kernel/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 #include <linux/debugfs.h>
  36
  37 #include <asm/reg.h>
  38 #include <asm/cputable.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 <asm/dbell.h>
  54 #include <linux/gfp.h>
  55 #include <linux/vmalloc.h>
  56 #include <linux/highmem.h>
  57 #include <linux/hugetlb.h>
  58 #include <linux/module.h>
  59
  60 #include "book3s.h"
  61
  62 #define CREATE_TRACE_POINTS
  63 #include "trace_hv.h"
  64
  65 /* #define EXIT_DEBUG */
  66 /* #define EXIT_DEBUG_SIMPLE */
  67 /* #define EXIT_DEBUG_INT */
  68
  69 /* Used to indicate that a guest page fault needs to be handled */
  70 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  71
  72 /* Used as a "null" value for timebase values */
  73 #define TB_NIL  (~(u64)0)
  74
  75 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
  76
  77 static int dynamic_mt_modes = 6;
  78 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
  79 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
  80 static int target_smt_mode;
  81 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
  82 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
  83
  84 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
  85 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
  86
  87 static bool kvmppc_ipi_thread(int cpu)
  88 {
  89         /* On POWER8 for IPIs to threads in the same core, use msgsnd */
  90         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
  91                 preempt_disable();
  92                 if (cpu_first_thread_sibling(cpu) ==
  93                     cpu_first_thread_sibling(smp_processor_id())) {
  94                         unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
  95                         msg |= cpu_thread_in_core(cpu);
  96                         smp_mb();
  97                         __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
  98                         preempt_enable();
  99                         return true;
 100                 }
 101                 preempt_enable();
 102         }
 103
 104 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
 105         if (cpu >= 0 && cpu < nr_cpu_ids && paca[cpu].kvm_hstate.xics_phys) {
 106                 xics_wake_cpu(cpu);
 107                 return true;
 108         }
 109 #endif
 110
 111         return false;
 112 }
 113
 114 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
 115 {
 116         int cpu;
 117         struct swait_queue_head *wqp;
 118
 119         wqp = kvm_arch_vcpu_wq(vcpu);
 120         if (swait_active(wqp)) {
 121                 swake_up(wqp);
 122                 ++vcpu->stat.halt_wakeup;
 123         }
 124
 125         if (kvmppc_ipi_thread(vcpu->arch.thread_cpu))
 126                 return;
 127
 128         /* CPU points to the first thread of the core */
 129         cpu = vcpu->cpu;
 130         if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
 131                 smp_send_reschedule(cpu);
 132 }
 133
 134 /*
 135  * We use the vcpu_load/put functions to measure stolen time.
 136  * Stolen time is counted as time when either the vcpu is able to
 137  * run as part of a virtual core, but the task running the vcore
 138  * is preempted or sleeping, or when the vcpu needs something done
 139  * in the kernel by the task running the vcpu, but that task is
 140  * preempted or sleeping.  Those two things have to be counted
 141  * separately, since one of the vcpu tasks will take on the job
 142  * of running the core, and the other vcpu tasks in the vcore will
 143  * sleep waiting for it to do that, but that sleep shouldn't count
 144  * as stolen time.
 145  *
 146  * Hence we accumulate stolen time when the vcpu can run as part of
 147  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 148  * needs its task to do other things in the kernel (for example,
 149  * service a page fault) in busy_stolen.  We don't accumulate
 150  * stolen time for a vcore when it is inactive, or for a vcpu
 151  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 152  * a misnomer; it means that the vcpu task is not executing in
 153  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 154  * the kernel.  We don't have any way of dividing up that time
 155  * between time that the vcpu is genuinely stopped, time that
 156  * the task is actively working on behalf of the vcpu, and time
 157  * that the task is preempted, so we don't count any of it as
 158  * stolen.
 159  *
 160  * Updates to busy_stolen are protected by arch.tbacct_lock;
 161  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
 162  * lock.  The stolen times are measured in units of timebase ticks.
 163  * (Note that the != TB_NIL checks below are purely defensive;
 164  * they should never fail.)
 165  */
 166
 167 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
 168 {
 169         unsigned long flags;
 170
 171         spin_lock_irqsave(&vc->stoltb_lock, flags);
 172         vc->preempt_tb = mftb();
 173         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 174 }
 175
 176 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
 177 {
 178         unsigned long flags;
 179
 180         spin_lock_irqsave(&vc->stoltb_lock, flags);
 181         if (vc->preempt_tb != TB_NIL) {
 182                 vc->stolen_tb += mftb() - vc->preempt_tb;
 183                 vc->preempt_tb = TB_NIL;
 184         }
 185         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 186 }
 187
 188 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 189 {
 190         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 191         unsigned long flags;
 192
 193         /*
 194          * We can test vc->runner without taking the vcore lock,
 195          * because only this task ever sets vc->runner to this
 196          * vcpu, and once it is set to this vcpu, only this task
 197          * ever sets it to NULL.
 198          */
 199         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 200                 kvmppc_core_end_stolen(vc);
 201
 202         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 203         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 204             vcpu->arch.busy_preempt != TB_NIL) {
 205                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 206                 vcpu->arch.busy_preempt = TB_NIL;
 207         }
 208         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 209 }
 210
 211 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 212 {
 213         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 214         unsigned long flags;
 215
 216         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 217                 kvmppc_core_start_stolen(vc);
 218
 219         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 220         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 221                 vcpu->arch.busy_preempt = mftb();
 222         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 223 }
 224
 225 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
 226 {
 227         /*
 228          * Check for illegal transactional state bit combination
 229          * and if we find it, force the TS field to a safe state.
 230          */
 231         if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
 232                 msr &= ~MSR_TS_MASK;
 233         vcpu->arch.shregs.msr = msr;
 234         kvmppc_end_cede(vcpu);
 235 }
 236
 237 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 238 {
 239         vcpu->arch.pvr = pvr;
 240 }
 241
 242 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 243 {
 244         unsigned long pcr = 0;
 245         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 246
 247         if (arch_compat) {
 248                 switch (arch_compat) {
 249                 case PVR_ARCH_205:
 250                         /*
 251                          * If an arch bit is set in PCR, all the defined
 252                          * higher-order arch bits also have to be set.
 253                          */
 254                         pcr = PCR_ARCH_206 | PCR_ARCH_205;
 255                         break;
 256                 case PVR_ARCH_206:
 257                 case PVR_ARCH_206p:
 258                         pcr = PCR_ARCH_206;
 259                         break;
 260                 case PVR_ARCH_207:
 261                         break;
 262                 default:
 263                         return -EINVAL;
 264                 }
 265
 266                 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
 267                         /* POWER7 can't emulate POWER8 */
 268                         if (!(pcr & PCR_ARCH_206))
 269                                 return -EINVAL;
 270                         pcr &= ~PCR_ARCH_206;
 271                 }
 272         }
 273
 274         spin_lock(&vc->lock);
 275         vc->arch_compat = arch_compat;
 276         vc->pcr = pcr;
 277         spin_unlock(&vc->lock);
 278
 279         return 0;
 280 }
 281
 282 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 283 {
 284         int r;
 285
 286         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 287         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 288                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
 289         for (r = 0; r < 16; ++r)
 290                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 291                        r, kvmppc_get_gpr(vcpu, r),
 292                        r+16, kvmppc_get_gpr(vcpu, r+16));
 293         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 294                vcpu->arch.ctr, vcpu->arch.lr);
 295         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 296                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 297         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 298                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 299         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 300                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 301         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 302                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 303         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 304         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 305                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 306         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 307         for (r = 0; r < vcpu->arch.slb_max; ++r)
 308                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 309                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 310         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 311                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 312                vcpu->arch.last_inst);
 313 }
 314
 315 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 316 {
 317         int r;
 318         struct kvm_vcpu *v, *ret = NULL;
 319
 320         mutex_lock(&kvm->lock);
 321         kvm_for_each_vcpu(r, v, kvm) {
 322                 if (v->vcpu_id == id) {
 323                         ret = v;
 324                         break;
 325                 }
 326         }
 327         mutex_unlock(&kvm->lock);
 328         return ret;
 329 }
 330
 331 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 332 {
 333         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 334         vpa->yield_count = cpu_to_be32(1);
 335 }
 336
 337 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 338                    unsigned long addr, unsigned long len)
 339 {
 340         /* check address is cacheline aligned */
 341         if (addr & (L1_CACHE_BYTES - 1))
 342                 return -EINVAL;
 343         spin_lock(&vcpu->arch.vpa_update_lock);
 344         if (v->next_gpa != addr || v->len != len) {
 345                 v->next_gpa = addr;
 346                 v->len = addr ? len : 0;
 347                 v->update_pending = 1;
 348         }
 349         spin_unlock(&vcpu->arch.vpa_update_lock);
 350         return 0;
 351 }
 352
 353 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 354 struct reg_vpa {
 355         u32 dummy;
 356         union {
 357                 __be16 hword;
 358                 __be32 word;
 359         } length;
 360 };
 361
 362 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 363 {
 364         if (vpap->update_pending)
 365                 return vpap->next_gpa != 0;
 366         return vpap->pinned_addr != NULL;
 367 }
 368
 369 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 370                                        unsigned long flags,
 371                                        unsigned long vcpuid, unsigned long vpa)
 372 {
 373         struct kvm *kvm = vcpu->kvm;
 374         unsigned long len, nb;
 375         void *va;
 376         struct kvm_vcpu *tvcpu;
 377         int err;
 378         int subfunc;
 379         struct kvmppc_vpa *vpap;
 380
 381         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 382         if (!tvcpu)
 383                 return H_PARAMETER;
 384
 385         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 386         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 387             subfunc == H_VPA_REG_SLB) {
 388                 /* Registering new area - address must be cache-line aligned */
 389                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 390                         return H_PARAMETER;
 391
 392                 /* convert logical addr to kernel addr and read length */
 393                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 394                 if (va == NULL)
 395                         return H_PARAMETER;
 396                 if (subfunc == H_VPA_REG_VPA)
 397                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 398                 else
 399                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 400                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 401
 402                 /* Check length */
 403                 if (len > nb || len < sizeof(struct reg_vpa))
 404                         return H_PARAMETER;
 405         } else {
 406                 vpa = 0;
 407                 len = 0;
 408         }
 409
 410         err = H_PARAMETER;
 411         vpap = NULL;
 412         spin_lock(&tvcpu->arch.vpa_update_lock);
 413
 414         switch (subfunc) {
 415         case H_VPA_REG_VPA:             /* register VPA */
 416                 if (len < sizeof(struct lppaca))
 417                         break;
 418                 vpap = &tvcpu->arch.vpa;
 419                 err = 0;
 420                 break;
 421
 422         case H_VPA_REG_DTL:             /* register DTL */
 423                 if (len < sizeof(struct dtl_entry))
 424                         break;
 425                 len -= len % sizeof(struct dtl_entry);
 426
 427                 /* Check that they have previously registered a VPA */
 428                 err = H_RESOURCE;
 429                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 430                         break;
 431
 432                 vpap = &tvcpu->arch.dtl;
 433                 err = 0;
 434                 break;
 435
 436         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 437                 /* Check that they have previously registered a VPA */
 438                 err = H_RESOURCE;
 439                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 440                         break;
 441
 442                 vpap = &tvcpu->arch.slb_shadow;
 443                 err = 0;
 444                 break;
 445
 446         case H_VPA_DEREG_VPA:           /* deregister VPA */
 447                 /* Check they don't still have a DTL or SLB buf registered */
 448                 err = H_RESOURCE;
 449                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 450                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 451                         break;
 452
 453                 vpap = &tvcpu->arch.vpa;
 454                 err = 0;
 455                 break;
 456
 457         case H_VPA_DEREG_DTL:           /* deregister DTL */
 458                 vpap = &tvcpu->arch.dtl;
 459                 err = 0;
 460                 break;
 461
 462         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 463                 vpap = &tvcpu->arch.slb_shadow;
 464                 err = 0;
 465                 break;
 466         }
 467
 468         if (vpap) {
 469                 vpap->next_gpa = vpa;
 470                 vpap->len = len;
 471                 vpap->update_pending = 1;
 472         }
 473
 474         spin_unlock(&tvcpu->arch.vpa_update_lock);
 475
 476         return err;
 477 }
 478
 479 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 480 {
 481         struct kvm *kvm = vcpu->kvm;
 482         void *va;
 483         unsigned long nb;
 484         unsigned long gpa;
 485
 486         /*
 487          * We need to pin the page pointed to by vpap->next_gpa,
 488          * but we can't call kvmppc_pin_guest_page under the lock
 489          * as it does get_user_pages() and down_read().  So we
 490          * have to drop the lock, pin the page, then get the lock
 491          * again and check that a new area didn't get registered
 492          * in the meantime.
