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         wait_queue_head_t *wqp;
 118
 119         wqp = kvm_arch_vcpu_wq(vcpu);
 120         if (waitqueue_active(wqp)) {
 121                 wake_up_interruptible(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         struct kvm_vcpu *ret;
 318
 319         mutex_lock(&kvm->lock);
 320         ret = kvm_get_vcpu_by_id(kvm, id);
 321         mutex_unlock(&kvm->lock);
 322         return ret;
 323 }
 324
 325 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 326 {
 327         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 328         vpa->yield_count = cpu_to_be32(1);
 329 }
 330
 331 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 332                    unsigned long addr, unsigned long len)
 333 {
 334         /* check address is cacheline aligned */
 335         if (addr & (L1_CACHE_BYTES - 1))
 336                 return -EINVAL;
 337         spin_lock(&vcpu->arch.vpa_update_lock);
 338         if (v->next_gpa != addr || v->len != len) {
 339                 v->next_gpa = addr;
 340                 v->len = addr ? len : 0;
 341                 v->update_pending = 1;
 342         }
 343         spin_unlock(&vcpu->arch.vpa_update_lock);
 344         return 0;
 345 }
 346
 347 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 348 struct reg_vpa {
 349         u32 dummy;
 350         union {
 351                 __be16 hword;
 352                 __be32 word;
 353         } length;
 354 };
 355
 356 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 357 {
 358         if (vpap->update_pending)
 359                 return vpap->next_gpa != 0;
 360         return vpap->pinned_addr != NULL;
 361 }
 362
 363 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 364                                        unsigned long flags,
 365                                        unsigned long vcpuid, unsigned long vpa)
 366 {
 367         struct kvm *kvm = vcpu->kvm;
 368         unsigned long len, nb;
 369         void *va;
 370         struct kvm_vcpu *tvcpu;
 371         int err;
 372         int subfunc;
 373         struct kvmppc_vpa *vpap;
 374
 375         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 376         if (!tvcpu)
 377                 return H_PARAMETER;
 378
 379         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 380         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 381             subfunc == H_VPA_REG_SLB) {
 382                 /* Registering new area - address must be cache-line aligned */
 383                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 384                         return H_PARAMETER;
 385
 386                 /* convert logical addr to kernel addr and read length */
 387                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 388                 if (va == NULL)
 389                         return H_PARAMETER;
 390                 if (subfunc == H_VPA_REG_VPA)
 391                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 392                 else
 393                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 394                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 395
 396                 /* Check length */
 397                 if (len > nb || len < sizeof(struct reg_vpa))
 398                         return H_PARAMETER;
 399         } else {
 400                 vpa = 0;
 401                 len = 0;
 402         }
 403
 404         err = H_PARAMETER;
 405         vpap = NULL;
 406         spin_lock(&tvcpu->arch.vpa_update_lock);
 407
 408         switch (subfunc) {
 409         case H_VPA_REG_VPA:             /* register VPA */
 410                 if (len < sizeof(struct lppaca))
 411                         break;
 412                 vpap = &tvcpu->arch.vpa;
 413                 err = 0;
 414                 break;
 415
 416         case H_VPA_REG_DTL:             /* register DTL */
 417                 if (len < sizeof(struct dtl_entry))
 418                         break;
 419                 len -= len % sizeof(struct dtl_entry);
 420
 421                 /* Check that they have previously registered a VPA */
 422                 err = H_RESOURCE;
 423                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 424                         break;
 425
 426                 vpap = &tvcpu->arch.dtl;
 427                 err = 0;
 428                 break;
 429
 430         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 431                 /* Check that they have previously registered a VPA */
 432                 err = H_RESOURCE;
 433                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 434                         break;
 435
 436                 vpap = &tvcpu->arch.slb_shadow;
 437                 err = 0;
 438                 break;
 439
 440         case H_VPA_DEREG_VPA:           /* deregister VPA */
 441                 /* Check they don't still have a DTL or SLB buf registered */
 442                 err = H_RESOURCE;
 443                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 444                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 445                         break;
 446
 447                 vpap = &tvcpu->arch.vpa;
 448                 err = 0;
 449                 break;
 450
 451         case H_VPA_DEREG_DTL:           /* deregister DTL */
 452                 vpap = &tvcpu->arch.dtl;
 453                 err = 0;
 454                 break;
 455
 456         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 457                 vpap = &tvcpu->arch.slb_shadow;
 458                 err = 0;
 459                 break;
 460         }
 461
 462         if (vpap) {
 463                 vpap->next_gpa = vpa;
 464                 vpap->len = len;
 465                 vpap->update_pending = 1;
 466         }
 467
 468         spin_unlock(&tvcpu->arch.vpa_update_lock);
 469
 470         return err;
 471 }
 472
 473 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 474 {
 475         struct kvm *kvm = vcpu->kvm;
 476         void *va;
 477         unsigned long nb;
 478         unsigned long gpa;
 479
 480         /*
 481          * We need to pin the page pointed to by vpap->next_gpa,
 482          * but we can't call kvmppc_pin_guest_page under the lock
 483          * as it does get_user_pages() and down_read().  So we
 484          * have to drop the lock, pin the page, then get the lock
 485          * again and check that a new area didn't get registered
 486          * in the meantime.
 487          */
 488         for (;;) {
 489                 gpa = vpap->next_gpa;
 490                 spin_unlock(&vcpu->arch.vpa_update_lock);
 491                 va = NULL;
 492                 nb = 0;
 493                 if (gpa)
 494                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 495                 spin_lock(&vcpu->arch.vpa_update_lock);
 496                 if (gpa == vpap->next_gpa)
 497                         break;
 498                 /* sigh... unpin that one and try again */
 499                 if (va)
 500                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 501         }
 502
 503         vpap->update_pending = 0;
 504         if (va && nb < vpap->len) {
 505                 /*
 506                  * If it's now too short, it must be that userspace
 507                  * has changed the mappings underlying guest memory,
 508                  * so unregister the region.
 509                  */
 510                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 511                 va = NULL;
 512         }
 513         if (vpap->pinned_addr)
 514                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 515                                         vpap->dirty);
 516         vpap->gpa = gpa;
 517         vpap->pinned_addr = va;
 518         vpap->dirty = false;
 519         if (va)
 520                 vpap->pinned_end = va + vpap->len;
 521 }
 522
 523 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 524 {
 525         if (!(vcpu->arch.vpa.update_pending ||
 526               vcpu->arch.slb_shadow.update_pending ||
 527               vcpu->arch.dtl.update_pending))
 528                 return;
 529
 530         spin_lock(&vcpu->arch.vpa_update_lock);
 531         if (vcpu->arch.vpa.update_pending) {
 532                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 533                 if (vcpu->arch.vpa.pinned_addr)
 534                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 535         }
 536         if (vcpu->arch.dtl.update_pending) {
 537                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 538                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 539                 vcpu->arch.dtl_index = 0;
 540         }
 541         if (vcpu->arch.slb_shadow.update_pending)
 542                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 543         spin_unlock(&vcpu->arch.vpa_update_lock);
 544 }
 545
 546 /*
 547  * Return the accumulated stolen time for the vcore up until `now'.
 548  * The caller should hold the vcore lock.
