now;
1512         int i, need_vpa_update;
1513         int srcu_idx;
1514         struct kvm_vcpu *vcpus_to_update[threads_per_core];
1515
1516         /* don't start if any threads have a signal pending */
1517         need_vpa_update = 0;
1518         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1519                 if (signal_pending(vcpu->arch.run_task))
1520                         return;
1521                 if (vcpu->arch.vpa.update_pending ||
1522                     vcpu->arch.slb_shadow.update_pending ||
1523                     vcpu->arch.dtl.update_pending)
1524                         vcpus_to_update[need_vpa_update++] = vcpu;
1525         }
1526
1527         /*
1528          * Initialize *vc, in particular vc->vcore_state, so we can
1529          * drop the vcore lock if necessary.
1530          */
1531         vc->n_woken = 0;
1532         vc->nap_count = 0;
1533         vc->entry_exit_count = 0;
1534         vc->vcore_state = VCORE_STARTING;
1535         vc->in_guest = 0;
1536         vc->napping_threads = 0;
1537
1538         /*
1539          * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1540          * which can't be called with any spinlocks held.
1541          */
1542         if (need_vpa_update) {
1543                 spin_unlock(&vc->lock);
1544                 for (i = 0; i < need_vpa_update; ++i)
1545                         kvmppc_update_vpas(vcpus_to_update[i]);
1546                 spin_lock(&vc->lock);
1547         }
1548
1549         /*
1550          * Make sure we are running on primary threads, and that secondary
1551          * threads are offline.  Also check if the number of threads in this
1552          * guest are greater than the current system threads per guest.
1553          */
1554         if ((threads_per_core > 1) &&
1555             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
1556                 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1557                         vcpu->arch.ret = -EBUSY;
1558                 goto out;
1559         }
1560
1561
1562         vc->pcpu = smp_processor_id();
1563         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1564                 kvmppc_start_thread(vcpu);
1565                 kvmppc_create_dtl_entry(vcpu, vc);
1566         }
1567
1568         /* Set this explicitly in case thread 0 doesn't have a vcpu */
1569         get_paca()->kvm_hstate.kvm_vcore = vc;
1570         get_paca()->kvm_hstate.ptid = 0;
1571
1572         vc->vcore_state = VCORE_RUNNING;
1573         preempt_disable();
1574         spin_unlock(&vc->lock);
1575
1576         kvm_guest_enter();
1577
1578         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
1579
1580         __kvmppc_vcore_entry();
1581
1582         spin_lock(&vc->lock);
1583         /* disable sending of IPIs on virtual external irqs */
1584         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1585                 vcpu->cpu = -1;
1586         /* wait for secondary threads to finish writing their state to memory */
1587         if (vc->nap_count < vc->n_woken)
1588                 kvmppc_wait_for_nap(vc);
1589         for (i = 0; i < threads_per_subcore; ++i)
1590                 kvmppc_release_hwthread(vc->pcpu + i);
1591         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
1592         vc->vcore_state = VCORE_EXITING;
1593         spin_unlock(&vc->lock);
1594
1595         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
1596
1597         /* make sure updates to secondary vcpu structs are visible now */
1598         smp_mb();
1599         kvm_guest_exit();
1600
1601         preempt_enable();
1602         cond_resched();
1603
1604         spin_lock(&vc->lock);
1605         now = get_tb();
1606         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1607                 /* cancel pending dec exception if dec is positive */
1608                 if (now < vcpu->arch.dec_expires &&
1609                     kvmppc_core_pending_dec(vcpu))
1610                         kvmppc_core_dequeue_dec(vcpu);
1611
1612                 ret = RESUME_GUEST;
1613                 if (vcpu->arch.trap)
1614                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1615                                                     vcpu->arch.run_task);
1616
1617                 vcpu->arch.ret = ret;
1618                 vcpu->arch.trap = 0;
1619
1620                 if (vcpu->arch.ceded) {
1621                         if (!is_kvmppc_resume_guest(ret))
1622                                 kvmppc_end_cede(vcpu);
1623                         else
1624                                 kvmppc_set_timer(vcpu);
1625                 }
1626         }
1627
1628  out:
1629         vc->vcore_state = VCORE_INACTIVE;
1630         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1631                                  arch.run_list) {
1632                 if (!is_kvmppc_resume_guest(vcpu->arch.ret)) {
1633                         kvmppc_remove_runnable(vc, vcpu);
1634                         wake_up(&vcpu->arch.cpu_run);
1635                 }
1636         }
1637 }
1638
1639 /*
1640  * Wait for some other vcpu thread to execute us, and
1641  * wake us up when we need to handle something in the host.