 493          */
 494         for (;;) {
 495                 gpa = vpap->next_gpa;
 496                 spin_unlock(&vcpu->arch.vpa_update_lock);
 497                 va = NULL;
 498                 nb = 0;
 499                 if (gpa)
 500                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 501                 spin_lock(&vcpu->arch.vpa_update_lock);
 502                 if (gpa == vpap->next_gpa)
 503                         break;
 504                 /* sigh... unpin that one and try again */
 505                 if (va)
 506                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 507         }
 508
 509         vpap->update_pending = 0;
 510         if (va && nb < vpap->len) {
 511                 /*
 512                  * If it's now too short, it must be that userspace
 513                  * has changed the mappings underlying guest memory,
 514                  * so unregister the region.
 515                  */
 516                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 517                 va = NULL;
 518         }
 519         if (vpap->pinned_addr)
 520                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 521                                         vpap->dirty);
 522         vpap->gpa = gpa;
 523         vpap->pinned_addr = va;
 524         vpap->dirty = false;
 525         if (va)
 526                 vpap->pinned_end = va + vpap->len;
 527 }
 528
 529 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 530 {
 531         if (!(vcpu->arch.vpa.update_pending ||
 532               vcpu->arch.slb_shadow.update_pending ||
 533               vcpu->arch.dtl.update_pending))
 534                 return;
 535
 536         spin_lock(&vcpu->arch.vpa_update_lock);
 537         if (vcpu->arch.vpa.update_pending) {
 538                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 539                 if (vcpu->arch.vpa.pinned_addr)
 540                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 541         }
 542         if (vcpu->arch.dtl.update_pending) {
 543                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 544                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 545                 vcpu->arch.dtl_index = 0;
 546         }
 547         if (vcpu->arch.slb_shadow.update_pending)
 548                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 549         spin_unlock(&vcpu->arch.vpa_update_lock);
 550 }
 551
 552 /*
 553  * Return the accumulated stolen time for the vcore up until `now'.
 554  * The caller should hold the vcore lock.
 555  */
 556 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 557 {
 558         u64 p;
 559         unsigned long flags;
 560
 561         spin_lock_irqsave(&vc->stoltb_lock, flags);
 562         p = vc->stolen_tb;
 563         if (vc->vcore_state != VCORE_INACTIVE &&
 564             vc->preempt_tb != TB_NIL)
 565                 p += now - vc->preempt_tb;
 566         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 567         return p;
 568 }
 569
 570 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 571                                     struct kvmppc_vcore *vc)
 572 {
 573         struct dtl_entry *dt;
 574         struct lppaca *vpa;
 575         unsigned long stolen;
 576         unsigned long core_stolen;
 577         u64 now;
 578
 579         dt = vcpu->arch.dtl_ptr;
 580         vpa = vcpu->arch.vpa.pinned_addr;
 581         now = mftb();
 582         core_stolen = vcore_stolen_time(vc, now);
 583         stolen = core_stolen - vcpu->arch.stolen_logged;
 584         vcpu->arch.stolen_logged = core_stolen;
 585         spin_lock_irq(&vcpu->arch.tbacct_lock);
 586         stolen += vcpu->arch.busy_stolen;
 587         vcpu->arch.busy_stolen = 0;
 588         spin_unlock_irq(&vcpu->arch.tbacct_lock);
 589         if (!dt || !vpa)
 590                 return;
 591         memset(dt, 0, sizeof(struct dtl_entry));
 592         dt->dispatch_reason = 7;
 593         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 594         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 595         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 596         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 597         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 598         ++dt;
 599         if (dt == vcpu->arch.dtl.pinned_end)
 600                 dt = vcpu->arch.dtl.pinned_addr;
 601         vcpu->arch.dtl_ptr = dt;
 602         /* order writing *dt vs. writing vpa->dtl_idx */
 603         smp_wmb();
 604         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 605         vcpu->arch.dtl.dirty = true;
 606 }
 607
 608 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 609 {
 610         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 611                 return true;
 612         if ((!vcpu->arch.vcore->arch_compat) &&
 613             cpu_has_feature(CPU_FTR_ARCH_207S))
 614                 return true;
 615         return false;
 616 }
 617
 618 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 619                              unsigned long resource, unsigned long value1,
 620                              unsigned long value2)
 621 {
 622         switch (resource) {
 623         case H_SET_MODE_RESOURCE_SET_CIABR:
 624                 if (!kvmppc_power8_compatible(vcpu))
 625                         return H_P2;
 626                 if (value2)
 627                         return H_P4;
 628                 if (mflags)
 629                         return H_UNSUPPORTED_FLAG_START;
 630                 /* Guests can't breakpoint the hypervisor */
 631                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 632                         return H_P3;
 633                 vcpu->arch.ciabr  = value1;
 634                 return H_SUCCESS;
 635         case H_SET_MODE_RESOURCE_SET_DAWR:
 636                 if (!kvmppc_power8_compatible(vcpu))
 637                         return H_P2;
 638                 if (mflags)
 639                         return H_UNSUPPORTED_FLAG_START;
 640                 if (value2 & DABRX_HYP)
 641                         return H_P4;
 642                 vcpu->arch.dawr  = value1;
 643                 vcpu->arch.dawrx = value2;
 644                 return H_SUCCESS;
 645         default:
 646                 return H_TOO_HARD;
 647         }
 648 }
 649
 650 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 651 {
 652         struct kvmppc_vcore *vcore = target->arch.vcore;
 653
 654         /*
 655          * We expect to have been called by the real mode handler
 656          * (kvmppc_rm_h_confer()) which would have directly returned
 657          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 658          * have useful work to do and should not confer) so we don't
 659          * recheck that here.
 660          */
 661
 662         spin_lock(&vcore->lock);
 663         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 664             vcore->vcore_state != VCORE_INACTIVE &&
 665             vcore->runner)
 666                 target = vcore->runner;
 667         spin_unlock(&vcore->lock);
 668
 669         return kvm_vcpu_yield_to(target);
 670 }
 671
 672 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 673 {
 674         int yield_count = 0;
 675         struct lppaca *lppaca;
 676
 677         spin_lock(&vcpu->arch.vpa_update_lock);
 678         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 679         if (lppaca)
 680                 yield_count = be32_to_cpu(lppaca->yield_count);
 681         spin_unlock(&vcpu->arch.vpa_update_lock);
 682         return yield_count;
 683 }
 684
 685 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 686 {
 687         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 688         unsigned long target, ret = H_SUCCESS;
 689         int yield_count;
 690         struct kvm_vcpu *tvcpu;
 691         int idx, rc;
 692
 693         if (req <= MAX_HCALL_OPCODE &&
 694             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 695                 return RESUME_HOST;
 696
 697         switch (req) {
 698         case H_CEDE:
 699                 break;
 700         case H_PROD:
 701                 target = kvmppc_get_gpr(vcpu, 4);
 702                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 703                 if (!tvcpu) {
 704                         ret = H_PARAMETER;
 705                         break;
 706                 }
 707                 tvcpu->arch.prodded = 1;
 708                 smp_mb();
 709                 if (vcpu->arch.ceded) {
 710                         if (swait_active(&vcpu->wq)) {
 711                                 swake_up(&vcpu->wq);
 712                                 vcpu->stat.halt_wakeup++;
 713                         }
 714                 }
 715                 break;
 716         case H_CONFER:
 717                 target = kvmppc_get_gpr(vcpu, 4);
 718                 if (target == -1)
 719                         break;
 720                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 721                 if (!tvcpu) {
 722                         ret = H_PARAMETER;
 723                         break;
 724                 }
 725                 yield_count = kvmppc_get_gpr(vcpu, 5);
 726                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 727                         break;
 728                 kvm_arch_vcpu_yield_to(tvcpu);
 729                 break;
 730         case H_REGISTER_VPA:
 731                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 732                                         kvmppc_get_gpr(vcpu, 5),
 733                                         kvmppc_get_gpr(vcpu, 6));
 734                 break;
 735         case H_RTAS:
 736                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 737                         return RESUME_HOST;
 738
 739                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 740                 rc = kvmppc_rtas_hcall(vcpu);
 741                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 742
 743                 if (rc == -ENOENT)
 744                         return RESUME_HOST;
 745                 else if (rc == 0)
 746                         break;
 747
 748                 /* Send the error out to userspace via KVM_RUN */
 749                 return rc;
 750         case H_LOGICAL_CI_LOAD:
 751                 ret = kvmppc_h_logical_ci_load(vcpu);
 752                 if (ret == H_TOO_HARD)
 753                         return RESUME_HOST;
 754                 break;
 755         case H_LOGICAL_CI_STORE:
 756                 ret = kvmppc_h_logical_ci_store(vcpu);
 757                 if (ret == H_TOO_HARD)
 758                         return RESUME_HOST;
 759                 break;
 760         case H_SET_MODE:
 761                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 762                                         kvmppc_get_gpr(vcpu, 5),
 763                                         kvmppc_get_gpr(vcpu, 6),
 764                                         kvmppc_get_gpr(vcpu, 7));
 765                 if (ret == H_TOO_HARD)
 766                         return RESUME_HOST;
 767                 break;
 768         case H_XIRR:
 769         case H_CPPR:
 770         case H_EOI:
 771         case H_IPI:
 772         case H_IPOLL:
 773         case H_XIRR_X:
 774                 if (kvmppc_xics_enabled(vcpu)) {
 775                         ret = kvmppc_xics_hcall(vcpu, req);
 776                         break;
 777                 } /* fallthrough */
 778         default:
 779                 return RESUME_HOST;
 780         }
 781         kvmppc_set_gpr(vcpu, 3, ret);
 782         vcpu->arch.hcall_needed = 0;
 783         return RESUME_GUEST;
 784 }
 785
 786 static int kvmppc_hcall_impl_hv(unsigned long cmd)
 787 {
 788         switch (cmd) {
 789         case H_CEDE:
 790         case H_PROD:
 791         case H_CONFER:
 792         case H_REGISTER_VPA:
 793         case H_SET_MODE:
 794         case H_LOGICAL_CI_LOAD:
 795         case H_LOGICAL_CI_STORE:
 796 #ifdef CONFIG_KVM_XICS
 797         case H_XIRR:
 798         case H_CPPR:
 799         case H_EOI:
 800         case H_IPI:
 801         case H_IPOLL:
 802         case H_XIRR_X:
 803 #endif
 804                 return 1;
 805         }
 806
 807         /* See if it's in the real-mode table */
 808         return kvmppc_hcall_impl_hv_realmode(cmd);
 809 }
 810
 811 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
 812                                         struct kvm_vcpu *vcpu)
 813 {
 814         u32 last_inst;
 815
 816         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
 817                                         EMULATE_DONE) {
 818                 /*
 819                  * Fetch failed, so return to guest and
 820                  * try executing it again.
 821                  */
 822                 return RESUME_GUEST;
 823         }
 824
 825         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
 826                 run->exit_reason = KVM_EXIT_DEBUG;
 827                 run->debug.arch.address = kvmppc_get_pc(vcpu);
 828                 return RESUME_HOST;
 829         } else {
 830                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 831                 return RESUME_GUEST;
 832         }
 833 }
 834
 835 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
 836                                  struct task_struct *tsk)
 837 {
 838         int r = RESUME_HOST;
 839
 840         vcpu->stat.sum_exits++;
 841
 842         run->exit_reason = KVM_EXIT_UNKNOWN;
 843         run->ready_for_interrupt_injection = 1;
 844         switch (vcpu->arch.trap) {
 845         /* We're good on these - the host merely wanted to get our attention */
 846         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 847                 vcpu->stat.dec_exits++;
 848                 r = RESUME_GUEST;
 849                 break;
 850         case BOOK3S_INTERRUPT_EXTERNAL:
 851         case BOOK3S_INTERRUPT_H_DOORBELL:
 852                 vcpu->stat.ext_intr_exits++;
 853                 r = RESUME_GUEST;
 854                 break;
 855         /* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
 856         case BOOK3S_INTERRUPT_HMI:
 857         case BOOK3S_INTERRUPT_PERFMON:
 858                 r = RESUME_GUEST;
 859                 break;
 860         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 861                 /*
 862                  * Deliver a machine check interrupt to the guest.
 863                  * We have to do this, even if the host has handled the
 864                  * machine check, because machine checks use SRR0/1 and
 865                  * the interrupt might have trashed guest state in them.
 866                  */
 867                 kvmppc_book3s_queue_irqprio(vcpu,
 868                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 869                 r = RESUME_GUEST;
 870                 break;
 871         case BOOK3S_INTERRUPT_PROGRAM:
 872         {
 873                 ulong flags;
 874                 /*
 875                  * Normally program interrupts are delivered directly
 876                  * to the guest by the hardware, but we can get here
 877                  * as a result of a hypervisor emulation interrupt
 878                  * (e40) getting turned into a 700 by BML RTAS.
 879                  */
 880                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 881                 kvmppc_core_queue_program(vcpu, flags);
 882                 r = RESUME_GUEST;
 883                 break;
 884         }
 885         case BOOK3S_INTERRUPT_SYSCALL:
 886         {
 887                 /* hcall - punt to userspace */
 888                 int i;
 889
 890                 /* hypercall with MSR_PR has already been handled in rmode,
 891                  * and never reaches here.