 549  */
 550 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 551 {
 552         u64 p;
 553         unsigned long flags;
 554
 555         spin_lock_irqsave(&vc->stoltb_lock, flags);
 556         p = vc->stolen_tb;
 557         if (vc->vcore_state != VCORE_INACTIVE &&
 558             vc->preempt_tb != TB_NIL)
 559                 p += now - vc->preempt_tb;
 560         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 561         return p;
 562 }
 563
 564 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 565                                     struct kvmppc_vcore *vc)
 566 {
 567         struct dtl_entry *dt;
 568         struct lppaca *vpa;
 569         unsigned long stolen;
 570         unsigned long core_stolen;
 571         u64 now;
 572
 573         dt = vcpu->arch.dtl_ptr;
 574         vpa = vcpu->arch.vpa.pinned_addr;
 575         now = mftb();
 576         core_stolen = vcore_stolen_time(vc, now);
 577         stolen = core_stolen - vcpu->arch.stolen_logged;
 578         vcpu->arch.stolen_logged = core_stolen;
 579         spin_lock_irq(&vcpu->arch.tbacct_lock);
 580         stolen += vcpu->arch.busy_stolen;
 581         vcpu->arch.busy_stolen = 0;
 582         spin_unlock_irq(&vcpu->arch.tbacct_lock);
 583         if (!dt || !vpa)
 584                 return;
 585         memset(dt, 0, sizeof(struct dtl_entry));
 586         dt->dispatch_reason = 7;
 587         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 588         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 589         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 590         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 591         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 592         ++dt;
 593         if (dt == vcpu->arch.dtl.pinned_end)
 594                 dt = vcpu->arch.dtl.pinned_addr;
 595         vcpu->arch.dtl_ptr = dt;
 596         /* order writing *dt vs. writing vpa->dtl_idx */
 597         smp_wmb();
 598         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 599         vcpu->arch.dtl.dirty = true;
 600 }
 601
 602 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 603 {
 604         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 605                 return true;
 606         if ((!vcpu->arch.vcore->arch_compat) &&
 607             cpu_has_feature(CPU_FTR_ARCH_207S))
 608                 return true;
 609         return false;
 610 }
 611
 612 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 613                              unsigned long resource, unsigned long value1,
 614                              unsigned long value2)
 615 {
 616         switch (resource) {
 617         case H_SET_MODE_RESOURCE_SET_CIABR:
 618                 if (!kvmppc_power8_compatible(vcpu))
 619                         return H_P2;
 620                 if (value2)
 621                         return H_P4;
 622                 if (mflags)
 623                         return H_UNSUPPORTED_FLAG_START;
 624                 /* Guests can't breakpoint the hypervisor */
 625                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 626                         return H_P3;
 627                 vcpu->arch.ciabr  = value1;
 628                 return H_SUCCESS;
 629         case H_SET_MODE_RESOURCE_SET_DAWR:
 630                 if (!kvmppc_power8_compatible(vcpu))
 631                         return H_P2;
 632                 if (mflags)
 633                         return H_UNSUPPORTED_FLAG_START;
 634                 if (value2 & DABRX_HYP)
 635                         return H_P4;
 636                 vcpu->arch.dawr  = value1;
 637                 vcpu->arch.dawrx = value2;
 638                 return H_SUCCESS;
 639         default:
 640                 return H_TOO_HARD;
 641         }
 642 }
 643
 644 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 645 {
 646         struct kvmppc_vcore *vcore = target->arch.vcore;
 647
 648         /*
 649          * We expect to have been called by the real mode handler
 650          * (kvmppc_rm_h_confer()) which would have directly returned
 651          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 652          * have useful work to do and should not confer) so we don't
 653          * recheck that here.
 654          */
 655
 656         spin_lock(&vcore->lock);
 657         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 658             vcore->vcore_state != VCORE_INACTIVE &&
 659             vcore->runner)
 660                 target = vcore->runner;
 661         spin_unlock(&vcore->lock);
 662
 663         return kvm_vcpu_yield_to(target);
 664 }
 665
 666 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 667 {
 668         int yield_count = 0;
 669         struct lppaca *lppaca;
 670
 671         spin_lock(&vcpu->arch.vpa_update_lock);
 672         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 673         if (lppaca)
 674                 yield_count = be32_to_cpu(lppaca->yield_count);
 675         spin_unlock(&vcpu->arch.vpa_update_lock);
 676         return yield_count;
 677 }
 678
 679 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 680 {
 681         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 682         unsigned long target, ret = H_SUCCESS;
 683         int yield_count;
 684         struct kvm_vcpu *tvcpu;
 685         int idx, rc;
 686
 687         if (req <= MAX_HCALL_OPCODE &&
 688             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 689                 return RESUME_HOST;
 690
 691         switch (req) {
 692         case H_CEDE:
 693                 break;
 694         case H_PROD:
 695                 target = kvmppc_get_gpr(vcpu, 4);
 696                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 697                 if (!tvcpu) {
 698                         ret = H_PARAMETER;
 699                         break;
 700                 }
 701                 tvcpu->arch.prodded = 1;
 702                 smp_mb();
 703                 if (vcpu->arch.ceded) {
 704                         if (waitqueue_active(&vcpu->wq)) {
 705                                 wake_up_interruptible(&vcpu->wq);
 706                                 vcpu->stat.halt_wakeup++;
 707                         }
 708                 }
 709                 break;
 710         case H_CONFER:
 711                 target = kvmppc_get_gpr(vcpu, 4);
 712                 if (target == -1)
 713                         break;
 714                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 715                 if (!tvcpu) {
 716                         ret = H_PARAMETER;
 717                         break;
 718                 }
 719                 yield_count = kvmppc_get_gpr(vcpu, 5);
 720                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 721                         break;
 722                 kvm_arch_vcpu_yield_to(tvcpu);
 723                 break;
 724         case H_REGISTER_VPA:
 725                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 726                                         kvmppc_get_gpr(vcpu, 5),
 727                                         kvmppc_get_gpr(vcpu, 6));
 728                 break;
 729         case H_RTAS:
 730                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 731                         return RESUME_HOST;
 732
 733                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 734                 rc = kvmppc_rtas_hcall(vcpu);
 735                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 736
 737                 if (rc == -ENOENT)
 738                         return RESUME_HOST;
 739                 else if (rc == 0)
 740                         break;
 741
 742                 /* Send the error out to userspace via KVM_RUN */
 743                 return rc;
 744         case H_LOGICAL_CI_LOAD:
 745                 ret = kvmppc_h_logical_ci_load(vcpu);
 746                 if (ret == H_TOO_HARD)
 747                         return RESUME_HOST;
 748                 break;
 749         case H_LOGICAL_CI_STORE:
 750                 ret = kvmppc_h_logical_ci_store(vcpu);
 751                 if (ret == H_TOO_HARD)
 752                         return RESUME_HOST;
 753                 break;
 754         case H_SET_MODE:
 755                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 756                                         kvmppc_get_gpr(vcpu, 5),
 757                                         kvmppc_get_gpr(vcpu, 6),
 758                                         kvmppc_get_gpr(vcpu, 7));
 759                 if (ret == H_TOO_HARD)
 760                         return RESUME_HOST;
 761                 break;
 762         case H_XIRR:
 763         case H_CPPR:
 764         case H_EOI:
 765         case H_IPI:
 766         case H_IPOLL:
 767         case H_XIRR_X:
 768                 if (kvmppc_xics_enabled(vcpu)) {
 769                         ret = kvmppc_xics_hcall(vcpu, req);
 770                         break;
 771                 }
 772                 return RESUME_HOST;
 773         case H_PUT_TCE:
 774                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 775                                                 kvmppc_get_gpr(vcpu, 5),
 776                                                 kvmppc_get_gpr(vcpu, 6));
 777                 if (ret == H_TOO_HARD)
 778                         return RESUME_HOST;
 779                 break;
 780         case H_PUT_TCE_INDIRECT:
 781                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
 782                                                 kvmppc_get_gpr(vcpu, 5),
 783                                                 kvmppc_get_gpr(vcpu, 6),
 784                                                 kvmppc_get_gpr(vcpu, 7));
 785                 if (ret == H_TOO_HARD)
 786                         return RESUME_HOST;
 787                 break;
 788         case H_STUFF_TCE:
 789                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 790                                                 kvmppc_get_gpr(vcpu, 5),
 791                                                 kvmppc_get_gpr(vcpu, 6),
 792                                                 kvmppc_get_gpr(vcpu, 7));
 793                 if (ret == H_TOO_HARD)
 794                         return RESUME_HOST;
 795                 break;
 796         default:
 797                 return RESUME_HOST;
 798         }
 799         kvmppc_set_gpr(vcpu, 3, ret);
 800         vcpu->arch.hcall_needed = 0;
 801         return RESUME_GUEST;
 802 }
 803
 804 static int kvmppc_hcall_impl_hv(unsigned long cmd)
 805 {
 806         switch (cmd) {
 807         case H_CEDE:
 808         case H_PROD:
 809         case H_CONFER:
 810         case H_REGISTER_VPA:
 811         case H_SET_MODE:
 812         case H_LOGICAL_CI_LOAD:
 813         case H_LOGICAL_CI_STORE:
 814 #ifdef CONFIG_KVM_XICS
 815         case H_XIRR:
 816         case H_CPPR:
 817         case H_EOI:
 818         case H_IPI:
 819         case H_IPOLL:
 820         case H_XIRR_X:
 821 #endif
 822                 return 1;
 823         }
 824
 825         /* See if it's in the real-mode table */
 826         return kvmppc_hcall_impl_hv_realmode(cmd);
 827 }
 828
 829 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
 830                                         struct kvm_vcpu *vcpu)
 831 {
 832         u32 last_inst;
 833
 834         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
 835                                         EMULATE_DONE) {
 836                 /*
 837                  * Fetch failed, so return to guest and
 838                  * try executing it again.
