arch/powerpc/kvm/book3s_hv.c

   1 /*
   2  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
   3  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
   4  *
   5  * Authors:
   6  *    Paul Mackerras <paulus@au1.ibm.com>
   7  *    Alexander Graf <agraf@suse.de>
   8  *    Kevin Wolf <mail@kevin-wolf.de>
   9  *
  10  * Description: KVM functions specific to running on Book 3S
  11  * processors in hypervisor mode (specifically POWER7 and later).
  12  *
  13  * This file is derived from arch/powerpc/kvm/book3s.c,
  14  * by Alexander Graf <agraf@suse.de>.
  15  *
  16  * This program is free software; you can redistribute it and/or modify
  17  * it under the terms of the GNU General Public License, version 2, as
  18  * published by the Free Software Foundation.
  19  */
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/err.h>
  23 #include <linux/slab.h>
  24 #include <linux/preempt.h>
  25 #include <linux/sched.h>
  26 #include <linux/delay.h>
  27 #include <linux/export.h>
  28 #include <linux/fs.h>
  29 #include <linux/anon_inodes.h>
  30 #include <linux/cpumask.h>
  31 #include <linux/spinlock.h>
  32 #include <linux/page-flags.h>
  33 #include <linux/srcu.h>
  34
  35 #include <asm/reg.h>
  36 #include <asm/cputable.h>
  37 #include <asm/cacheflush.h>
  38 #include <asm/tlbflush.h>
  39 #include <asm/uaccess.h>
  40 #include <asm/io.h>
  41 #include <asm/kvm_ppc.h>
  42 #include <asm/kvm_book3s.h>
  43 #include <asm/mmu_context.h>
  44 #include <asm/lppaca.h>
  45 #include <asm/processor.h>
  46 #include <asm/cputhreads.h>
  47 #include <asm/page.h>
  48 #include <asm/hvcall.h>
  49 #include <asm/switch_to.h>
  50 #include <asm/smp.h>
  51 #include <linux/gfp.h>
  52 #include <linux/vmalloc.h>
  53 #include <linux/highmem.h>
  54 #include <linux/hugetlb.h>
  55
  56 /* #define EXIT_DEBUG */
  57 /* #define EXIT_DEBUG_SIMPLE */
  58 /* #define EXIT_DEBUG_INT */
  59
  60 /* Used to indicate that a guest page fault needs to be handled */
  61 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  62
  63 /* Used as a "null" value for timebase values */
  64 #define TB_NIL  (~(u64)0)
  65
  66 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
  67 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
  68
  69 void kvmppc_fast_vcpu_kick(struct kvm_vcpu *vcpu)
  70 {
  71         int me;
  72         int cpu = vcpu->cpu;
  73         wait_queue_head_t *wqp;
  74
  75         wqp = kvm_arch_vcpu_wq(vcpu);
  76         if (waitqueue_active(wqp)) {
  77                 wake_up_interruptible(wqp);
  78                 ++vcpu->stat.halt_wakeup;
  79         }
  80
  81         me = get_cpu();
  82
  83         /* CPU points to the first thread of the core */
  84         if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
  85 #ifdef CONFIG_KVM_XICS
  86                 int real_cpu = cpu + vcpu->arch.ptid;
  87                 if (paca[real_cpu].kvm_hstate.xics_phys)
  88                         xics_wake_cpu(real_cpu);
  89                 else
  90 #endif
  91                 if (cpu_online(cpu))
  92                         smp_send_reschedule(cpu);
  93         }
  94         put_cpu();
  95 }
  96
  97 /*
  98  * We use the vcpu_load/put functions to measure stolen time.
  99  * Stolen time is counted as time when either the vcpu is able to
 100  * run as part of a virtual core, but the task running the vcore
 101  * is preempted or sleeping, or when the vcpu needs something done
 102  * in the kernel by the task running the vcpu, but that task is
 103  * preempted or sleeping.  Those two things have to be counted
 104  * separately, since one of the vcpu tasks will take on the job
 105  * of running the core, and the other vcpu tasks in the vcore will
 106  * sleep waiting for it to do that, but that sleep shouldn't count
 107  * as stolen time.
 108  *
 109  * Hence we accumulate stolen time when the vcpu can run as part of
 110  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 111  * needs its task to do other things in the kernel (for example,
 112  * service a page fault) in busy_stolen.  We don't accumulate
 113  * stolen time for a vcore when it is inactive, or for a vcpu
 114  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 115  * a misnomer; it means that the vcpu task is not executing in
 116  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 117  * the kernel.  We don't have any way of dividing up that time
 118  * between time that the vcpu is genuinely stopped, time that
 119  * the task is actively working on behalf of the vcpu, and time
 120  * that the task is preempted, so we don't count any of it as
 121  * stolen.
 122  *
 123  * Updates to busy_stolen are protected by arch.tbacct_lock;
 124  * updates to vc->stolen_tb are protected by the arch.tbacct_lock
 125  * of the vcpu that has taken responsibility for running the vcore
 126  * (i.e. vc->runner).  The stolen times are measured in units of
 127  * timebase ticks.  (Note that the != TB_NIL checks below are
 128  * purely defensive; they should never fail.)
 129  */
 130
 131 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 132 {
 133         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 134
 135         spin_lock(&vcpu->arch.tbacct_lock);
 136         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
 137             vc->preempt_tb != TB_NIL) {
 138                 vc->stolen_tb += mftb() - vc->preempt_tb;
 139                 vc->preempt_tb = TB_NIL;
 140         }
 141         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 142             vcpu->arch.busy_preempt != TB_NIL) {
 143                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 144                 vcpu->arch.busy_preempt = TB_NIL;
 145         }
 146         spin_unlock(&vcpu->arch.tbacct_lock);
 147 }
 148
 149 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
 150 {
 151         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 152
 153         spin_lock(&vcpu->arch.tbacct_lock);
 154         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
 155                 vc->preempt_tb = mftb();
 156         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 157                 vcpu->arch.busy_preempt = mftb();
 158         spin_unlock(&vcpu->arch.tbacct_lock);
 159 }
 160
 161 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
 162 {
 163         vcpu->arch.shregs.msr = msr;
 164         kvmppc_end_cede(vcpu);
 165 }
 166
 167 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
 168 {
 169         vcpu->arch.pvr = pvr;
 170 }
 171
 172 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 173 {
 174         int r;
 175
 176         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 177         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 178                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
 179         for (r = 0; r < 16; ++r)
 180                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 181                        r, kvmppc_get_gpr(vcpu, r),
 182                        r+16, kvmppc_get_gpr(vcpu, r+16));
 183         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 184                vcpu->arch.ctr, vcpu->arch.lr);
 185         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 186                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 187         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 188                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 189         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 190                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 191         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 192                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 193         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 194         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 195                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 196         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 197         for (r = 0; r < vcpu->arch.slb_max; ++r)
 198                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 199                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 200         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 201                vcpu->kvm->arch.lpcr, vcpu->kvm->arch.sdr1,
 202                vcpu->arch.last_inst);
 203 }
 204
 205 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 206 {
 207         int r;
 208         struct kvm_vcpu *v, *ret = NULL;
 209
 210         mutex_lock(&kvm->lock);
 211         kvm_for_each_vcpu(r, v, kvm) {
 212                 if (v->vcpu_id == id) {
 213                         ret = v;
 214                         break;
 215                 }
 216         }
 217         mutex_unlock(&kvm->lock);
 218         return ret;
 219 }
 220
 221 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 222 {