 892                  */
 893
 894                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 895                 for (i = 0; i < 9; ++i)
 896                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 897                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 898                 vcpu->arch.hcall_needed = 1;
 899                 r = RESUME_HOST;
 900                 break;
 901         }
 902         /*
 903          * We get these next two if the guest accesses a page which it thinks
 904          * it has mapped but which is not actually present, either because
 905          * it is for an emulated I/O device or because the corresonding
 906          * host page has been paged out.  Any other HDSI/HISI interrupts
 907          * have been handled already.
 908          */
 909         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 910                 r = RESUME_PAGE_FAULT;
 911                 break;
 912         case BOOK3S_INTERRUPT_H_INST_STORAGE:
 913                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
 914                 vcpu->arch.fault_dsisr = 0;
 915                 r = RESUME_PAGE_FAULT;
 916                 break;
 917         /*
 918          * This occurs if the guest executes an illegal instruction.
 919          * If the guest debug is disabled, generate a program interrupt
 920          * to the guest. If guest debug is enabled, we need to check
 921          * whether the instruction is a software breakpoint instruction.
 922          * Accordingly return to Guest or Host.
 923          */
 924         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 925                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
 926                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
 927                                 swab32(vcpu->arch.emul_inst) :
 928                                 vcpu->arch.emul_inst;
 929                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
 930                         r = kvmppc_emulate_debug_inst(run, vcpu);
 931                 } else {
 932                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 933                         r = RESUME_GUEST;
 934                 }
 935                 break;
 936         /*
 937          * This occurs if the guest (kernel or userspace), does something that
 938          * is prohibited by HFSCR.  We just generate a program interrupt to
 939          * the guest.
 940          */
 941         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
 942                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 943                 r = RESUME_GUEST;
 944                 break;
 945         default:
 946                 kvmppc_dump_regs(vcpu);
 947                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 948                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 949                         vcpu->arch.shregs.msr);
 950                 run->hw.hardware_exit_reason = vcpu->arch.trap;
 951                 r = RESUME_HOST;
 952                 break;
 953         }
 954
 955         return r;
 956 }
 957
 958 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
 959                                             struct kvm_sregs *sregs)
 960 {
 961         int i;
 962
 963         memset(sregs, 0, sizeof(struct kvm_sregs));
 964         sregs->pvr = vcpu->arch.pvr;
 965         for (i = 0; i < vcpu->arch.slb_max; i++) {
 966                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
 967                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 968         }
 969
 970         return 0;
 971 }
 972
 973 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
 974                                             struct kvm_sregs *sregs)
 975 {
 976         int i, j;
 977
 978         /* Only accept the same PVR as the host's, since we can't spoof it */
 979         if (sregs->pvr != vcpu->arch.pvr)
 980                 return -EINVAL;
 981
 982         j = 0;
 983         for (i = 0; i < vcpu->arch.slb_nr; i++) {
 984                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
 985                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
 986                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
 987                         ++j;
 988                 }
 989         }
 990         vcpu->arch.slb_max = j;
 991
 992         return 0;
 993 }
 994
 995 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
 996                 bool preserve_top32)
 997 {
 998         struct kvm *kvm = vcpu->kvm;
 999         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1000         u64 mask;
1001
1002         mutex_lock(&kvm->lock);
1003         spin_lock(&vc->lock);
1004         /*
1005          * If ILE (interrupt little-endian) has changed, update the
1006          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1007          */
1008         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1009                 struct kvm_vcpu *vcpu;
1010                 int i;
1011
1012                 kvm_for_each_vcpu(i, vcpu, kvm) {
1013                         if (vcpu->arch.vcore != vc)
1014                                 continue;
1015                         if (new_lpcr & LPCR_ILE)
1016                                 vcpu->arch.intr_msr |= MSR_LE;
1017                         else
1018                                 vcpu->arch.intr_msr &= ~MSR_LE;
1019                 }
1020         }
1021
1022         /*
1023          * Userspace can only modify DPFD (default prefetch depth),
1024          * ILE (interrupt little-endian) and TC (translation control).
1025          * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
1026          */
1027         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1028         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1029                 mask |= LPCR_AIL;
1030
1031         /* Broken 32-bit version of LPCR must not clear top bits */
1032         if (preserve_top32)
1033                 mask &= 0xFFFFFFFF;
1034         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1035         spin_unlock(&vc->lock);
1036         mutex_unlock(&kvm->lock);
1037 }
1038
1039 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1040                                  union kvmppc_one_reg *val)
1041 {
1042         int r = 0;
1043         long int i;
1044
1045         switch (id) {
1046         case KVM_REG_PPC_DEBUG_INST:
1047                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1048                 break;
1049         case KVM_REG_PPC_HIOR:
1050                 *val = get_reg_val(id, 0);
1051                 break;
1052         case KVM_REG_PPC_DABR:
1053                 *val = get_reg_val(id, vcpu->arch.dabr);
1054                 break;
1055         case KVM_REG_PPC_DABRX:
1056                 *val = get_reg_val(id, vcpu->arch.dabrx);
1057                 break;
1058         case KVM_REG_PPC_DSCR:
1059                 *val = get_reg_val(id, vcpu->arch.dscr);
1060                 break;
1061         case KVM_REG_PPC_PURR:
1062                 *val = get_reg_val(id, vcpu->arch.purr);
1063                 break;
1064         case KVM_REG_PPC_SPURR:
1065                 *val = get_reg_val(id, vcpu->arch.spurr);
1066                 break;
1067         case KVM_REG_PPC_AMR:
1068                 *val = get_reg_val(id, vcpu->arch.amr);
1069                 break;
1070         case KVM_REG_PPC_UAMOR:
1071                 *val = get_reg_val(id, vcpu->arch.uamor);
1072                 break;
1073         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1074                 i = id - KVM_REG_PPC_MMCR0;
1075                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1076                 break;
1077         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1078                 i = id - KVM_REG_PPC_PMC1;
1079                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1080                 break;
1081         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1082                 i = id - KVM_REG_PPC_SPMC1;
1083                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1084                 break;
1085         case KVM_REG_PPC_SIAR:
1086                 *val = get_reg_val(id, vcpu->arch.siar);
1087                 break;
1088         case KVM_REG_PPC_SDAR:
1089                 *val = get_reg_val(id, vcpu->arch.sdar);
1090                 break;
1091         case KVM_REG_PPC_SIER:
1092                 *val = get_reg_val(id, vcpu->arch.sier);
1093                 break;
1094         case KVM_REG_PPC_IAMR:
1095                 *val = get_reg_val(id, vcpu->arch.iamr);
1096                 break;
1097         case KVM_REG_PPC_PSPB:
1098                 *val = get_reg_val(id, vcpu->arch.pspb);
1099                 break;
1100         case KVM_REG_PPC_DPDES:
1101                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1102                 break;
1103         case KVM_REG_PPC_DAWR:
1104                 *val = get_reg_val(id, vcpu->arch.dawr);
1105                 break;
1106         case KVM_REG_PPC_DAWRX:
1107                 *val = get_reg_val(id, vcpu->arch.dawrx);
1108                 break;
1109         case KVM_REG_PPC_CIABR:
1110                 *val = get_reg_val(id, vcpu->arch.ciabr);
1111                 break;
1112         case KVM_REG_PPC_CSIGR:
1113                 *val = get_reg_val(id, vcpu->arch.csigr);
1114                 break;
1115         case KVM_REG_PPC_TACR:
1116                 *val = get_reg_val(id, vcpu->arch.tacr);
1117                 break;
1118         case KVM_REG_PPC_TCSCR:
1119                 *val = get_reg_val(id, vcpu->arch.tcscr);
1120                 break;
1121         case KVM_REG_PPC_PID:
1122                 *val = get_reg_val(id, vcpu->arch.pid);
1123                 break;
1124         case KVM_REG_PPC_ACOP:
1125                 *val = get_reg_val(id, vcpu->arch.acop);
1126                 break;
1127         case KVM_REG_PPC_WORT:
1128                 *val = get_reg_val(id, vcpu->arch.wort);
1129                 break;
1130         case KVM_REG_PPC_VPA_ADDR:
1131                 spin_lock(&vcpu->arch.vpa_update_lock);
1132                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1133                 spin_unlock(&vcpu->arch.vpa_update_lock);
1134                 break;
1135         case KVM_REG_PPC_VPA_SLB:
1136                 spin_lock(&vcpu->arch.vpa_update_lock);
1137                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1138                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1139                 spin_unlock(&vcpu->arch.vpa_update_lock);
1140                 break;
1141         case KVM_REG_PPC_VPA_DTL:
1142                 spin_lock(&vcpu->arch.vpa_update_lock);
1143                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1144                 val->vpaval.length = vcpu->arch.dtl.len;
1145                 spin_unlock(&vcpu->arch.vpa_update_lock);
1146                 break;
1147         case KVM_REG_PPC_TB_OFFSET:
1148                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1149                 break;
1150         case KVM_REG_PPC_LPCR:
1151         case KVM_REG_PPC_LPCR_64:
1152                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1153                 break;
1154         case KVM_REG_PPC_PPR:
1155                 *val = get_reg_val(id, vcpu->arch.ppr);
1156                 break;
1157 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1158         case KVM_REG_PPC_TFHAR:
1159                 *val = get_reg_val(id, vcpu->arch.tfhar);
1160                 break;
1161         case KVM_REG_PPC_TFIAR:
1162                 *val = get_reg_val(id, vcpu->arch.tfiar);
1163                 break;
1164         case KVM_REG_PPC_TEXASR:
1165                 *val = get_reg_val(id, vcpu->arch.texasr);
1166                 break;
1167         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1168                 i = id - KVM_REG_PPC_TM_GPR0;
1169                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1170                 break;
1171         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1172         {
1173                 int j;
1174                 i = id - KVM_REG_PPC_TM_VSR0;
1175                 if (i < 32)
1176                         for (j = 0; j < TS_FPRWIDTH; j++)
1177                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1178                 else {
1179                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1180                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1181                         else
1182                                 r = -ENXIO;
1183                 }
1184                 break;
1185         }
1186         case KVM_REG_PPC_TM_CR:
1187                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1188                 break;
1189         case KVM_REG_PPC_TM_XER:
1190                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1191                 break;
1192         case KVM_REG_PPC_TM_LR:
1193                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1194                 break;
1195         case KVM_REG_PPC_TM_CTR:
1196                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1197                 break;
1198         case KVM_REG_PPC_TM_FPSCR:
1199                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1200                 break;
1201         case KVM_REG_PPC_TM_AMR:
1202                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1203                 break;
1204         case KVM_REG_PPC_TM_PPR:
1205                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1206                 break;
1207         case KVM_REG_PPC_TM_VRSAVE:
1208                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1209                 break;
1210         case KVM_REG_PPC_TM_VSCR:
1211                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1212                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1213                 else
1214                         r = -ENXIO;
1215                 break;
1216         case KVM_REG_PPC_TM_DSCR:
1217                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1218                 break;
1219         case KVM_REG_PPC_TM_TAR:
1220                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1221                 break;
1222 #endif
1223         case KVM_REG_PPC_ARCH_COMPAT:
1224                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1225                 break;
1226         default:
1227                 r = -EINVAL;
1228                 break;
1229         }
1230
1231         return r;
1232 }
1233
1234 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1235                                  union kvmppc_one_reg *val)
1236 {
1237         int r = 0;
1238         long int i;
1239         unsigned long addr, len;
1240
1241         switch (id) {
1242         case KVM_REG_PPC_HIOR:
1243                 /* Only allow this to be set to zero */
1244                 if (set_reg_val(id, *val))
1245                         r = -EINVAL;
1246                 break;
1247         case KVM_REG_PPC_DABR:
1248                 vcpu->arch.dabr = set_reg_val(id, *val);
1249                 break;
1250         case KVM_REG_PPC_DABRX:
1251                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1252                 break;
1253         case KVM_REG_PPC_DSCR:
1254                 vcpu->arch.dscr = set_reg_val(id, *val);
1255                 break;
1256         case KVM_REG_PPC_PURR:
1257                 vcpu->arch.purr = set_reg_val(id, *val);
1258                 break;
1259         case KVM_REG_PPC_SPURR:
1260                 vcpu->arch.spurr = set_reg_val(id, *val);
1261                 break;
1262         case KVM_REG_PPC_AMR:
1263                 vcpu->arch.amr = set_reg_val(id, *val);
1264                 break;
1265         case KVM_REG_PPC_UAMOR:
1266                 vcpu->arch.uamor = set_reg_val(id, *val);
1267                 break;
1268         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1269                 i = id - KVM_REG_PPC_MMCR0;
1270                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1271                 break;
1272         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1273                 i = id - KVM_REG_PPC_PMC1;
1274                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1275                 break;
1276         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1277                 i = id - KVM_REG_PPC_SPMC1;
1278                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1279                 break;
1280         case KVM_REG_PPC_SIAR:
1281                 vcpu->arch.siar = set_reg_val(id, *val);
1282                 break;
1283         case KVM_REG_PPC_SDAR:
1284                 vcpu->arch.sdar = set_reg_val(id, *val);
1285                 break;
1286         case KVM_REG_PPC_SIER:
1287                 vcpu->arch.sier = set_reg_val(id, *val);
1288                 break;
1289         case KVM_REG_PPC_IAMR:
1290                 vcpu->arch.iamr = set_reg_val(id, *val);
1291                 break;
1292         case KVM_REG_PPC_PSPB:
1293                 vcpu->arch.pspb = set_reg_val(id, *val);
1294                 break;
1295         case KVM_REG_PPC_DPDES:
1296                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1297                 break;
1298         case KVM_REG_PPC_DAWR:
1299                 vcpu->arch.dawr = set_reg_val(id, *val);
1300                 break;
1301         case KVM_REG_PPC_DAWRX:
1302                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1303                 break;
1304         case KVM_REG_PPC_CIABR:
1305                 vcpu->arch.ciabr = set_reg_val(id, *val);
1306                 /* Don't allow setting breakpoints in hypervisor code */
1307                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1308                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1309                 break;