1642  */
1643 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1644 {
1645         DEFINE_WAIT(wait);
1646
1647         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1648         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1649                 schedule();
1650         finish_wait(&vcpu->arch.cpu_run, &wait);
1651 }
1652
1653 /*
1654  * All the vcpus in this vcore are idle, so wait for a decrementer
1655  * or external interrupt to one of the vcpus.  vc->lock is held.
1656  */
1657 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1658 {
1659         DEFINE_WAIT(wait);
1660
1661         prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1662         vc->vcore_state = VCORE_SLEEPING;
1663         spin_unlock(&vc->lock);
1664         schedule();
1665         finish_wait(&vc->wq, &wait);
1666         spin_lock(&vc->lock);
1667         vc->vcore_state = VCORE_INACTIVE;
1668 }
1669
1670 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1671 {
1672         int n_ceded;
1673         struct kvmppc_vcore *vc;
1674         struct kvm_vcpu *v, *vn;
1675
1676         kvm_run->exit_reason = 0;
1677         vcpu->arch.ret = RESUME_GUEST;
1678         vcpu->arch.trap = 0;
1679         kvmppc_update_vpas(vcpu);
1680
1681         /*
1682          * Synchronize with other threads in this virtual core
1683          */
1684         vc = vcpu->arch.vcore;
1685         spin_lock(&vc->lock);
1686         vcpu->arch.ceded = 0;
1687         vcpu->arch.run_task = current;
1688         vcpu->arch.kvm_run = kvm_run;
1689         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
1690         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1691         vcpu->arch.busy_preempt = TB_NIL;
1692         list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1693         ++vc->n_runnable;
1694
1695         /*
1696          * This happens the first time this is called for a vcpu.
1697          * If the vcore is already running, we may be able to start
1698          * this thread straight away and have it join in.
1699          */
1700         if (!signal_pending(current)) {
1701                 if (vc->vcore_state == VCORE_RUNNING &&
1702                     VCORE_EXIT_COUNT(vc) == 0) {
1703                         kvmppc_create_dtl_entry(vcpu, vc);
1704                         kvmppc_start_thread(vcpu);
1705                 } else if (vc->vcore_state == VCORE_SLEEPING) {
1706                         wake_up(&vc->wq);
1707                 }
1708
1709         }
1710
1711         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1712                !signal_pending(current)) {
1713                 if (vc->vcore_state != VCORE_INACTIVE) {
1714                         spin_unlock(&vc->lock);
1715                         kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1716                         spin_lock(&vc->lock);
1717                         continue;
1718                 }
1719                 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1720                                          arch.run_list) {
1721                         kvmppc_core_prepare_to_enter(v);
1722                         if (signal_pending(v->arch.run_task)) {
1723                                 kvmppc_remove_runnable(vc, v);
1724                                 v->stat.signal_exits++;
1725                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1726                                 v->arch.ret = -EINTR;
1727                                 wake_up(&v->arch.cpu_run);
1728                         }
1729                 }
1730                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1731                         break;
1732                 vc->runner = vcpu;
1733                 n_ceded = 0;
1734                 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
1735                         if (!v->arch.pending_exceptions)
1736                                 n_ceded += v->arch.ceded;
1737                         else
1738                                 v->arch.ceded = 0;
1739                 }
1740                 if (n_ceded == vc->n_runnable)
1741                         kvmppc_vcore_blocked(vc);