 223         vpa->shared_proc = 1;
 224         vpa->yield_count = 1;
 225 }
 226
 227 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 228                    unsigned long addr, unsigned long len)
 229 {
 230         /* check address is cacheline aligned */
 231         if (addr & (L1_CACHE_BYTES - 1))
 232                 return -EINVAL;
 233         spin_lock(&vcpu->arch.vpa_update_lock);
 234         if (v->next_gpa != addr || v->len != len) {
 235                 v->next_gpa = addr;
 236                 v->len = addr ? len : 0;
 237                 v->update_pending = 1;
 238         }
 239         spin_unlock(&vcpu->arch.vpa_update_lock);
 240         return 0;
 241 }
 242
 243 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 244 struct reg_vpa {
 245         u32 dummy;
 246         union {
 247                 u16 hword;
 248                 u32 word;
 249         } length;
 250 };
 251
 252 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 253 {
 254         if (vpap->update_pending)
 255                 return vpap->next_gpa != 0;
 256         return vpap->pinned_addr != NULL;
 257 }
 258
 259 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 260                                        unsigned long flags,
 261                                        unsigned long vcpuid, unsigned long vpa)
 262 {
 263         struct kvm *kvm = vcpu->kvm;
 264         unsigned long len, nb;
 265         void *va;
 266         struct kvm_vcpu *tvcpu;
 267         int err;
 268         int subfunc;
 269         struct kvmppc_vpa *vpap;
 270
 271         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 272         if (!tvcpu)
 273                 return H_PARAMETER;
 274
 275         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 276         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 277             subfunc == H_VPA_REG_SLB) {
 278                 /* Registering new area - address must be cache-line aligned */
 279                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 280                         return H_PARAMETER;
 281
 282                 /* convert logical addr to kernel addr and read length */
 283                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 284                 if (va == NULL)
 285                         return H_PARAMETER;
 286                 if (subfunc == H_VPA_REG_VPA)
 287                         len = ((struct reg_vpa *)va)->length.hword;
 288                 else
 289                         len = ((struct reg_vpa *)va)->length.word;
 290                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 291
 292                 /* Check length */
 293                 if (len > nb || len < sizeof(struct reg_vpa))
 294                         return H_PARAMETER;
 295         } else {
 296                 vpa = 0;
 297                 len = 0;
 298         }
 299
 300         err = H_PARAMETER;
 301         vpap = NULL;
 302         spin_lock(&tvcpu->arch.vpa_update_lock);
 303
 304         switch (subfunc) {
 305         case H_VPA_REG_VPA:             /* register VPA */
 306                 if (len < sizeof(struct lppaca))
 307                         break;
 308                 vpap = &tvcpu->arch.vpa;
 309                 err = 0;
 310                 break;
 311
 312         case H_VPA_REG_DTL:             /* register DTL */
 313                 if (len < sizeof(struct dtl_entry))
 314                         break;
 315                 len -= len % sizeof(struct dtl_entry);
 316
 317                 /* Check that they have previously registered a VPA */
 318                 err = H_RESOURCE;
 319                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 320                         break;
 321
 322                 vpap = &tvcpu->arch.dtl;
 323                 err = 0;
 324                 break;
 325
 326         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 327                 /* Check that they have previously registered a VPA */
 328                 err = H_RESOURCE;
 329                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 330                         break;
 331
 332                 vpap = &tvcpu->arch.slb_shadow;
 333                 err = 0;
 334                 break;
 335
 336         case H_VPA_DEREG_VPA:           /* deregister VPA */
 337                 /* Check they don't still have a DTL or SLB buf registered */
 338                 err = H_RESOURCE;
 339                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 340                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 341                         break;
 342
 343                 vpap = &tvcpu->arch.vpa;
 344                 err = 0;
 345                 break;
 346
 347         case H_VPA_DEREG_DTL:           /* deregister DTL */
 348                 vpap = &tvcpu->arch.dtl;
 349                 err = 0;
 350                 break;
 351
 352         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 353                 vpap = &tvcpu->arch.slb_shadow;
 354                 err = 0;
 355                 break;
 356         }
 357
 358         if (vpap) {
 359                 vpap->next_gpa = vpa;
 360                 vpap->len = len;
 361                 vpap->update_pending = 1;
 362         }
 363
 364         spin_unlock(&tvcpu->arch.vpa_update_lock);
 365
 366         return err;
 367 }
 368
 369 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 370 {
 371         struct kvm *kvm = vcpu->kvm;
 372         void *va;
 373         unsigned long nb;
 374         unsigned long gpa;
 375
 376         /*
 377          * We need to pin the page pointed to by vpap->next_gpa,
 378          * but we can't call kvmppc_pin_guest_page under the lock
 379          * as it does get_user_pages() and down_read().  So we
 380          * have to drop the lock, pin the page, then get the lock
 381          * again and check that a new area didn't get registered
 382          * in the meantime.
 383          */
 384         for (;;) {
 385                 gpa = vpap->next_gpa;
 386                 spin_unlock(&vcpu->arch.vpa_update_lock);
 387                 va = NULL;
 388                 nb = 0;
 389                 if (gpa)
 390                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 391                 spin_lock(&vcpu->arch.vpa_update_lock);
 392                 if (gpa == vpap->next_gpa)
 393                         break;
 394                 /* sigh... unpin that one and try again */
 395                 if (va)
 396                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 397         }
 398
 399         vpap->update_pending = 0;
 400         if (va && nb < vpap->len) {
 401                 /*
 402                  * If it's now too short, it must be that userspace
 403                  * has changed the mappings underlying guest memory,
 404                  * so unregister the region.
 405                  */
 406                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 407                 va = NULL;
 408         }
 409         if (vpap->pinned_addr)
 410                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 411                                         vpap->dirty);
 412         vpap->gpa = gpa;
 413         vpap->pinned_addr = va;
 414         vpap->dirty = false;
 415         if (va)
 416                 vpap->pinned_end = va + vpap->len;
 417 }
 418
 419 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 420 {
 421         if (!(vcpu->arch.vpa.update_pending ||
 422               vcpu->arch.slb_shadow.update_pending ||
 423               vcpu->arch.dtl.update_pending))
 424                 return;
 425
 426         spin_lock(&vcpu->arch.vpa_update_lock);
 427         if (vcpu->arch.vpa.update_pending) {
 428                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 429                 if (vcpu->arch.vpa.pinned_addr)
 430                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 431         }
 432         if (vcpu->arch.dtl.update_pending) {
 433                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 434                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 435                 vcpu->arch.dtl_index = 0;
 436         }
 437         if (vcpu->arch.slb_shadow.update_pending)
 438                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 439         spin_unlock(&vcpu->arch.vpa_update_lock);
 440 }
 441
 442 /*
 443  * Return the accumulated stolen time for the vcore up until `now'.
 444  * The caller should hold the vcore lock.
 445  */
 446 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 447 {
 448         u64 p;
 449
 450         /*
 451          * If we are the task running the vcore, then since we hold
 452          * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
 453          * can't be updated, so we don't need the tbacct_lock.