1310         case KVM_REG_PPC_CSIGR:
1311                 vcpu->arch.csigr = set_reg_val(id, *val);
1312                 break;
1313         case KVM_REG_PPC_TACR:
1314                 vcpu->arch.tacr = set_reg_val(id, *val);
1315                 break;
1316         case KVM_REG_PPC_TCSCR:
1317                 vcpu->arch.tcscr = set_reg_val(id, *val);
1318                 break;
1319         case KVM_REG_PPC_PID:
1320                 vcpu->arch.pid = set_reg_val(id, *val);
1321                 break;
1322         case KVM_REG_PPC_ACOP:
1323                 vcpu->arch.acop = set_reg_val(id, *val);
1324                 break;
1325         case KVM_REG_PPC_WORT:
1326                 vcpu->arch.wort = set_reg_val(id, *val);
1327                 break;
1328         case KVM_REG_PPC_VPA_ADDR:
1329                 addr = set_reg_val(id, *val);
1330                 r = -EINVAL;
1331                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1332                               vcpu->arch.dtl.next_gpa))
1333                         break;
1334                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1335                 break;
1336         case KVM_REG_PPC_VPA_SLB:
1337                 addr = val->vpaval.addr;
1338                 len = val->vpaval.length;
1339                 r = -EINVAL;
1340                 if (addr && !vcpu->arch.vpa.next_gpa)
1341                         break;
1342                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1343                 break;
1344         case KVM_REG_PPC_VPA_DTL:
1345                 addr = val->vpaval.addr;
1346                 len = val->vpaval.length;
1347                 r = -EINVAL;
1348                 if (addr && (len < sizeof(struct dtl_entry) ||
1349                              !vcpu->arch.vpa.next_gpa))
1350                         break;
1351                 len -= len % sizeof(struct dtl_entry);
1352                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1353                 break;
1354         case KVM_REG_PPC_TB_OFFSET:
1355                 /* round up to multiple of 2^24 */
1356                 vcpu->arch.vcore->tb_offset =
1357                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1358                 break;
1359         case KVM_REG_PPC_LPCR:
1360                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1361                 break;
1362         case KVM_REG_PPC_LPCR_64:
1363                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1364                 break;
1365         case KVM_REG_PPC_PPR:
1366                 vcpu->arch.ppr = set_reg_val(id, *val);
1367                 break;
1368 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1369         case KVM_REG_PPC_TFHAR:
1370                 vcpu->arch.tfhar = set_reg_val(id, *val);
1371                 break;
1372         case KVM_REG_PPC_TFIAR:
1373                 vcpu->arch.tfiar = set_reg_val(id, *val);
1374                 break;
1375         case KVM_REG_PPC_TEXASR:
1376                 vcpu->arch.texasr = set_reg_val(id, *val);
1377                 break;
1378         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1379                 i = id - KVM_REG_PPC_TM_GPR0;
1380                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1381                 break;
1382         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1383         {
1384                 int j;
1385                 i = id - KVM_REG_PPC_TM_VSR0;
1386                 if (i < 32)
1387                         for (j = 0; j < TS_FPRWIDTH; j++)
1388                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1389                 else
1390                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1391                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1392                         else
1393                                 r = -ENXIO;
1394                 break;
1395         }
1396         case KVM_REG_PPC_TM_CR:
1397                 vcpu->arch.cr_tm = set_reg_val(id, *val);
1398                 break;
1399         case KVM_REG_PPC_TM_XER:
1400                 vcpu->arch.xer_tm = set_reg_val(id, *val);
1401                 break;
1402         case KVM_REG_PPC_TM_LR:
1403                 vcpu->arch.lr_tm = set_reg_val(id, *val);
1404                 break;
1405         case KVM_REG_PPC_TM_CTR:
1406                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1407                 break;
1408         case KVM_REG_PPC_TM_FPSCR:
1409                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1410                 break;
1411         case KVM_REG_PPC_TM_AMR:
1412                 vcpu->arch.amr_tm = set_reg_val(id, *val);
1413                 break;
1414         case KVM_REG_PPC_TM_PPR:
1415                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1416                 break;
1417         case KVM_REG_PPC_TM_VRSAVE:
1418                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1419                 break;
1420         case KVM_REG_PPC_TM_VSCR:
1421                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1422                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1423                 else
1424                         r = - ENXIO;
1425                 break;
1426         case KVM_REG_PPC_TM_DSCR:
1427                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1428                 break;
1429         case KVM_REG_PPC_TM_TAR:
1430                 vcpu->arch.tar_tm = set_reg_val(id, *val);
1431                 break;
1432 #endif
1433         case KVM_REG_PPC_ARCH_COMPAT:
1434                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1435                 break;
1436         default:
1437                 r = -EINVAL;
1438                 break;
1439         }
1440
1441         return r;
1442 }
1443
1444 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1445 {
1446         struct kvmppc_vcore *vcore;
1447
1448         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1449
1450         if (vcore == NULL)
1451                 return NULL;
1452
1453         INIT_LIST_HEAD(&vcore->runnable_threads);
1454         spin_lock_init(&vcore->lock);
1455         spin_lock_init(&vcore->stoltb_lock);
1456         init_swait_queue_head(&vcore->wq);
1457         vcore->preempt_tb = TB_NIL;
1458         vcore->lpcr = kvm->arch.lpcr;
1459         vcore->first_vcpuid = core * threads_per_subcore;
1460         vcore->kvm = kvm;
1461         INIT_LIST_HEAD(&vcore->preempt_list);
1462
1463         return vcore;
1464 }
1465
1466 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1467 static struct debugfs_timings_element {
1468         const char *name;
1469         size_t offset;
1470 } timings[] = {
1471         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
1472         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
1473         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
1474         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
1475         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
1476 };
1477
1478 #define N_TIMINGS       (sizeof(timings) / sizeof(timings[0]))
1479
1480 struct debugfs_timings_state {
1481         struct kvm_vcpu *vcpu;
1482         unsigned int    buflen;
1483         char            buf[N_TIMINGS * 100];
1484 };
1485
1486 static int debugfs_timings_open(struct inode *inode, struct file *file)
1487 {
1488         struct kvm_vcpu *vcpu = inode->i_private;
1489         struct debugfs_timings_state *p;
1490
1491         p = kzalloc(sizeof(*p), GFP_KERNEL);
1492         if (!p)
1493                 return -ENOMEM;
1494
1495         kvm_get_kvm(vcpu->kvm);
1496         p->vcpu = vcpu;
1497         file->private_data = p;
1498
1499         return nonseekable_open(inode, file);
1500 }
1501
1502 static int debugfs_timings_release(struct inode *inode, struct file *file)
1503 {
1504         struct debugfs_timings_state *p = file->private_data;
1505
1506         kvm_put_kvm(p->vcpu->kvm);
1507         kfree(p);
1508         return 0;
1509 }
1510
1511 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1512                                     size_t len, loff_t *ppos)
1513 {
1514         struct debugfs_timings_state *p = file->private_data;
1515         struct kvm_vcpu *vcpu = p->vcpu;
1516         char *s, *buf_end;
1517         struct kvmhv_tb_accumulator tb;
1518         u64 count;
1519         loff_t pos;
1520         ssize_t n;
1521         int i, loops;
1522         bool ok;
1523
1524         if (!p->buflen) {
1525                 s = p->buf;
1526                 buf_end = s + sizeof(p->buf);
1527                 for (i = 0; i < N_TIMINGS; ++i) {
1528                         struct kvmhv_tb_accumulator *acc;
1529
1530                         acc = (struct kvmhv_tb_accumulator *)
1531                                 ((unsigned long)vcpu + timings[i].offset);
1532                         ok = false;
1533                         for (loops = 0; loops < 1000; ++loops) {
1534                                 count = acc->seqcount;
1535                                 if (!(count & 1)) {
1536                                         smp_rmb();
1537                                         tb = *acc;
1538                                         smp_rmb();
1539                                         if (count == acc->seqcount) {
1540                                                 ok = true;
1541                                                 break;
1542                                         }
1543                                 }
1544                                 udelay(1);
1545                         }
1546                         if (!ok)
1547                                 snprintf(s, buf_end - s, "%s: stuck\n",
1548                                         timings[i].name);
1549                         else
1550                                 snprintf(s, buf_end - s,
1551                                         "%s: %llu %llu %llu %llu\n",
1552                                         timings[i].name, count / 2,
1553                                         tb_to_ns(tb.tb_total),
1554                                         tb_to_ns(tb.tb_min),
1555                                         tb_to_ns(tb.tb_max));
1556                         s += strlen(s);
1557                 }
1558                 p->buflen = s - p->buf;
1559         }
1560
1561         pos = *ppos;
1562         if (pos >= p->buflen)
1563                 return 0;
1564         if (len > p->buflen - pos)
1565                 len = p->buflen - pos;
1566         n = copy_to_user(buf, p->buf + pos, len);
1567         if (n) {
1568                 if (n == len)
1569                         return -EFAULT;
1570                 len -= n;
1571         }
1572         *ppos = pos + len;
1573         return len;
1574 }
1575
1576 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1577                                      size_t len, loff_t *ppos)
1578 {
1579         return -EACCES;
1580 }
1581
1582 static const struct file_operations debugfs_timings_ops = {
1583         .owner   = THIS_MODULE,
1584         .open    = debugfs_timings_open,
1585         .release = debugfs_timings_release,
1586         .read    = debugfs_timings_read,
1587         .write   = debugfs_timings_write,
1588         .llseek  = generic_file_llseek,
1589 };
1590
1591 /* Create a debugfs directory for the vcpu */
1592 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1593 {
1594         char buf[16];
1595         struct kvm *kvm = vcpu->kvm;
1596
1597         snprintf(buf, sizeof(buf), "vcpu%u", id);
1598         if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1599                 return;
1600         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1601         if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1602                 return;
1603         vcpu->arch.debugfs_timings =
1604                 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1605                                     vcpu, &debugfs_timings_ops);
1606 }
1607
1608 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1609 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1610 {
1611 }
1612 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1613
1614 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1615                                                    unsigned int id)
1616 {
1617         struct kvm_vcpu *vcpu;
1618         int err = -EINVAL;
1619         int core;
1620         struct kvmppc_vcore *vcore;
1621
1622         core = id / threads_per_subcore;
1623         if (core >= KVM_MAX_VCORES)
1624                 goto out;
1625
1626         err = -ENOMEM;
1627         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1628         if (!vcpu)
1629                 goto out;
1630
1631         err = kvm_vcpu_init(vcpu, kvm, id);
1632         if (err)
1633                 goto free_vcpu;
1634
1635         vcpu->arch.shared = &vcpu->arch.shregs;
1636 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1637         /*
1638          * The shared struct is never shared on HV,
1639          * so we can always use host endianness
1640          */
1641 #ifdef __BIG_ENDIAN__
1642         vcpu->arch.shared_big_endian = true;
1643 #else
1644         vcpu->arch.shared_big_endian = false;
1645 #endif
1646 #endif
1647         vcpu->arch.mmcr[0] = MMCR0_FC;
1648         vcpu->arch.ctrl = CTRL_RUNLATCH;
1649         /* default to host PVR, since we can't spoof it */
1650         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1651         spin_lock_init(&vcpu->arch.vpa_update_lock);
1652         spin_lock_init(&vcpu->arch.tbacct_lock);
1653         vcpu->arch.busy_preempt = TB_NIL;
1654         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1655
1656         kvmppc_mmu_book3s_hv_init(vcpu);
1657
1658         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1659
1660         init_waitqueue_head(&vcpu->arch.cpu_run);
1661
1662         mutex_lock(&kvm->lock);
1663         vcore = kvm->arch.vcores[core];
1664         if (!vcore) {
1665                 vcore = kvmppc_vcore_create(kvm, core);
1666                 kvm->arch.vcores[core] = vcore;
1667                 kvm->arch.online_vcores++;
1668         }
1669         mutex_unlock(&kvm->lock);
1670
1671         if (!vcore)
1672                 goto free_vcpu;
1673
1674         spin_lock(&vcore->lock);
1675         ++vcore->num_threads;
1676         spin_unlock(&vcore->lock);
1677         vcpu->arch.vcore = vcore;
1678         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1679         vcpu->arch.thread_cpu = -1;
1680
1681         vcpu->arch.cpu_type = KVM_CPU_3S_64;
1682         kvmppc_sanity_check(vcpu);
1683
1684         debugfs_vcpu_init(vcpu, id);
1685
1686         return vcpu;
1687
1688 free_vcpu:
1689         kmem_cache_free(kvm_vcpu_cache, vcpu);
1690 out:
1691         return ERR_PTR(err);
1692 }
1693
1694 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1695 {
1696         if (vpa->pinned_addr)
1697                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1698                                         vpa->dirty);
1699 }
1700
1701 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1702 {
1703         spin_lock(&vcpu->arch.vpa_update_lock);
1704         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1705         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1706         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1707         spin_unlock(&vcpu->arch.vpa_update_lock);
1708         kvm_vcpu_uninit(vcpu);
1709         kmem_cache_free(kvm_vcpu_cache, vcpu);
1710 }
1711
1712 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1713 {
1714         /* Indicate we want to get back into the guest */
1715         return 1;
1716 }
1717
1718 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1719 {
1720         unsigned long dec_nsec, now;
1721
1722         now = get_tb();
1723         if (now > vcpu->arch.dec_expires) {
1724                 /* decrementer has already gone negative */
1725                 kvmppc_core_queue_dec(vcpu);
1726                 kvmppc_core_prepare_to_enter(vcpu);
1727                 return;
1728         }
1729         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1730                    / tb_ticks_per_sec;
1731         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1732                       HRTIMER_MODE_REL);
1733         vcpu->arch.timer_running = 1;
1734 }
1735
1736 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1737 {
1738         vcpu->arch.ceded = 0;
1739         if (vcpu->arch.timer_running) {
1740                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1741                 vcpu->arch.timer_running = 0;
1742         }
1743 }
1744
1745 extern void __kvmppc_vcore_entry(void);
1746
1747 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1748                                    struct kvm_vcpu *vcpu)
1749 {
1750         u64 now;
1751
1752         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1753                 return;
1754         spin_lock_irq(&vcpu->arch.tbacct_lock);
1755         now = mftb();
1756         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1757                 vcpu->arch.stolen_logged;
1758         vcpu->arch.busy_preempt = now;
1759         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1760         spin_unlock_irq(&vcpu->arch.tbacct_lock);
1761         --vc->n_runnable;
1762         list_del(&vcpu->arch.run_list);
1763 }
1764
1765 static int kvmppc_grab_hwthread(int cpu)
1766 {
1767         struct paca_struct *tpaca;
1768         long timeout = 10000;
1769
1770         tpaca = &paca[cpu];
1771
1772         /* Ensure the thread won't go into the kernel if it wakes */
1773         tpaca->kvm_hstate.kvm_vcpu = NULL;
1774         tpaca->kvm_hstate.kvm_vcore = NULL;
1775         tpaca->kvm_hstate.napping = 0;
1776         smp_wmb();
1777         tpaca->kvm_hstate.hwthread_req = 1;
1778
1779         /*
1780          * If the thread is already executing in the kernel (e.g. handling
1781          * a stray interrupt), wait for it to get back to nap mode.
