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/cache.h>
  39 #include <asm/cacheflush.h>
  40 #include <asm/tlbflush.h>
  41 #include <asm/uaccess.h>
  42 #include <asm/io.h>
  43 #include <asm/kvm_ppc.h>
  44 #include <asm/kvm_book3s.h>
  45 #include <asm/mmu_context.h>
  46 #include <asm/lppaca.h>
  47 #include <asm/processor.h>
  48 #include <asm/cputhreads.h>
  49 #include <asm/page.h>
  50 #include <asm/hvcall.h>
  51 #include <asm/switch_to.h>
  52 #include <asm/smp.h>
  53 #include <linux/gfp.h>
  54 #include <linux/vmalloc.h>
  55 #include <linux/highmem.h>
  56 #include <linux/hugetlb.h>
  57 #include <linux/module.h>
  58
  59 #include "book3s.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include "trace_hv.h"
  63
  64 /* #define EXIT_DEBUG */
  65 /* #define EXIT_DEBUG_SIMPLE */
  66 /* #define EXIT_DEBUG_INT */
  67
  68 /* Used to indicate that a guest page fault needs to be handled */
  69 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  70
  71 /* Used as a "null" value for timebase values */
  72 #define TB_NIL  (~(u64)0)
  73
  74 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
  75
  76 #if defined(CONFIG_PPC_64K_PAGES)
  77 #define MPP_BUFFER_ORDER        0
  78 #elif defined(CONFIG_PPC_4K_PAGES)
  79 #define MPP_BUFFER_ORDER        3
  80 #endif
  81
  82
  83 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
  84 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
  85
  86 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
  87 {
  88         int me;
  89         int cpu = vcpu->cpu;
  90         wait_queue_head_t *wqp;
  91
  92         wqp = kvm_arch_vcpu_wq(vcpu);
  93         if (waitqueue_active(wqp)) {
  94                 wake_up_interruptible(wqp);
  95                 ++vcpu->stat.halt_wakeup;
  96         }
  97
  98         me = get_cpu();
  99
 100         /* CPU points to the first thread of the core */
 101         if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
 102 #ifdef CONFIG_PPC_ICP_NATIVE
 103                 int real_cpu = cpu + vcpu->arch.ptid;
 104                 if (paca[real_cpu].kvm_hstate.xics_phys)
 105                         xics_wake_cpu(real_cpu);
 106                 else
 107 #endif
 108                 if (cpu_online(cpu))
 109                         smp_send_reschedule(cpu);
 110         }
 111         put_cpu();
 112 }
 113
 114 /*
 115  * We use the vcpu_load/put functions to measure stolen time.
 116  * Stolen time is counted as time when either the vcpu is able to
 117  * run as part of a virtual core, but the task running the vcore
 118  * is preempted or sleeping, or when the vcpu needs something done
 119  * in the kernel by the task running the vcpu, but that task is
 120  * preempted or sleeping.  Those two things have to be counted
 121  * separately, since one of the vcpu tasks will take on the job
 122  * of running the core, and the other vcpu tasks in the vcore will
 123  * sleep waiting for it to do that, but that sleep shouldn't count
 124  * as stolen time.
 125  *
 126  * Hence we accumulate stolen time when the vcpu can run as part of
 127  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 128  * needs its task to do other things in the kernel (for example,
 129  * service a page fault) in busy_stolen.  We don't accumulate
 130  * stolen time for a vcore when it is inactive, or for a vcpu
 131  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 132  * a misnomer; it means that the vcpu task is not executing in
 133  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 134  * the kernel.  We don't have any way of dividing up that time
 135  * between time that the vcpu is genuinely stopped, time that
 136  * the task is actively working on behalf of the vcpu, and time
 137  * that the task is preempted, so we don't count any of it as
 138  * stolen.
 139  *
 140  * Updates to busy_stolen are protected by arch.tbacct_lock;
 141  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
 142  * lock.  The stolen times are measured in units of timebase ticks.
 143  * (Note that the != TB_NIL checks below are purely defensive;
 144  * they should never fail.)
 145  */
 146
 147 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 148 {
 149         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 150         unsigned long flags;
 151
 152         /*
 153          * We can test vc->runner without taking the vcore lock,
 154          * because only this task ever sets vc->runner to this
 155          * vcpu, and once it is set to this vcpu, only this task
 156          * ever sets it to NULL.
 157          */
 158         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE) {
 159                 spin_lock_irqsave(&vc->stoltb_lock, flags);
 160                 if (vc->preempt_tb != TB_NIL) {
 161                         vc->stolen_tb += mftb() - vc->preempt_tb;
 162                         vc->preempt_tb = TB_NIL;
 163                 }
 164                 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 165         }
 166         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 167         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 168             vcpu->arch.busy_preempt != TB_NIL) {
 169                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 170                 vcpu->arch.busy_preempt = TB_NIL;
 171         }
 172         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 173 }
 174
 175 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 176 {
 177         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 178         unsigned long flags;
 179
 180         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE) {
 181                 spin_lock_irqsave(&vc->stoltb_lock, flags);
 182                 vc->preempt_tb = mftb();
 183                 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 184         }
 185         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 186         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 187                 vcpu->arch.busy_preempt = mftb();
 188         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 189 }
 190
 191 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
 192 {
 193         vcpu->arch.shregs.msr = msr;
 194         kvmppc_end_cede(vcpu);
 195 }
 196
 197 void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 198 {
 199         vcpu->arch.pvr = pvr;
 200 }
 201
 202 int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 203 {
 204         unsigned long pcr = 0;
 205         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 206
 207         if (arch_compat) {
 208                 switch (arch_compat) {
 209                 case PVR_ARCH_205:
 210                         /*
 211                          * If an arch bit is set in PCR, all the defined
 212                          * higher-order arch bits also have to be set.
 213                          */
 214                         pcr = PCR_ARCH_206 | PCR_ARCH_205;
 215                         break;
 216                 case PVR_ARCH_206:
 217                 case PVR_ARCH_206p:
 218                         pcr = PCR_ARCH_206;
 219                         break;
 220                 case PVR_ARCH_207:
 221                         break;
 222                 default:
 223                         return -EINVAL;
 224                 }
 225
 226                 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
 227                         /* POWER7 can't emulate POWER8 */
 228                         if (!(pcr & PCR_ARCH_206))
 229                                 return -EINVAL;
 230                         pcr &= ~PCR_ARCH_206;
 231                 }
 232         }
 233
 234         spin_lock(&vc->lock);
 235         vc->arch_compat = arch_compat;
 236         vc->pcr = pcr;
 237         spin_unlock(&vc->lock);
 238
 239         return 0;
 240 }
 241
 242 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 243 {
 244         int r;
 245
 246         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 247         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 248                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
 249         for (r = 0; r < 16; ++r)
 250                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 251                        r, kvmppc_get_gpr(vcpu, r),
 252                        r+16, kvmppc_get_gpr(vcpu, r+16));
 253         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 254                vcpu->arch.ctr, vcpu->arch.lr);
 255         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 256                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 257         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 258                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 259         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 260                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 261         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 262                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 263         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 264         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 265                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 266         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 267         for (r = 0; r < vcpu->arch.slb_max; ++r)
 268                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 269                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 270         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 271                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 272                vcpu->arch.last_inst);
 273 }
 274
 275 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 276 {
 277         int r;
 278         struct kvm_vcpu *v, *ret = NULL;
 279
 280         mutex_lock(&kvm->lock);
 281         kvm_for_each_vcpu(r, v, kvm) {
 282                 if (v->vcpu_id == id) {
 283                         ret = v;
 284                         break;
 285                 }
 286         }
 287         mutex_unlock(&kvm->lock);
 288         return ret;
 289 }
 290
 291 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 292 {
 293         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 294         vpa->yield_count = cpu_to_be32(1);
 295 }
 296
 297 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 298                    unsigned long addr, unsigned long len)
 299 {
 300         /* check address is cacheline aligned */
 301         if (addr & (L1_CACHE_BYTES - 1))
 302                 return -EINVAL;
 303         spin_lock(&vcpu->arch.vpa_update_lock);
 304         if (v->next_gpa != addr || v->len != len) {
 305                 v->next_gpa = addr;
 306                 v->len = addr ? len : 0;
 307                 v->update_pending = 1;
 308         }
 309         spin_unlock(&vcpu->arch.vpa_update_lock);
 310         return 0;
 311 }
 312
 313 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 314 struct reg_vpa {
 315         u32 dummy;
 316         union {
 317                 __be16 hword;
 318                 __be32 word;
 319         } length;
 320 };
 321
 322 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 323 {
 324         if (vpap->update_pending)
 325                 return vpap->next_gpa != 0;
 326         return vpap->pinned_addr != NULL;
 327 }
 328
 329 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 330                                        unsigned long flags,
 331                                        unsigned long vcpuid, unsigned long vpa)
 332 {
 333         struct kvm *kvm = vcpu->kvm;
 334         unsigned long len, nb;
 335         void *va;
 336         struct kvm_vcpu *tvcpu;
 337         int err;
 338         int subfunc;
 339         struct kvmppc_vpa *vpap;
 340
 341         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 342         if (!tvcpu)
 343                 return H_PARAMETER;
 344
 345         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 346         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 347             subfunc == H_VPA_REG_SLB) {
 348                 /* Registering new area - address must be cache-line aligned */
 349                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 350                         return H_PARAMETER;
 351
 352                 /* convert logical addr to kernel addr and read length */
 353                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 354                 if (va == NULL)
 355                         return H_PARAMETER;
 356                 if (subfunc == H_VPA_REG_VPA)
 357                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 358                 else
 359                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 360                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 361
 362                 /* Check length */
 363                 if (len > nb || len < sizeof(struct reg_vpa))
 364                         return H_PARAMETER;
 365         } else {
 366                 vpa = 0;
 367                 len = 0;
 368         }
 369
 370         err = H_PARAMETER;
 371         vpap = NULL;
 372         spin_lock(&tvcpu->arch.vpa_update_lock);
 373
 374         switch (subfunc) {
 375         case H_VPA_REG_VPA:             /* register VPA */
 376                 if (len < sizeof(struct lppaca))
 377                         break;
 378                 vpap = &tvcpu->arch.vpa;
 379                 err = 0;
 380                 break;
 381
 382         case H_VPA_REG_DTL:             /* register DTL */
 383                 if (len < sizeof(struct dtl_entry))
 384                         break;
 385                 len -= len % sizeof(struct dtl_entry);
 386
 387                 /* Check that they have previously registered a VPA */
 388                 err = H_RESOURCE;
 389                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 390                         break;
 391
 392                 vpap = &tvcpu->arch.dtl;
 393                 err = 0;
 394                 break;
 395
 396         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 397                 /* Check that they have previously registered a VPA */
 398                 err = H_RESOURCE;
 399                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 400                         break;
 401
 402                 vpap = &tvcpu->arch.slb_shadow;
 403                 err = 0;
 404                 break;
 405
 406         case H_VPA_DEREG_VPA:           /* deregister VPA */
 407                 /* Check they don't still have a DTL or SLB buf registered */
 408                 err = H_RESOURCE;
 409                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 410                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 411                         break;
 412
 413                 vpap = &tvcpu->arch.vpa;
 414                 err = 0;
 415                 break;
 416
 417         case H_VPA_DEREG_DTL:           /* deregister DTL */
 418                 vpap = &tvcpu->arch.dtl;
 419                 err = 0;
 420                 break;
 421
 422         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 423                 vpap = &tvcpu->arch.slb_shadow;
 424                 err = 0;
 425                 break;
 426         }
 427
 428         if (vpap) {
 429                 vpap->next_gpa = vpa;
 430                 vpap->len = len;
 431                 vpap->update_pending = 1;
 432         }
 433
 434         spin_unlock(&tvcpu->arch.vpa_update_lock);
 435
 436         return err;
 437 }
 438
 439 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 440 {
 441         struct kvm *kvm = vcpu->kvm;
 442         void *va;
 443         unsigned long nb;
 444         unsigned long gpa;
 445
 446         /*
 447          * We need to pin the page pointed to by vpap->next_gpa,
 448          * but we can't call kvmppc_pin_guest_page under the lock
 449          * as it does get_user_pages() and down_read().  So we
 450          * have to drop the lock, pin the page, then get the lock
 451          * again and check that a new area didn't get registered
 452          * in the meantime.