 839                  */
 840                 return RESUME_GUEST;
 841         }
 842
 843         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
 844                 run->exit_reason = KVM_EXIT_DEBUG;
 845                 run->debug.arch.address = kvmppc_get_pc(vcpu);
 846                 return RESUME_HOST;
 847         } else {
 848                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 849                 return RESUME_GUEST;
 850         }
 851 }
 852
 853 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
 854                                  struct task_struct *tsk)
 855 {
 856         int r = RESUME_HOST;
 857
 858         vcpu->stat.sum_exits++;
 859
 860         /*
 861          * This can happen if an interrupt occurs in the last stages
 862          * of guest entry or the first stages of guest exit (i.e. after
 863          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
 864          * and before setting it to KVM_GUEST_MODE_HOST_HV).
 865          * That can happen due to a bug, or due to a machine check
 866          * occurring at just the wrong time.
 867          */
 868         if (vcpu->arch.shregs.msr & MSR_HV) {
 869                 printk(KERN_EMERG "KVM trap in HV mode!\n");
 870                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 871                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 872                         vcpu->arch.shregs.msr);
 873                 kvmppc_dump_regs(vcpu);
 874                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
 875                 run->hw.hardware_exit_reason = vcpu->arch.trap;
 876                 return RESUME_HOST;
 877         }
 878         run->exit_reason = KVM_EXIT_UNKNOWN;
 879         run->ready_for_interrupt_injection = 1;
 880         switch (vcpu->arch.trap) {
 881         /* We're good on these - the host merely wanted to get our attention */
 882         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 883                 vcpu->stat.dec_exits++;
 884                 r = RESUME_GUEST;
 885                 break;
 886         case BOOK3S_INTERRUPT_EXTERNAL:
 887         case BOOK3S_INTERRUPT_H_DOORBELL:
 888                 vcpu->stat.ext_intr_exits++;
 889                 r = RESUME_GUEST;
 890                 break;
 891         /* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
 892         case BOOK3S_INTERRUPT_HMI:
 893         case BOOK3S_INTERRUPT_PERFMON:
 894                 r = RESUME_GUEST;
 895                 break;
 896         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 897                 /*
 898                  * Deliver a machine check interrupt to the guest.
 899                  * We have to do this, even if the host has handled the
 900                  * machine check, because machine checks use SRR0/1 and
 901                  * the interrupt might have trashed guest state in them.
 902                  */
 903                 kvmppc_book3s_queue_irqprio(vcpu,
 904                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 905                 r = RESUME_GUEST;
 906                 break;
 907         case BOOK3S_INTERRUPT_PROGRAM:
 908         {
 909                 ulong flags;
 910                 /*
 911                  * Normally program interrupts are delivered directly
 912                  * to the guest by the hardware, but we can get here
 913                  * as a result of a hypervisor emulation interrupt
 914                  * (e40) getting turned into a 700 by BML RTAS.
 915                  */
 916                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 917                 kvmppc_core_queue_program(vcpu, flags);
 918                 r = RESUME_GUEST;
 919                 break;
 920         }
 921         case BOOK3S_INTERRUPT_SYSCALL:
 922         {
 923                 /* hcall - punt to userspace */
 924                 int i;
 925
 926                 /* hypercall with MSR_PR has already been handled in rmode,
 927                  * and never reaches here.
 928                  */
 929
 930                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 931                 for (i = 0; i < 9; ++i)
 932                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 933                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 934                 vcpu->arch.hcall_needed = 1;
 935                 r = RESUME_HOST;
 936                 break;
 937         }
 938         /*
 939          * We get these next two if the guest accesses a page which it thinks
 940          * it has mapped but which is not actually present, either because
 941          * it is for an emulated I/O device or because the corresonding
 942          * host page has been paged out.  Any other HDSI/HISI interrupts
 943          * have been handled already.
 944          */
 945         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 946                 r = RESUME_PAGE_FAULT;
 947                 break;
 948         case BOOK3S_INTERRUPT_H_INST_STORAGE:
 949                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
 950                 vcpu->arch.fault_dsisr = 0;
 951                 r = RESUME_PAGE_FAULT;
 952                 break;
 953         /*
 954          * This occurs if the guest executes an illegal instruction.
 955          * If the guest debug is disabled, generate a program interrupt
 956          * to the guest. If guest debug is enabled, we need to check
 957          * whether the instruction is a software breakpoint instruction.
 958          * Accordingly return to Guest or Host.
 959          */
 960         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 961                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
 962                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
 963                                 swab32(vcpu->arch.emul_inst) :
 964                                 vcpu->arch.emul_inst;
 965                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
 966                         r = kvmppc_emulate_debug_inst(run, vcpu);
 967                 } else {
 968                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 969                         r = RESUME_GUEST;
 970                 }
 971                 break;
 972         /*
 973          * This occurs if the guest (kernel or userspace), does something that
 974          * is prohibited by HFSCR.  We just generate a program interrupt to
 975          * the guest.
 976          */
 977         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
 978                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 979                 r = RESUME_GUEST;
 980                 break;
 981         default:
 982                 kvmppc_dump_regs(vcpu);
 983                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 984                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 985                         vcpu->arch.shregs.msr);
 986                 run->hw.hardware_exit_reason = vcpu->arch.trap;
 987                 r = RESUME_HOST;
 988                 break;
 989         }
 990
 991         return r;
 992 }
 993
 994 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
 995                                             struct kvm_sregs *sregs)
 996 {
 997         int i;
 998
 999         memset(sregs, 0, sizeof(struct kvm_sregs));
1000         sregs->pvr = vcpu->arch.pvr;
1001         for (i = 0; i < vcpu->arch.slb_max; i++) {
1002                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1003                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1004         }
1005
1006         return 0;
1007 }
1008
1009 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1010                                             struct kvm_sregs *sregs)
1011 {
1012         int i, j;
1013
1014         /* Only accept the same PVR as the host's, since we can't spoof it */
1015         if (sregs->pvr != vcpu->arch.pvr)
1016                 return -EINVAL;
1017
1018         j = 0;
1019         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1020                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1021                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1022                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1023                         ++j;
1024                 }
1025         }
1026         vcpu->arch.slb_max = j;
1027
1028         return 0;
1029 }
1030
1031 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1032                 bool preserve_top32)
1033 {
1034         struct kvm *kvm = vcpu->kvm;
1035         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1036         u64 mask;
1037
1038         mutex_lock(&kvm->lock);
1039         spin_lock(&vc->lock);
1040         /*
1041          * If ILE (interrupt little-endian) has changed, update the
1042          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1043          */
1044         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1045                 struct kvm_vcpu *vcpu;
1046                 int i;
1047
1048                 kvm_for_each_vcpu(i, vcpu, kvm) {
1049                         if (vcpu->arch.vcore != vc)
1050                                 continue;
1051                         if (new_lpcr & LPCR_ILE)
1052                                 vcpu->arch.intr_msr |= MSR_LE;
1053                         else
1054                                 vcpu->arch.intr_msr &= ~MSR_LE;
1055                 }
1056         }
1057
1058         /*
1059          * Userspace can only modify DPFD (default prefetch depth),
1060          * ILE (interrupt little-endian) and TC (translation control).