 454          * If the vcore is inactive, it can't become active (since we
 455          * hold the vcore lock), so the vcpu load/put functions won't
 456          * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
 457          */
 458         if (vc->vcore_state != VCORE_INACTIVE &&
 459             vc->runner->arch.run_task != current) {
 460                 spin_lock(&vc->runner->arch.tbacct_lock);
 461                 p = vc->stolen_tb;
 462                 if (vc->preempt_tb != TB_NIL)
 463                         p += now - vc->preempt_tb;
 464                 spin_unlock(&vc->runner->arch.tbacct_lock);
 465         } else {
 466                 p = vc->stolen_tb;
 467         }
 468         return p;
 469 }
 470
 471 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 472                                     struct kvmppc_vcore *vc)
 473 {
 474         struct dtl_entry *dt;
 475         struct lppaca *vpa;
 476         unsigned long stolen;
 477         unsigned long core_stolen;
 478         u64 now;
 479
 480         dt = vcpu->arch.dtl_ptr;
 481         vpa = vcpu->arch.vpa.pinned_addr;
 482         now = mftb();
 483         core_stolen = vcore_stolen_time(vc, now);
 484         stolen = core_stolen - vcpu->arch.stolen_logged;
 485         vcpu->arch.stolen_logged = core_stolen;
 486         spin_lock(&vcpu->arch.tbacct_lock);
 487         stolen += vcpu->arch.busy_stolen;
 488         vcpu->arch.busy_stolen = 0;
 489         spin_unlock(&vcpu->arch.tbacct_lock);
 490         if (!dt || !vpa)
 491                 return;
 492         memset(dt, 0, sizeof(struct dtl_entry));
 493         dt->dispatch_reason = 7;
 494         dt->processor_id = vc->pcpu + vcpu->arch.ptid;
 495         dt->timebase = now;
 496         dt->enqueue_to_dispatch_time = stolen;
 497         dt->srr0 = kvmppc_get_pc(vcpu);
 498         dt->srr1 = vcpu->arch.shregs.msr;
 499         ++dt;
 500         if (dt == vcpu->arch.dtl.pinned_end)
 501                 dt = vcpu->arch.dtl.pinned_addr;
 502         vcpu->arch.dtl_ptr = dt;
 503         /* order writing *dt vs. writing vpa->dtl_idx */
 504         smp_wmb();
 505         vpa->dtl_idx = ++vcpu->arch.dtl_index;
 506         vcpu->arch.dtl.dirty = true;
 507 }
 508
 509 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 510 {
 511         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 512         unsigned long target, ret = H_SUCCESS;
 513         struct kvm_vcpu *tvcpu;
 514         int idx, rc;
 515
 516         switch (req) {
 517         case H_ENTER:
 518                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 519                 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
 520                                               kvmppc_get_gpr(vcpu, 5),
 521                                               kvmppc_get_gpr(vcpu, 6),
 522                                               kvmppc_get_gpr(vcpu, 7));
 523                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 524                 break;
 525         case H_CEDE:
 526                 break;
 527         case H_PROD:
 528                 target = kvmppc_get_gpr(vcpu, 4);
 529                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 530                 if (!tvcpu) {
 531                         ret = H_PARAMETER;
 532                         break;
 533                 }
 534                 tvcpu->arch.prodded = 1;
 535                 smp_mb();
 536                 if (vcpu->arch.ceded) {
 537                         if (waitqueue_active(&vcpu->wq)) {
 538                                 wake_up_interruptible(&vcpu->wq);
 539                                 vcpu->stat.halt_wakeup++;
 540                         }
 541                 }
 542                 break;
 543         case H_CONFER:
 544                 break;
 545         case H_REGISTER_VPA:
 546                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 547                                         kvmppc_get_gpr(vcpu, 5),
 548                                         kvmppc_get_gpr(vcpu, 6));
 549                 break;
 550         case H_RTAS:
 551                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 552                         return RESUME_HOST;
 553
 554                 rc = kvmppc_rtas_hcall(vcpu);
 555
 556                 if (rc == -ENOENT)
 557                         return RESUME_HOST;
 558                 else if (rc == 0)
 559                         break;
 560
 561                 /* Send the error out to userspace via KVM_RUN */
 562                 return rc;
 563
 564         case H_XIRR:
 565         case H_CPPR:
 566         case H_EOI:
 567         case H_IPI:
 568         case H_IPOLL:
 569         case H_XIRR_X:
 570                 if (kvmppc_xics_enabled(vcpu)) {
 571                         ret = kvmppc_xics_hcall(vcpu, req);
 572                         break;
 573                 } /* fallthrough */
 574         default:
 575                 return RESUME_HOST;
 576         }
 577         kvmppc_set_gpr(vcpu, 3, ret);
 578         vcpu->arch.hcall_needed = 0;
 579         return RESUME_GUEST;
 580 }
 581
 582 static int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 583                               struct task_struct *tsk)
 584 {
 585         int r = RESUME_HOST;
 586
 587         vcpu->stat.sum_exits++;
 588
 589         run->exit_reason = KVM_EXIT_UNKNOWN;
 590         run->ready_for_interrupt_injection = 1;
 591         switch (vcpu->arch.trap) {
 592         /* We're good on these - the host merely wanted to get our attention */
 593         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 594                 vcpu->stat.dec_exits++;
 595                 r = RESUME_GUEST;
 596                 break;
 597         case BOOK3S_INTERRUPT_EXTERNAL:
 598                 vcpu->stat.ext_intr_exits++;
 599                 r = RESUME_GUEST;
 600                 break;
 601         case BOOK3S_INTERRUPT_PERFMON:
 602                 r = RESUME_GUEST;
 603                 break;
 604         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 605                 /*
 606                  * Deliver a machine check interrupt to the guest.
 607                  * We have to do this, even if the host has handled the
 608                  * machine check, because machine checks use SRR0/1 and
 609                  * the interrupt might have trashed guest state in them.
 610                  */
 611                 kvmppc_book3s_queue_irqprio(vcpu,
 612                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 613                 r = RESUME_GUEST;
 614                 break;
 615         case BOOK3S_INTERRUPT_PROGRAM:
 616         {
 617                 ulong flags;
 618                 /*
 619                  * Normally program interrupts are delivered directly
 620                  * to the guest by the hardware, but we can get here
 621                  * as a result of a hypervisor emulation interrupt
 622                  * (e40) getting turned into a 700 by BML RTAS.
 623                  */
 624                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 625                 kvmppc_core_queue_program(vcpu, flags);
 626                 r = RESUME_GUEST;
 627                 break;
 628         }
 629         case BOOK3S_INTERRUPT_SYSCALL:
 630         {
 631                 /* hcall - punt to userspace */
 632                 int i;
 633
 634                 if (vcpu->arch.shregs.msr & MSR_PR) {
 635                         /* sc 1 from userspace - reflect to guest syscall */
 636                         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_SYSCALL);
 637                         r = RESUME_GUEST;
 638                         break;
 639                 }
 640                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 641                 for (i = 0; i < 9; ++i)
 642                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 643                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 644                 vcpu->arch.hcall_needed = 1;
 645                 r = RESUME_HOST;
 646                 break;
 647         }
 648         /*
 649          * We get these next two if the guest accesses a page which it thinks
 650          * it has mapped but which is not actually present, either because
 651          * it is for an emulated I/O device or because the corresonding
 652          * host page has been paged out.  Any other HDSI/HISI interrupts
 653          * have been handled already.
 654          */
 655         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 656                 r = RESUME_PAGE_FAULT;
 657                 break;
 658         case BOOK3S_INTERRUPT_H_INST_STORAGE:
 659                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
 660                 vcpu->arch.fault_dsisr = 0;
 661                 r = RESUME_PAGE_FAULT;
 662                 break;
 663         /*
 664          * This occurs if the guest executes an illegal instruction.
 665          * We just generate a program interrupt to the guest, since
 666          * we don't emulate any guest instructions at this stage.