 453          */
 454         for (;;) {
 455                 gpa = vpap->next_gpa;
 456                 spin_unlock(&vcpu->arch.vpa_update_lock);
 457                 va = NULL;
 458                 nb = 0;
 459                 if (gpa)
 460                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 461                 spin_lock(&vcpu->arch.vpa_update_lock);
 462                 if (gpa == vpap->next_gpa)
 463                         break;
 464                 /* sigh... unpin that one and try again */
 465                 if (va)
 466                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 467         }
 468
 469         vpap->update_pending = 0;
 470         if (va && nb < vpap->len) {
 471                 /*
 472                  * If it's now too short, it must be that userspace
 473                  * has changed the mappings underlying guest memory,
 474                  * so unregister the region.
 475                  */
 476                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 477                 va = NULL;
 478         }
 479         if (vpap->pinned_addr)
 480                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 481                                         vpap->dirty);
 482         vpap->gpa = gpa;
 483         vpap->pinned_addr = va;
 484         vpap->dirty = false;
 485         if (va)
 486                 vpap->pinned_end = va + vpap->len;
 487 }
 488
 489 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 490 {
 491         if (!(vcpu->arch.vpa.update_pending ||
 492               vcpu->arch.slb_shadow.update_pending ||
 493               vcpu->arch.dtl.update_pending))
 494                 return;
 495
 496         spin_lock(&vcpu->arch.vpa_update_lock);
 497         if (vcpu->arch.vpa.update_pending) {
 498                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 499                 if (vcpu->arch.vpa.pinned_addr)
 500                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 501         }
 502         if (vcpu->arch.dtl.update_pending) {
 503                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 504                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 505                 vcpu->arch.dtl_index = 0;
 506         }
 507         if (vcpu->arch.slb_shadow.update_pending)
 508                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 509         spin_unlock(&vcpu->arch.vpa_update_lock);
 510 }
 511
 512 /*
 513  * Return the accumulated stolen time for the vcore up until `now'.
 514  * The caller should hold the vcore lock.
 515  */
 516 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 517 {
 518         u64 p;
 519         unsigned long flags;
 520
 521         spin_lock_irqsave(&vc->stoltb_lock, flags);
 522         p = vc->stolen_tb;
 523         if (vc->vcore_state != VCORE_INACTIVE &&
 524             vc->preempt_tb != TB_NIL)
 525                 p += now - vc->preempt_tb;
 526         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 527         return p;
 528 }
 529
 530 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 531                                     struct kvmppc_vcore *vc)
 532 {
 533         struct dtl_entry *dt;
 534         struct lppaca *vpa;
 535         unsigned long stolen;
 536         unsigned long core_stolen;
 537         u64 now;
 538
 539         dt = vcpu->arch.dtl_ptr;
 540         vpa = vcpu->arch.vpa.pinned_addr;
 541         now = mftb();
 542         core_stolen = vcore_stolen_time(vc, now);
 543         stolen = core_stolen - vcpu->arch.stolen_logged;
 544         vcpu->arch.stolen_logged = core_stolen;
 545         spin_lock_irq(&vcpu->arch.tbacct_lock);
 546         stolen += vcpu->arch.busy_stolen;
 547         vcpu->arch.busy_stolen = 0;
 548         spin_unlock_irq(&vcpu->arch.tbacct_lock);
 549         if (!dt || !vpa)
 550                 return;
 551         memset(dt, 0, sizeof(struct dtl_entry));
 552         dt->dispatch_reason = 7;
 553         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 554         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 555         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 556         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 557         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 558         ++dt;
 559         if (dt == vcpu->arch.dtl.pinned_end)
 560                 dt = vcpu->arch.dtl.pinned_addr;
 561         vcpu->arch.dtl_ptr = dt;
 562         /* order writing *dt vs. writing vpa->dtl_idx */
 563         smp_wmb();
 564         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 565         vcpu->arch.dtl.dirty = true;
 566 }
 567
 568 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 569 {
 570         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 571                 return true;
 572         if ((!vcpu->arch.vcore->arch_compat) &&
 573             cpu_has_feature(CPU_FTR_ARCH_207S))
 574                 return true;
 575         return false;
 576 }
 577
 578 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 579                              unsigned long resource, unsigned long value1,
 580                              unsigned long value2)
 581 {
 582         switch (resource) {
 583         case H_SET_MODE_RESOURCE_SET_CIABR:
 584                 if (!kvmppc_power8_compatible(vcpu))
 585                         return H_P2;
 586                 if (value2)
 587                         return H_P4;
 588                 if (mflags)
 589                         return H_UNSUPPORTED_FLAG_START;
 590                 /* Guests can't breakpoint the hypervisor */
 591                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 592                         return H_P3;
 593                 vcpu->arch.ciabr  = value1;
 594                 return H_SUCCESS;
 595         case H_SET_MODE_RESOURCE_SET_DAWR:
 596                 if (!kvmppc_power8_compatible(vcpu))
 597                         return H_P2;
 598                 if (mflags)
 599                         return H_UNSUPPORTED_FLAG_START;
 600                 if (value2 & DABRX_HYP)
 601                         return H_P4;
 602                 vcpu->arch.dawr  = value1;
 603                 vcpu->arch.dawrx = value2;
 604                 return H_SUCCESS;
 605         default:
 606                 return H_TOO_HARD;
 607         }
 608 }
 609
 610 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 611 {
 612         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 613         unsigned long target, ret = H_SUCCESS;
 614         struct kvm_vcpu *tvcpu;
 615         int idx, rc;
 616
 617         if (req <= MAX_HCALL_OPCODE &&
 618             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 619                 return RESUME_HOST;
 620
 621         switch (req) {
 622         case H_CEDE:
 623                 break;
 624         case H_PROD:
 625                 target = kvmppc_get_gpr(vcpu, 4);
 626                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 627                 if (!tvcpu) {
 628                         ret = H_PARAMETER;
 629                         break;
 630                 }
 631                 tvcpu->arch.prodded = 1;
 632                 smp_mb();
 633                 if (vcpu->arch.ceded) {
 634                         if (waitqueue_active(&vcpu->wq)) {
 635                                 wake_up_interruptible(&vcpu->wq);
 636                                 vcpu->stat.halt_wakeup++;
 637                         }
 638                 }
 639                 break;
 640         case H_CONFER:
 641                 target = kvmppc_get_gpr(vcpu, 4);
 642                 if (target == -1)
 643                         break;
 644                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 645                 if (!tvcpu) {
 646                         ret = H_PARAMETER;
 647                         break;
 648                 }
 649                 kvm_vcpu_yield_to(tvcpu);
 650                 break;
 651         case H_REGISTER_VPA:
 652                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 653                                         kvmppc_get_gpr(vcpu, 5),
 654                                         kvmppc_get_gpr(vcpu, 6));
 655                 break;
 656         case H_RTAS:
 657                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 658                         return RESUME_HOST;
 659
 660                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 661                 rc = kvmppc_rtas_hcall(vcpu);
 662                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 663
 664                 if (rc == -ENOENT)
 665                         return RESUME_HOST;
 666                 else if (rc == 0)
 667                         break;
 668
 669                 /* Send the error out to userspace via KVM_RUN */
 670                 return rc;
 671         case H_SET_MODE:
 672                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 673                                         kvmppc_get_gpr(vcpu, 5),
 674                                         kvmppc_get_gpr(vcpu, 6),
 675                                         kvmppc_get_gpr(vcpu, 7));
 676                 if (ret == H_TOO_HARD)
 677                         return RESUME_HOST;
 678                 break;
 679         case H_XIRR:
 680         case H_CPPR:
 681         case H_EOI:
 682         case H_IPI:
 683         case H_IPOLL:
 684         case H_XIRR_X:
 685                 if (kvmppc_xics_enabled(vcpu)) {
 686                         ret = kvmppc_xics_hcall(vcpu, req);
 687                         break;
 688                 } /* fallthrough */
 689         default:
 690                 return RESUME_HOST;
 691         }
 692         kvmppc_set_gpr(vcpu, 3, ret);
 693         vcpu->arch.hcall_needed = 0;
 694         return RESUME_GUEST;
 695 }
 696
 697 static int kvmppc_hcall_impl_hv(unsigned long cmd)
 698 {
 699         switch (cmd) {
 700         case H_CEDE:
 701         case H_PROD:
 702         case H_CONFER:
 703         case H_REGISTER_VPA:
 704         case H_SET_MODE:
 705 #ifdef CONFIG_KVM_XICS
 706         case H_XIRR:
 707         case H_CPPR:
 708         case H_EOI:
 709         case H_IPI:
 710         case H_IPOLL:
 711         case H_XIRR_X:
 712 #endif
 713                 return 1;
 714         }
 715
 716         /* See if it's in the real-mode table */
 717         return kvmppc_hcall_impl_hv_realmode(cmd);
 718 }
 719
 720 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
 721                                         struct kvm_vcpu *vcpu)
 722 {
 723         u32 last_inst;
 724
 725         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
 726                                         EMULATE_DONE) {
 727                 /*
 728                  * Fetch failed, so return to guest and
 729                  * try executing it again.