1782          * The smp_mb() is to ensure that our setting of hwthread_req
1783          * is visible before we look at hwthread_state, so if this
1784          * races with the code at system_reset_pSeries and the thread
1785          * misses our setting of hwthread_req, we are sure to see its
1786          * setting of hwthread_state, and vice versa.
1787          */
1788         smp_mb();
1789         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1790                 if (--timeout <= 0) {
1791                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1792                         return -EBUSY;
1793                 }
1794                 udelay(1);
1795         }
1796         return 0;
1797 }
1798
1799 static void kvmppc_release_hwthread(int cpu)
1800 {
1801         struct paca_struct *tpaca;
1802
1803         tpaca = &paca[cpu];
1804         tpaca->kvm_hstate.hwthread_req = 0;
1805         tpaca->kvm_hstate.kvm_vcpu = NULL;
1806         tpaca->kvm_hstate.kvm_vcore = NULL;
1807         tpaca->kvm_hstate.kvm_split_mode = NULL;
1808 }
1809
1810 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
1811 {
1812         int cpu;
1813         struct paca_struct *tpaca;
1814         struct kvmppc_vcore *mvc = vc->master_vcore;
1815
1816         cpu = vc->pcpu;
1817         if (vcpu) {
1818                 if (vcpu->arch.timer_running) {
1819                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1820                         vcpu->arch.timer_running = 0;
1821                 }
1822                 cpu += vcpu->arch.ptid;
1823                 vcpu->cpu = mvc->pcpu;
1824                 vcpu->arch.thread_cpu = cpu;
1825         }
1826         tpaca = &paca[cpu];
1827         tpaca->kvm_hstate.kvm_vcpu = vcpu;
1828         tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
1829         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
1830         smp_wmb();
1831         tpaca->kvm_hstate.kvm_vcore = mvc;
1832         if (cpu != smp_processor_id())
1833                 kvmppc_ipi_thread(cpu);
1834 }
1835
1836 static void kvmppc_wait_for_nap(void)
1837 {
1838         int cpu = smp_processor_id();
1839         int i, loops;
1840
1841         for (loops = 0; loops < 1000000; ++loops) {
1842                 /*
1843                  * Check if all threads are finished.
1844                  * We set the vcore pointer when starting a thread
1845                  * and the thread clears it when finished, so we look
1846                  * for any threads that still have a non-NULL vcore ptr.
1847                  */
1848                 for (i = 1; i < threads_per_subcore; ++i)
1849                         if (paca[cpu + i].kvm_hstate.kvm_vcore)
1850                                 break;
1851                 if (i == threads_per_subcore) {
1852                         HMT_medium();
1853                         return;
1854                 }
1855                 HMT_low();
1856         }
1857         HMT_medium();
1858         for (i = 1; i < threads_per_subcore; ++i)
1859                 if (paca[cpu + i].kvm_hstate.kvm_vcore)
1860                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
1861 }
1862
1863 /*
1864  * Check that we are on thread 0 and that any other threads in
1865  * this core are off-line.  Then grab the threads so they can't
1866  * enter the kernel.
1867  */
1868 static int on_primary_thread(void)
1869 {
1870         int cpu = smp_processor_id();
1871         int thr;
1872
1873         /* Are we on a primary subcore? */
1874         if (cpu_thread_in_subcore(cpu))
1875                 return 0;
1876
1877         thr = 0;
1878         while (++thr < threads_per_subcore)
1879                 if (cpu_online(cpu + thr))
1880                         return 0;
1881
1882         /* Grab all hw threads so they can't go into the kernel */
1883         for (thr = 1; thr < threads_per_subcore; ++thr) {
1884                 if (kvmppc_grab_hwthread(cpu + thr)) {
1885                         /* Couldn't grab one; let the others go */
1886                         do {
1887                                 kvmppc_release_hwthread(cpu + thr);
1888                         } while (--thr > 0);
1889                         return 0;
1890                 }
1891         }
1892         return 1;
1893 }
1894
1895 /*
1896  * A list of virtual cores for each physical CPU.
1897  * These are vcores that could run but their runner VCPU tasks are
1898  * (or may be) preempted.
1899  */
1900 struct preempted_vcore_list {
1901         struct list_head        list;
1902         spinlock_t              lock;
1903 };
1904
1905 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
1906
1907 static void init_vcore_lists(void)
1908 {
1909         int cpu;
1910
1911         for_each_possible_cpu(cpu) {
1912                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
1913                 spin_lock_init(&lp->lock);
1914                 INIT_LIST_HEAD(&lp->list);
1915         }
1916 }
1917
1918 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
1919 {
1920         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
1921
1922         vc->vcore_state = VCORE_PREEMPT;
1923         vc->pcpu = smp_processor_id();
1924         if (vc->num_threads < threads_per_subcore) {
1925                 spin_lock(&lp->lock);
1926                 list_add_tail(&vc->preempt_list, &lp->list);
1927                 spin_unlock(&lp->lock);
1928         }
1929
1930         /* Start accumulating stolen time */
1931         kvmppc_core_start_stolen(vc);
1932 }
1933
1934 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
1935 {
1936         struct preempted_vcore_list *lp;
1937
1938         kvmppc_core_end_stolen(vc);
1939         if (!list_empty(&vc->preempt_list)) {
1940                 lp = &per_cpu(preempted_vcores, vc->pcpu);
1941                 spin_lock(&lp->lock);
1942                 list_del_init(&vc->preempt_list);
1943                 spin_unlock(&lp->lock);
1944         }
1945         vc->vcore_state = VCORE_INACTIVE;
1946 }
1947
1948 /*
1949  * This stores information about the virtual cores currently
1950  * assigned to a physical core.
1951  */
1952 struct core_info {
1953         int             n_subcores;
1954         int             max_subcore_threads;
1955         int             total_threads;
1956         int             subcore_threads[MAX_SUBCORES];
1957         struct kvm      *subcore_vm[MAX_SUBCORES];
1958         struct list_head vcs[MAX_SUBCORES];
1959 };
1960
1961 /*
1962  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
1963  * respectively in 2-way micro-threading (split-core) mode.
1964  */
1965 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
1966
1967 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
1968 {
1969         int sub;
1970
1971         memset(cip, 0, sizeof(*cip));
1972         cip->n_subcores = 1;
1973         cip->max_subcore_threads = vc->num_threads;
1974         cip->total_threads = vc->num_threads;
1975         cip->subcore_threads[0] = vc->num_threads;
1976         cip->subcore_vm[0] = vc->kvm;
1977         for (sub = 0; sub < MAX_SUBCORES; ++sub)
1978                 INIT_LIST_HEAD(&cip->vcs[sub]);
1979         list_add_tail(&vc->preempt_list, &cip->vcs[0]);
1980 }
1981
1982 static bool subcore_config_ok(int n_subcores, int n_threads)
1983 {
1984         /* Can only dynamically split if unsplit to begin with */
1985         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
1986                 return false;
1987         if (n_subcores > MAX_SUBCORES)
1988                 return false;
1989         if (n_subcores > 1) {
1990                 if (!(dynamic_mt_modes & 2))
1991                         n_subcores = 4;
1992                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
1993                         return false;
1994         }
1995
1996         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
1997 }
1998
1999 static void init_master_vcore(struct kvmppc_vcore *vc)
2000 {
2001         vc->master_vcore = vc;
2002         vc->entry_exit_map = 0;
2003         vc->in_guest = 0;
2004         vc->napping_threads = 0;
2005         vc->conferring_threads = 0;
2006 }
2007
2008 /*
2009  * See if the existing subcores can be split into 3 (or fewer) subcores
2010  * of at most two threads each, so we can fit in another vcore.  This
2011  * assumes there are at most two subcores and at most 6 threads in total.
2012  */
2013 static bool can_split_piggybacked_subcores(struct core_info *cip)
2014 {
2015         int sub, new_sub;
2016         int large_sub = -1;
2017         int thr;
2018         int n_subcores = cip->n_subcores;
2019         struct kvmppc_vcore *vc, *vcnext;
2020         struct kvmppc_vcore *master_vc = NULL;
2021
2022         for (sub = 0; sub < cip->n_subcores; ++sub) {
2023                 if (cip->subcore_threads[sub] <= 2)
2024                         continue;
2025                 if (large_sub >= 0)
2026                         return false;
2027                 large_sub = sub;
2028                 vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2029                                       preempt_list);
2030                 if (vc->num_threads > 2)
2031                         return false;
2032                 n_subcores += (cip->subcore_threads[sub] - 1) >> 1;
2033         }
2034         if (large_sub < 0 || !subcore_config_ok(n_subcores + 1, 2))
2035                 return false;
2036
2037         /*
2038          * Seems feasible, so go through and move vcores to new subcores.
2039          * Note that when we have two or more vcores in one subcore,
2040          * all those vcores must have only one thread each.