1742                 else
1743                         kvmppc_run_core(vc);
1744                 vc->runner = NULL;
1745         }
1746
1747         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1748                (vc->vcore_state == VCORE_RUNNING ||
1749                 vc->vcore_state == VCORE_EXITING)) {
1750                 spin_unlock(&vc->lock);
1751                 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1752                 spin_lock(&vc->lock);
1753         }
1754
1755         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1756                 kvmppc_remove_runnable(vc, vcpu);
1757                 vcpu->stat.signal_exits++;
1758                 kvm_run->exit_reason = KVM_EXIT_INTR;
1759                 vcpu->arch.ret = -EINTR;
1760         }
1761
1762         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1763                 /* Wake up some vcpu to run the core */
1764                 v = list_first_entry(&vc->runnable_threads,
1765                                      struct kvm_vcpu, arch.run_list);
1766                 wake_up(&v->arch.cpu_run);
1767         }
1768
1769         spin_unlock(&vc->lock);
1770         return vcpu->arch.ret;
1771 }
1772
1773 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
1774 {
1775         int r;
1776         int srcu_idx;
1777
1778         if (!vcpu->arch.sane) {
1779                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1780                 return -EINVAL;
1781         }
1782
1783         kvmppc_core_prepare_to_enter(vcpu);
1784
1785         /* No need to go into the guest when all we'll do is come back out */
1786         if (signal_pending(current)) {
1787                 run->exit_reason = KVM_EXIT_INTR;
1788                 return -EINTR;
1789         }
1790
1791         atomic_inc(&vcpu->kvm->arch.vcpus_running);
1792         /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1793         smp_mb();
1794
1795         /* On the first time here, set up HTAB and VRMA or RMA */
1796         if (!vcpu->kvm->arch.rma_setup_done) {
1797                 r = kvmppc_hv_setup_htab_rma(vcpu);
1798                 if (r)
1799                         goto out;
1800         }
1801
1802         flush_fp_to_thread(current);
1803         flush_altivec_to_thread(current);
1804         flush_vsx_to_thread(current);
1805         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1806         vcpu->arch.pgdir = current->mm->pgd;
1807         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1808
1809         do {
1810                 r = kvmppc_run_vcpu(run, vcpu);
1811
1812                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1813                     !(vcpu->arch.shregs.msr & MSR_PR)) {
1814                         r = kvmppc_pseries_do_hcall(vcpu);
1815                         kvmppc_core_prepare_to_enter(vcpu);
1816                 } else if (r == RESUME_PAGE_FAULT) {
1817                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1818                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
1819                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1820                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1821                 }
1822         } while (is_kvmppc_resume_guest(r));
1823
1824  out:
1825         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1826         atomic_dec(&vcpu->kvm->arch.vcpus_running);
1827         return r;
1828 }
1829
1830
1831 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1832    Assumes POWER7 or PPC970. */
1833 static inline int lpcr_rmls(unsigned long rma_size)
1834 {
1835         switch (rma_size) {
1836         case 32ul << 20:        /* 32 MB */
1837                 if (cpu_has_feature(CPU_FTR_ARCH_206))
1838                         return 8;       /* only supported on POWER7 */
1839                 return -1;
1840         case 64ul << 20:        /* 64 MB */
1841                 return 3;
1842         case 128ul << 20:       /* 128 MB */
1843                 return 7;
1844         case 256ul << 20:       /* 256 MB */
1845                 return 4;
1846         case 1ul << 30:         /* 1 GB */