 730                  */
 731                 return RESUME_GUEST;
 732         }
 733
 734         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
 735                 run->exit_reason = KVM_EXIT_DEBUG;
 736                 run->debug.arch.address = kvmppc_get_pc(vcpu);
 737                 return RESUME_HOST;
 738         } else {
 739                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 740                 return RESUME_GUEST;
 741         }
 742 }
 743
 744 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
 745                                  struct task_struct *tsk)
 746 {
 747         int r = RESUME_HOST;
 748
 749         vcpu->stat.sum_exits++;
 750
 751         run->exit_reason = KVM_EXIT_UNKNOWN;
 752         run->ready_for_interrupt_injection = 1;
 753         switch (vcpu->arch.trap) {
 754         /* We're good on these - the host merely wanted to get our attention */
 755         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 756                 vcpu->stat.dec_exits++;
 757                 r = RESUME_GUEST;
 758                 break;
 759         case BOOK3S_INTERRUPT_EXTERNAL:
 760         case BOOK3S_INTERRUPT_H_DOORBELL:
 761                 vcpu->stat.ext_intr_exits++;
 762                 r = RESUME_GUEST;
 763                 break;
 764         /* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
 765         case BOOK3S_INTERRUPT_HMI:
 766         case BOOK3S_INTERRUPT_PERFMON:
 767                 r = RESUME_GUEST;
 768                 break;
 769         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 770                 /*
 771                  * Deliver a machine check interrupt to the guest.
 772                  * We have to do this, even if the host has handled the
 773                  * machine check, because machine checks use SRR0/1 and
 774                  * the interrupt might have trashed guest state in them.
 775                  */
 776                 kvmppc_book3s_queue_irqprio(vcpu,
 777                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 778                 r = RESUME_GUEST;
 779                 break;
 780         case BOOK3S_INTERRUPT_PROGRAM:
 781         {
 782                 ulong flags;
 783                 /*
 784                  * Normally program interrupts are delivered directly
 785                  * to the guest by the hardware, but we can get here
 786                  * as a result of a hypervisor emulation interrupt
 787                  * (e40) getting turned into a 700 by BML RTAS.
 788                  */
 789                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 790                 kvmppc_core_queue_program(vcpu, flags);
 791                 r = RESUME_GUEST;
 792                 break;
 793         }
 794         case BOOK3S_INTERRUPT_SYSCALL:
 795         {
 796                 /* hcall - punt to userspace */
 797                 int i;
 798
 799                 /* hypercall with MSR_PR has already been handled in rmode,
 800                  * and never reaches here.
 801                  */
 802
 803                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 804                 for (i = 0; i < 9; ++i)
 805                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 806                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 807                 vcpu->arch.hcall_needed = 1;
 808                 r = RESUME_HOST;
 809                 break;
 810         }
 811         /*
 812          * We get these next two if the guest accesses a page which it thinks
 813          * it has mapped but which is not actually present, either because
 814          * it is for an emulated I/O device or because the corresonding
 815          * host page has been paged out.  Any other HDSI/HISI interrupts
 816          * have been handled already.
 817          */
 818         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 819                 r = RESUME_PAGE_FAULT;
 820                 break;
 821         case BOOK3S_INTERRUPT_H_INST_STORAGE:
 822                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
 823                 vcpu->arch.fault_dsisr = 0;
 824                 r = RESUME_PAGE_FAULT;
 825                 break;
 826         /*
 827          * This occurs if the guest executes an illegal instruction.
 828          * If the guest debug is disabled, generate a program interrupt
 829          * to the guest. If guest debug is enabled, we need to check
 830          * whether the instruction is a software breakpoint instruction.
 831          * Accordingly return to Guest or Host.
 832          */
 833         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 834                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
 835                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
 836                                 swab32(vcpu->arch.emul_inst) :
 837                                 vcpu->arch.emul_inst;
 838                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
 839                         r = kvmppc_emulate_debug_inst(run, vcpu);
 840                 } else {
 841                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 842                         r = RESUME_GUEST;
 843                 }
 844                 break;
 845         /*
 846          * This occurs if the guest (kernel or userspace), does something that
 847          * is prohibited by HFSCR.  We just generate a program interrupt to
 848          * the guest.
 849          */
 850         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
 851                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 852                 r = RESUME_GUEST;
 853                 break;
 854         default:
 855                 kvmppc_dump_regs(vcpu);
 856                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 857                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 858                         vcpu->arch.shregs.msr);
 859                 run->hw.hardware_exit_reason = vcpu->arch.trap;
 860                 r = RESUME_HOST;
 861                 break;
 862         }
 863
 864         return r;
 865 }
 866
 867 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
 868                                             struct kvm_sregs *sregs)
 869 {
 870         int i;
 871
 872         memset(sregs, 0, sizeof(struct kvm_sregs));
 873         sregs->pvr = vcpu->arch.pvr;
 874         for (i = 0; i < vcpu->arch.slb_max; i++) {
 875                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
 876                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 877         }
 878
 879         return 0;
 880 }
 881
 882 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
 883                                             struct kvm_sregs *sregs)
 884 {
 885         int i, j;
 886
 887         /* Only accept the same PVR as the host's, since we can't spoof it */
 888         if (sregs->pvr != vcpu->arch.pvr)
 889                 return -EINVAL;
 890
 891         j = 0;
 892         for (i = 0; i < vcpu->arch.slb_nr; i++) {
 893                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
 894                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
 895                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
 896                         ++j;
 897                 }
 898         }
 899         vcpu->arch.slb_max = j;
 900
 901         return 0;
 902 }
 903
 904 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
 905                 bool preserve_top32)
 906 {
 907         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 908         u64 mask;
 909
 910         spin_lock(&vc->lock);
 911         /*
 912          * If ILE (interrupt little-endian) has changed, update the
 913          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
 914          */
 915         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
 916                 struct kvm *kvm = vcpu->kvm;
 917                 struct kvm_vcpu *vcpu;
 918                 int i;
 919
 920                 mutex_lock(&kvm->lock);
 921                 kvm_for_each_vcpu(i, vcpu, kvm) {
 922                         if (vcpu->arch.vcore != vc)
 923                                 continue;
 924                         if (new_lpcr & LPCR_ILE)
 925                                 vcpu->arch.intr_msr |= MSR_LE;
 926                         else
 927                                 vcpu->arch.intr_msr &= ~MSR_LE;
 928                 }
 929                 mutex_unlock(&kvm->lock);
 930         }
 931
 932         /*
 933          * Userspace can only modify DPFD (default prefetch depth),
 934          * ILE (interrupt little-endian) and TC (translation control).