 667          */
 668         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 669                 kvmppc_core_queue_program(vcpu, 0x80000);
 670                 r = RESUME_GUEST;
 671                 break;
 672         default:
 673                 kvmppc_dump_regs(vcpu);
 674                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 675                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 676                         vcpu->arch.shregs.msr);
 677                 r = RESUME_HOST;
 678                 BUG();
 679                 break;
 680         }
 681
 682         return r;
 683 }
 684
 685 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 686                                   struct kvm_sregs *sregs)
 687 {
 688         int i;
 689
 690         sregs->pvr = vcpu->arch.pvr;
 691
 692         memset(sregs, 0, sizeof(struct kvm_sregs));
 693         for (i = 0; i < vcpu->arch.slb_max; i++) {
 694                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
 695                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 696         }
 697
 698         return 0;
 699 }
 700
 701 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 702                                   struct kvm_sregs *sregs)
 703 {
 704         int i, j;
 705
 706         kvmppc_set_pvr(vcpu, sregs->pvr);
 707
 708         j = 0;
 709         for (i = 0; i < vcpu->arch.slb_nr; i++) {
 710                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
 711                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
 712                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
 713                         ++j;
 714                 }
 715         }
 716         vcpu->arch.slb_max = j;
 717
 718         return 0;
 719 }
 720
 721 int kvmppc_get_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
 722 {
 723         int r = 0;
 724         long int i;
 725
 726         switch (id) {
 727         case KVM_REG_PPC_HIOR:
 728                 *val = get_reg_val(id, 0);
 729                 break;
 730         case KVM_REG_PPC_DABR:
 731                 *val = get_reg_val(id, vcpu->arch.dabr);
 732                 break;
 733         case KVM_REG_PPC_DSCR:
 734                 *val = get_reg_val(id, vcpu->arch.dscr);
 735                 break;
 736         case KVM_REG_PPC_PURR:
 737                 *val = get_reg_val(id, vcpu->arch.purr);
 738                 break;
 739         case KVM_REG_PPC_SPURR:
 740                 *val = get_reg_val(id, vcpu->arch.spurr);
 741                 break;
 742         case KVM_REG_PPC_AMR:
 743                 *val = get_reg_val(id, vcpu->arch.amr);
 744                 break;
 745         case KVM_REG_PPC_UAMOR:
 746                 *val = get_reg_val(id, vcpu->arch.uamor);
 747                 break;
 748         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
 749                 i = id - KVM_REG_PPC_MMCR0;
 750                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
 751                 break;
 752         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
 753                 i = id - KVM_REG_PPC_PMC1;
 754                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
 755                 break;
 756 #ifdef CONFIG_VSX
 757         case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
 758                 if (cpu_has_feature(CPU_FTR_VSX)) {
 759                         /* VSX => FP reg i is stored in arch.vsr[2*i] */
 760                         long int i = id - KVM_REG_PPC_FPR0;
 761                         *val = get_reg_val(id, vcpu->arch.vsr[2 * i]);
 762                 } else {
 763                         /* let generic code handle it */
 764                         r = -EINVAL;
 765                 }
 766                 break;
 767         case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
 768                 if (cpu_has_feature(CPU_FTR_VSX)) {
 769                         long int i = id - KVM_REG_PPC_VSR0;
 770                         val->vsxval[0] = vcpu->arch.vsr[2 * i];
 771                         val->vsxval[1] = vcpu->arch.vsr[2 * i + 1];
 772                 } else {
 773                         r = -ENXIO;
 774                 }
 775                 break;
 776 #endif /* CONFIG_VSX */
 777         case KVM_REG_PPC_VPA_ADDR:
 778                 spin_lock(&vcpu->arch.vpa_update_lock);
 779                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
 780                 spin_unlock(&vcpu->arch.vpa_update_lock);
 781                 break;
 782         case KVM_REG_PPC_VPA_SLB:
 783                 spin_lock(&vcpu->arch.vpa_update_lock);
 784                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
 785                 val->vpaval.length = vcpu->arch.slb_shadow.len;
 786                 spin_unlock(&vcpu->arch.vpa_update_lock);
 787                 break;
 788         case KVM_REG_PPC_VPA_DTL:
 789                 spin_lock(&vcpu->arch.vpa_update_lock);
 790                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
 791                 val->vpaval.length = vcpu->arch.dtl.len;
 792                 spin_unlock(&vcpu->arch.vpa_update_lock);
 793                 break;
 794         default:
 795                 r = -EINVAL;
 796                 break;
 797         }
 798
 799         return r;
 800 }
 801
 802 int kvmppc_set_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
 803 {
 804         int r = 0;
 805         long int i;
 806         unsigned long addr, len;
 807
 808         switch (id) {
 809         case KVM_REG_PPC_HIOR:
 810                 /* Only allow this to be set to zero */
 811                 if (set_reg_val(id, *val))
 812                         r = -EINVAL;
 813                 break;
 814         case KVM_REG_PPC_DABR:
 815                 vcpu->arch.dabr = set_reg_val(id, *val);
 816                 break;
 817         case KVM_REG_PPC_DSCR:
 818                 vcpu->arch.dscr = set_reg_val(id, *val);
 819                 break;
 820         case KVM_REG_PPC_PURR:
 821                 vcpu->arch.purr = set_reg_val(id, *val);
 822                 break;
 823         case KVM_REG_PPC_SPURR:
 824                 vcpu->arch.spurr = set_reg_val(id, *val);
 825                 break;
 826         case KVM_REG_PPC_AMR:
 827                 vcpu->arch.amr = set_reg_val(id, *val);
 828                 break;
 829         case KVM_REG_PPC_UAMOR:
 830                 vcpu->arch.uamor = set_reg_val(id, *val);
 831                 break;
 832         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
 833                 i = id - KVM_REG_PPC_MMCR0;
 834                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
 835                 break;
 836         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
 837                 i = id - KVM_REG_PPC_PMC1;
 838                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
 839                 break;
 840 #ifdef CONFIG_VSX
 841         case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
 842                 if (cpu_has_feature(CPU_FTR_VSX)) {
 843                         /* VSX => FP reg i is stored in arch.vsr[2*i] */
 844                         long int i = id - KVM_REG_PPC_FPR0;
 845                         vcpu->arch.vsr[2 * i] = set_reg_val(id, *val);
 846                 } else {
 847                         /* let generic code handle it */
 848                         r = -EINVAL;
 849                 }
 850                 break;
 851         case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
 852                 if (cpu_has_feature(CPU_FTR_VSX)) {
 853                         long int i = id - KVM_REG_PPC_VSR0;
 854                         vcpu->arch.vsr[2 * i] = val->vsxval[0];
 855                         vcpu->arch.vsr[2 * i + 1] = val->vsxval[1];
 856                 } else {
 857                         r = -ENXIO;
 858                 }
 859                 break;
 860 #endif /* CONFIG_VSX */
 861         case KVM_REG_PPC_VPA_ADDR:
 862                 addr = set_reg_val(id, *val);
 863                 r = -EINVAL;
 864                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
 865                               vcpu->arch.dtl.next_gpa))
 866                         break;
 867                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
 868                 break;
 869         case KVM_REG_PPC_VPA_SLB:
 870                 addr = val->vpaval.addr;
 871                 len = val->vpaval.length;
 872                 r = -EINVAL;
 873                 if (addr && !vcpu->arch.vpa.next_gpa)
 874                         break;
 875                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
 876                 break;
 877         case KVM_REG_PPC_VPA_DTL:
 878                 addr = val->vpaval.addr;
 879                 len = val->vpaval.length;
 880                 r = -EINVAL;
 881                 if (addr && (len < sizeof(struct dtl_entry) ||
 882                              !vcpu->arch.vpa.next_gpa))
 883                         break;
 884                 len -= len % sizeof(struct dtl_entry);
 885                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
 886                 break;
 887         default:
 888                 r = -EINVAL;
 889                 break;
 890         }
 891
 892         return r;
 893 }
 894
 895 int kvmppc_core_check_processor_compat(void)
 896 {
 897         if (cpu_has_feature(CPU_FTR_HVMODE))
 898                 return 0;
 899         return -EIO;
 900 }
 901
 902 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
 903 {
 904         struct kvm_vcpu *vcpu;
 905         int err = -EINVAL;
 906         int core;
 907         struct kvmppc_vcore *vcore;
 908
 909         core = id / threads_per_core;
 910         if (core >= KVM_MAX_VCORES)
 911                 goto out;
 912
 913         err = -ENOMEM;
 914         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 915         if (!vcpu)
 916                 goto out;
 917
 918         err = kvm_vcpu_init(vcpu, kvm, id);
 919         if (err)
 920                 goto free_vcpu;
 921
 922         vcpu->arch.shared = &vcpu->arch.shregs;
 923         vcpu->arch.mmcr[0] = MMCR0_FC;
 924         vcpu->arch.ctrl = CTRL_RUNLATCH;
 925         /* default to host PVR, since we can't spoof it */
 926         vcpu->arch.pvr = mfspr(SPRN_PVR);
 927         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
 928         spin_lock_init(&vcpu->arch.vpa_update_lock);
 929         spin_lock_init(&vcpu->arch.tbacct_lock);
 930         vcpu->arch.busy_preempt = TB_NIL;
 931
 932         kvmppc_mmu_book3s_hv_init(vcpu);
 933
 934         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
 935
 936         init_waitqueue_head(&vcpu->arch.cpu_run);
 937
 938         mutex_lock(&kvm->lock);
 939         vcore = kvm->arch.vcores[core];
 940         if (!vcore) {
 941                 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
 942                 if (vcore) {
 943                         INIT_LIST_HEAD(&vcore->runnable_threads);
 944                         spin_lock_init(&vcore->lock);
 945                         init_waitqueue_head(&vcore->wq);
 946                         vcore->preempt_tb = TB_NIL;
 947                 }
 948                 kvm->arch.vcores[core] = vcore;
 949                 kvm->arch.online_vcores++;
 950         }
 951         mutex_unlock(&kvm->lock);
 952
 953         if (!vcore)
 954                 goto free_vcpu;
 955
 956         spin_lock(&vcore->lock);
 957         ++vcore->num_threads;
 958         spin_unlock(&vcore->lock);
 959         vcpu->arch.vcore = vcore;
 960
 961         vcpu->arch.cpu_type = KVM_CPU_3S_64;
 962         kvmppc_sanity_check(vcpu);
 963
 964         return vcpu;
 965
 966 free_vcpu:
 967         kmem_cache_free(kvm_vcpu_cache, vcpu);
 968 out:
 969         return ERR_PTR(err);
 970 }
 971
 972 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
 973 {
 974         if (vpa->pinned_addr)
 975                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
 976                                         vpa->dirty);
 977 }
 978
 979 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
 980 {
 981         spin_lock(&vcpu->arch.vpa_update_lock);
 982         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
 983         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
 984         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
 985         spin_unlock(&vcpu->arch.vpa_update_lock);
 986         kvm_vcpu_uninit(vcpu);
 987         kmem_cache_free(kvm_vcpu_cache, vcpu);
 988 }
 989
 990 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
 991 {
 992         unsigned long dec_nsec, now;
 993
 994         now = get_tb();
 995         if (now > vcpu->arch.dec_expires) {
 996                 /* decrementer has already gone negative */
 997                 kvmppc_core_queue_dec(vcpu);
 998                 kvmppc_core_prepare_to_enter(vcpu);
 999                 return;
1000         }
1001         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1002                    / tb_ticks_per_sec;
1003         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1004                       HRTIMER_MODE_REL);
1005         vcpu->arch.timer_running = 1;
1006 }
1007
1008 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1009 {
1010         vcpu->arch.ceded = 0;
1011         if (vcpu->arch.timer_running) {
1012                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1013                 vcpu->arch.timer_running = 0;
1014         }
1015 }
1016
1017 extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
1018
1019 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1020                                    struct kvm_vcpu *vcpu)
1021 {
1022         u64 now;
1023
1024         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1025                 return;
1026         spin_lock(&vcpu->arch.tbacct_lock);
1027         now = mftb();
1028         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1029                 vcpu->arch.stolen_logged;
1030         vcpu->arch.busy_preempt = now;
1031         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1032         spin_unlock(&vcpu->arch.tbacct_lock);
1033         --vc->n_runnable;
1034         list_del(&vcpu->arch.run_list);
1035 }
1036
1037 static int kvmppc_grab_hwthread(int cpu)
1038 {
1039         struct paca_struct *tpaca;
1040         long timeout = 1000;
1041
1042         tpaca = &paca[cpu];
1043
1044         /* Ensure the thread won't go into the kernel if it wakes */
1045         tpaca->kvm_hstate.hwthread_req = 1;
1046         tpaca->kvm_hstate.kvm_vcpu = NULL;
1047
1048         /*
1049          * If the thread is already executing in the kernel (e.g. handling
1050          * a stray interrupt), wait for it to get back to nap mode.