1847                 return 2;
1848         case 16ul << 30:        /* 16 GB */
1849                 return 1;
1850         case 256ul << 30:       /* 256 GB */
1851                 return 0;
1852         default:
1853                 return -1;
1854         }
1855 }
1856
1857 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1858 {
1859         struct page *page;
1860         struct kvm_rma_info *ri = vma->vm_file->private_data;
1861
1862         if (vmf->pgoff >= kvm_rma_pages)
1863                 return VM_FAULT_SIGBUS;
1864
1865         page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1866         get_page(page);
1867         vmf->page = page;
1868         return 0;
1869 }
1870
1871 static const struct vm_operations_struct kvm_rma_vm_ops = {
1872         .fault = kvm_rma_fault,
1873 };
1874
1875 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1876 {
1877         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1878         vma->vm_ops = &kvm_rma_vm_ops;
1879         return 0;
1880 }
1881
1882 static int kvm_rma_release(struct inode *inode, struct file *filp)
1883 {
1884         struct kvm_rma_info *ri = filp->private_data;
1885
1886         kvm_release_rma(ri);
1887         return 0;
1888 }
1889
1890 static const struct file_operations kvm_rma_fops = {
1891         .mmap           = kvm_rma_mmap,
1892         .release        = kvm_rma_release,
1893 };
1894
1895 static long kvm_vm_ioctl_allocate_rma(struct kvm *kvm,
1896                                       struct kvm_allocate_rma *ret)
1897 {
1898         long fd;
1899         struct kvm_rma_info *ri;
1900         /*
1901          * Only do this on PPC970 in HV mode
1902          */
1903         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
1904             !cpu_has_feature(CPU_FTR_ARCH_201))
1905                 return -EINVAL;
1906
1907         if (!kvm_rma_pages)
1908                 return -EINVAL;
1909
1910         ri = kvm_alloc_rma();
1911         if (!ri)
1912                 return -ENOMEM;
1913
1914         fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC);
1915         if (fd < 0)
1916                 kvm_release_rma(ri);
1917
1918         ret->rma_size = kvm_rma_pages << PAGE_SHIFT;
1919         return fd;
1920 }
1921
1922 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1923                                      int linux_psize)
1924 {
1925         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1926
1927         if (!def->shift)
1928                 return;
1929         (*sps)->page_shift = def->shift;
1930         (*sps)->slb_enc = def->sllp;
1931         (*sps)->enc[0].page_shift = def->shift;
1932         /*
1933          * Only return base page encoding. We don't want to return
1934          * all the supporting pte_enc, because our H_ENTER doesn't
1935          * support MPSS yet. Once they do, we can start passing all
1936          * support pte_enc here
1937          */
1938         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
1939         /*
1940          * Add 16MB MPSS support if host supports it
1941          */
1942         if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
1943                 (*sps)->enc[1].page_shift = 24;
1944                 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
1945         }
1946         (*sps)++;
1947 }
1948
1949 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
1950                                          struct kvm_ppc_smmu_info *info)
1951 {
1952         struct kvm_ppc_one_seg_page_size *sps;
1953
1954         info->flags = KVM_PPC_PAGE_SIZES_REAL;
1955         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1956                 info->flags |= KVM_PPC_1T_SEGMENTS;
1957         info->slb_size = mmu_slb_size;
1958
1959         /* We only support these sizes for now, and no muti-size segments */
1960         sps = &info->sps[0];
1961         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1962         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1963         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1964
1965         return 0;
1966 }
1967
1968 /*
1969  * Get (and clear) the dirty memory log for a memory slot.