 935          * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
 936          */
 937         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
 938         if (cpu_has_feature(CPU_FTR_ARCH_207S))
 939                 mask |= LPCR_AIL;
 940
 941         /* Broken 32-bit version of LPCR must not clear top bits */
 942         if (preserve_top32)
 943                 mask &= 0xFFFFFFFF;
 944         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
 945         spin_unlock(&vc->lock);
 946 }
 947
 948 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
 949                                  union kvmppc_one_reg *val)
 950 {
 951         int r = 0;
 952         long int i;
 953
 954         switch (id) {
 955         case KVM_REG_PPC_DEBUG_INST:
 956                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
 957                 break;
 958         case KVM_REG_PPC_HIOR:
 959                 *val = get_reg_val(id, 0);
 960                 break;
 961         case KVM_REG_PPC_DABR:
 962                 *val = get_reg_val(id, vcpu->arch.dabr);
 963                 break;
 964         case KVM_REG_PPC_DABRX:
 965                 *val = get_reg_val(id, vcpu->arch.dabrx);
 966                 break;
 967         case KVM_REG_PPC_DSCR:
 968                 *val = get_reg_val(id, vcpu->arch.dscr);
 969                 break;
 970         case KVM_REG_PPC_PURR:
 971                 *val = get_reg_val(id, vcpu->arch.purr);
 972                 break;
 973         case KVM_REG_PPC_SPURR:
 974                 *val = get_reg_val(id, vcpu->arch.spurr);
 975                 break;
 976         case KVM_REG_PPC_AMR:
 977                 *val = get_reg_val(id, vcpu->arch.amr);
 978                 break;
 979         case KVM_REG_PPC_UAMOR:
 980                 *val = get_reg_val(id, vcpu->arch.uamor);
 981                 break;
 982         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
 983                 i = id - KVM_REG_PPC_MMCR0;
 984                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
 985                 break;
 986         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
 987                 i = id - KVM_REG_PPC_PMC1;
 988                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
 989                 break;
 990         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
 991                 i = id - KVM_REG_PPC_SPMC1;
 992                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
 993                 break;
 994         case KVM_REG_PPC_SIAR:
 995                 *val = get_reg_val(id, vcpu->arch.siar);
 996                 break;
 997         case KVM_REG_PPC_SDAR:
 998                 *val = get_reg_val(id, vcpu->arch.sdar);
 999                 break;
1000         case KVM_REG_PPC_SIER:
1001                 *val = get_reg_val(id, vcpu->arch.sier);
1002                 break;
1003         case KVM_REG_PPC_IAMR:
1004                 *val = get_reg_val(id, vcpu->arch.iamr);
1005                 break;
1006         case KVM_REG_PPC_PSPB:
1007                 *val = get_reg_val(id, vcpu->arch.pspb);
1008                 break;
1009         case KVM_REG_PPC_DPDES:
1010                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1011                 break;
1012         case KVM_REG_PPC_DAWR:
1013                 *val = get_reg_val(id, vcpu->arch.dawr);
1014                 break;
1015         case KVM_REG_PPC_DAWRX:
1016                 *val = get_reg_val(id, vcpu->arch.dawrx);
1017                 break;
1018         case KVM_REG_PPC_CIABR:
1019                 *val = get_reg_val(id, vcpu->arch.ciabr);
1020                 break;
1021         case KVM_REG_PPC_CSIGR:
1022                 *val = get_reg_val(id, vcpu->arch.csigr);
1023                 break;
1024         case KVM_REG_PPC_TACR:
1025                 *val = get_reg_val(id, vcpu->arch.tacr);
1026                 break;
1027         case KVM_REG_PPC_TCSCR:
1028                 *val = get_reg_val(id, vcpu->arch.tcscr);
1029                 break;
1030         case KVM_REG_PPC_PID:
1031                 *val = get_reg_val(id, vcpu->arch.pid);
1032                 break;
1033         case KVM_REG_PPC_ACOP:
1034                 *val = get_reg_val(id, vcpu->arch.acop);
1035                 break;
1036         case KVM_REG_PPC_WORT:
1037                 *val = get_reg_val(id, vcpu->arch.wort);
1038                 break;
1039         case KVM_REG_PPC_VPA_ADDR:
1040                 spin_lock(&vcpu->arch.vpa_update_lock);
1041                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1042                 spin_unlock(&vcpu->arch.vpa_update_lock);
1043                 break;
1044         case KVM_REG_PPC_VPA_SLB:
1045                 spin_lock(&vcpu->arch.vpa_update_lock);
1046                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1047                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1048                 spin_unlock(&vcpu->arch.vpa_update_lock);
1049                 break;
1050         case KVM_REG_PPC_VPA_DTL:
1051                 spin_lock(&vcpu->arch.vpa_update_lock);
1052                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1053                 val->vpaval.length = vcpu->arch.dtl.len;
1054                 spin_unlock(&vcpu->arch.vpa_update_lock);
1055                 break;
1056         case KVM_REG_PPC_TB_OFFSET:
1057                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1058                 break;
1059         case KVM_REG_PPC_LPCR:
1060         case KVM_REG_PPC_LPCR_64:
1061                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1062                 break;
1063         case KVM_REG_PPC_PPR:
1064                 *val = get_reg_val(id, vcpu->arch.ppr);
1065                 break;
1066 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1067         case KVM_REG_PPC_TFHAR:
1068                 *val = get_reg_val(id, vcpu->arch.tfhar);
1069                 break;
1070         case KVM_REG_PPC_TFIAR:
1071                 *val = get_reg_val(id, vcpu->arch.tfiar);
1072                 break;
1073         case KVM_REG_PPC_TEXASR:
1074                 *val = get_reg_val(id, vcpu->arch.texasr);
1075                 break;
1076         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1077                 i = id - KVM_REG_PPC_TM_GPR0;
1078                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1079                 break;
1080         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1081         {
1082                 int j;
1083                 i = id - KVM_REG_PPC_TM_VSR0;
1084                 if (i < 32)
1085                         for (j = 0; j < TS_FPRWIDTH; j++)
1086                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1087                 else {
1088                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1089                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1090                         else
1091                                 r = -ENXIO;
1092                 }
1093                 break;
1094         }
1095         case KVM_REG_PPC_TM_CR:
1096                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1097                 break;
1098         case KVM_REG_PPC_TM_LR:
1099                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1100                 break;
1101         case KVM_REG_PPC_TM_CTR:
1102                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1103                 break;
1104         case KVM_REG_PPC_TM_FPSCR:
1105                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1106                 break;
1107         case KVM_REG_PPC_TM_AMR:
1108                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1109                 break;
1110         case KVM_REG_PPC_TM_PPR:
1111                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1112                 break;
1113         case KVM_REG_PPC_TM_VRSAVE:
1114                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1115                 break;
1116         case KVM_REG_PPC_TM_VSCR:
1117                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1118                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1119                 else
1120                         r = -ENXIO;
1121                 break;
1122         case KVM_REG_PPC_TM_DSCR:
1123                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1124                 break;
1125         case KVM_REG_PPC_TM_TAR:
1126                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1127                 break;
1128 #endif
1129         case KVM_REG_PPC_ARCH_COMPAT:
1130                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1131                 break;
1132         default:
1133                 r = -EINVAL;
1134                 break;
1135         }
1136
1137         return r;
1138 }
1139
1140 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1141                                  union kvmppc_one_reg *val)
1142 {
1143         int r = 0;
1144         long int i;
1145         unsigned long addr, len;
1146
1147         switch (id) {
1148         case KVM_REG_PPC_HIOR:
1149                 /* Only allow this to be set to zero */
1150                 if (set_reg_val(id, *val))
1151                         r = -EINVAL;
1152                 break;
1153         case KVM_REG_PPC_DABR:
1154                 vcpu->arch.dabr = set_reg_val(id, *val);
1155                 break;
1156         case KVM_REG_PPC_DABRX:
1157                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1158                 break;
1159         case KVM_REG_PPC_DSCR:
1160                 vcpu->arch.dscr = set_reg_val(id, *val);
1161                 break;
1162         case KVM_REG_PPC_PURR:
1163                 vcpu->arch.purr = set_reg_val(id, *val);
1164                 break;
1165         case KVM_REG_PPC_SPURR:
1166                 vcpu->arch.spurr = set_reg_val(id, *val);
1167                 break;
1168         case KVM_REG_PPC_AMR:
1169                 vcpu->arch.amr = set_reg_val(id, *val);
1170                 break;
1171         case KVM_REG_PPC_UAMOR:
1172                 vcpu->arch.uamor = set_reg_val(id, *val);
1173                 break;
1174         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1175                 i = id - KVM_REG_PPC_MMCR0;
1176                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1177                 break;
1178         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1179                 i = id - KVM_REG_PPC_PMC1;
1180                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1181                 break;
1182         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1183                 i = id - KVM_REG_PPC_SPMC1;
1184                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1185                 break;
1186         case KVM_REG_PPC_SIAR:
1187                 vcpu->arch.siar = set_reg_val(id, *val);
1188                 break;
1189         case KVM_REG_PPC_SDAR:
1190                 vcpu->arch.sdar = set_reg_val(id, *val);
1191                 break;
1192         case KVM_REG_PPC_SIER:
1193                 vcpu->arch.sier = set_reg_val(id, *val);
1194                 break;
1195         case KVM_REG_PPC_IAMR:
1196                 vcpu->arch.iamr = set_reg_val(id, *val);
1197                 break;
1198         case KVM_REG_PPC_PSPB:
1199                 vcpu->arch.pspb = set_reg_val(id, *val);
1200                 break;
1201         case KVM_REG_PPC_DPDES:
1202                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1203                 break;
1204         case KVM_REG_PPC_DAWR:
1205                 vcpu->arch.dawr = set_reg_val(id, *val);
1206                 break;
1207         case KVM_REG_PPC_DAWRX:
1208                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1209                 break;
1210         case KVM_REG_PPC_CIABR:
1211                 vcpu->arch.ciabr = set_reg_val(id, *val);
1212                 /* Don't allow setting breakpoints in hypervisor code */
1213                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1214                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1215                 break;
1216         case KVM_REG_PPC_CSIGR:
1217                 vcpu->arch.csigr = set_reg_val(id, *val);
1218                 break;
1219         case KVM_REG_PPC_TACR:
1220                 vcpu->arch.tacr = set_reg_val(id, *val);
1221                 break;
1222         case KVM_REG_PPC_TCSCR:
1223                 vcpu->arch.tcscr = set_reg_val(id, *val);
1224                 break;
1225         case KVM_REG_PPC_PID:
1226                 vcpu->arch.pid = set_reg_val(id, *val);
1227                 break;
1228         case KVM_REG_PPC_ACOP:
1229                 vcpu->arch.acop = set_reg_val(id, *val);
1230                 break;
1231         case KVM_REG_PPC_WORT:
1232                 vcpu->arch.wort = set_reg_val(id, *val);
1233                 break;
1234         case KVM_REG_PPC_VPA_ADDR:
1235                 addr = set_reg_val(id, *val);
1236                 r = -EINVAL;
1237                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1238                               vcpu->arch.dtl.next_gpa))
1239                         break;
1240                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1241                 break;
1242         case KVM_REG_PPC_VPA_SLB:
1243                 addr = val->vpaval.addr;
1244                 len = val->vpaval.length;
1245                 r = -EINVAL;