1061          * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
1062          */
1063         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1064         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1065                 mask |= LPCR_AIL;
1066
1067         /* Broken 32-bit version of LPCR must not clear top bits */
1068         if (preserve_top32)
1069                 mask &= 0xFFFFFFFF;
1070         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1071         spin_unlock(&vc->lock);
1072         mutex_unlock(&kvm->lock);
1073 }
1074
1075 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1076                                  union kvmppc_one_reg *val)
1077 {
1078         int r = 0;
1079         long int i;
1080
1081         switch (id) {
1082         case KVM_REG_PPC_DEBUG_INST:
1083                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1084                 break;
1085         case KVM_REG_PPC_HIOR:
1086                 *val = get_reg_val(id, 0);
1087                 break;
1088         case KVM_REG_PPC_DABR:
1089                 *val = get_reg_val(id, vcpu->arch.dabr);
1090                 break;
1091         case KVM_REG_PPC_DABRX:
1092                 *val = get_reg_val(id, vcpu->arch.dabrx);
1093                 break;
1094         case KVM_REG_PPC_DSCR:
1095                 *val = get_reg_val(id, vcpu->arch.dscr);
1096                 break;
1097         case KVM_REG_PPC_PURR:
1098                 *val = get_reg_val(id, vcpu->arch.purr);
1099                 break;
1100         case KVM_REG_PPC_SPURR:
1101                 *val = get_reg_val(id, vcpu->arch.spurr);
1102                 break;
1103         case KVM_REG_PPC_AMR:
1104                 *val = get_reg_val(id, vcpu->arch.amr);
1105                 break;
1106         case KVM_REG_PPC_UAMOR:
1107                 *val = get_reg_val(id, vcpu->arch.uamor);
1108                 break;
1109         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1110                 i = id - KVM_REG_PPC_MMCR0;
1111                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1112                 break;
1113         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1114                 i = id - KVM_REG_PPC_PMC1;
1115                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1116                 break;
1117         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1118                 i = id - KVM_REG_PPC_SPMC1;
1119                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1120                 break;
1121         case KVM_REG_PPC_SIAR:
1122                 *val = get_reg_val(id, vcpu->arch.siar);
1123                 break;
1124         case KVM_REG_PPC_SDAR:
1125                 *val = get_reg_val(id, vcpu->arch.sdar);
1126                 break;
1127         case KVM_REG_PPC_SIER:
1128                 *val = get_reg_val(id, vcpu->arch.sier);
1129                 break;
1130         case KVM_REG_PPC_IAMR:
1131                 *val = get_reg_val(id, vcpu->arch.iamr);
1132                 break;
1133         case KVM_REG_PPC_PSPB:
1134                 *val = get_reg_val(id, vcpu->arch.pspb);
1135                 break;
1136         case KVM_REG_PPC_DPDES:
1137                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1138                 break;
1139         case KVM_REG_PPC_DAWR:
1140                 *val = get_reg_val(id, vcpu->arch.dawr);
1141                 break;
1142         case KVM_REG_PPC_DAWRX:
1143                 *val = get_reg_val(id, vcpu->arch.dawrx);
1144                 break;
1145         case KVM_REG_PPC_CIABR:
1146                 *val = get_reg_val(id, vcpu->arch.ciabr);
1147                 break;
1148         case KVM_REG_PPC_CSIGR:
1149                 *val = get_reg_val(id, vcpu->arch.csigr);
1150                 break;
1151         case KVM_REG_PPC_TACR:
1152                 *val = get_reg_val(id, vcpu->arch.tacr);
1153                 break;
1154         case KVM_REG_PPC_TCSCR:
1155                 *val = get_reg_val(id, vcpu->arch.tcscr);
1156                 break;
1157         case KVM_REG_PPC_PID:
1158                 *val = get_reg_val(id, vcpu->arch.pid);
1159                 break;
1160         case KVM_REG_PPC_ACOP:
1161                 *val = get_reg_val(id, vcpu->arch.acop);
1162                 break;
1163         case KVM_REG_PPC_WORT:
1164                 *val = get_reg_val(id, vcpu->arch.wort);
1165                 break;
1166         case KVM_REG_PPC_VPA_ADDR:
1167                 spin_lock(&vcpu->arch.vpa_update_lock);
1168                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1169                 spin_unlock(&vcpu->arch.vpa_update_lock);
1170                 break;
1171         case KVM_REG_PPC_VPA_SLB:
1172                 spin_lock(&vcpu->arch.vpa_update_lock);
1173                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1174                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1175                 spin_unlock(&vcpu->arch.vpa_update_lock);
1176                 break;
1177         case KVM_REG_PPC_VPA_DTL:
1178                 spin_lock(&vcpu->arch.vpa_update_lock);
1179                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1180                 val->vpaval.length = vcpu->arch.dtl.len;
1181                 spin_unlock(&vcpu->arch.vpa_update_lock);
1182                 break;
1183         case KVM_REG_PPC_TB_OFFSET:
1184                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1185                 break;
1186         case KVM_REG_PPC_LPCR:
1187         case KVM_REG_PPC_LPCR_64:
1188                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1189                 break;
1190         case KVM_REG_PPC_PPR:
1191                 *val = get_reg_val(id, vcpu->arch.ppr);
1192                 break;
1193 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1194         case KVM_REG_PPC_TFHAR:
1195                 *val = get_reg_val(id, vcpu->arch.tfhar);
1196                 break;
1197         case KVM_REG_PPC_TFIAR:
1198                 *val = get_reg_val(id, vcpu->arch.tfiar);
1199                 break;
1200         case KVM_REG_PPC_TEXASR:
1201                 *val = get_reg_val(id, vcpu->arch.texasr);
1202                 break;
1203         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1204                 i = id - KVM_REG_PPC_TM_GPR0;
1205                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1206                 break;
1207         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1208         {
1209                 int j;
1210                 i = id - KVM_REG_PPC_TM_VSR0;
1211                 if (i < 32)
1212                         for (j = 0; j < TS_FPRWIDTH; j++)
1213                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1214                 else {
1215                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1216                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1217                         else
1218                                 r = -ENXIO;
1219                 }
1220                 break;
1221         }
1222         case KVM_REG_PPC_TM_CR:
1223                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1224                 break;
1225         case KVM_REG_PPC_TM_LR:
1226                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1227                 break;
1228         case KVM_REG_PPC_TM_CTR:
1229                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1230                 break;
1231         case KVM_REG_PPC_TM_FPSCR:
1232                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1233                 break;
1234         case KVM_REG_PPC_TM_AMR:
1235                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1236                 break;
1237         case KVM_REG_PPC_TM_PPR:
1238                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1239                 break;
1240         case KVM_REG_PPC_TM_VRSAVE:
1241                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1242                 break;
1243         case KVM_REG_PPC_TM_VSCR:
1244                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1245                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1246                 else
1247                         r = -ENXIO;
1248                 break;
1249         case KVM_REG_PPC_TM_DSCR:
1250                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1251                 break;
1252         case KVM_REG_PPC_TM_TAR:
1253                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1254                 break;
1255 #endif
1256         case KVM_REG_PPC_ARCH_COMPAT:
1257                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1258                 break;
1259         default:
1260                 r = -EINVAL;
1261                 break;
1262         }
1263
1264         return r;
1265 }
1266
1267 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1268                                  union kvmppc_one_reg *val)
1269 {
1270         int r = 0;
1271         long int i;
1272         unsigned long addr, len;
1273
1274         switch (id) {
1275         case KVM_REG_PPC_HIOR:
1276                 /* Only allow this to be set to zero */
1277                 if (set_reg_val(id, *val))
1278                         r = -EINVAL;
1279                 break;
1280         case KVM_REG_PPC_DABR:
1281                 vcpu->arch.dabr = set_reg_val(id, *val);
1282                 break;
1283         case KVM_REG_PPC_DABRX:
1284                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1285                 break;
1286         case KVM_REG_PPC_DSCR:
1287                 vcpu->arch.dscr = set_reg_val(id, *val);
1288                 break;
1289         case KVM_REG_PPC_PURR:
1290                 vcpu->arch.purr = set_reg_val(id, *val);
1291                 break;
1292         case KVM_REG_PPC_SPURR:
1293                 vcpu->arch.spurr = set_reg_val(id, *val);
1294                 break;
1295         case KVM_REG_PPC_AMR:
1296                 vcpu->arch.amr = set_reg_val(id, *val);
1297                 break;
1298         case KVM_REG_PPC_UAMOR:
1299                 vcpu->arch.uamor = set_reg_val(id, *val);
1300                 break;
1301         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1302                 i = id - KVM_REG_PPC_MMCR0;
1303                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1304                 break;
1305         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1306                 i = id - KVM_REG_PPC_PMC1;
1307                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1308                 break;
1309         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1310                 i = id - KVM_REG_PPC_SPMC1;
1311                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1312                 break;
1313         case KVM_REG_PPC_SIAR:
1314                 vcpu->arch.siar = set_reg_val(id, *val);
1315                 break;
1316         case KVM_REG_PPC_SDAR:
1317                 vcpu->arch.sdar = set_reg_val(id, *val);
1318                 break;
1319         case KVM_REG_PPC_SIER:
1320                 vcpu->arch.sier = set_reg_val(id, *val);
1321                 break;
1322         case KVM_REG_PPC_IAMR:
1323                 vcpu->arch.iamr = set_reg_val(id, *val);
1324                 break;
1325         case KVM_REG_PPC_PSPB:
1326                 vcpu->arch.pspb = set_reg_val(id, *val);
1327                 break;
1328         case KVM_REG_PPC_DPDES:
1329                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1330                 break;
1331         case KVM_REG_PPC_DAWR:
1332                 vcpu->arch.dawr = set_reg_val(id, *val);
1333                 break;
1334         case KVM_REG_PPC_DAWRX:
1335                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1336                 break;
1337         case KVM_REG_PPC_CIABR:
1338                 vcpu->arch.ciabr = set_reg_val(id, *val);
1339                 /* Don't allow setting breakpoints in hypervisor code */
1340                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1341                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1342                 break;
1343         case KVM_REG_PPC_CSIGR:
1344                 vcpu->arch.csigr = set_reg_val(id, *val);
1345                 break;
1346         case KVM_REG_PPC_TACR:
1347                 vcpu->arch.tacr = set_reg_val(id, *val);
1348                 break;
1349         case KVM_REG_PPC_TCSCR:
1350                 vcpu->arch.tcscr = set_reg_val(id, *val);
1351                 break;
1352         case KVM_REG_PPC_PID:
1353                 vcpu->arch.pid = set_reg_val(id, *val);
1354                 break;
1355         case KVM_REG_PPC_ACOP:
1356                 vcpu->arch.acop = set_reg_val(id, *val);
1357                 break;
1358         case KVM_REG_PPC_WORT:
1359                 vcpu->arch.wort = set_reg_val(id, *val);
1360                 break;
1361         case KVM_REG_PPC_VPA_ADDR:
1362                 addr = set_reg_val(id, *val);
1363                 r = -EINVAL;
1364                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1365                               vcpu->arch.dtl.next_gpa))
1366                         break;
1367                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1368                 break;
1369         case KVM_REG_PPC_VPA_SLB:
1370                 addr = val->vpaval.addr;
1371                 len = val->vpaval.length;
1372                 r = -EINVAL;
1373                 if (addr && !vcpu->arch.vpa.next_gpa)
1374                         break;
1375                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1376                 break;
1377         case KVM_REG_PPC_VPA_DTL:
1378                 addr = val->vpaval.addr;
1379                 len = val->vpaval.length;
1380                 r = -EINVAL;
1381                 if (addr && (len < sizeof(struct dtl_entry) ||
1382                              !vcpu->arch.vpa.next_gpa))
1383                         break;
1384                 len -= len % sizeof(struct dtl_entry);
1385                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1386                 break;
1387         case KVM_REG_PPC_TB_OFFSET:
1388                 /* round up to multiple of 2^24 */
1389                 vcpu->arch.vcore->tb_offset =
1390                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1391                 break;
1392         case KVM_REG_PPC_LPCR:
1393                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1394                 break;
1395         case KVM_REG_PPC_LPCR_64:
1396                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1397                 break;
1398         case KVM_REG_PPC_PPR:
1399                 vcpu->arch.ppr = set_reg_val(id, *val);
1400                 break;
1401 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1402         case KVM_REG_PPC_TFHAR:
1403                 vcpu->arch.tfhar = set_reg_val(id, *val);
1404                 break;
1405         case KVM_REG_PPC_TFIAR:
1406                 vcpu->arch.tfiar = set_reg_val(id, *val);
1407                 break;
1408         case KVM_REG_PPC_TEXASR:
1409                 vcpu->arch.texasr = set_reg_val(id, *val);
1410                 break;
1411         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1412                 i = id - KVM_REG_PPC_TM_GPR0;
1413                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1414                 break;
1415         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1416         {
1417                 int j;
1418                 i = id - KVM_REG_PPC_TM_VSR0;
1419                 if (i < 32)
1420                         for (j = 0; j < TS_FPRWIDTH; j++)
1421                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1422                 else
1423                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1424                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1425                         else
1426                                 r = -ENXIO;
1427                 break;
1428         }
1429         case KVM_REG_PPC_TM_CR:
1430                 vcpu->arch.cr_tm = set_reg_val(id, *val);
1431                 break;
1432         case KVM_REG_PPC_TM_LR:
1433                 vcpu->arch.lr_tm = set_reg_val(id, *val);
1434                 break;
1435         case KVM_REG_PPC_TM_CTR:
1436                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1437                 break;
1438         case KVM_REG_PPC_TM_FPSCR:
1439                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1440                 break;
1441         case KVM_REG_PPC_TM_AMR:
1442                 vcpu->arch.amr_tm = set_reg_val(id, *val);
1443                 break;
1444         case KVM_REG_PPC_TM_PPR:
1445                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1446                 break;
1447         case KVM_REG_PPC_TM_VRSAVE:
1448                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1449                 break;
1450         case KVM_REG_PPC_TM_VSCR:
1451                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1452                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1453                 else
1454                         r = - ENXIO;
1455                 break;
1456         case KVM_REG_PPC_TM_DSCR:
1457                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1458                 break;
1459         case KVM_REG_PPC_TM_TAR:
1460                 vcpu->arch.tar_tm = set_reg_val(id, *val);
1461                 break;
1462 #endif
1463         case KVM_REG_PPC_ARCH_COMPAT:
1464                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1465                 break;
1466         default:
1467                 r = -EINVAL;
1468                 break;
1469         }
1470
1471         return r;
1472 }
1473
1474 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1475 {
1476         struct kvmppc_vcore *vcore;
1477
1478         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1479
1480         if (vcore == NULL)
1481                 return NULL;
1482
1483         INIT_LIST_HEAD(&vcore->runnable_threads);
1484         spin_lock_init(&vcore->lock);
1485         spin_lock_init(&vcore->stoltb_lock);
1486         init_waitqueue_head(&vcore->wq);
1487         vcore->preempt_tb = TB_NIL;
1488         vcore->lpcr = kvm->arch.lpcr;
1489         vcore->first_vcpuid = core * threads_per_subcore;
1490         vcore->kvm = kvm;
1491         INIT_LIST_HEAD(&vcore->preempt_list);
1492
1493         return vcore;
1494 }
1495
1496 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1497 static struct debugfs_timings_element {
1498         const char *name;
1499         size_t offset;
1500 } timings[] = {
1501         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
1502         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
1503         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
1504         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
1505         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
1506 };
1507
1508 #define N_TIMINGS       (sizeof(timings) / sizeof(timings[0]))
1509
1510 struct debugfs_timings_state {
1511         struct kvm_vcpu *vcpu;
1512         unsigned int    buflen;
1513         char            buf[N_TIMINGS * 100];
1514 };
1515
1516 static int debugfs_timings_open(struct inode *inode, struct file *file)
1517 {
1518         struct kvm_vcpu *vcpu = inode->i_private;
1519         struct debugfs_timings_state *p;
1520
1521         p = kzalloc(sizeof(*p), GFP_KERNEL);
1522         if (!p)
1523                 return -ENOMEM;
1524
1525         kvm_get_kvm(vcpu->kvm);
1526         p->vcpu = vcpu;
1527         file->private_data = p;
1528
1529         return nonseekable_open(inode, file);
1530 }
1531
1532 static int debugfs_timings_release(struct inode *inode, struct file *file)
1533 {
1534         struct debugfs_timings_state *p = file->private_data;
1535
1536         kvm_put_kvm(p->vcpu->kvm);
1537         kfree(p);