1970  */
1971 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
1972                                          struct kvm_dirty_log *log)
1973 {
1974         struct kvm_memory_slot *memslot;
1975         int r;
1976         unsigned long n;
1977
1978         mutex_lock(&kvm->slots_lock);
1979
1980         r = -EINVAL;
1981         if (log->slot >= KVM_USER_MEM_SLOTS)
1982                 goto out;
1983
1984         memslot = id_to_memslot(kvm->memslots, log->slot);
1985         r = -ENOENT;
1986         if (!memslot->dirty_bitmap)
1987                 goto out;
1988
1989         n = kvm_dirty_bitmap_bytes(memslot);
1990         memset(memslot->dirty_bitmap, 0, n);
1991
1992         r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
1993         if (r)
1994                 goto out;
1995
1996         r = -EFAULT;
1997         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1998                 goto out;
1999
2000         r = 0;
2001 out:
2002         mutex_unlock(&kvm->slots_lock);
2003         return r;
2004 }
2005
2006 static void unpin_slot(struct kvm_memory_slot *memslot)
2007 {
2008         unsigned long *physp;
2009         unsigned long j, npages, pfn;
2010         struct page *page;
2011
2012         physp = memslot->arch.slot_phys;
2013         npages = memslot->npages;
2014         if (!physp)
2015                 return;
2016         for (j = 0; j < npages; j++) {
2017                 if (!(physp[j] & KVMPPC_GOT_PAGE))
2018                         continue;
2019                 pfn = physp[j] >> PAGE_SHIFT;
2020                 page = pfn_to_page(pfn);
2021                 SetPageDirty(page);
2022                 put_page(page);
2023         }
2024 }
2025
2026 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2027                                         struct kvm_memory_slot *dont)
2028 {
2029         if (!dont || free->arch.rmap != dont->arch.rmap) {
2030                 vfree(free->arch.rmap);
2031                 free->arch.rmap = NULL;
2032         }
2033         if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
2034                 unpin_slot(free);
2035                 vfree(free->arch.slot_phys);
2036                 free->arch.slot_phys = NULL;
2037         }
2038 }
2039
2040 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2041                                          unsigned long npages)
2042 {
2043         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2044         if (!slot->arch.rmap)
2045                 return -ENOMEM;
2046         slot->arch.slot_phys = NULL;
2047
2048         return 0;
2049 }
2050
2051 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2052                                         struct kvm_memory_slot *memslot,
2053                                         struct kvm_userspace_memory_region *mem)
2054 {
2055         unsigned long *phys;
2056
2057         /* Allocate a slot_phys array if needed */
2058         phys = memslot->arch.slot_phys;
2059         if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
2060                 phys = vzalloc(memslot->npages * sizeof(unsigned long));
2061                 if (!phys)
2062                         return -ENOMEM;
2063                 memslot->arch.slot_phys = phys;
2064         }
2065
2066         return 0;
2067 }
2068
2069 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2070                                 struct kvm_userspace_memory_region *mem,
2071                                 const struct kvm_memory_slot *old)
2072 {
2073         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2074         struct kvm_memory_slot *memslot;
2075
2076         if (npages && old->npages) {
2077                 /*
2078                  * If modifying a memslot, reset all the rmap dirty bits.
2079                  * If this is a new memslot, we don't need to do anything
2080                  * since the rmap array starts out as all zeroes,
2081                  * i.e. no pages are dirty.
2082                  */
2083                 memslot = id_to_memslot(kvm->memslots, mem->slot);
2084                 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2085         }
2086 }
2087
2088 /*
2089  * Update LPCR values in kvm->arch and in vcores.
2090  * Caller must hold kvm->lock.
2091  */
2092 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2093 {
2094         long int i;
2095         u32 cores_done = 0;
2096
2097         if ((kvm->arch.lpcr & mask) == lpcr)
2098                 return;
2099
2100         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2101
2102         for (i = 0; i < KVM_MAX_VCORES; ++i) {
2103                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2104                 if (!vc)
2105                         continue;
2106                 spin_lock(&vc->lock);
2107                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2108                 spin_unlock(&vc->lock);