1246                 if (addr && !vcpu->arch.vpa.next_gpa)
1247                         break;
1248                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1249                 break;
1250         case KVM_REG_PPC_VPA_DTL:
1251                 addr = val->vpaval.addr;
1252                 len = val->vpaval.length;
1253                 r = -EINVAL;
1254                 if (addr && (len < sizeof(struct dtl_entry) ||
1255                              !vcpu->arch.vpa.next_gpa))
1256                         break;
1257                 len -= len % sizeof(struct dtl_entry);
1258                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1259                 break;
1260         case KVM_REG_PPC_TB_OFFSET:
1261                 /* round up to multiple of 2^24 */
1262                 vcpu->arch.vcore->tb_offset =
1263                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1264                 break;
1265         case KVM_REG_PPC_LPCR:
1266                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1267                 break;
1268         case KVM_REG_PPC_LPCR_64:
1269                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1270                 break;
1271         case KVM_REG_PPC_PPR:
1272                 vcpu->arch.ppr = set_reg_val(id, *val);
1273                 break;
1274 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1275         case KVM_REG_PPC_TFHAR:
1276                 vcpu->arch.tfhar = set_reg_val(id, *val);
1277                 break;
1278         case KVM_REG_PPC_TFIAR:
1279                 vcpu->arch.tfiar = set_reg_val(id, *val);
1280                 break;
1281         case KVM_REG_PPC_TEXASR:
1282                 vcpu->arch.texasr = set_reg_val(id, *val);
1283                 break;
1284         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1285                 i = id - KVM_REG_PPC_TM_GPR0;
1286                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1287                 break;
1288         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1289         {
1290                 int j;
1291                 i = id - KVM_REG_PPC_TM_VSR0;
1292                 if (i < 32)
1293                         for (j = 0; j < TS_FPRWIDTH; j++)
1294                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1295                 else
1296                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1297                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1298                         else
1299                                 r = -ENXIO;
1300                 break;
1301         }
1302         case KVM_REG_PPC_TM_CR:
1303                 vcpu->arch.cr_tm = set_reg_val(id, *val);
1304                 break;
1305         case KVM_REG_PPC_TM_LR:
1306                 vcpu->arch.lr_tm = set_reg_val(id, *val);
1307                 break;
1308         case KVM_REG_PPC_TM_CTR:
1309                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1310                 break;
1311         case KVM_REG_PPC_TM_FPSCR:
1312                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1313                 break;
1314         case KVM_REG_PPC_TM_AMR:
1315                 vcpu->arch.amr_tm = set_reg_val(id, *val);
1316                 break;
1317         case KVM_REG_PPC_TM_PPR:
1318                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1319                 break;
1320         case KVM_REG_PPC_TM_VRSAVE:
1321                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1322                 break;
1323         case KVM_REG_PPC_TM_VSCR:
1324                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1325                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1326                 else
1327                         r = - ENXIO;
1328                 break;
1329         case KVM_REG_PPC_TM_DSCR:
1330                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1331                 break;
1332         case KVM_REG_PPC_TM_TAR:
1333                 vcpu->arch.tar_tm = set_reg_val(id, *val);
1334                 break;
1335 #endif
1336         case KVM_REG_PPC_ARCH_COMPAT:
1337                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1338                 break;
1339         default:
1340                 r = -EINVAL;
1341                 break;
1342         }
1343
1344         return r;
1345 }
1346
1347 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1348 {
1349         struct kvmppc_vcore *vcore;
1350
1351         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1352
1353         if (vcore == NULL)
1354                 return NULL;
1355
1356         INIT_LIST_HEAD(&vcore->runnable_threads);
1357         spin_lock_init(&vcore->lock);
1358         spin_lock_init(&vcore->stoltb_lock);
1359         init_waitqueue_head(&vcore->wq);
1360         vcore->preempt_tb = TB_NIL;
1361         vcore->lpcr = kvm->arch.lpcr;
1362         vcore->first_vcpuid = core * threads_per_subcore;
1363         vcore->kvm = kvm;
1364
1365         vcore->mpp_buffer_is_valid = false;
1366
1367         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1368                 vcore->mpp_buffer = (void *)__get_free_pages(
1369                         GFP_KERNEL|__GFP_ZERO,
1370                         MPP_BUFFER_ORDER);
1371
1372         return vcore;
1373 }
1374
1375 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1376                                                    unsigned int id)
1377 {
1378         struct kvm_vcpu *vcpu;
1379         int err = -EINVAL;
1380         int core;
1381         struct kvmppc_vcore *vcore;
1382
1383         core = id / threads_per_subcore;
1384         if (core >= KVM_MAX_VCORES)
1385                 goto out;
1386
1387         err = -ENOMEM;
1388         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1389         if (!vcpu)
1390                 goto out;
1391
1392         err = kvm_vcpu_init(vcpu, kvm, id);
1393         if (err)
1394                 goto free_vcpu;
1395
1396         vcpu->arch.shared = &vcpu->arch.shregs;
1397 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1398         /*
1399          * The shared struct is never shared on HV,
1400          * so we can always use host endianness
1401          */
1402 #ifdef __BIG_ENDIAN__
1403         vcpu->arch.shared_big_endian = true;
1404 #else
1405         vcpu->arch.shared_big_endian = false;
1406 #endif
1407 #endif
1408         vcpu->arch.mmcr[0] = MMCR0_FC;
1409         vcpu->arch.ctrl = CTRL_RUNLATCH;
1410         /* default to host PVR, since we can't spoof it */
1411         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1412         spin_lock_init(&vcpu->arch.vpa_update_lock);
1413         spin_lock_init(&vcpu->arch.tbacct_lock);
1414         vcpu->arch.busy_preempt = TB_NIL;
1415         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1416
1417         kvmppc_mmu_book3s_hv_init(vcpu);
1418
1419         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1420
1421         init_waitqueue_head(&vcpu->arch.cpu_run);
1422
1423         mutex_lock(&kvm->lock);
1424         vcore = kvm->arch.vcores[core];
1425         if (!vcore) {
1426                 vcore = kvmppc_vcore_create(kvm, core);
1427                 kvm->arch.vcores[core] = vcore;
1428                 kvm->arch.online_vcores++;
1429         }
1430         mutex_unlock(&kvm->lock);
1431
1432         if (!vcore)
1433                 goto free_vcpu;
1434
1435         spin_lock(&vcore->lock);
1436         ++vcore->num_threads;
1437         spin_unlock(&vcore->lock);
1438         vcpu->arch.vcore = vcore;
1439         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1440
1441         vcpu->arch.cpu_type = KVM_CPU_3S_64;
1442         kvmppc_sanity_check(vcpu);
1443
1444         return vcpu;
1445
1446 free_vcpu:
1447         kmem_cache_free(kvm_vcpu_cache, vcpu);
1448 out:
1449         return ERR_PTR(err);
1450 }
1451
1452 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1453 {
1454         if (vpa->pinned_addr)
1455                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1456                                         vpa->dirty);
1457 }
1458
1459 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1460 {
1461         spin_lock(&vcpu->arch.vpa_update_lock);
1462         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1463         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1464         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1465         spin_unlock(&vcpu->arch.vpa_update_lock);
1466         kvm_vcpu_uninit(vcpu);
1467         kmem_cache_free(kvm_vcpu_cache, vcpu);
1468 }
1469
1470 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1471 {
1472         /* Indicate we want to get back into the guest */
1473         return 1;
1474 }
1475
1476 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1477 {
1478         unsigned long dec_nsec, now;
1479
1480         now = get_tb();
1481         if (now > vcpu->arch.dec_expires) {
1482                 /* decrementer has already gone negative */
1483                 kvmppc_core_queue_dec(vcpu);
1484                 kvmppc_core_prepare_to_enter(vcpu);
1485                 return;
1486         }
1487         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1488                    / tb_ticks_per_sec;
1489         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1490                       HRTIMER_MODE_REL);
1491         vcpu->arch.timer_running = 1;
1492 }
1493
1494 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1495 {
1496         vcpu->arch.ceded = 0;
1497         if (vcpu->arch.timer_running) {
1498                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1499                 vcpu->arch.timer_running = 0;
1500         }
1501 }
1502
1503 extern void __kvmppc_vcore_entry(void);
1504
1505 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1506                                    struct kvm_vcpu *vcpu)
1507 {
1508         u64 now;
1509
1510         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1511                 return;
1512         spin_lock_irq(&vcpu->arch.tbacct_lock);
1513         now = mftb();
1514         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1515                 vcpu->arch.stolen_logged;
1516         vcpu->arch.busy_preempt = now;
1517         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1518         spin_unlock_irq(&vcpu->arch.tbacct_lock);
1519         --vc->n_runnable;
1520         list_del(&vcpu->arch.run_list);
1521 }
1522
1523 static int kvmppc_grab_hwthread(int cpu)
1524 {
1525         struct paca_struct *tpaca;
1526         long timeout = 10000;
1527
1528         tpaca = &paca[cpu];
1529
1530         /* Ensure the thread won't go into the kernel if it wakes */
1531         tpaca->kvm_hstate.hwthread_req = 1;
1532         tpaca->kvm_hstate.kvm_vcpu = NULL;
1533
1534         /*
1535          * If the thread is already executing in the kernel (e.g. handling
1536          * a stray interrupt), wait for it to get back to nap mode.
1537          * The smp_mb() is to ensure that our setting of hwthread_req
1538          * is visible before we look at hwthread_state, so if this
1539          * races with the code at system_reset_pSeries and the thread
1540          * misses our setting of hwthread_req, we are sure to see its
1541          * setting of hwthread_state, and vice versa.
1542          */
1543         smp_mb();
1544         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1545                 if (--timeout <= 0) {
1546                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1547                         return -EBUSY;
1548                 }
1549                 udelay(1);
1550         }
1551         return 0;
1552 }
1553
1554 static void kvmppc_release_hwthread(int cpu)
1555 {
1556         struct paca_struct *tpaca;
1557
1558         tpaca = &paca[cpu];
1559         tpaca->kvm_hstate.hwthread_req = 0;
1560         tpaca->kvm_hstate.kvm_vcpu = NULL;
1561 }
1562
1563 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1564 {
1565         int cpu;
1566         struct paca_struct *tpaca;
1567         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1568
1569         if (vcpu->arch.timer_running) {
1570                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1571                 vcpu->arch.timer_running = 0;
1572         }
1573         cpu = vc->pcpu + vcpu->arch.ptid;
1574         tpaca = &paca[cpu];
1575         tpaca->kvm_hstate.kvm_vcpu = vcpu;
1576         tpaca->kvm_hstate.kvm_vcore = vc;
1577         tpaca->kvm_hstate.ptid = vcpu->arch.ptid;
1578         vcpu->cpu = vc->pcpu;
1579         smp_wmb();
1580 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1581         if (cpu != smp_processor_id()) {
1582                 xics_wake_cpu(cpu);
1583                 if (vcpu->arch.ptid)
1584                         ++vc->n_woken;
1585         }
1586 #endif
1587 }
1588
1589 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1590 {
1591         int i;
1592
1593         HMT_low();
1594         i = 0;
1595         while (vc->nap_count < vc->n_woken) {
1596                 if (++i >= 1000000) {
1597                         pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1598                                vc->nap_count, vc->n_woken);
1599                         break;
1600                 }
1601                 cpu_relax();
1602         }
1603         HMT_medium();
1604 }
1605
1606 /*
1607  * Check that we are on thread 0 and that any other threads in
1608  * this core are off-line.  Then grab the threads so they can't
1609  * enter the kernel.