1051          * The smp_mb() is to ensure that our setting of hwthread_req
1052          * is visible before we look at hwthread_state, so if this
1053          * races with the code at system_reset_pSeries and the thread
1054          * misses our setting of hwthread_req, we are sure to see its
1055          * setting of hwthread_state, and vice versa.
1056          */
1057         smp_mb();
1058         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1059                 if (--timeout <= 0) {
1060                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1061                         return -EBUSY;
1062                 }
1063                 udelay(1);
1064         }
1065         return 0;
1066 }
1067
1068 static void kvmppc_release_hwthread(int cpu)
1069 {
1070         struct paca_struct *tpaca;
1071
1072         tpaca = &paca[cpu];
1073         tpaca->kvm_hstate.hwthread_req = 0;
1074         tpaca->kvm_hstate.kvm_vcpu = NULL;
1075 }
1076
1077 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1078 {
1079         int cpu;
1080         struct paca_struct *tpaca;
1081         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1082
1083         if (vcpu->arch.timer_running) {
1084                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1085                 vcpu->arch.timer_running = 0;
1086         }
1087         cpu = vc->pcpu + vcpu->arch.ptid;
1088         tpaca = &paca[cpu];
1089         tpaca->kvm_hstate.kvm_vcpu = vcpu;
1090         tpaca->kvm_hstate.kvm_vcore = vc;
1091         tpaca->kvm_hstate.napping = 0;
1092         vcpu->cpu = vc->pcpu;
1093         smp_wmb();
1094 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1095         if (vcpu->arch.ptid) {
1096 #ifdef CONFIG_KVM_XICS
1097                 xics_wake_cpu(cpu);
1098 #endif
1099                 ++vc->n_woken;
1100         }
1101 #endif
1102 }
1103
1104 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1105 {
1106         int i;
1107
1108         HMT_low();
1109         i = 0;
1110         while (vc->nap_count < vc->n_woken) {
1111                 if (++i >= 1000000) {
1112                         pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1113                                vc->nap_count, vc->n_woken);
1114                         break;
1115                 }
1116                 cpu_relax();
1117         }
1118         HMT_medium();
1119 }
1120
1121 /*
1122  * Check that we are on thread 0 and that any other threads in
1123  * this core are off-line.  Then grab the threads so they can't
1124  * enter the kernel.
1125  */
1126 static int on_primary_thread(void)
1127 {
1128         int cpu = smp_processor_id();
1129         int thr = cpu_thread_in_core(cpu);
1130
1131         if (thr)
1132                 return 0;
1133         while (++thr < threads_per_core)
1134                 if (cpu_online(cpu + thr))
1135                         return 0;
1136
1137         /* Grab all hw threads so they can't go into the kernel */
1138         for (thr = 1; thr < threads_per_core; ++thr) {
1139                 if (kvmppc_grab_hwthread(cpu + thr)) {
1140                         /* Couldn't grab one; let the others go */
1141                         do {
1142                                 kvmppc_release_hwthread(cpu + thr);
1143                         } while (--thr > 0);
1144                         return 0;
1145                 }
1146         }
1147         return 1;
1148 }
1149
1150 /*
1151  * Run a set of guest threads on a physical core.
1152  * Called with vc->lock held.
1153  */
1154 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1155 {
1156         struct kvm_vcpu *vcpu, *vcpu0, *vnext;
1157         long ret;
1158         u64 now;
1159         int ptid, i, need_vpa_update;
1160         int srcu_idx;
1161         struct kvm_vcpu *vcpus_to_update[threads_per_core];
1162
1163         /* don't start if any threads have a signal pending */
1164         need_vpa_update = 0;
1165         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1166                 if (signal_pending(vcpu->arch.run_task))
1167                         return;
1168                 if (vcpu->arch.vpa.update_pending ||
1169                     vcpu->arch.slb_shadow.update_pending ||
1170                     vcpu->arch.dtl.update_pending)
1171                         vcpus_to_update[need_vpa_update++] = vcpu;
1172         }
1173
1174         /*
1175          * Initialize *vc, in particular vc->vcore_state, so we can
1176          * drop the vcore lock if necessary.
1177          */
1178         vc->n_woken = 0;
1179         vc->nap_count = 0;
1180         vc->entry_exit_count = 0;
1181         vc->vcore_state = VCORE_STARTING;
1182         vc->in_guest = 0;
1183         vc->napping_threads = 0;
1184
1185         /*
1186          * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1187          * which can't be called with any spinlocks held.
1188          */
1189         if (need_vpa_update) {
1190                 spin_unlock(&vc->lock);
1191                 for (i = 0; i < need_vpa_update; ++i)
1192                         kvmppc_update_vpas(vcpus_to_update[i]);
1193                 spin_lock(&vc->lock);
1194         }
1195
1196         /*
1197          * Assign physical thread IDs, first to non-ceded vcpus
1198          * and then to ceded ones.
1199          */
1200         ptid = 0;
1201         vcpu0 = NULL;
1202         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1203                 if (!vcpu->arch.ceded) {
1204                         if (!ptid)
1205                                 vcpu0 = vcpu;
1206                         vcpu->arch.ptid = ptid++;
1207                 }
1208         }
1209         if (!vcpu0)
1210                 goto out;       /* nothing to run; should never happen */
1211         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1212                 if (vcpu->arch.ceded)
1213                         vcpu->arch.ptid = ptid++;
1214
1215         /*
1216          * Make sure we are running on thread 0, and that
1217          * secondary threads are offline.