2041          */
2042         new_sub = cip->n_subcores;
2043         thr = 0;
2044         sub = large_sub;
2045         list_for_each_entry_safe(vc, vcnext, &cip->vcs[sub], preempt_list) {
2046                 if (thr >= 2) {
2047                         list_del(&vc->preempt_list);
2048                         list_add_tail(&vc->preempt_list, &cip->vcs[new_sub]);
2049                         /* vc->num_threads must be 1 */
2050                         if (++cip->subcore_threads[new_sub] == 1) {
2051                                 cip->subcore_vm[new_sub] = vc->kvm;
2052                                 init_master_vcore(vc);
2053                                 master_vc = vc;
2054                                 ++cip->n_subcores;
2055                         } else {
2056                                 vc->master_vcore = master_vc;
2057                                 ++new_sub;
2058                         }
2059                 }
2060                 thr += vc->num_threads;
2061         }
2062         cip->subcore_threads[large_sub] = 2;
2063         cip->max_subcore_threads = 2;
2064
2065         return true;
2066 }
2067
2068 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2069 {
2070         int n_threads = vc->num_threads;
2071         int sub;
2072
2073         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2074                 return false;
2075
2076         if (n_threads < cip->max_subcore_threads)
2077                 n_threads = cip->max_subcore_threads;
2078         if (subcore_config_ok(cip->n_subcores + 1, n_threads)) {
2079                 cip->max_subcore_threads = n_threads;
2080         } else if (cip->n_subcores <= 2 && cip->total_threads <= 6 &&
2081                    vc->num_threads <= 2) {
2082                 /*
2083                  * We may be able to fit another subcore in by
2084                  * splitting an existing subcore with 3 or 4
2085                  * threads into two 2-thread subcores, or one
2086                  * with 5 or 6 threads into three subcores.
2087                  * We can only do this if those subcores have
2088                  * piggybacked virtual cores.
2089                  */
2090                 if (!can_split_piggybacked_subcores(cip))
2091                         return false;
2092         } else {
2093                 return false;
2094         }
2095
2096         sub = cip->n_subcores;
2097         ++cip->n_subcores;
2098         cip->total_threads += vc->num_threads;
2099         cip->subcore_threads[sub] = vc->num_threads;
2100         cip->subcore_vm[sub] = vc->kvm;
2101         init_master_vcore(vc);
2102         list_del(&vc->preempt_list);
2103         list_add_tail(&vc->preempt_list, &cip->vcs[sub]);
2104
2105         return true;
2106 }
2107
2108 static bool can_piggyback_subcore(struct kvmppc_vcore *pvc,
2109                                   struct core_info *cip, int sub)
2110 {
2111         struct kvmppc_vcore *vc;
2112         int n_thr;
2113
2114         vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2115                               preempt_list);
2116
2117         /* require same VM and same per-core reg values */
2118         if (pvc->kvm != vc->kvm ||
2119             pvc->tb_offset != vc->tb_offset ||
2120             pvc->pcr != vc->pcr ||
2121             pvc->lpcr != vc->lpcr)
2122                 return false;
2123
2124         /* P8 guest with > 1 thread per core would see wrong TIR value */
2125         if (cpu_has_feature(CPU_FTR_ARCH_207S) &&
2126             (vc->num_threads > 1 || pvc->num_threads > 1))
2127                 return false;
2128
2129         n_thr = cip->subcore_threads[sub] + pvc->num_threads;
2130         if (n_thr > cip->max_subcore_threads) {
2131                 if (!subcore_config_ok(cip->n_subcores, n_thr))
2132                         return false;
2133                 cip->max_subcore_threads = n_thr;
2134         }
2135
2136         cip->total_threads += pvc->num_threads;
2137         cip->subcore_threads[sub] = n_thr;
2138         pvc->master_vcore = vc;
2139         list_del(&pvc->preempt_list);
2140         list_add_tail(&pvc->preempt_list, &cip->vcs[sub]);
2141
2142         return true;
2143 }
2144
2145 /*
2146  * Work out whether it is possible to piggyback the execution of
2147  * vcore *pvc onto the execution of the other vcores described in *cip.
2148  */
2149 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2150                           int target_threads)
2151 {
2152         int sub;
2153
2154         if (cip->total_threads + pvc->num_threads > target_threads)
2155                 return false;
2156         for (sub = 0; sub < cip->n_subcores; ++sub)
2157                 if (cip->subcore_threads[sub] &&
2158                     can_piggyback_subcore(pvc, cip, sub))
2159                         return true;
2160
2161         if (can_dynamic_split(pvc, cip))
2162                 return true;
2163
2164         return false;
2165 }
2166
2167 static void prepare_threads(struct kvmppc_vcore *vc)
2168 {
2169         struct kvm_vcpu *vcpu, *vnext;
2170
2171         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2172                                  arch.run_list) {
2173                 if (signal_pending(vcpu->arch.run_task))
2174                         vcpu->arch.ret = -EINTR;
2175                 else if (vcpu->arch.vpa.update_pending ||
2176                          vcpu->arch.slb_shadow.update_pending ||
2177                          vcpu->arch.dtl.update_pending)
2178                         vcpu->arch.ret = RESUME_GUEST;
2179                 else
2180                         continue;
2181                 kvmppc_remove_runnable(vc, vcpu);
2182                 wake_up(&vcpu->arch.cpu_run);
2183         }
2184 }
2185
2186 static void collect_piggybacks(struct core_info *cip, int target_threads)
2187 {
2188         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2189         struct kvmppc_vcore *pvc, *vcnext;
2190
2191         spin_lock(&lp->lock);
2192         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2193                 if (!spin_trylock(&pvc->lock))
2194                         continue;
2195                 prepare_threads(pvc);
2196                 if (!pvc->n_runnable) {
2197                         list_del_init(&pvc->preempt_list);
2198                         if (pvc->runner == NULL) {
2199                                 pvc->vcore_state = VCORE_INACTIVE;
2200                                 kvmppc_core_end_stolen(pvc);
2201                         }
2202                         spin_unlock(&pvc->lock);
2203                         continue;
2204                 }
2205                 if (!can_piggyback(pvc, cip, target_threads)) {
2206                         spin_unlock(&pvc->lock);
2207                         continue;
2208                 }
2209                 kvmppc_core_end_stolen(pvc);
2210                 pvc->vcore_state = VCORE_PIGGYBACK;
2211                 if (cip->total_threads >= target_threads)
2212                         break;
2213         }
2214         spin_unlock(&lp->lock);
2215 }
2216
2217 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2218 {
2219         int still_running = 0;
2220         u64 now;
2221         long ret;
2222         struct kvm_vcpu *vcpu, *vnext;
2223
2224         spin_lock(&vc->lock);
2225         now = get_tb();
2226         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2227                                  arch.run_list) {
2228                 /* cancel pending dec exception if dec is positive */
2229                 if (now < vcpu->arch.dec_expires &&
2230                     kvmppc_core_pending_dec(vcpu))
2231                         kvmppc_core_dequeue_dec(vcpu);
2232
2233                 trace_kvm_guest_exit(vcpu);
2234
2235                 ret = RESUME_GUEST;
2236                 if (vcpu->arch.trap)
2237                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2238                                                     vcpu->arch.run_task);
2239
2240                 vcpu->arch.ret = ret;
2241                 vcpu->arch.trap = 0;
2242
2243                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2244                         if (vcpu->arch.pending_exceptions)
2245                                 kvmppc_core_prepare_to_enter(vcpu);
2246                         if (vcpu->arch.ceded)
2247                                 kvmppc_set_timer(vcpu);
2248                         else
2249                                 ++still_running;
2250                 } else {
2251                         kvmppc_remove_runnable(vc, vcpu);
2252                         wake_up(&vcpu->arch.cpu_run);
2253                 }
2254         }
2255         list_del_init(&vc->preempt_list);
2256         if (!is_master) {
2257                 if (still_running > 0) {
2258                         kvmppc_vcore_preempt(vc);
2259                 } else if (vc->runner) {
2260                         vc->vcore_state = VCORE_PREEMPT;
2261                         kvmppc_core_start_stolen(vc);
2262                 } else {
2263                         vc->vcore_state = VCORE_INACTIVE;
2264                 }
2265                 if (vc->n_runnable > 0 && vc->runner == NULL) {
2266                         /* make sure there's a candidate runner awake */
2267                         vcpu = list_first_entry(&vc->runnable_threads,
2268                                                 struct kvm_vcpu, arch.run_list);
2269                         wake_up(&vcpu->arch.cpu_run);
2270                 }
2271         }
2272         spin_unlock(&vc->lock);
2273 }
2274
2275 /*
2276  * Run a set of guest threads on a physical core.
2277  * Called with vc->lock held.
2278  */
2279 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2280 {
2281         struct kvm_vcpu *vcpu, *vnext;
2282         int i;
2283         int srcu_idx;
2284         struct core_info core_info;
2285         struct kvmppc_vcore *pvc, *vcnext;
2286         struct kvm_split_mode split_info, *sip;
2287         int split, subcore_size, active;
2288         int sub;
2289         bool thr0_done;
2290         unsigned long cmd_bit, stat_bit;
2291         int pcpu, thr;
2292         int target_threads;
2293
2294         /*
2295          * Remove from the list any threads that have a signal pending
2296          * or need a VPA update done
2297          */
2298         prepare_threads(vc);
2299
2300         /* if the runner is no longer runnable, let the caller pick a new one */
2301         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2302                 return;
2303
2304         /*
2305          * Initialize *vc.
2306          */
2307         init_master_vcore(vc);
2308         vc->preempt_tb = TB_NIL;
2309
2310         /*
2311          * Make sure we are running on primary threads, and that secondary
2312          * threads are offline.  Also check if the number of threads in this
2313          * guest are greater than the current system threads per guest.
2314          */
2315         if ((threads_per_core > 1) &&
2316             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2317                 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2318                                          arch.run_list) {
2319                         vcpu->arch.ret = -EBUSY;
2320                         kvmppc_remove_runnable(vc, vcpu);
2321                         wake_up(&vcpu->arch.cpu_run);
2322                 }
2323                 goto out;
2324         }
2325
2326         /*
2327          * See if we could run any other vcores on the physical core
2328          * along with this one.
2329          */
2330         init_core_info(&core_info, vc);
2331         pcpu = smp_processor_id();
2332         target_threads = threads_per_subcore;
2333         if (target_smt_mode && target_smt_mode < target_threads)
2334                 target_threads = target_smt_mode;
2335         if (vc->num_threads < target_threads)
2336                 collect_piggybacks(&core_info, target_threads);
2337
2338         /* Decide on micro-threading (split-core) mode */
2339         subcore_size = threads_per_subcore;
2340         cmd_bit = stat_bit = 0;
2341         split = core_info.n_subcores;
2342         sip = NULL;
2343         if (split > 1) {
2344                 /* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2345                 if (split == 2 && (dynamic_mt_modes & 2)) {
2346                         cmd_bit = HID0_POWER8_1TO2LPAR;
2347                         stat_bit = HID0_POWER8_2LPARMODE;
2348                 } else {
2349                         split = 4;
2350                         cmd_bit = HID0_POWER8_1TO4LPAR;
2351                         stat_bit = HID0_POWER8_4LPARMODE;
2352                 }
2353                 subcore_size = MAX_SMT_THREADS / split;
2354                 sip = &split_info;
2355                 memset(&split_info, 0, sizeof(split_info));
2356                 split_info.rpr = mfspr(SPRN_RPR);
2357                 split_info.pmmar = mfspr(SPRN_PMMAR);
2358                 split_info.ldbar = mfspr(SPRN_LDBAR);
2359                 split_info.subcore_size = subcore_size;
2360                 for (sub = 0; sub < core_info.n_subcores; ++sub)
2361                         split_info.master_vcs[sub] =
2362                                 list_first_entry(&core_info.vcs[sub],
2363                                         struct kvmppc_vcore, preempt_list);
2364                 /* order writes to split_info before kvm_split_mode pointer */
2365                 smp_wmb();
2366         }
2367         pcpu = smp_processor_id();
2368         for (thr = 0; thr < threads_per_subcore; ++thr)
2369                 paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2370
2371         /* Initiate micro-threading (split-core) if required */
2372         if (cmd_bit) {
2373                 unsigned long hid0 = mfspr(SPRN_HID0);
2374
2375                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2376                 mb();
2377                 mtspr(SPRN_HID0, hid0);
2378                 isync();
2379                 for (;;) {
2380                         hid0 = mfspr(SPRN_HID0);
2381                         if (hid0 & stat_bit)
2382                                 break;
2383                         cpu_relax();
2384                 }
2385         }
2386
2387         /* Start all the threads */
2388         active = 0;
2389         for (sub = 0; sub < core_info.n_subcores; ++sub) {
2390                 thr = subcore_thread_map[sub];
2391                 thr0_done = false;
2392                 active |= 1 << thr;
2393                 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2394                         pvc->pcpu = pcpu + thr;
2395                         list_for_each_entry(vcpu, &pvc->runnable_threads,
2396                                             arch.run_list) {
2397                                 kvmppc_start_thread(vcpu, pvc);
2398                                 kvmppc_create_dtl_entry(vcpu, pvc);
2399                                 trace_kvm_guest_enter(vcpu);
2400                                 if (!vcpu->arch.ptid)
2401                                         thr0_done = true;
2402                                 active |= 1 << (thr + vcpu->arch.ptid);
2403                         }
2404                         /*
2405                          * We need to start the first thread of each subcore
2406                          * even if it doesn't have a vcpu.
2407                          */
2408                         if (pvc->master_vcore == pvc && !thr0_done)
2409                                 kvmppc_start_thread(NULL, pvc);
2410                         thr += pvc->num_threads;
2411                 }
2412         }
2413
2414         /*
2415          * Ensure that split_info.do_nap is set after setting
2416          * the vcore pointer in the PACA of the secondaries.