1610  */
1611 static int on_primary_thread(void)
1612 {
1613         int cpu = smp_processor_id();
1614         int thr;
1615
1616         /* Are we on a primary subcore? */
1617         if (cpu_thread_in_subcore(cpu))
1618                 return 0;
1619
1620         thr = 0;
1621         while (++thr < threads_per_subcore)
1622                 if (cpu_online(cpu + thr))
1623                         return 0;
1624
1625         /* Grab all hw threads so they can't go into the kernel */
1626         for (thr = 1; thr < threads_per_subcore; ++thr) {
1627                 if (kvmppc_grab_hwthread(cpu + thr)) {
1628                         /* Couldn't grab one; let the others go */
1629                         do {
1630                                 kvmppc_release_hwthread(cpu + thr);
1631                         } while (--thr > 0);
1632                         return 0;
1633                 }
1634         }
1635         return 1;
1636 }
1637
1638 static void kvmppc_start_saving_l2_cache(struct kvmppc_vcore *vc)
1639 {
1640         phys_addr_t phy_addr, mpp_addr;
1641
1642         phy_addr = (phys_addr_t)virt_to_phys(vc->mpp_buffer);
1643         mpp_addr = phy_addr & PPC_MPPE_ADDRESS_MASK;
1644
1645         mtspr(SPRN_MPPR, mpp_addr | PPC_MPPR_FETCH_ABORT);
1646         logmpp(mpp_addr | PPC_LOGMPP_LOG_L2);
1647
1648         vc->mpp_buffer_is_valid = true;
1649 }
1650
1651 static void kvmppc_start_restoring_l2_cache(const struct kvmppc_vcore *vc)
1652 {
1653         phys_addr_t phy_addr, mpp_addr;
1654
1655         phy_addr = virt_to_phys(vc->mpp_buffer);
1656         mpp_addr = phy_addr & PPC_MPPE_ADDRESS_MASK;
1657
1658         /* We must abort any in-progress save operations to ensure
1659          * the table is valid so that prefetch engine knows when to
1660          * stop prefetching. */
1661         logmpp(mpp_addr | PPC_LOGMPP_LOG_ABORT);
1662         mtspr(SPRN_MPPR, mpp_addr | PPC_MPPR_FETCH_WHOLE_TABLE);
1663 }
1664
1665 /*
1666  * Run a set of guest threads on a physical core.
1667  * Called with vc->lock held.
1668  */
1669 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1670 {
1671         struct kvm_vcpu *vcpu, *vnext;
1672         long ret;
1673         u64 now;
1674         int i, need_vpa_update;
1675         int srcu_idx;
1676         struct kvm_vcpu *vcpus_to_update[threads_per_core];
1677
1678         /* don't start if any threads have a signal pending */
1679         need_vpa_update = 0;
1680         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1681                 if (signal_pending(vcpu->arch.run_task))
1682                         return;
1683                 if (vcpu->arch.vpa.update_pending ||
1684                     vcpu->arch.slb_shadow.update_pending ||
1685                     vcpu->arch.dtl.update_pending)
1686                         vcpus_to_update[need_vpa_update++] = vcpu;
1687         }
1688
1689         /*
1690          * Initialize *vc, in particular vc->vcore_state, so we can
1691          * drop the vcore lock if necessary.
1692          */
1693         vc->n_woken = 0;
1694         vc->nap_count = 0;
1695         vc->entry_exit_count = 0;
1696         vc->preempt_tb = TB_NIL;
1697         vc->vcore_state = VCORE_STARTING;
1698         vc->in_guest = 0;
1699         vc->napping_threads = 0;
1700
1701         /*
1702          * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1703          * which can't be called with any spinlocks held.
1704          */
1705         if (need_vpa_update) {
1706                 spin_unlock(&vc->lock);
1707                 for (i = 0; i < need_vpa_update; ++i)
1708                         kvmppc_update_vpas(vcpus_to_update[i]);
1709                 spin_lock(&vc->lock);
1710         }
1711
1712         /*
1713          * Make sure we are running on primary threads, and that secondary
1714          * threads are offline.  Also check if the number of threads in this
1715          * guest are greater than the current system threads per guest.
1716          */
1717         if ((threads_per_core > 1) &&
1718             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
1719                 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1720                         vcpu->arch.ret = -EBUSY;
1721                 goto out;
1722         }
1723
1724
1725         vc->pcpu = smp_processor_id();
1726         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1727                 kvmppc_start_thread(vcpu);
1728                 kvmppc_create_dtl_entry(vcpu, vc);
1729                 trace_kvm_guest_enter(vcpu);
1730         }
1731
1732         /* Set this explicitly in case thread 0 doesn't have a vcpu */
1733         get_paca()->kvm_hstate.kvm_vcore = vc;
1734         get_paca()->kvm_hstate.ptid = 0;
1735
1736         vc->vcore_state = VCORE_RUNNING;
1737         preempt_disable();
1738
1739         trace_kvmppc_run_core(vc, 0);
1740
1741         spin_unlock(&vc->lock);
1742
1743         kvm_guest_enter();
1744
1745         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
1746
1747         if (vc->mpp_buffer_is_valid)
1748                 kvmppc_start_restoring_l2_cache(vc);
1749
1750         __kvmppc_vcore_entry();
1751
1752         spin_lock(&vc->lock);
1753
1754         if (vc->mpp_buffer)
1755                 kvmppc_start_saving_l2_cache(vc);
1756
1757         /* disable sending of IPIs on virtual external irqs */
1758         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1759                 vcpu->cpu = -1;
1760         /* wait for secondary threads to finish writing their state to memory */
1761         if (vc->nap_count < vc->n_woken)
1762                 kvmppc_wait_for_nap(vc);
1763         for (i = 0; i < threads_per_subcore; ++i)
1764                 kvmppc_release_hwthread(vc->pcpu + i);
1765         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
1766         vc->vcore_state = VCORE_EXITING;
1767         spin_unlock(&vc->lock);
1768
1769         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
1770
1771         /* make sure updates to secondary vcpu structs are visible now */
1772         smp_mb();
1773         kvm_guest_exit();
1774
1775         preempt_enable();
1776         cond_resched();
1777
1778         spin_lock(&vc->lock);
1779         now = get_tb();
1780         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1781                 /* cancel pending dec exception if dec is positive */
1782                 if (now < vcpu->arch.dec_expires &&
1783                     kvmppc_core_pending_dec(vcpu))
1784                         kvmppc_core_dequeue_dec(vcpu);
1785
1786                 trace_kvm_guest_exit(vcpu);
1787
1788                 ret = RESUME_GUEST;
1789                 if (vcpu->arch.trap)
1790                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1791                                                     vcpu->arch.run_task);
1792
1793                 vcpu->arch.ret = ret;
1794                 vcpu->arch.trap = 0;
1795
1796                 if (vcpu->arch.ceded) {
1797                         if (!is_kvmppc_resume_guest(ret))
1798                                 kvmppc_end_cede(vcpu);
1799                         else
1800                                 kvmppc_set_timer(vcpu);
1801                 }
1802         }
1803
1804  out:
1805         vc->vcore_state = VCORE_INACTIVE;
1806         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1807                                  arch.run_list) {
1808                 if (!is_kvmppc_resume_guest(vcpu->arch.ret)) {
1809                         kvmppc_remove_runnable(vc, vcpu);
1810                         wake_up(&vcpu->arch.cpu_run);
1811                 }
1812         }
1813
1814         trace_kvmppc_run_core(vc, 1);
1815 }
1816
1817 /*
1818  * Wait for some other vcpu thread to execute us, and
1819  * wake us up when we need to handle something in the host.
1820  */
1821 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1822 {
1823         DEFINE_WAIT(wait);
1824
1825         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1826         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1827                 schedule();
1828         finish_wait(&vcpu->arch.cpu_run, &wait);
1829 }
1830
1831 /*
1832  * All the vcpus in this vcore are idle, so wait for a decrementer
1833  * or external interrupt to one of the vcpus.  vc->lock is held.
1834  */
1835 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1836 {
1837         struct kvm_vcpu *vcpu;
1838         int do_sleep = 1;
1839
1840         DEFINE_WAIT(wait);
1841
1842         prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1843
1844         /*
1845          * Check one last time for pending exceptions and ceded state after
1846          * we put ourselves on the wait queue
1847          */
1848         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1849                 if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded) {
1850                         do_sleep = 0;
1851                         break;
1852                 }
1853         }
1854
1855         if (!do_sleep) {
1856                 finish_wait(&vc->wq, &wait);
1857                 return;
1858         }
1859
1860         vc->vcore_state = VCORE_SLEEPING;
1861         trace_kvmppc_vcore_blocked(vc, 0);
1862         spin_unlock(&vc->lock);
1863         schedule();
1864         finish_wait(&vc->wq, &wait);
1865         spin_lock(&vc->lock);
1866         vc->vcore_state = VCORE_INACTIVE;
1867         trace_kvmppc_vcore_blocked(vc, 1);
1868 }
1869
1870 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1871 {
1872         int n_ceded;
1873         struct kvmppc_vcore *vc;
1874         struct kvm_vcpu *v, *vn;
1875
1876         trace_kvmppc_run_vcpu_enter(vcpu);
1877
1878         kvm_run->exit_reason = 0;
1879         vcpu->arch.ret = RESUME_GUEST;
1880         vcpu->arch.trap = 0;
1881         kvmppc_update_vpas(vcpu);
1882
1883         /*
1884          * Synchronize with other threads in this virtual core
1885          */
1886         vc = vcpu->arch.vcore;
1887         spin_lock(&vc->lock);
1888         vcpu->arch.ceded = 0;
1889         vcpu->arch.run_task = current;
1890         vcpu->arch.kvm_run = kvm_run;
1891         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
1892         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1893         vcpu->arch.busy_preempt = TB_NIL;
1894         list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1895         ++vc->n_runnable;
1896
1897         /*
1898          * This happens the first time this is called for a vcpu.