2109                 if (++cores_done >= kvm->arch.online_vcores)
2110                         break;
2111         }
2112 }
2113
2114 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2115 {
2116         return;
2117 }
2118
2119 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2120 {
2121         int err = 0;
2122         struct kvm *kvm = vcpu->kvm;
2123         struct kvm_rma_info *ri = NULL;
2124         unsigned long hva;
2125         struct kvm_memory_slot *memslot;
2126         struct vm_area_struct *vma;
2127         unsigned long lpcr = 0, senc;
2128         unsigned long lpcr_mask = 0;
2129         unsigned long psize, porder;
2130         unsigned long rma_size;
2131         unsigned long rmls;
2132         unsigned long *physp;
2133         unsigned long i, npages;
2134         int srcu_idx;
2135
2136         mutex_lock(&kvm->lock);
2137         if (kvm->arch.rma_setup_done)
2138                 goto out;       /* another vcpu beat us to it */
2139
2140         /* Allocate hashed page table (if not done already) and reset it */
2141         if (!kvm->arch.hpt_virt) {
2142                 err = kvmppc_alloc_hpt(kvm, NULL);
2143                 if (err) {
2144                         pr_err("KVM: Couldn't alloc HPT\n");
2145                         goto out;
2146                 }
2147         }
2148
2149         /* Look up the memslot for guest physical address 0 */
2150         srcu_idx = srcu_read_lock(&kvm->srcu);
2151         memslot = gfn_to_memslot(kvm, 0);
2152
2153         /* We must have some memory at 0 by now */
2154         err = -EINVAL;
2155         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2156                 goto out_srcu;
2157
2158         /* Look up the VMA for the start of this memory slot */
2159         hva = memslot->userspace_addr;
2160         down_read(&current->mm->mmap_sem);
2161         vma = find_vma(current->mm, hva);
2162         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2163                 goto up_out;
2164
2165         psize = vma_kernel_pagesize(vma);
2166         porder = __ilog2(psize);
2167
2168         /* Is this one of our preallocated RMAs? */
2169         if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
2170             hva == vma->vm_start)
2171                 ri = vma->vm_file->private_data;
2172
2173         up_read(&current->mm->mmap_sem);
2174
2175         if (!ri) {
2176                 /* On POWER7, use VRMA; on PPC970, give up */
2177                 err = -EPERM;
2178                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2179                         pr_err("KVM: CPU requires an RMO\n");
2180                         goto out_srcu;
2181                 }
2182
2183                 /* We can handle 4k, 64k or 16M pages in the VRMA */
2184                 err = -EINVAL;
2185                 if (!(psize == 0x1000 || psize == 0x10000 ||
2186                       psize == 0x1000000))
2187                         goto out_srcu;
2188
2189                 /* Update VRMASD field in the LPCR */
2190                 senc = slb_pgsize_encoding(psize);
2191                 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2192                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
2193                 lpcr_mask = LPCR_VRMASD;
2194                 /* the -4 is to account for senc values starting at 0x10 */
2195                 lpcr = senc << (LPCR_VRMASD_SH - 4);
2196
2197                 /* Create HPTEs in the hash page table for the VRMA */
2198                 kvmppc_map_vrma(vcpu, memslot, porder);
2199
2200         } else {
2201                 /* Set up to use an RMO region */
2202                 rma_size = kvm_rma_pages;
2203                 if (rma_size > memslot->npages)
2204                         rma_size = memslot->npages;
2205                 rma_size <<= PAGE_SHIFT;
2206                 rmls = lpcr_rmls(rma_size);
2207                 err = -EINVAL;
2208                 if ((long)rmls < 0) {
2209                         pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
2210                         goto out_srcu;
2211                 }
2212                 atomic_inc(&ri->use_count);
2213                 kvm->arch.rma = ri;
2214
2215                 /* Update LPCR and RMOR */
2216                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2217                         /* PPC970; insert RMLS value (split field) in HID4 */
2218                         lpcr_mask = (1ul << HID4_RMLS0_SH) |
2219                                 (3ul << HID4_RMLS2_SH) | HID4_RMOR;
2220                         lpcr = ((rmls >> 2) << HID4_RMLS0_SH) |
2221                                 ((rmls & 3) << HID4_RMLS2_SH);
2222                         /* RMOR is also in HID4 */
2223                         lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
2224                                 << HID4_RMOR_SH;