2417          */
2418         smp_mb();
2419         if (cmd_bit)
2420                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
2421
2422         /*
2423          * When doing micro-threading, poke the inactive threads as well.
2424          * This gets them to the nap instruction after kvm_do_nap,
2425          * which reduces the time taken to unsplit later.
2426          */
2427         if (split > 1)
2428                 for (thr = 1; thr < threads_per_subcore; ++thr)
2429                         if (!(active & (1 << thr)))
2430                                 kvmppc_ipi_thread(pcpu + thr);
2431
2432         vc->vcore_state = VCORE_RUNNING;
2433         preempt_disable();
2434
2435         trace_kvmppc_run_core(vc, 0);
2436
2437         for (sub = 0; sub < core_info.n_subcores; ++sub)
2438                 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2439                         spin_unlock(&pvc->lock);
2440
2441         kvm_guest_enter();
2442
2443         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2444
2445         __kvmppc_vcore_entry();
2446
2447         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2448
2449         spin_lock(&vc->lock);
2450         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2451         vc->vcore_state = VCORE_EXITING;
2452
2453         /* wait for secondary threads to finish writing their state to memory */
2454         kvmppc_wait_for_nap();
2455
2456         /* Return to whole-core mode if we split the core earlier */
2457         if (split > 1) {
2458                 unsigned long hid0 = mfspr(SPRN_HID0);
2459                 unsigned long loops = 0;
2460
2461                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2462                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2463                 mb();
2464                 mtspr(SPRN_HID0, hid0);
2465                 isync();
2466                 for (;;) {
2467                         hid0 = mfspr(SPRN_HID0);
2468                         if (!(hid0 & stat_bit))
2469                                 break;
2470                         cpu_relax();
2471                         ++loops;
2472                 }
2473                 split_info.do_nap = 0;
2474         }
2475
2476         /* Let secondaries go back to the offline loop */
2477         for (i = 0; i < threads_per_subcore; ++i) {
2478                 kvmppc_release_hwthread(pcpu + i);
2479                 if (sip && sip->napped[i])
2480                         kvmppc_ipi_thread(pcpu + i);
2481         }
2482
2483         spin_unlock(&vc->lock);
2484
2485         /* make sure updates to secondary vcpu structs are visible now */
2486         smp_mb();
2487         kvm_guest_exit();
2488
2489         for (sub = 0; sub < core_info.n_subcores; ++sub)
2490                 list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2491                                          preempt_list)
2492                         post_guest_process(pvc, pvc == vc);
2493
2494         spin_lock(&vc->lock);
2495         preempt_enable();
2496
2497  out:
2498         vc->vcore_state = VCORE_INACTIVE;
2499         trace_kvmppc_run_core(vc, 1);
2500 }
2501
2502 /*
2503  * Wait for some other vcpu thread to execute us, and
2504  * wake us up when we need to handle something in the host.
2505  */
2506 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2507                                  struct kvm_vcpu *vcpu, int wait_state)
2508 {
2509         DEFINE_WAIT(wait);
2510
2511         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2512         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2513                 spin_unlock(&vc->lock);
2514                 schedule();
2515                 spin_lock(&vc->lock);
2516         }
2517         finish_wait(&vcpu->arch.cpu_run, &wait);
2518 }
2519
2520 /*
2521  * All the vcpus in this vcore are idle, so wait for a decrementer
2522  * or external interrupt to one of the vcpus.  vc->lock is held.
2523  */
2524 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2525 {
2526         struct kvm_vcpu *vcpu;
2527         int do_sleep = 1;
2528         DECLARE_SWAITQUEUE(wait);
2529
2530         prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2531
2532         /*
2533          * Check one last time for pending exceptions and ceded state after
2534          * we put ourselves on the wait queue
2535          */
2536         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
2537                 if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded) {
2538                         do_sleep = 0;
2539                         break;
2540                 }
2541         }
2542
2543         if (!do_sleep) {
2544                 finish_swait(&vc->wq, &wait);
2545                 return;
2546         }
2547
2548         vc->vcore_state = VCORE_SLEEPING;
2549         trace_kvmppc_vcore_blocked(vc, 0);
2550         spin_unlock(&vc->lock);
2551         schedule();
2552         finish_swait(&vc->wq, &wait);
2553         spin_lock(&vc->lock);
2554         vc->vcore_state = VCORE_INACTIVE;
2555         trace_kvmppc_vcore_blocked(vc, 1);
2556 }
2557
2558 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2559 {
2560         int n_ceded;
2561         struct kvmppc_vcore *vc;
2562         struct kvm_vcpu *v, *vn;
2563
2564         trace_kvmppc_run_vcpu_enter(vcpu);
2565
2566         kvm_run->exit_reason = 0;
2567         vcpu->arch.ret = RESUME_GUEST;
2568         vcpu->arch.trap = 0;
2569         kvmppc_update_vpas(vcpu);
2570
2571         /*
2572          * Synchronize with other threads in this virtual core
2573          */
2574         vc = vcpu->arch.vcore;
2575         spin_lock(&vc->lock);
2576         vcpu->arch.ceded = 0;
2577         vcpu->arch.run_task = current;
2578         vcpu->arch.kvm_run = kvm_run;
2579         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2580         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2581         vcpu->arch.busy_preempt = TB_NIL;
2582         list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
2583         ++vc->n_runnable;
2584
2585         /*
2586          * This happens the first time this is called for a vcpu.
2587          * If the vcore is already running, we may be able to start
2588          * this thread straight away and have it join in.
2589          */
2590         if (!signal_pending(current)) {
2591                 if (vc->vcore_state == VCORE_PIGGYBACK) {
2592                         struct kvmppc_vcore *mvc = vc->master_vcore;
2593                         if (spin_trylock(&mvc->lock)) {
2594                                 if (mvc->vcore_state == VCORE_RUNNING &&
2595                                     !VCORE_IS_EXITING(mvc)) {
2596                                         kvmppc_create_dtl_entry(vcpu, vc);
2597                                         kvmppc_start_thread(vcpu, vc);
2598                                         trace_kvm_guest_enter(vcpu);
2599                                 }
2600                                 spin_unlock(&mvc->lock);
2601                         }
2602                 } else if (vc->vcore_state == VCORE_RUNNING &&
2603                            !VCORE_IS_EXITING(vc)) {
2604                         kvmppc_create_dtl_entry(vcpu, vc);
2605                         kvmppc_start_thread(vcpu, vc);
2606                         trace_kvm_guest_enter(vcpu);
2607                 } else if (vc->vcore_state == VCORE_SLEEPING) {
2608                         swake_up(&vc->wq);
2609                 }
2610
2611         }
2612
2613         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2614                !signal_pending(current)) {
2615                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2616                         kvmppc_vcore_end_preempt(vc);
2617
2618                 if (vc->vcore_state != VCORE_INACTIVE) {
2619                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
2620                         continue;
2621                 }
2622                 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
2623                                          arch.run_list) {
2624                         kvmppc_core_prepare_to_enter(v);
2625                         if (signal_pending(v->arch.run_task)) {
2626                                 kvmppc_remove_runnable(vc, v);
2627                                 v->stat.signal_exits++;
2628                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2629                                 v->arch.ret = -EINTR;
2630                                 wake_up(&v->arch.cpu_run);
2631                         }
2632                 }
2633                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2634                         break;
2635                 n_ceded = 0;
2636                 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
2637                         if (!v->arch.pending_exceptions)
2638                                 n_ceded += v->arch.ceded;
2639                         else
2640                                 v->arch.ceded = 0;
2641                 }
2642                 vc->runner = vcpu;
2643                 if (n_ceded == vc->n_runnable) {
2644                         kvmppc_vcore_blocked(vc);
2645                 } else if (need_resched()) {
2646                         kvmppc_vcore_preempt(vc);
2647                         /* Let something else run */
2648                         cond_resched_lock(&vc->lock);
2649                         if (vc->vcore_state == VCORE_PREEMPT)
2650                                 kvmppc_vcore_end_preempt(vc);
2651                 } else {
2652                         kvmppc_run_core(vc);
2653                 }
2654                 vc->runner = NULL;
2655         }
2656
2657         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2658                (vc->vcore_state == VCORE_RUNNING ||
2659                 vc->vcore_state == VCORE_EXITING ||
2660                 vc->vcore_state == VCORE_PIGGYBACK))
2661                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
2662
2663         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2664                 kvmppc_vcore_end_preempt(vc);
2665
2666         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2667                 kvmppc_remove_runnable(vc, vcpu);
2668                 vcpu->stat.signal_exits++;
2669                 kvm_run->exit_reason = KVM_EXIT_INTR;
2670                 vcpu->arch.ret = -EINTR;
2671         }
2672
2673         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2674                 /* Wake up some vcpu to run the core */
2675                 v = list_first_entry(&vc->runnable_threads,
2676                                      struct kvm_vcpu, arch.run_list);
2677                 wake_up(&v->arch.cpu_run);
2678         }
2679
2680         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2681         spin_unlock(&vc->lock);
2682         return vcpu->arch.ret;
2683 }
2684
2685 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2686 {
2687         int r;
2688         int srcu_idx;
2689
2690         if (!vcpu->arch.sane) {
2691                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2692                 return -EINVAL;
2693         }
2694
2695         kvmppc_core_prepare_to_enter(vcpu);
2696
2697         /* No need to go into the guest when all we'll do is come back out */
2698         if (signal_pending(current)) {
2699                 run->exit_reason = KVM_EXIT_INTR;
2700                 return -EINTR;
2701         }
2702
2703         atomic_inc(&vcpu->kvm->arch.vcpus_running);
2704         /* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2705         smp_mb();
2706
2707         /* On the first time here, set up HTAB and VRMA */
2708         if (!vcpu->kvm->arch.hpte_setup_done) {
2709                 r = kvmppc_hv_setup_htab_rma(vcpu);
2710                 if (r)
2711                         goto out;
2712         }
2713
2714         flush_fp_to_thread(current);
2715         flush_altivec_to_thread(current);
2716         flush_vsx_to_thread(current);
2717         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2718         vcpu->arch.pgdir = current->mm->pgd;
2719         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2720
2721         do {
2722                 r = kvmppc_run_vcpu(run, vcpu);
2723
2724                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2725                     !(vcpu->arch.shregs.msr & MSR_PR)) {
2726                         trace_kvm_hcall_enter(vcpu);
2727                         r = kvmppc_pseries_do_hcall(vcpu);
2728                         trace_kvm_hcall_exit(vcpu, r);
2729                         kvmppc_core_prepare_to_enter(vcpu);
2730                 } else if (r == RESUME_PAGE_FAULT) {
2731                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2732                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
2733                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2734                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2735                 }
2736         } while (is_kvmppc_resume_guest(r));
2737
2738  out:
2739         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2740         atomic_dec(&vcpu->kvm->arch.vcpus_running);
2741         return r;
2742 }
2743
2744 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2745                                      int linux_psize)
2746 {
2747         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2748
2749         if (!def->shift)
2750                 return;
2751         (*sps)->page_shift = def->shift;
2752         (*sps)->slb_enc = def->sllp;
2753         (*sps)->enc[0].page_shift = def->shift;
2754         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
2755         /*
2756          * Add 16MB MPSS support if host supports it
2757          */
2758         if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2759                 (*sps)->enc[1].page_shift = 24;
2760                 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2761         }
2762         (*sps)++;
2763 }
2764
2765 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2766                                          struct kvm_ppc_smmu_info *info)
2767 {
2768         struct kvm_ppc_one_seg_page_size *sps;
2769
2770         info->flags = KVM_PPC_PAGE_SIZES_REAL;
2771         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2772                 info->flags |= KVM_PPC_1T_SEGMENTS;
2773         info->slb_size = mmu_slb_size;
2774
2775         /* We only support these sizes for now, and no muti-size segments */
2776         sps = &info->sps[0];
2777         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2778         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2779         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2780
2781         return 0;
2782 }
2783
2784 /*
2785  * Get (and clear) the dirty memory log for a memory slot.