1218          */
1219         if (threads_per_core > 1 && !on_primary_thread()) {
1220                 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1221                         vcpu->arch.ret = -EBUSY;
1222                 goto out;
1223         }
1224
1225         vc->pcpu = smp_processor_id();
1226         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1227                 kvmppc_start_thread(vcpu);
1228                 kvmppc_create_dtl_entry(vcpu, vc);
1229         }
1230
1231         vc->vcore_state = VCORE_RUNNING;
1232         preempt_disable();
1233         spin_unlock(&vc->lock);
1234
1235         kvm_guest_enter();
1236
1237         srcu_idx = srcu_read_lock(&vcpu0->kvm->srcu);
1238
1239         __kvmppc_vcore_entry(NULL, vcpu0);
1240
1241         spin_lock(&vc->lock);
1242         /* disable sending of IPIs on virtual external irqs */
1243         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1244                 vcpu->cpu = -1;
1245         /* wait for secondary threads to finish writing their state to memory */
1246         if (vc->nap_count < vc->n_woken)
1247                 kvmppc_wait_for_nap(vc);
1248         for (i = 0; i < threads_per_core; ++i)
1249                 kvmppc_release_hwthread(vc->pcpu + i);
1250         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
1251         vc->vcore_state = VCORE_EXITING;
1252         spin_unlock(&vc->lock);
1253
1254         srcu_read_unlock(&vcpu0->kvm->srcu, srcu_idx);
1255
1256         /* make sure updates to secondary vcpu structs are visible now */
1257         smp_mb();
1258         kvm_guest_exit();
1259
1260         preempt_enable();
1261         kvm_resched(vcpu);
1262
1263         spin_lock(&vc->lock);
1264         now = get_tb();
1265         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1266                 /* cancel pending dec exception if dec is positive */
1267                 if (now < vcpu->arch.dec_expires &&
1268                     kvmppc_core_pending_dec(vcpu))
1269                         kvmppc_core_dequeue_dec(vcpu);
1270
1271                 ret = RESUME_GUEST;
1272                 if (vcpu->arch.trap)
1273                         ret = kvmppc_handle_exit(vcpu->arch.kvm_run, vcpu,
1274                                                  vcpu->arch.run_task);
1275
1276                 vcpu->arch.ret = ret;
1277                 vcpu->arch.trap = 0;
1278
1279                 if (vcpu->arch.ceded) {
1280                         if (ret != RESUME_GUEST)
1281                                 kvmppc_end_cede(vcpu);
1282                         else
1283                                 kvmppc_set_timer(vcpu);
1284                 }
1285         }
1286
1287  out:
1288         vc->vcore_state = VCORE_INACTIVE;
1289         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1290                                  arch.run_list) {
1291                 if (vcpu->arch.ret != RESUME_GUEST) {
1292                         kvmppc_remove_runnable(vc, vcpu);
1293                         wake_up(&vcpu->arch.cpu_run);
1294                 }
1295         }
1296 }
1297
1298 /*
1299  * Wait for some other vcpu thread to execute us, and
1300  * wake us up when we need to handle something in the host.
1301  */
1302 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1303 {
1304         DEFINE_WAIT(wait);
1305
1306         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1307         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1308                 schedule();
1309         finish_wait(&vcpu->arch.cpu_run, &wait);
1310 }
1311
1312 /*
1313  * All the vcpus in this vcore are idle, so wait for a decrementer
1314  * or external interrupt to one of the vcpus.  vc->lock is held.
1315  */
1316 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1317 {
1318         DEFINE_WAIT(wait);
1319
1320         prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1321         vc->vcore_state = VCORE_SLEEPING;
1322         spin_unlock(&vc->lock);
1323         schedule();
1324         finish_wait(&vc->wq, &wait);
1325         spin_lock(&vc->lock);
1326         vc->vcore_state = VCORE_INACTIVE;
1327 }
1328
1329 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1330 {
1331         int n_ceded;
1332         struct kvmppc_vcore *vc;
1333         struct kvm_vcpu *v, *vn;
1334
1335         kvm_run->exit_reason = 0;
1336         vcpu->arch.ret = RESUME_GUEST;
1337         vcpu->arch.trap = 0;
1338         kvmppc_update_vpas(vcpu);
1339
1340         /*
1341          * Synchronize with other threads in this virtual core
1342          */
1343         vc = vcpu->arch.vcore;
1344         spin_lock(&vc->lock);
1345         vcpu->arch.ceded = 0;
1346         vcpu->arch.run_task = current;
1347         vcpu->arch.kvm_run = kvm_run;
1348         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
1349         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1350         vcpu->arch.busy_preempt = TB_NIL;
1351         list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1352         ++vc->n_runnable;
1353
1354         /*
1355          * This happens the first time this is called for a vcpu.
1356          * If the vcore is already running, we may be able to start
1357          * this thread straight away and have it join in.
1358          */
1359         if (!signal_pending(current)) {
1360                 if (vc->vcore_state == VCORE_RUNNING &&
1361                     VCORE_EXIT_COUNT(vc) == 0) {
1362                         vcpu->arch.ptid = vc->n_runnable - 1;
1363                         kvmppc_create_dtl_entry(vcpu, vc);
1364                         kvmppc_start_thread(vcpu);
1365                 } else if (vc->vcore_state == VCORE_SLEEPING) {
1366                         wake_up(&vc->wq);
1367                 }
1368
1369         }
1370
1371         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1372                !signal_pending(current)) {
1373                 if (vc->vcore_state != VCORE_INACTIVE) {
1374                         spin_unlock(&vc->lock);
1375                         kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1376                         spin_lock(&vc->lock);
1377                         continue;
1378                 }
1379                 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1380                                          arch.run_list) {
1381                         kvmppc_core_prepare_to_enter(v);
1382                         if (signal_pending(v->arch.run_task)) {
1383                                 kvmppc_remove_runnable(vc, v);
1384                                 v->stat.signal_exits++;
1385                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1386                                 v->arch.ret = -EINTR;
1387                                 wake_up(&v->arch.cpu_run);
1388                         }
1389                 }
1390                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1391                         break;
1392                 vc->runner = vcpu;
1393                 n_ceded = 0;
1394                 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
1395                         if (!v->arch.pending_exceptions)
1396                                 n_ceded += v->arch.ceded;
1397                         else
1398                                 v->arch.ceded = 0;
1399                 }
1400                 if (n_ceded == vc->n_runnable)
1401                         kvmppc_vcore_blocked(vc);
1402                 else
1403                         kvmppc_run_core(vc);
1404                 vc->runner = NULL;
1405         }
1406
1407         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1408                (vc->vcore_state == VCORE_RUNNING ||
1409                 vc->vcore_state == VCORE_EXITING)) {
1410                 spin_unlock(&vc->lock);
1411                 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1412                 spin_lock(&vc->lock);
1413         }
1414
1415         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1416                 kvmppc_remove_runnable(vc, vcpu);
1417                 vcpu->stat.signal_exits++;
1418                 kvm_run->exit_reason = KVM_EXIT_INTR;
1419                 vcpu->arch.ret = -EINTR;
1420         }
1421
1422         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1423                 /* Wake up some vcpu to run the core */
1424                 v = list_first_entry(&vc->runnable_threads,
1425                                      struct kvm_vcpu, arch.run_list);
1426                 wake_up(&v->arch.cpu_run);
1427         }
1428
1429         spin_unlock(&vc->lock);
1430         return vcpu->arch.ret;
1431 }
1432
1433 int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
1434 {
1435         int r;
1436         int srcu_idx;
1437
1438         if (!vcpu->arch.sane) {
1439                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1440                 return -EINVAL;
1441         }
1442
1443         kvmppc_core_prepare_to_enter(vcpu);
1444
1445         /* No need to go into the guest when all we'll do is come back out */
1446         if (signal_pending(current)) {
1447                 run->exit_reason = KVM_EXIT_INTR;
1448                 return -EINTR;
1449         }
1450
1451         atomic_inc(&vcpu->kvm->arch.vcpus_running);
1452         /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1453         smp_mb();
1454
1455         /* On the first time here, set up HTAB and VRMA or RMA */