1538         return 0;
1539 }
1540
1541 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1542                                     size_t len, loff_t *ppos)
1543 {
1544         struct debugfs_timings_state *p = file->private_data;
1545         struct kvm_vcpu *vcpu = p->vcpu;
1546         char *s, *buf_end;
1547         struct kvmhv_tb_accumulator tb;
1548         u64 count;
1549         loff_t pos;
1550         ssize_t n;
1551         int i, loops;
1552         bool ok;
1553
1554         if (!p->buflen) {
1555                 s = p->buf;
1556                 buf_end = s + sizeof(p->buf);
1557                 for (i = 0; i < N_TIMINGS; ++i) {
1558                         struct kvmhv_tb_accumulator *acc;
1559
1560                         acc = (struct kvmhv_tb_accumulator *)
1561                                 ((unsigned long)vcpu + timings[i].offset);
1562                         ok = false;
1563                         for (loops = 0; loops < 1000; ++loops) {
1564                                 count = acc->seqcount;
1565                                 if (!(count & 1)) {
1566                                         smp_rmb();
1567                                         tb = *acc;
1568                                         smp_rmb();
1569                                         if (count == acc->seqcount) {
1570                                                 ok = true;
1571                                                 break;
1572                                         }
1573                                 }
1574                                 udelay(1);
1575                         }
1576                         if (!ok)
1577                                 snprintf(s, buf_end - s, "%s: stuck\n",
1578                                         timings[i].name);
1579                         else
1580                                 snprintf(s, buf_end - s,
1581                                         "%s: %llu %llu %llu %llu\n",
1582                                         timings[i].name, count / 2,
1583                                         tb_to_ns(tb.tb_total),
1584                                         tb_to_ns(tb.tb_min),
1585                                         tb_to_ns(tb.tb_max));
1586                         s += strlen(s);
1587                 }
1588                 p->buflen = s - p->buf;
1589         }
1590
1591         pos = *ppos;
1592         if (pos >= p->buflen)
1593                 return 0;
1594         if (len > p->buflen - pos)
1595                 len = p->buflen - pos;
1596         n = copy_to_user(buf, p->buf + pos, len);
1597         if (n) {
1598                 if (n == len)
1599                         return -EFAULT;
1600                 len -= n;
1601         }
1602         *ppos = pos + len;
1603         return len;
1604 }
1605
1606 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1607                                      size_t len, loff_t *ppos)
1608 {
1609         return -EACCES;
1610 }
1611
1612 static const struct file_operations debugfs_timings_ops = {
1613         .owner   = THIS_MODULE,
1614         .open    = debugfs_timings_open,
1615         .release = debugfs_timings_release,
1616         .read    = debugfs_timings_read,
1617         .write   = debugfs_timings_write,
1618         .llseek  = generic_file_llseek,
1619 };
1620
1621 /* Create a debugfs directory for the vcpu */
1622 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1623 {
1624         char buf[16];
1625         struct kvm *kvm = vcpu->kvm;
1626
1627         snprintf(buf, sizeof(buf), "vcpu%u", id);
1628         if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1629                 return;
1630         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1631         if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1632                 return;
1633         vcpu->arch.debugfs_timings =
1634                 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1635                                     vcpu, &debugfs_timings_ops);
1636 }
1637
1638 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1639 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1640 {
1641 }
1642 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1643
1644 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1645                                                    unsigned int id)
1646 {
1647         struct kvm_vcpu *vcpu;
1648         int err = -EINVAL;
1649         int core;
1650         struct kvmppc_vcore *vcore;
1651
1652         core = id / threads_per_subcore;
1653         if (core >= KVM_MAX_VCORES)
1654                 goto out;
1655
1656         err = -ENOMEM;
1657         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1658         if (!vcpu)
1659                 goto out;
1660
1661         err = kvm_vcpu_init(vcpu, kvm, id);
1662         if (err)
1663                 goto free_vcpu;
1664
1665         vcpu->arch.shared = &vcpu->arch.shregs;
1666 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1667         /*
1668          * The shared struct is never shared on HV,
1669          * so we can always use host endianness
1670          */
1671 #ifdef __BIG_ENDIAN__
1672         vcpu->arch.shared_big_endian = true;
1673 #else
1674         vcpu->arch.shared_big_endian = false;
1675 #endif
1676 #endif
1677         vcpu->arch.mmcr[0] = MMCR0_FC;
1678         vcpu->arch.ctrl = CTRL_RUNLATCH;
1679         /* default to host PVR, since we can't spoof it */
1680         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1681         spin_lock_init(&vcpu->arch.vpa_update_lock);
1682         spin_lock_init(&vcpu->arch.tbacct_lock);
1683         vcpu->arch.busy_preempt = TB_NIL;
1684         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1685
1686         kvmppc_mmu_book3s_hv_init(vcpu);
1687
1688         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1689
1690         init_waitqueue_head(&vcpu->arch.cpu_run);
1691
1692         mutex_lock(&kvm->lock);
1693         vcore = kvm->arch.vcores[core];
1694         if (!vcore) {
1695                 vcore = kvmppc_vcore_create(kvm, core);
1696                 kvm->arch.vcores[core] = vcore;
1697                 kvm->arch.online_vcores++;
1698         }
1699         mutex_unlock(&kvm->lock);
1700
1701         if (!vcore)
1702                 goto free_vcpu;
1703
1704         spin_lock(&vcore->lock);
1705         ++vcore->num_threads;
1706         spin_unlock(&vcore->lock);
1707         vcpu->arch.vcore = vcore;
1708         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1709         vcpu->arch.thread_cpu = -1;
1710
1711         vcpu->arch.cpu_type = KVM_CPU_3S_64;
1712         kvmppc_sanity_check(vcpu);
1713
1714         debugfs_vcpu_init(vcpu, id);
1715
1716         return vcpu;
1717
1718 free_vcpu:
1719         kmem_cache_free(kvm_vcpu_cache, vcpu);
1720 out:
1721         return ERR_PTR(err);
1722 }
1723
1724 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1725 {
1726         if (vpa->pinned_addr)
1727                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1728                                         vpa->dirty);
1729 }
1730
1731 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1732 {
1733         spin_lock(&vcpu->arch.vpa_update_lock);
1734         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1735         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1736         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1737         spin_unlock(&vcpu->arch.vpa_update_lock);
1738         kvm_vcpu_uninit(vcpu);
1739         kmem_cache_free(kvm_vcpu_cache, vcpu);
1740 }
1741
1742 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1743 {
1744         /* Indicate we want to get back into the guest */
1745         return 1;
1746 }
1747
1748 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1749 {
1750         unsigned long dec_nsec, now;
1751
1752         now = get_tb();
1753         if (now > vcpu->arch.dec_expires) {
1754                 /* decrementer has already gone negative */
1755                 kvmppc_core_queue_dec(vcpu);
1756                 kvmppc_core_prepare_to_enter(vcpu);
1757                 return;
1758         }
1759         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1760                    / tb_ticks_per_sec;
1761         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1762                       HRTIMER_MODE_REL);
1763         vcpu->arch.timer_running = 1;
1764 }
1765
1766 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1767 {
1768         vcpu->arch.ceded = 0;
1769         if (vcpu->arch.timer_running) {
1770                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1771                 vcpu->arch.timer_running = 0;
1772         }
1773 }
1774
1775 extern void __kvmppc_vcore_entry(void);
1776
1777 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1778                                    struct kvm_vcpu *vcpu)
1779 {
1780         u64 now;
1781
1782         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1783                 return;
1784         spin_lock_irq(&vcpu->arch.tbacct_lock);
1785         now = mftb();
1786         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1787                 vcpu->arch.stolen_logged;
1788         vcpu->arch.busy_preempt = now;
1789         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1790         spin_unlock_irq(&vcpu->arch.tbacct_lock);
1791         --vc->n_runnable;
1792         list_del(&vcpu->arch.run_list);
1793 }
1794
1795 static int kvmppc_grab_hwthread(int cpu)
1796 {
1797         struct paca_struct *tpaca;
1798         long timeout = 10000;
1799
1800         tpaca = &paca[cpu];
1801
1802         /* Ensure the thread won't go into the kernel if it wakes */
1803         tpaca->kvm_hstate.kvm_vcpu = NULL;
1804         tpaca->kvm_hstate.kvm_vcore = NULL;
1805         tpaca->kvm_hstate.napping = 0;
1806         smp_wmb();
1807         tpaca->kvm_hstate.hwthread_req = 1;
1808
1809         /*
1810          * If the thread is already executing in the kernel (e.g. handling
1811          * a stray interrupt), wait for it to get back to nap mode.