2786  */
2787 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2788                                          struct kvm_dirty_log *log)
2789 {
2790         struct kvm_memslots *slots;
2791         struct kvm_memory_slot *memslot;
2792         int r;
2793         unsigned long n;
2794
2795         mutex_lock(&kvm->slots_lock);
2796
2797         r = -EINVAL;
2798         if (log->slot >= KVM_USER_MEM_SLOTS)
2799                 goto out;
2800
2801         slots = kvm_memslots(kvm);
2802         memslot = id_to_memslot(slots, log->slot);
2803         r = -ENOENT;
2804         if (!memslot->dirty_bitmap)
2805                 goto out;
2806
2807         n = kvm_dirty_bitmap_bytes(memslot);
2808         memset(memslot->dirty_bitmap, 0, n);
2809
2810         r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2811         if (r)
2812                 goto out;
2813
2814         r = -EFAULT;
2815         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2816                 goto out;
2817
2818         r = 0;
2819 out:
2820         mutex_unlock(&kvm->slots_lock);
2821         return r;
2822 }
2823
2824 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2825                                         struct kvm_memory_slot *dont)
2826 {
2827         if (!dont || free->arch.rmap != dont->arch.rmap) {
2828                 vfree(free->arch.rmap);
2829                 free->arch.rmap = NULL;
2830         }
2831 }
2832
2833 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2834                                          unsigned long npages)
2835 {
2836         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2837         if (!slot->arch.rmap)
2838                 return -ENOMEM;
2839
2840         return 0;
2841 }
2842
2843 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2844                                         struct kvm_memory_slot *memslot,
2845                                         const struct kvm_userspace_memory_region *mem)
2846 {
2847         return 0;
2848 }
2849
2850 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2851                                 const struct kvm_userspace_memory_region *mem,
2852                                 const struct kvm_memory_slot *old,
2853                                 const struct kvm_memory_slot *new)
2854 {
2855         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2856         struct kvm_memslots *slots;
2857         struct kvm_memory_slot *memslot;
2858
2859         if (npages && old->npages) {
2860                 /*
2861                  * If modifying a memslot, reset all the rmap dirty bits.
2862                  * If this is a new memslot, we don't need to do anything
2863                  * since the rmap array starts out as all zeroes,
2864                  * i.e. no pages are dirty.
2865                  */
2866                 slots = kvm_memslots(kvm);
2867                 memslot = id_to_memslot(slots, mem->slot);
2868                 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2869         }
2870 }
2871
2872 /*
2873  * Update LPCR values in kvm->arch and in vcores.
2874  * Caller must hold kvm->lock.
2875  */
2876 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2877 {
2878         long int i;
2879         u32 cores_done = 0;
2880
2881         if ((kvm->arch.lpcr & mask) == lpcr)
2882                 return;
2883
2884         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2885
2886         for (i = 0; i < KVM_MAX_VCORES; ++i) {
2887                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2888                 if (!vc)
2889                         continue;
2890                 spin_lock(&vc->lock);
2891                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2892                 spin_unlock(&vc->lock);
2893                 if (++cores_done >= kvm->arch.online_vcores)
2894                         break;
2895         }
2896 }
2897
2898 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2899 {
2900         return;
2901 }
2902
2903 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2904 {
2905         int err = 0;
2906         struct kvm *kvm = vcpu->kvm;
2907         unsigned long hva;
2908         struct kvm_memory_slot *memslot;
2909         struct vm_area_struct *vma;
2910         unsigned long lpcr = 0, senc;
2911         unsigned long psize, porder;
2912         int srcu_idx;
2913
2914         mutex_lock(&kvm->lock);
2915         if (kvm->arch.hpte_setup_done)
2916                 goto out;       /* another vcpu beat us to it */
2917
2918         /* Allocate hashed page table (if not done already) and reset it */
2919         if (!kvm->arch.hpt_virt) {
2920                 err = kvmppc_alloc_hpt(kvm, NULL);
2921                 if (err) {
2922                         pr_err("KVM: Couldn't alloc HPT\n");
2923                         goto out;
2924                 }
2925         }
2926
2927         /* Look up the memslot for guest physical address 0 */
2928         srcu_idx = srcu_read_lock(&kvm->srcu);
2929         memslot = gfn_to_memslot(kvm, 0);
2930
2931         /* We must have some memory at 0 by now */
2932         err = -EINVAL;
2933         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2934                 goto out_srcu;
2935
2936         /* Look up the VMA for the start of this memory slot */
2937         hva = memslot->userspace_addr;
2938         down_read(&current->mm->mmap_sem);
2939         vma = find_vma(current->mm, hva);
2940         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2941                 goto up_out;
2942
2943         psize = vma_kernel_pagesize(vma);
2944         porder = __ilog2(psize);
2945
2946         up_read(&current->mm->mmap_sem);
2947
2948         /* We can handle 4k, 64k or 16M pages in the VRMA */
2949         err = -EINVAL;
2950         if (!(psize == 0x1000 || psize == 0x10000 ||
2951               psize == 0x1000000))
2952                 goto out_srcu;
2953
2954         /* Update VRMASD field in the LPCR */
2955         senc = slb_pgsize_encoding(psize);
2956         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2957                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
2958         /* the -4 is to account for senc values starting at 0x10 */
2959         lpcr = senc << (LPCR_VRMASD_SH - 4);
2960
2961         /* Create HPTEs in the hash page table for the VRMA */
2962         kvmppc_map_vrma(vcpu, memslot, porder);
2963
2964         kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
2965
2966         /* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
2967         smp_wmb();
2968         kvm->arch.hpte_setup_done = 1;
2969         err = 0;
2970  out_srcu:
2971         srcu_read_unlock(&kvm->srcu, srcu_idx);
2972  out:
2973         mutex_unlock(&kvm->lock);
2974         return err;
2975
2976  up_out:
2977         up_read(&current->mm->mmap_sem);
2978         goto out_srcu;
2979 }
2980
2981 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2982 {
2983         unsigned long lpcr, lpid;
2984         char buf[32];
2985
2986         /* Allocate the guest's logical partition ID */
2987
2988         lpid = kvmppc_alloc_lpid();
2989         if ((long)lpid < 0)
2990                 return -ENOMEM;
2991         kvm->arch.lpid = lpid;
2992
2993         /*
2994          * Since we don't flush the TLB when tearing down a VM,
2995          * and this lpid might have previously been used,
2996          * make sure we flush on each core before running the new VM.
2997          */
2998         cpumask_setall(&kvm->arch.need_tlb_flush);
2999
3000         /* Start out with the default set of hcalls enabled */
3001         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3002                sizeof(kvm->arch.enabled_hcalls));
3003
3004         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3005
3006         /* Init LPCR for virtual RMA mode */
3007         kvm->arch.host_lpid = mfspr(SPRN_LPID);
3008         kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3009         lpcr &= LPCR_PECE | LPCR_LPES;
3010         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3011                 LPCR_VPM0 | LPCR_VPM1;
3012         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3013                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3014         /* On POWER8 turn on online bit to enable PURR/SPURR */
3015         if (cpu_has_feature(CPU_FTR_ARCH_207S))
3016                 lpcr |= LPCR_ONL;
3017         kvm->arch.lpcr = lpcr;
3018
3019         /*
3020          * Track that we now have a HV mode VM active. This blocks secondary
3021          * CPU threads from coming online.
3022          */
3023         kvm_hv_vm_activated();
3024
3025         /*
3026          * Create a debugfs directory for the VM
3027          */
3028         snprintf(buf, sizeof(buf), "vm%d", current->pid);
3029         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3030         if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3031                 kvmppc_mmu_debugfs_init(kvm);
3032
3033         return 0;
3034 }
3035
3036 static void kvmppc_free_vcores(struct kvm *kvm)
3037 {
3038         long int i;
3039
3040         for (i = 0; i < KVM_MAX_VCORES; ++i)
3041                 kfree(kvm->arch.vcores[i]);
3042         kvm->arch.online_vcores = 0;
3043 }
3044
3045 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3046 {
3047         debugfs_remove_recursive(kvm->arch.debugfs_dir);
3048
3049         kvm_hv_vm_deactivated();
3050
3051         kvmppc_free_vcores(kvm);
3052
3053         kvmppc_free_hpt(kvm);
3054 }
3055
3056 /* We don't need to emulate any privileged instructions or dcbz */
3057 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3058                                      unsigned int inst, int *advance)
3059 {
3060         return EMULATE_FAIL;
3061 }
3062
3063 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3064                                         ulong spr_val)
3065 {
3066         return EMULATE_FAIL;
3067 }
3068
3069 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3070                                         ulong *spr_val)
3071 {
3072         return EMULATE_FAIL;
3073 }
3074
3075 static int kvmppc_core_check_processor_compat_hv(void)
3076 {
3077         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3078             !cpu_has_feature(CPU_FTR_ARCH_206))
3079                 return -EIO;
3080         return 0;
3081 }
3082
3083 static long kvm_arch_vm_ioctl_hv(struct file *filp,
3084                                  unsigned int ioctl, unsigned long arg)
3085 {
3086         struct kvm *kvm __maybe_unused = filp->private_data;
3087         void __user *argp = (void __user *)arg;
3088         long r;
3089
3090         switch (ioctl) {
3091
3092         case KVM_PPC_ALLOCATE_HTAB: {
3093                 u32 htab_order;
3094
3095                 r = -EFAULT;
3096                 if (get_user(htab_order, (u32 __user *)argp))
3097                         break;
3098                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
3099                 if (r)
3100                         break;
3101                 r = -EFAULT;
3102                 if (put_user(htab_order, (u32 __user *)argp))
3103                         break;
3104                 r = 0;
3105                 break;
3106         }
3107
3108         case KVM_PPC_GET_HTAB_FD: {
3109                 struct kvm_get_htab_fd ghf;
3110
3111                 r = -EFAULT;
3112                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
3113                         break;
3114                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3115                 break;
3116         }
3117
3118         default:
3119                 r = -ENOTTY;
3120         }
3121
3122         return r;
3123 }
3124
3125 /*
3126  * List of hcall numbers to enable by default.
3127  * For compatibility with old userspace, we enable by default
3128  * all hcalls that were implemented before the hcall-enabling
3129  * facility was added.  Note this list should not include H_RTAS.
3130  */
3131 static unsigned int default_hcall_list[] = {
3132         H_REMOVE,
3133         H_ENTER,
3134         H_READ,
3135         H_PROTECT,
3136         H_BULK_REMOVE,
3137         H_GET_TCE,
3138         H_PUT_TCE,
3139         H_SET_DABR,
3140         H_SET_XDABR,
3141         H_CEDE,
3142         H_PROD,
3143         H_CONFER,
3144         H_REGISTER_VPA,
3145 #ifdef CONFIG_KVM_XICS
3146         H_EOI,
3147         H_CPPR,
3148         H_IPI,
3149         H_IPOLL,
3150         H_XIRR,
3151         H_XIRR_X,
3152 #endif
3153         0
3154 };
3155
3156 static void init_default_hcalls(void)
3157 {
3158         int i;
3159         unsigned int hcall;
3160
3161         for (i = 0; default_hcall_list[i]; ++i) {
3162                 hcall = default_hcall_list[i];
3163                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3164                 __set_bit(hcall / 4, default_enabled_hcalls);
3165         }
3166 }
3167
3168 static struct kvmppc_ops kvm_ops_hv = {
3169         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3170         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3171         .get_one_reg = kvmppc_get_one_reg_hv,
3172         .set_one_reg = kvmppc_set_one_reg_hv,
3173         .vcpu_load   = kvmppc_core_vcpu_load_hv,
3174         .vcpu_put    = kvmppc_core_vcpu_put_hv,
3175         .set_msr     = kvmppc_set_msr_hv,
3176         .vcpu_run    = kvmppc_vcpu_run_hv,
3177         .vcpu_create = kvmppc_core_vcpu_create_hv,
3178         .vcpu_free   = kvmppc_core_vcpu_free_hv,
3179         .check_requests = kvmppc_core_check_requests_hv,
3180         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
3181         .flush_memslot  = kvmppc_core_flush_memslot_hv,
3182         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3183         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
3184         .unmap_hva = kvm_unmap_hva_hv,
3185         .unmap_hva_range = kvm_unmap_hva_range_hv,
3186         .age_hva  = kvm_age_hva_hv,
3187         .test_age_hva = kvm_test_age_hva_hv,
3188         .set_spte_hva = kvm_set_spte_hva_hv,
3189         .mmu_destroy  = kvmppc_mmu_destroy_hv,
3190         .free_memslot = kvmppc_core_free_memslot_hv,
3191         .create_memslot = kvmppc_core_create_memslot_hv,
3192         .init_vm =  kvmppc_core_init_vm_hv,
3193         .destroy_vm = kvmppc_core_destroy_vm_hv,
3194         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3195         .emulate_op = kvmppc_core_emulate_op_hv,
3196         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3197         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3198         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3199         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
3200         .hcall_implemented = kvmppc_hcall_impl_hv,
3201 };
3202
3203 static int kvmppc_book3s_init_hv(void)
3204 {
3205         int r;
3206         /*
3207          * FIXME!! Do we need to check on all cpus ?
3208          */
3209         r = kvmppc_core_check_processor_compat_hv();
3210         if (r < 0)
3211                 return -ENODEV;
3212
3213         kvm_ops_hv.owner = THIS_MODULE;
3214         kvmppc_hv_ops = &kvm_ops_hv;
3215
3216         init_default_hcalls();
3217
3218         init_vcore_lists();
3219
3220         r = kvmppc_mmu_hv_init();
3221         return r;
3222 }
3223
3224 static void kvmppc_book3s_exit_hv(void)
3225 {
3226         kvmppc_hv_ops = NULL;
3227 }
3228
3229 module_init(kvmppc_book3s_init_hv);
3230 module_exit(kvmppc_book3s_exit_hv);
3231 MODULE_LICENSE("GPL");
3232 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3233 MODULE_ALIAS("devname:kvm");