1456         if (!vcpu->kvm->arch.rma_setup_done) {
1457                 r = kvmppc_hv_setup_htab_rma(vcpu);
1458                 if (r)
1459                         goto out;
1460         }
1461
1462         flush_fp_to_thread(current);
1463         flush_altivec_to_thread(current);
1464         flush_vsx_to_thread(current);
1465         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1466         vcpu->arch.pgdir = current->mm->pgd;
1467         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1468
1469         do {
1470                 r = kvmppc_run_vcpu(run, vcpu);
1471
1472                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1473                     !(vcpu->arch.shregs.msr & MSR_PR)) {
1474                         r = kvmppc_pseries_do_hcall(vcpu);
1475                         kvmppc_core_prepare_to_enter(vcpu);
1476                 } else if (r == RESUME_PAGE_FAULT) {
1477                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1478                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
1479                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1480                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1481                 }
1482         } while (r == RESUME_GUEST);
1483
1484  out:
1485         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1486         atomic_dec(&vcpu->kvm->arch.vcpus_running);
1487         return r;
1488 }
1489
1490
1491 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1492    Assumes POWER7 or PPC970. */
1493 static inline int lpcr_rmls(unsigned long rma_size)
1494 {
1495         switch (rma_size) {
1496         case 32ul << 20:        /* 32 MB */
1497                 if (cpu_has_feature(CPU_FTR_ARCH_206))
1498                         return 8;       /* only supported on POWER7 */
1499                 return -1;
1500         case 64ul << 20:        /* 64 MB */
1501                 return 3;
1502         case 128ul << 20:       /* 128 MB */
1503                 return 7;
1504         case 256ul << 20:       /* 256 MB */
1505                 return 4;
1506         case 1ul << 30:         /* 1 GB */
1507                 return 2;
1508         case 16ul << 30:        /* 16 GB */
1509                 return 1;
1510         case 256ul << 30:       /* 256 GB */
1511                 return 0;
1512         default:
1513                 return -1;
1514         }
1515 }
1516
1517 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1518 {
1519         struct kvmppc_linear_info *ri = vma->vm_file->private_data;
1520         struct page *page;
1521
1522         if (vmf->pgoff >= ri->npages)
1523                 return VM_FAULT_SIGBUS;
1524
1525         page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1526         get_page(page);
1527         vmf->page = page;
1528         return 0;
1529 }
1530
1531 static const struct vm_operations_struct kvm_rma_vm_ops = {
1532         .fault = kvm_rma_fault,
1533 };
1534
1535 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1536 {
1537         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1538         vma->vm_ops = &kvm_rma_vm_ops;
1539         return 0;
1540 }
1541
1542 static int kvm_rma_release(struct inode *inode, struct file *filp)
1543 {
1544         struct kvmppc_linear_info *ri = filp->private_data;
1545
1546         kvm_release_rma(ri);
1547         return 0;
1548 }
1549
1550 static const struct file_operations kvm_rma_fops = {
1551         .mmap           = kvm_rma_mmap,
1552         .release        = kvm_rma_release,
1553 };
1554
1555 long kvm_vm_ioctl_allocate_rma(struct kvm *kvm, struct kvm_allocate_rma *ret)
1556 {
1557         struct kvmppc_linear_info *ri;
1558         long fd;
1559
1560         ri = kvm_alloc_rma();
1561         if (!ri)
1562                 return -ENOMEM;
1563
1564         fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR);
1565         if (fd < 0)
1566                 kvm_release_rma(ri);
1567
1568         ret->rma_size = ri->npages << PAGE_SHIFT;
1569         return fd;
1570 }
1571
1572 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1573                                      int linux_psize)
1574 {
1575         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1576
1577         if (!def->shift)
1578                 return;
1579         (*sps)->page_shift = def->shift;
1580         (*sps)->slb_enc = def->sllp;
1581         (*sps)->enc[0].page_shift = def->shift;
1582         /*
1583          * Only return base page encoding. We don't want to return
1584          * all the supporting pte_enc, because our H_ENTER doesn't
1585          * support MPSS yet. Once they do, we can start passing all
1586          * support pte_enc here
1587          */
1588         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
1589         (*sps)++;
1590 }
1591
1592 int kvm_vm_ioctl_get_smmu_info(struct kvm *kvm, struct kvm_ppc_smmu_info *info)
1593 {
1594         struct kvm_ppc_one_seg_page_size *sps;
1595
1596         info->flags = KVM_PPC_PAGE_SIZES_REAL;
1597         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1598                 info->flags |= KVM_PPC_1T_SEGMENTS;
1599         info->slb_size = mmu_slb_size;
1600
1601         /* We only support these sizes for now, and no muti-size segments */
1602         sps = &info->sps[0];
1603         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1604         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1605         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1606
1607         return 0;
1608 }
1609
1610 /*
1611  * Get (and clear) the dirty memory log for a memory slot.
1612  */
1613 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1614 {
1615         struct kvm_memory_slot *memslot;
1616         int r;
1617         unsigned long n;
1618
1619         mutex_lock(&kvm->slots_lock);
1620
1621         r = -EINVAL;
1622         if (log->slot >= KVM_USER_MEM_SLOTS)
1623                 goto out;
1624
1625         memslot = id_to_memslot(kvm->memslots, log->slot);
1626         r = -ENOENT;
1627         if (!memslot->dirty_bitmap)
1628                 goto out;
1629
1630         n = kvm_dirty_bitmap_bytes(memslot);
1631         memset(memslot->dirty_bitmap, 0, n);
1632
1633         r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
1634         if (r)
1635                 goto out;
1636
1637         r = -EFAULT;
1638         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1639                 goto out;
1640
1641         r = 0;
1642 out:
1643         mutex_unlock(&kvm->slots_lock);
1644         return r;
1645 }
1646
1647 static void unpin_slot(struct kvm_memory_slot *memslot)
1648 {
1649         unsigned long *physp;
1650         unsigned long j, npages, pfn;
1651         struct page *page;
1652
1653         physp = memslot->arch.slot_phys;
1654         npages = memslot->npages;
1655         if (!physp)
1656                 return;
1657         for (j = 0; j < npages; j++) {
1658                 if (!(physp[j] & KVMPPC_GOT_PAGE))
1659                         continue;
1660                 pfn = physp[j] >> PAGE_SHIFT;
1661                 page = pfn_to_page(pfn);
1662                 SetPageDirty(page);
1663                 put_page(page);
1664         }
1665 }
1666
1667 void kvmppc_core_free_memslot(struct kvm_memory_slot *free,
1668                               struct kvm_memory_slot *dont)
1669 {
1670         if (!dont || free->arch.rmap != dont->arch.rmap) {
1671                 vfree(free->arch.rmap);
1672                 free->arch.rmap = NULL;
1673         }
1674         if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
1675                 unpin_slot(free);
1676                 vfree(free->arch.slot_phys);
1677                 free->arch.slot_phys = NULL;
1678         }
1679 }
1680
1681 int kvmppc_core_create_memslot(struct kvm_memory_slot *slot,
1682                                unsigned long npages)
1683 {
1684         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
1685         if (!slot->arch.rmap)
1686                 return -ENOMEM;
1687         slot->arch.slot_phys = NULL;
1688
1689         return 0;
1690 }
1691
1692 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1693                                       struct kvm_memory_slot *memslot,
1694                                       struct kvm_userspace_memory_region *mem)
1695 {
1696         unsigned long *phys;
1697
1698         /* Allocate a slot_phys array if needed */
1699         phys = memslot->arch.slot_phys;
1700         if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
1701                 phys = vzalloc(memslot->npages * sizeof(unsigned long));
1702                 if (!phys)
1703                         return -ENOMEM;
1704                 memslot->arch.slot_phys = phys;
1705         }
1706
1707         return 0;
1708 }
1709
1710 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1711                                       struct kvm_userspace_memory_region *mem,
1712                                       const struct kvm_memory_slot *old)
1713 {
1714         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
1715         struct kvm_memory_slot *memslot;
1716
1717         if (npages && old->npages) {
1718                 /*
1719                  * If modifying a memslot, reset all the rmap dirty bits.
1720                  * If this is a new memslot, we don't need to do anything
1721                  * since the rmap array starts out as all zeroes,
1722                  * i.e. no pages are dirty.