1812          * The smp_mb() is to ensure that our setting of hwthread_req
1813          * is visible before we look at hwthread_state, so if this
1814          * races with the code at system_reset_pSeries and the thread
1815          * misses our setting of hwthread_req, we are sure to see its
1816          * setting of hwthread_state, and vice versa.
1817          */
1818         smp_mb();
1819         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1820                 if (--timeout <= 0) {
1821                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1822                         return -EBUSY;
1823                 }
1824                 udelay(1);
1825         }
1826         return 0;
1827 }
1828
1829 static void kvmppc_release_hwthread(int cpu)
1830 {
1831         struct paca_struct *tpaca;
1832
1833         tpaca = &paca[cpu];
1834         tpaca->kvm_hstate.hwthread_req = 0;
1835         tpaca->kvm_hstate.kvm_vcpu = NULL;
1836         tpaca->kvm_hstate.kvm_vcore = NULL;
1837         tpaca->kvm_hstate.kvm_split_mode = NULL;
1838 }
1839
1840 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
1841 {
1842         int cpu;
1843         struct paca_struct *tpaca;
1844         struct kvmppc_vcore *mvc = vc->master_vcore;
1845
1846         cpu = vc->pcpu;
1847         if (vcpu) {
1848                 if (vcpu->arch.timer_running) {
1849                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1850                         vcpu->arch.timer_running = 0;
1851                 }
1852                 cpu += vcpu->arch.ptid;
1853                 vcpu->cpu = mvc->pcpu;
1854                 vcpu->arch.thread_cpu = cpu;
1855         }
1856         tpaca = &paca[cpu];
1857         tpaca->kvm_hstate.kvm_vcpu = vcpu;
1858         tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
1859         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
1860         smp_wmb();
1861         tpaca->kvm_hstate.kvm_vcore = mvc;
1862         if (cpu != smp_processor_id())
1863                 kvmppc_ipi_thread(cpu);
1864 }
1865
1866 static void kvmppc_wait_for_nap(void)
1867 {
1868         int cpu = smp_processor_id();
1869         int i, loops;
1870
1871         for (loops = 0; loops < 1000000; ++loops) {
1872                 /*
1873                  * Check if all threads are finished.
1874                  * We set the vcore pointer when starting a thread
1875                  * and the thread clears it when finished, so we look
1876                  * for any threads that still have a non-NULL vcore ptr.
1877                  */
1878                 for (i = 1; i < threads_per_subcore; ++i)
1879                         if (paca[cpu + i].kvm_hstate.kvm_vcore)
1880                                 break;
1881                 if (i == threads_per_subcore) {
1882                         HMT_medium();
1883                         return;
1884                 }
1885                 HMT_low();
1886         }
1887         HMT_medium();
1888         for (i = 1; i < threads_per_subcore; ++i)
1889                 if (paca[cpu + i].kvm_hstate.kvm_vcore)
1890                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
1891 }
1892
1893 /*
1894  * Check that we are on thread 0 and that any other threads in
1895  * this core are off-line.  Then grab the threads so they can't
1896  * enter the kernel.
1897  */
1898 static int on_primary_thread(void)
1899 {
1900         int cpu = smp_processor_id();
1901         int thr;
1902
1903         /* Are we on a primary subcore? */
1904         if (cpu_thread_in_subcore(cpu))
1905                 return 0;
1906
1907         thr = 0;
1908         while (++thr < threads_per_subcore)
1909                 if (cpu_online(cpu + thr))
1910                         return 0;
1911
1912         /* Grab all hw threads so they can't go into the kernel */
1913         for (thr = 1; thr < threads_per_subcore; ++thr) {
1914                 if (kvmppc_grab_hwthread(cpu + thr)) {
1915                         /* Couldn't grab one; let the others go */
1916                         do {
1917                                 kvmppc_release_hwthread(cpu + thr);
1918                         } while (--thr > 0);
1919                         return 0;
1920                 }
1921         }
1922         return 1;
1923 }
1924
1925 /*
1926  * A list of virtual cores for each physical CPU.
1927  * These are vcores that could run but their runner VCPU tasks are
1928  * (or may be) preempted.
1929  */
1930 struct preempted_vcore_list {
1931         struct list_head        list;
1932         spinlock_t              lock;
1933 };
1934
1935 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
1936
1937 static void init_vcore_lists(void)
1938 {
1939         int cpu;
1940
1941         for_each_possible_cpu(cpu) {
1942                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
1943                 spin_lock_init(&lp->lock);
1944                 INIT_LIST_HEAD(&lp->list);
1945         }
1946 }
1947
1948 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
1949 {
1950         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
1951
1952         vc->vcore_state = VCORE_PREEMPT;
1953         vc->pcpu = smp_processor_id();
1954         if (vc->num_threads < threads_per_subcore) {
1955                 spin_lock(&lp->lock);
1956                 list_add_tail(&vc->preempt_list, &lp->list);
1957                 spin_unlock(&lp->lock);
1958         }
1959
1960         /* Start accumulating stolen time */
1961         kvmppc_core_start_stolen(vc);
1962 }
1963
1964 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
1965 {
1966         struct preempted_vcore_list *lp;
1967
1968         kvmppc_core_end_stolen(vc);
1969         if (!list_empty(&vc->preempt_list)) {
1970                 lp = &per_cpu(preempted_vcores, vc->pcpu);
1971                 spin_lock(&lp->lock);
1972                 list_del_init(&vc->preempt_list);
1973                 spin_unlock(&lp->lock);
1974         }
1975         vc->vcore_state = VCORE_INACTIVE;
1976 }
1977
1978 /*
1979  * This stores information about the virtual cores currently
1980  * assigned to a physical core.
1981  */
1982 struct core_info {
1983         int             n_subcores;
1984         int             max_subcore_threads;
1985         int             total_threads;
1986         int             subcore_threads[MAX_SUBCORES];
1987         struct kvm      *subcore_vm[MAX_SUBCORES];
1988         struct list_head vcs[MAX_SUBCORES];
1989 };
1990
1991 /*
1992  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
1993  * respectively in 2-way micro-threading (split-core) mode.
1994  */
1995 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
1996
1997 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
1998 {
1999         int sub;
2000
2001         memset(cip, 0, sizeof(*cip));
2002         cip->n_subcores = 1;
2003         cip->max_subcore_threads = vc->num_threads;
2004         cip->total_threads = vc->num_threads;
2005         cip->subcore_threads[0] = vc->num_threads;
2006         cip->subcore_vm[0] = vc->kvm;
2007         for (sub = 0; sub < MAX_SUBCORES; ++sub)
2008                 INIT_LIST_HEAD(&cip->vcs[sub]);
2009         list_add_tail(&vc->preempt_list, &cip->vcs[0]);
2010 }
2011
2012 static bool subcore_config_ok(int n_subcores, int n_threads)
2013 {
2014         /* Can only dynamically split if unsplit to begin with */
2015         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2016                 return false;
2017         if (n_subcores > MAX_SUBCORES)
2018                 return false;
2019         if (n_subcores > 1) {
2020                 if (!(dynamic_mt_modes & 2))
2021                         n_subcores = 4;
2022                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2023                         return false;
2024         }
2025
2026         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2027 }
2028
2029 static void init_master_vcore(struct kvmppc_vcore *vc)
2030 {
2031         vc->master_vcore = vc;
2032         vc->entry_exit_map = 0;
2033         vc->in_guest = 0;
2034         vc->napping_threads = 0;
2035         vc->conferring_threads = 0;
2036 }
2037
2038 /*
2039  * See if the existing subcores can be split into 3 (or fewer) subcores
2040  * of at most two threads each, so we can fit in another vcore.  This
2041  * assumes there are at most two subcores and at most 6 threads in total.