2225                 } else {
2226                         /* POWER7 */
2227                         lpcr_mask = LPCR_VPM0 | LPCR_VRMA_L | LPCR_RMLS;
2228                         lpcr = rmls << LPCR_RMLS_SH;
2229                         kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT;
2230                 }
2231                 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
2232                         ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
2233
2234                 /* Initialize phys addrs of pages in RMO */
2235                 npages = kvm_rma_pages;
2236                 porder = __ilog2(npages);
2237                 physp = memslot->arch.slot_phys;
2238                 if (physp) {
2239                         if (npages > memslot->npages)
2240                                 npages = memslot->npages;
2241                         spin_lock(&kvm->arch.slot_phys_lock);
2242                         for (i = 0; i < npages; ++i)
2243                                 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
2244                                         porder;
2245                         spin_unlock(&kvm->arch.slot_phys_lock);
2246                 }
2247         }
2248
2249         kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
2250
2251         /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
2252         smp_wmb();
2253         kvm->arch.rma_setup_done = 1;
2254         err = 0;
2255  out_srcu:
2256         srcu_read_unlock(&kvm->srcu, srcu_idx);
2257  out:
2258         mutex_unlock(&kvm->lock);
2259         return err;
2260
2261  up_out:
2262         up_read(&current->mm->mmap_sem);
2263         goto out_srcu;
2264 }
2265
2266 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2267 {
2268         unsigned long lpcr, lpid;
2269
2270         /* Allocate the guest's logical partition ID */
2271
2272         lpid = kvmppc_alloc_lpid();
2273         if ((long)lpid < 0)
2274                 return -ENOMEM;
2275         kvm->arch.lpid = lpid;
2276
2277         /*
2278          * Since we don't flush the TLB when tearing down a VM,
2279          * and this lpid might have previously been used,
2280          * make sure we flush on each core before running the new VM.
2281          */
2282         cpumask_setall(&kvm->arch.need_tlb_flush);
2283
2284         kvm->arch.rma = NULL;
2285
2286         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
2287
2288         if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2289                 /* PPC970; HID4 is effectively the LPCR */
2290                 kvm->arch.host_lpid = 0;
2291                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
2292                 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
2293                 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
2294                         ((lpid & 0xf) << HID4_LPID5_SH);
2295         } else {
2296                 /* POWER7; init LPCR for virtual RMA mode */
2297                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
2298                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2299                 lpcr &= LPCR_PECE | LPCR_LPES;
2300                 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
2301                         LPCR_VPM0 | LPCR_VPM1;
2302                 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2303                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
2304                 /* On POWER8 turn on online bit to enable PURR/SPURR */
2305                 if (cpu_has_feature(CPU_FTR_ARCH_207S))
2306                         lpcr |= LPCR_ONL;
2307         }
2308         kvm->arch.lpcr = lpcr;
2309
2310         kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
2311         spin_lock_init(&kvm->arch.slot_phys_lock);
2312
2313         /*
2314          * Track that we now have a HV mode VM active. This blocks secondary
2315          * CPU threads from coming online.
2316          */
2317         kvm_hv_vm_activated();
2318
2319         return 0;
2320 }
2321
2322 static void kvmppc_free_vcores(struct kvm *kvm)
2323 {
2324         long int i;
2325
2326         for (i = 0; i < KVM_MAX_VCORES; ++i)
2327                 kfree(kvm->arch.vcores[i]);
2328         kvm->arch.online_vcores = 0;
2329 }
2330
2331 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
2332 {
2333         kvm_hv_vm_deactivated();
2334
2335         kvmppc_free_vcores(kvm);
2336         if (kvm->arch.rma) {
2337                 kvm_release_rma(kvm->arch.rma);
2338                 kvm->arch.rma = NULL;
2339         }
2340
2341         kvmppc_free_hpt(kvm);
2342 }
2343
2344 /* We don't need to emulate any privileged instructions or dcbz */
2345 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2346                                      unsigned int inst, int *advance)
2347 {