1723                  */
1724                 memslot = id_to_memslot(kvm->memslots, mem->slot);
1725                 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
1726         }
1727 }
1728
1729 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
1730 {
1731         int err = 0;
1732         struct kvm *kvm = vcpu->kvm;
1733         struct kvmppc_linear_info *ri = NULL;
1734         unsigned long hva;
1735         struct kvm_memory_slot *memslot;
1736         struct vm_area_struct *vma;
1737         unsigned long lpcr, senc;
1738         unsigned long psize, porder;
1739         unsigned long rma_size;
1740         unsigned long rmls;
1741         unsigned long *physp;
1742         unsigned long i, npages;
1743         int srcu_idx;
1744
1745         mutex_lock(&kvm->lock);
1746         if (kvm->arch.rma_setup_done)
1747                 goto out;       /* another vcpu beat us to it */
1748
1749         /* Allocate hashed page table (if not done already) and reset it */
1750         if (!kvm->arch.hpt_virt) {
1751                 err = kvmppc_alloc_hpt(kvm, NULL);
1752                 if (err) {
1753                         pr_err("KVM: Couldn't alloc HPT\n");
1754                         goto out;
1755                 }
1756         }
1757
1758         /* Look up the memslot for guest physical address 0 */
1759         srcu_idx = srcu_read_lock(&kvm->srcu);
1760         memslot = gfn_to_memslot(kvm, 0);
1761
1762         /* We must have some memory at 0 by now */
1763         err = -EINVAL;
1764         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1765                 goto out_srcu;
1766
1767         /* Look up the VMA for the start of this memory slot */
1768         hva = memslot->userspace_addr;
1769         down_read(&current->mm->mmap_sem);
1770         vma = find_vma(current->mm, hva);
1771         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
1772                 goto up_out;
1773
1774         psize = vma_kernel_pagesize(vma);
1775         porder = __ilog2(psize);
1776
1777         /* Is this one of our preallocated RMAs? */
1778         if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
1779             hva == vma->vm_start)
1780                 ri = vma->vm_file->private_data;
1781
1782         up_read(&current->mm->mmap_sem);
1783
1784         if (!ri) {
1785                 /* On POWER7, use VRMA; on PPC970, give up */
1786                 err = -EPERM;
1787                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1788                         pr_err("KVM: CPU requires an RMO\n");
1789                         goto out_srcu;
1790                 }
1791
1792                 /* We can handle 4k, 64k or 16M pages in the VRMA */
1793                 err = -EINVAL;
1794                 if (!(psize == 0x1000 || psize == 0x10000 ||
1795                       psize == 0x1000000))
1796                         goto out_srcu;
1797
1798                 /* Update VRMASD field in the LPCR */
1799                 senc = slb_pgsize_encoding(psize);
1800                 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1801                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
1802                 lpcr = kvm->arch.lpcr & ~LPCR_VRMASD;
1803                 lpcr |= senc << (LPCR_VRMASD_SH - 4);
1804                 kvm->arch.lpcr = lpcr;
1805
1806                 /* Create HPTEs in the hash page table for the VRMA */
1807                 kvmppc_map_vrma(vcpu, memslot, porder);
1808
1809         } else {
1810                 /* Set up to use an RMO region */
1811                 rma_size = ri->npages;
1812                 if (rma_size > memslot->npages)
1813                         rma_size = memslot->npages;
1814                 rma_size <<= PAGE_SHIFT;
1815                 rmls = lpcr_rmls(rma_size);
1816                 err = -EINVAL;
1817                 if (rmls < 0) {
1818                         pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
1819                         goto out_srcu;
1820                 }
1821                 atomic_inc(&ri->use_count);
1822                 kvm->arch.rma = ri;
1823
1824                 /* Update LPCR and RMOR */
1825                 lpcr = kvm->arch.lpcr;
1826                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1827                         /* PPC970; insert RMLS value (split field) in HID4 */
1828                         lpcr &= ~((1ul << HID4_RMLS0_SH) |
1829                                   (3ul << HID4_RMLS2_SH));
1830                         lpcr |= ((rmls >> 2) << HID4_RMLS0_SH) |
1831                                 ((rmls & 3) << HID4_RMLS2_SH);
1832                         /* RMOR is also in HID4 */
1833                         lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
1834                                 << HID4_RMOR_SH;
1835                 } else {
1836                         /* POWER7 */
1837                         lpcr &= ~(LPCR_VPM0 | LPCR_VRMA_L);
1838                         lpcr |= rmls << LPCR_RMLS_SH;
1839                         kvm->arch.rmor = kvm->arch.rma->base_pfn << PAGE_SHIFT;
1840                 }
1841                 kvm->arch.lpcr = lpcr;
1842                 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
1843                         ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
1844
1845                 /* Initialize phys addrs of pages in RMO */
1846                 npages = ri->npages;
1847                 porder = __ilog2(npages);
1848                 physp = memslot->arch.slot_phys;
1849                 if (physp) {
1850                         if (npages > memslot->npages)
1851                                 npages = memslot->npages;
1852                         spin_lock(&kvm->arch.slot_phys_lock);
1853                         for (i = 0; i < npages; ++i)
1854                                 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
1855                                         porder;
1856                         spin_unlock(&kvm->arch.slot_phys_lock);
1857                 }
1858         }
1859
1860         /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
1861         smp_wmb();
1862         kvm->arch.rma_setup_done = 1;
1863         err = 0;
1864  out_srcu:
1865         srcu_read_unlock(&kvm->srcu, srcu_idx);
1866  out:
1867         mutex_unlock(&kvm->lock);
1868         return err;
1869
1870  up_out:
1871         up_read(&current->mm->mmap_sem);
1872         goto out;
1873 }
1874
1875 int kvmppc_core_init_vm(struct kvm *kvm)
1876 {
1877         unsigned long lpcr, lpid;
1878
1879         /* Allocate the guest's logical partition ID */
1880
1881         lpid = kvmppc_alloc_lpid();
1882         if (lpid < 0)
1883                 return -ENOMEM;
1884         kvm->arch.lpid = lpid;
1885
1886         /*
1887          * Since we don't flush the TLB when tearing down a VM,
1888          * and this lpid might have previously been used,
1889          * make sure we flush on each core before running the new VM.
1890          */
1891         cpumask_setall(&kvm->arch.need_tlb_flush);
1892
1893         INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
1894         INIT_LIST_HEAD(&kvm->arch.rtas_tokens);
1895
1896         kvm->arch.rma = NULL;
1897
1898         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
1899
1900         if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1901                 /* PPC970; HID4 is effectively the LPCR */
1902                 kvm->arch.host_lpid = 0;
1903                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
1904                 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
1905                 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
1906                         ((lpid & 0xf) << HID4_LPID5_SH);
1907         } else {
1908                 /* POWER7; init LPCR for virtual RMA mode */
1909                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
1910                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
1911                 lpcr &= LPCR_PECE | LPCR_LPES;
1912                 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
1913                         LPCR_VPM0 | LPCR_VPM1;
1914                 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
1915                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
1916         }
1917         kvm->arch.lpcr = lpcr;
1918
1919         kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
1920         spin_lock_init(&kvm->arch.slot_phys_lock);
1921
1922         /*
1923          * Don't allow secondary CPU threads to come online
1924          * while any KVM VMs exist.
1925          */
1926         inhibit_secondary_onlining();
1927
1928         return 0;
1929 }
1930
1931 void kvmppc_core_destroy_vm(struct kvm *kvm)
1932 {
1933         uninhibit_secondary_onlining();
1934
1935         if (kvm->arch.rma) {
1936                 kvm_release_rma(kvm->arch.rma);
1937                 kvm->arch.rma = NULL;
1938         }
1939
1940         kvmppc_rtas_tokens_free(kvm);
1941
1942         kvmppc_free_hpt(kvm);
1943         WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1944 }
1945
1946 /* These are stubs for now */
1947 void kvmppc_mmu_pte_pflush(struct kvm_vcpu *vcpu, ulong pa_start, ulong pa_end)
1948 {
1949 }
1950
1951 /* We don't need to emulate any privileged instructions or dcbz */
1952 int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu,
1953                            unsigned int inst, int *advance)
1954 {
1955         return EMULATE_FAIL;
1956 }
1957
1958 int kvmppc_core_emulate_mtspr(struct kvm_vcpu *vcpu, int sprn, ulong spr_val)
1959 {
1960         return EMULATE_FAIL;
1961 }
1962
1963 int kvmppc_core_emulate_mfspr(struct kvm_vcpu *vcpu, int sprn, ulong *spr_val)
1964 {
1965         return EMULATE_FAIL;
1966 }
1967
1968 static int kvmppc_book3s_hv_init(void)
1969 {
1970         int r;
1971
1972         r = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1973
1974         if (r)
1975                 return r;
1976
1977         r = kvmppc_mmu_hv_init();
1978
1979         return r;
1980 }
1981
1982 static void kvmppc_book3s_hv_exit(void)
1983 {
1984         kvm_exit();
1985 }
1986
1987 module_init(kvmppc_book3s_hv_init);
1988 module_exit(kvmppc_book3s_hv_exit);