1899          * If the vcore is already running, we may be able to start
1900          * this thread straight away and have it join in.
1901          */
1902         if (!signal_pending(current)) {
1903                 if (vc->vcore_state == VCORE_RUNNING &&
1904                     VCORE_EXIT_COUNT(vc) == 0) {
1905                         kvmppc_create_dtl_entry(vcpu, vc);
1906                         kvmppc_start_thread(vcpu);
1907                         trace_kvm_guest_enter(vcpu);
1908                 } else if (vc->vcore_state == VCORE_SLEEPING) {
1909                         wake_up(&vc->wq);
1910                 }
1911
1912         }
1913
1914         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1915                !signal_pending(current)) {
1916                 if (vc->vcore_state != VCORE_INACTIVE) {
1917                         spin_unlock(&vc->lock);
1918                         kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1919                         spin_lock(&vc->lock);
1920                         continue;
1921                 }
1922                 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1923                                          arch.run_list) {
1924                         kvmppc_core_prepare_to_enter(v);
1925                         if (signal_pending(v->arch.run_task)) {
1926                                 kvmppc_remove_runnable(vc, v);
1927                                 v->stat.signal_exits++;
1928                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1929                                 v->arch.ret = -EINTR;
1930                                 wake_up(&v->arch.cpu_run);
1931                         }
1932                 }
1933                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1934                         break;
1935                 vc->runner = vcpu;
1936                 n_ceded = 0;
1937                 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
1938                         if (!v->arch.pending_exceptions)
1939                                 n_ceded += v->arch.ceded;
1940                         else
1941                                 v->arch.ceded = 0;
1942                 }
1943                 if (n_ceded == vc->n_runnable)
1944                         kvmppc_vcore_blocked(vc);
1945                 else
1946                         kvmppc_run_core(vc);
1947                 vc->runner = NULL;
1948         }
1949
1950         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1951                (vc->vcore_state == VCORE_RUNNING ||
1952                 vc->vcore_state == VCORE_EXITING)) {
1953                 spin_unlock(&vc->lock);
1954                 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1955                 spin_lock(&vc->lock);
1956         }
1957
1958         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1959                 kvmppc_remove_runnable(vc, vcpu);
1960                 vcpu->stat.signal_exits++;
1961                 kvm_run->exit_reason = KVM_EXIT_INTR;
1962                 vcpu->arch.ret = -EINTR;
1963         }
1964
1965         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1966                 /* Wake up some vcpu to run the core */
1967                 v = list_first_entry(&vc->runnable_threads,
1968                                      struct kvm_vcpu, arch.run_list);
1969                 wake_up(&v->arch.cpu_run);
1970         }
1971
1972         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
1973         spin_unlock(&vc->lock);
1974         return vcpu->arch.ret;
1975 }
1976
1977 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
1978 {
1979         int r;
1980         int srcu_idx;
1981
1982         if (!vcpu->arch.sane) {
1983                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1984                 return -EINVAL;
1985         }
1986
1987         kvmppc_core_prepare_to_enter(vcpu);
1988
1989         /* No need to go into the guest when all we'll do is come back out */
1990         if (signal_pending(current)) {
1991                 run->exit_reason = KVM_EXIT_INTR;
1992                 return -EINTR;
1993         }
1994
1995         atomic_inc(&vcpu->kvm->arch.vcpus_running);
1996         /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1997         smp_mb();
1998
1999         /* On the first time here, set up HTAB and VRMA */
2000         if (!vcpu->kvm->arch.rma_setup_done) {
2001                 r = kvmppc_hv_setup_htab_rma(vcpu);
2002                 if (r)
2003                         goto out;
2004         }
2005
2006         flush_fp_to_thread(current);
2007         flush_altivec_to_thread(current);
2008         flush_vsx_to_thread(current);
2009         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2010         vcpu->arch.pgdir = current->mm->pgd;
2011         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2012
2013         do {
2014                 r = kvmppc_run_vcpu(run, vcpu);
2015
2016                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2017                     !(vcpu->arch.shregs.msr & MSR_PR)) {
2018                         trace_kvm_hcall_enter(vcpu);
2019                         r = kvmppc_pseries_do_hcall(vcpu);
2020                         trace_kvm_hcall_exit(vcpu, r);
2021                         kvmppc_core_prepare_to_enter(vcpu);
2022                 } else if (r == RESUME_PAGE_FAULT) {
2023                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2024                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
2025                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2026                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2027                 }
2028         } while (is_kvmppc_resume_guest(r));
2029
2030  out:
2031         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2032         atomic_dec(&vcpu->kvm->arch.vcpus_running);
2033         return r;
2034 }
2035
2036 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2037                                      int linux_psize)
2038 {
2039         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2040
2041         if (!def->shift)
2042                 return;
2043         (*sps)->page_shift = def->shift;
2044         (*sps)->slb_enc = def->sllp;
2045         (*sps)->enc[0].page_shift = def->shift;
2046         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
2047         /*
2048          * Add 16MB MPSS support if host supports it
2049          */
2050         if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2051                 (*sps)->enc[1].page_shift = 24;
2052                 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2053         }
2054         (*sps)++;
2055 }
2056
2057 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2058                                          struct kvm_ppc_smmu_info *info)
2059 {
2060         struct kvm_ppc_one_seg_page_size *sps;
2061
2062         info->flags = KVM_PPC_PAGE_SIZES_REAL;
2063         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2064                 info->flags |= KVM_PPC_1T_SEGMENTS;
2065         info->slb_size = mmu_slb_size;
2066
2067         /* We only support these sizes for now, and no muti-size segments */
2068         sps = &info->sps[0];
2069         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2070         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2071         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2072
2073         return 0;
2074 }
2075
2076 /*
2077  * Get (and clear) the dirty memory log for a memory slot.
2078  */
2079 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2080                                          struct kvm_dirty_log *log)
2081 {
2082         struct kvm_memory_slot *memslot;
2083         int r;
2084         unsigned long n;
2085
2086         mutex_lock(&kvm->slots_lock);
2087
2088         r = -EINVAL;
2089         if (log->slot >= KVM_USER_MEM_SLOTS)
2090                 goto out;
2091
2092         memslot = id_to_memslot(kvm->memslots, log->slot);
2093         r = -ENOENT;
2094         if (!memslot->dirty_bitmap)
2095                 goto out;
2096
2097         n = kvm_dirty_bitmap_bytes(memslot);
2098         memset(memslot->dirty_bitmap, 0, n);
2099
2100         r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2101         if (r)
2102                 goto out;
2103
2104         r = -EFAULT;
2105         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2106                 goto out;
2107
2108         r = 0;
2109 out:
2110         mutex_unlock(&kvm->slots_lock);
2111         return r;
2112 }
2113
2114 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2115                                         struct kvm_memory_slot *dont)
2116 {
2117         if (!dont || free->arch.rmap != dont->arch.rmap) {
2118                 vfree(free->arch.rmap);
2119                 free->arch.rmap = NULL;
2120         }
2121 }
2122
2123 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2124                                          unsigned long npages)
2125 {
2126         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2127         if (!slot->arch.rmap)
2128                 return -ENOMEM;
2129
2130         return 0;
2131 }
2132
2133 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2134                                         struct kvm_memory_slot *memslot,
2135                                         struct kvm_userspace_memory_region *mem)
2136 {
2137         return 0;
2138 }
2139
2140 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2141                                 struct kvm_userspace_memory_region *mem,
2142                                 const struct kvm_memory_slot *old)
2143 {
2144         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2145         struct kvm_memory_slot *memslot;
2146
2147         if (npages && old->npages) {
2148                 /*
2149                  * If modifying a memslot, reset all the rmap dirty bits.
2150                  * If this is a new memslot, we don't need to do anything
2151                  * since the rmap array starts out as all zeroes,
2152                  * i.e. no pages are dirty.
2153                  */
2154                 memslot = id_to_memslot(kvm->memslots, mem->slot);
2155                 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2156         }
2157 }
2158
2159 /*
2160  * Update LPCR values in kvm->arch and in vcores.
2161  * Caller must hold kvm->lock.