2348         return EMULATE_FAIL;
2349 }
2350
2351 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2352                                         ulong spr_val)
2353 {
2354         return EMULATE_FAIL;
2355 }
2356
2357 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2358                                         ulong *spr_val)
2359 {
2360         return EMULATE_FAIL;
2361 }
2362
2363 static int kvmppc_core_check_processor_compat_hv(void)
2364 {
2365         if (!cpu_has_feature(CPU_FTR_HVMODE))
2366                 return -EIO;
2367         return 0;
2368 }
2369
2370 static long kvm_arch_vm_ioctl_hv(struct file *filp,
2371                                  unsigned int ioctl, unsigned long arg)
2372 {
2373         struct kvm *kvm __maybe_unused = filp->private_data;
2374         void __user *argp = (void __user *)arg;
2375         long r;
2376
2377         switch (ioctl) {
2378
2379         case KVM_ALLOCATE_RMA: {
2380                 struct kvm_allocate_rma rma;
2381                 struct kvm *kvm = filp->private_data;
2382
2383                 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
2384                 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
2385                         r = -EFAULT;
2386                 break;
2387         }
2388
2389         case KVM_PPC_ALLOCATE_HTAB: {
2390                 u32 htab_order;
2391
2392                 r = -EFAULT;
2393                 if (get_user(htab_order, (u32 __user *)argp))
2394                         break;
2395                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2396                 if (r)
2397                         break;
2398                 r = -EFAULT;
2399                 if (put_user(htab_order, (u32 __user *)argp))
2400                         break;
2401                 r = 0;
2402                 break;
2403         }
2404
2405         case KVM_PPC_GET_HTAB_FD: {
2406                 struct kvm_get_htab_fd ghf;
2407
2408                 r = -EFAULT;
2409                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
2410                         break;
2411                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2412                 break;
2413         }
2414
2415         default:
2416                 r = -ENOTTY;
2417         }
2418
2419         return r;
2420 }
2421
2422 static struct kvmppc_ops kvm_ops_hv = {
2423         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2424         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2425         .get_one_reg = kvmppc_get_one_reg_hv,
2426         .set_one_reg = kvmppc_set_one_reg_hv,
2427         .vcpu_load   = kvmppc_core_vcpu_load_hv,
2428         .vcpu_put    = kvmppc_core_vcpu_put_hv,
2429         .set_msr     = kvmppc_set_msr_hv,
2430         .vcpu_run    = kvmppc_vcpu_run_hv,
2431         .vcpu_create = kvmppc_core_vcpu_create_hv,
2432         .vcpu_free   = kvmppc_core_vcpu_free_hv,
2433         .check_requests = kvmppc_core_check_requests_hv,
2434         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
2435         .flush_memslot  = kvmppc_core_flush_memslot_hv,
2436         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2437         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
2438         .unmap_hva = kvm_unmap_hva_hv,
2439         .unmap_hva_range = kvm_unmap_hva_range_hv,
2440         .age_hva  = kvm_age_hva_hv,
2441         .test_age_hva = kvm_test_age_hva_hv,
2442         .set_spte_hva = kvm_set_spte_hva_hv,
2443         .mmu_destroy  = kvmppc_mmu_destroy_hv,
2444         .free_memslot = kvmppc_core_free_memslot_hv,
2445         .create_memslot = kvmppc_core_create_memslot_hv,
2446         .init_vm =  kvmppc_core_init_vm_hv,
2447         .destroy_vm = kvmppc_core_destroy_vm_hv,
2448         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2449         .emulate_op = kvmppc_core_emulate_op_hv,
2450         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2451         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2452         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2453         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
2454 };
2455
2456 static int kvmppc_book3s_init_hv(void)
2457 {
2458         int r;
2459         /*
2460          * FIXME!! Do we need to check on all cpus ?
2461          */
2462         r = kvmppc_core_check_processor_compat_hv();
2463         if (r < 0)
2464                 return -ENODEV;
2465
2466         kvm_ops_hv.owner = THIS_MODULE;
2467         kvmppc_hv_ops = &kvm_ops_hv;
2468
2469         r = kvmppc_mmu_hv_init();
2470         return r;
2471 }
2472
2473 static void kvmppc_book3s_exit_hv(void)
2474 {
2475         kvmppc_hv_ops = NULL;
2476 }
2477
2478 module_init(kvmppc_book3s_init_hv);
2479 module_exit(kvmppc_book3s_exit_hv);
2480 MODULE_LICENSE("GPL");
2481 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2482 MODULE_ALIAS("devname:kvm");