2162  */
2163 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2164 {
2165         long int i;
2166         u32 cores_done = 0;
2167
2168         if ((kvm->arch.lpcr & mask) == lpcr)
2169                 return;
2170
2171         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2172
2173         for (i = 0; i < KVM_MAX_VCORES; ++i) {
2174                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2175                 if (!vc)
2176                         continue;
2177                 spin_lock(&vc->lock);
2178                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2179                 spin_unlock(&vc->lock);
2180                 if (++cores_done >= kvm->arch.online_vcores)
2181                         break;
2182         }
2183 }
2184
2185 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2186 {
2187         return;
2188 }
2189
2190 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2191 {
2192         int err = 0;
2193         struct kvm *kvm = vcpu->kvm;
2194         unsigned long hva;
2195         struct kvm_memory_slot *memslot;
2196         struct vm_area_struct *vma;
2197         unsigned long lpcr = 0, senc;
2198         unsigned long psize, porder;
2199         int srcu_idx;
2200
2201         mutex_lock(&kvm->lock);
2202         if (kvm->arch.rma_setup_done)
2203                 goto out;       /* another vcpu beat us to it */
2204
2205         /* Allocate hashed page table (if not done already) and reset it */
2206         if (!kvm->arch.hpt_virt) {
2207                 err = kvmppc_alloc_hpt(kvm, NULL);
2208                 if (err) {
2209                         pr_err("KVM: Couldn't alloc HPT\n");
2210                         goto out;
2211                 }
2212         }
2213
2214         /* Look up the memslot for guest physical address 0 */
2215         srcu_idx = srcu_read_lock(&kvm->srcu);
2216         memslot = gfn_to_memslot(kvm, 0);
2217
2218         /* We must have some memory at 0 by now */
2219         err = -EINVAL;
2220         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2221                 goto out_srcu;
2222
2223         /* Look up the VMA for the start of this memory slot */
2224         hva = memslot->userspace_addr;
2225         down_read(&current->mm->mmap_sem);
2226         vma = find_vma(current->mm, hva);
2227         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2228                 goto up_out;
2229
2230         psize = vma_kernel_pagesize(vma);
2231         porder = __ilog2(psize);
2232
2233         up_read(&current->mm->mmap_sem);
2234
2235         /* We can handle 4k, 64k or 16M pages in the VRMA */
2236         err = -EINVAL;
2237         if (!(psize == 0x1000 || psize == 0x10000 ||
2238               psize == 0x1000000))
2239                 goto out_srcu;
2240
2241         /* Update VRMASD field in the LPCR */
2242         senc = slb_pgsize_encoding(psize);
2243         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2244                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
2245         /* the -4 is to account for senc values starting at 0x10 */
2246         lpcr = senc << (LPCR_VRMASD_SH - 4);
2247
2248         /* Create HPTEs in the hash page table for the VRMA */
2249         kvmppc_map_vrma(vcpu, memslot, porder);
2250
2251         kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
2252
2253         /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
2254         smp_wmb();
2255         kvm->arch.rma_setup_done = 1;
2256         err = 0;
2257  out_srcu:
2258         srcu_read_unlock(&kvm->srcu, srcu_idx);
2259  out:
2260         mutex_unlock(&kvm->lock);
2261         return err;
2262
2263  up_out:
2264         up_read(&current->mm->mmap_sem);
2265         goto out_srcu;
2266 }
2267
2268 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2269 {
2270         unsigned long lpcr, lpid;
2271
2272         /* Allocate the guest's logical partition ID */
2273
2274         lpid = kvmppc_alloc_lpid();
2275         if ((long)lpid < 0)
2276                 return -ENOMEM;
2277         kvm->arch.lpid = lpid;
2278
2279         /*
2280          * Since we don't flush the TLB when tearing down a VM,
2281          * and this lpid might have previously been used,
2282          * make sure we flush on each core before running the new VM.
2283          */
2284         cpumask_setall(&kvm->arch.need_tlb_flush);
2285
2286         /* Start out with the default set of hcalls enabled */
2287         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
2288                sizeof(kvm->arch.enabled_hcalls));
2289
2290         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
2291
2292         /* Init LPCR for virtual RMA mode */
2293         kvm->arch.host_lpid = mfspr(SPRN_LPID);
2294         kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2295         lpcr &= LPCR_PECE | LPCR_LPES;
2296         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
2297                 LPCR_VPM0 | LPCR_VPM1;
2298         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2299                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
2300         /* On POWER8 turn on online bit to enable PURR/SPURR */
2301         if (cpu_has_feature(CPU_FTR_ARCH_207S))
2302                 lpcr |= LPCR_ONL;
2303         kvm->arch.lpcr = lpcr;
2304
2305         /*
2306          * Track that we now have a HV mode VM active. This blocks secondary
2307          * CPU threads from coming online.
2308          */
2309         kvm_hv_vm_activated();
2310
2311         return 0;
2312 }
2313
2314 static void kvmppc_free_vcores(struct kvm *kvm)
2315 {
2316         long int i;
2317
2318         for (i = 0; i < KVM_MAX_VCORES; ++i) {
2319                 if (kvm->arch.vcores[i] && kvm->arch.vcores[i]->mpp_buffer) {
2320                         struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2321                         free_pages((unsigned long)vc->mpp_buffer,
2322                                    MPP_BUFFER_ORDER);
2323                 }
2324                 kfree(kvm->arch.vcores[i]);
2325         }
2326         kvm->arch.online_vcores = 0;
2327 }
2328
2329 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
2330 {
2331         kvm_hv_vm_deactivated();
2332
2333         kvmppc_free_vcores(kvm);
2334
2335         kvmppc_free_hpt(kvm);
2336 }
2337
2338 /* We don't need to emulate any privileged instructions or dcbz */
2339 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2340                                      unsigned int inst, int *advance)
2341 {
2342         return EMULATE_FAIL;
2343 }
2344
2345 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2346                                         ulong spr_val)
2347 {
2348         return EMULATE_FAIL;
2349 }
2350
2351 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2352                                         ulong *spr_val)
2353 {
2354         return EMULATE_FAIL;
2355 }
2356
2357 static int kvmppc_core_check_processor_compat_hv(void)
2358 {
2359         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
2360             !cpu_has_feature(CPU_FTR_ARCH_206))
2361                 return -EIO;
2362         return 0;
2363 }
2364
2365 static long kvm_arch_vm_ioctl_hv(struct file *filp,
2366                                  unsigned int ioctl, unsigned long arg)
2367 {
2368         struct kvm *kvm __maybe_unused = filp->private_data;
2369         void __user *argp = (void __user *)arg;
2370         long r;
2371
2372         switch (ioctl) {
2373
2374         case KVM_PPC_ALLOCATE_HTAB: {
2375                 u32 htab_order;
2376
2377                 r = -EFAULT;
2378                 if (get_user(htab_order, (u32 __user *)argp))
2379                         break;
2380                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2381                 if (r)
2382                         break;
2383                 r = -EFAULT;
2384                 if (put_user(htab_order, (u32 __user *)argp))
2385                         break;
2386                 r = 0;
2387                 break;
2388         }
2389
2390         case KVM_PPC_GET_HTAB_FD: {
2391                 struct kvm_get_htab_fd ghf;
2392
2393                 r = -EFAULT;
2394                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
2395                         break;
2396                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2397                 break;
2398         }
2399
2400         default:
2401                 r = -ENOTTY;
2402         }
2403
2404         return r;
2405 }
2406
2407 /*
2408  * List of hcall numbers to enable by default.
2409  * For compatibility with old userspace, we enable by default
2410  * all hcalls that were implemented before the hcall-enabling
2411  * facility was added.  Note this list should not include H_RTAS.
2412  */
2413 static unsigned int default_hcall_list[] = {
2414         H_REMOVE,
2415         H_ENTER,
2416         H_READ,
2417         H_PROTECT,
2418         H_BULK_REMOVE,
2419         H_GET_TCE,
2420         H_PUT_TCE,
2421         H_SET_DABR,
2422         H_SET_XDABR,
2423         H_CEDE,
2424         H_PROD,
2425         H_CONFER,
2426         H_REGISTER_VPA,
2427 #ifdef CONFIG_KVM_XICS
2428         H_EOI,
2429         H_CPPR,
2430         H_IPI,
2431         H_IPOLL,
2432         H_XIRR,
2433         H_XIRR_X,
2434 #endif
2435         0
2436 };
2437
2438 static void init_default_hcalls(void)
2439 {
2440         int i;
2441         unsigned int hcall;
2442
2443         for (i = 0; default_hcall_list[i]; ++i) {
2444                 hcall = default_hcall_list[i];
2445                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
2446                 __set_bit(hcall / 4, default_enabled_hcalls);
2447         }
2448 }
2449
2450 static struct kvmppc_ops kvm_ops_hv = {
2451         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2452         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2453         .get_one_reg = kvmppc_get_one_reg_hv,
2454         .set_one_reg = kvmppc_set_one_reg_hv,
2455         .vcpu_load   = kvmppc_core_vcpu_load_hv,
2456         .vcpu_put    = kvmppc_core_vcpu_put_hv,
2457         .set_msr     = kvmppc_set_msr_hv,
2458         .vcpu_run    = kvmppc_vcpu_run_hv,
2459         .vcpu_create = kvmppc_core_vcpu_create_hv,
2460         .vcpu_free   = kvmppc_core_vcpu_free_hv,
2461         .check_requests = kvmppc_core_check_requests_hv,
2462         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
2463         .flush_memslot  = kvmppc_core_flush_memslot_hv,
2464         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2465         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
2466         .unmap_hva = kvm_unmap_hva_hv,
2467         .unmap_hva_range = kvm_unmap_hva_range_hv,
2468         .age_hva  = kvm_age_hva_hv,
2469         .test_age_hva = kvm_test_age_hva_hv,
2470         .set_spte_hva = kvm_set_spte_hva_hv,
2471         .mmu_destroy  = kvmppc_mmu_destroy_hv,
2472         .free_memslot = kvmppc_core_free_memslot_hv,
2473         .create_memslot = kvmppc_core_create_memslot_hv,
2474         .init_vm =  kvmppc_core_init_vm_hv,
2475         .destroy_vm = kvmppc_core_destroy_vm_hv,
2476         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2477         .emulate_op = kvmppc_core_emulate_op_hv,
2478         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2479         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2480         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2481         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
2482         .hcall_implemented = kvmppc_hcall_impl_hv,
2483 };
2484
2485 static int kvmppc_book3s_init_hv(void)
2486 {
2487         int r;
2488         /*
2489          * FIXME!! Do we need to check on all cpus ?
2490          */
2491         r = kvmppc_core_check_processor_compat_hv();
2492         if (r < 0)
2493                 return -ENODEV;
2494
2495         kvm_ops_hv.owner = THIS_MODULE;
2496         kvmppc_hv_ops = &kvm_ops_hv;
2497
2498         init_default_hcalls();
2499
2500         r = kvmppc_mmu_hv_init();
2501         return r;
2502 }
2503
2504 static void kvmppc_book3s_exit_hv(void)
2505 {
2506         kvmppc_hv_ops = NULL;
2507 }
2508
2509 module_init(kvmppc_book3s_init_hv);
2510 module_exit(kvmppc_book3s_exit_hv);
2511 MODULE_LICENSE("GPL");
2512 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2513 MODULE_ALIAS("devname:kvm");