2042  */
2043 static bool can_split_piggybacked_subcores(struct core_info *cip)
2044 {
2045         int sub, new_sub;
2046         int large_sub = -1;
2047         int thr;
2048         int n_subcores = cip->n_subcores;
2049         struct kvmppc_vcore *vc, *vcnext;
2050         struct kvmppc_vcore *master_vc = NULL;
2051
2052         for (sub = 0; sub < cip->n_subcores; ++sub) {
2053                 if (cip->subcore_threads[sub] <= 2)
2054                         continue;
2055                 if (large_sub >= 0)
2056                         return false;
2057                 large_sub = sub;
2058                 vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2059                                       preempt_list);
2060                 if (vc->num_threads > 2)
2061                         return false;
2062                 n_subcores += (cip->subcore_threads[sub] - 1) >> 1;
2063         }
2064         if (large_sub < 0 || !subcore_config_ok(n_subcores + 1, 2))
2065                 return false;
2066
2067         /*
2068          * Seems feasible, so go through and move vcores to new subcores.
2069          * Note that when we have two or more vcores in one subcore,
2070          * all those vcores must have only one thread each.
2071          */
2072         new_sub = cip->n_subcores;
2073         thr = 0;
2074         sub = large_sub;
2075         list_for_each_entry_safe(vc, vcnext, &cip->vcs[sub], preempt_list) {
2076                 if (thr >= 2) {
2077                         list_del(&vc->preempt_list);
2078                         list_add_tail(&vc->preempt_list, &cip->vcs[new_sub]);
2079                         /* vc->num_threads must be 1 */
2080                         if (++cip->subcore_threads[new_sub] == 1) {
2081                                 cip->subcore_vm[new_sub] = vc->kvm;
2082                                 init_master_vcore(vc);
2083                                 master_vc = vc;
2084                                 ++cip->n_subcores;
2085                         } else {
2086                                 vc->master_vcore = master_vc;
2087                                 ++new_sub;
2088                         }
2089                 }
2090                 thr += vc->num_threads;
2091         }
2092         cip->subcore_threads[large_sub] = 2;
2093         cip->max_subcore_threads = 2;
2094
2095         return true;
2096 }
2097
2098 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2099 {
2100         int n_threads = vc->num_threads;
2101         int sub;
2102
2103         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2104                 return false;
2105
2106         if (n_threads < cip->max_subcore_threads)
2107                 n_threads = cip->max_subcore_threads;
2108         if (subcore_config_ok(cip->n_subcores + 1, n_threads)) {
2109                 cip->max_subcore_threads = n_threads;
2110         } else if (cip->n_subcores <= 2 && cip->total_threads <= 6 &&
2111                    vc->num_threads <= 2) {
2112                 /*
2113                  * We may be able to fit another subcore in by
2114                  * splitting an existing subcore with 3 or 4
2115                  * threads into two 2-thread subcores, or one
2116                  * with 5 or 6 threads into three subcores.
2117                  * We can only do this if those subcores have
2118                  * piggybacked virtual cores.
2119                  */
2120                 if (!can_split_piggybacked_subcores(cip))
2121                         return false;
2122         } else {
2123                 return false;
2124         }
2125
2126         sub = cip->n_subcores;
2127         ++cip->n_subcores;
2128         cip->total_threads += vc->num_threads;
2129         cip->subcore_threads[sub] = vc->num_threads;
2130         cip->subcore_vm[sub] = vc->kvm;
2131         init_master_vcore(vc);
2132         list_del(&vc->preempt_list);
2133         list_add_tail(&vc->preempt_list, &cip->vcs[sub]);
2134
2135         return true;
2136 }
2137
2138 static bool can_piggyback_subcore(struct kvmppc_vcore *pvc,
2139                                   struct core_info *cip, int sub)
2140 {
2141         struct kvmppc_vcore *vc;
2142         int n_thr;
2143
2144         vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2145                               preempt_list);
2146
2147         /* require same VM and same per-core reg values */
2148         if (pvc->kvm != vc->kvm ||
2149             pvc->tb_offset != vc->tb_offset ||
2150             pvc->pcr != vc->pcr ||
2151             pvc->lpcr != vc->lpcr)
2152                 return false;
2153
2154         /* P8 guest with > 1 thread per core would see wrong TIR value */
2155         if (cpu_has_feature(CPU_FTR_ARCH_207S) &&
2156             (vc->num_threads > 1 || pvc->num_threads > 1))
2157                 return false;
2158
2159         n_thr = cip->subcore_threads[sub] + pvc->num_threads;
2160         if (n_thr > cip->max_subcore_threads) {
2161                 if (!subcore_config_ok(cip->n_subcores, n_thr))
2162                         return false;
2163                 cip->max_subcore_threads = n_thr;
2164         }
2165
2166         cip->total_threads += pvc->num_threads;
2167         cip->subcore_threads[sub] = n_thr;
2168         pvc->master_vcore = vc;
2169         list_del(&pvc->preempt_list);
2170         list_add_tail(&pvc->preempt_list, &cip->vcs[sub]);
2171
2172         return true;
2173 }
2174
2175 /*
2176  * Work out whether it is possible to piggyback the execution of
2177  * vcore *pvc onto the execution of the other vcores described in *cip.
2178  */
2179 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2180                           int target_threads)
2181 {
2182         int sub;
2183
2184         if (cip->total_threads + pvc->num_threads > target_threads)
2185                 return false;
2186         for (sub = 0; sub < cip->n_subcores; ++sub)
2187                 if (cip->subcore_threads[sub] &&
2188                     can_piggyback_subcore(pvc, cip, sub))
2189                         return true;
2190
2191         if (can_dynamic_split(pvc, cip))
2192                 return true;
2193
2194         return false;
2195 }
2196
2197 static void prepare_threads(struct kvmppc_vcore *vc)
2198 {
2199         struct kvm_vcpu *vcpu, *vnext;
2200
2201         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2202                                  arch.run_list) {
2203                 if (signal_pending(vcpu->arch.run_task))
2204                         vcpu->arch.ret = -EINTR;
2205                 else if (vcpu->arch.vpa.update_pending ||
2206                          vcpu->arch.slb_shadow.update_pending ||
2207                          vcpu->arch.dtl.update_pending)
2208                         vcpu->arch.ret = RESUME_GUEST;
2209                 else
2210                         continue;
2211                 kvmppc_remove_runnable(vc, vcpu);
2212                 wake_up(&vcpu->arch.cpu_run);
2213         }
2214 }
2215
2216 static void collect_piggybacks(struct core_info *cip, int target_threads)
2217 {
2218         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2219         struct kvmppc_vcore *pvc, *vcnext;
2220
2221         spin_lock(&lp->lock);
2222         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2223                 if (!spin_trylock(&pvc->lock))
2224                         continue;
2225                 prepare_threads(pvc);
2226                 if (!pvc->n_runnable) {
2227                         list_del_init(&pvc->preempt_list);
2228                         if (pvc->runner == NULL) {
2229                                 pvc->vcore_state = VCORE_INACTIVE;
2230                                 kvmppc_core_end_stolen(pvc);
2231                         }
2232                         spin_unlock(&pvc->lock);
2233                         continue;
2234                 }
2235                 if (!can_piggyback(pvc, cip, target_threads)) {
2236                         spin_unlock(&pvc->lock);
2237                         continue;
2238                 }
2239                 kvmppc_core_end_stolen(pvc);
2240                 pvc->vcore_state = VCORE_PIGGYBACK;
2241                 if (cip->total_threads >= target_threads)
2242                         break;
2243         }
2244         spin_unlock(&lp->lock);
2245 }
2246
2247 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2248 {
2249         int still_running = 0;
2250         u64 now;
2251         long ret;
2252         struct kvm_vcpu *vcpu, *vnext;
2253
2254         spin_lock(&vc->lock);
2255         now = get_tb();
2256         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2257                                  arch.run_list) {
2258                 /* cancel pending dec exception if dec is positive */
2259                 if (now < vcpu->arch.dec_expires &&
2260                     kvmppc_core_pending_dec(vcpu))
2261                         kvmppc_core_dequeue_dec(vcpu);
2262
2263                 trace_kvm_guest_exit(vcpu);
2264
2265                 ret = RESUME_GUEST;
2266                 if (vcpu->arch.trap)
2267                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2268                                                     vcpu->arch.run_task);
2269
2270                 vcpu->arch.ret = ret;
2271                 vcpu->arch.trap = 0;
2272
2273                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2274                         if (vcpu->arch.pending_exceptions)
2275                                 kvmppc_core_prepare_to_enter(vcpu);
2276                         if (vcpu->arch.ceded)
2277                                 kvmppc_set_timer(vcpu);
2278                         else
2279                                 ++still_running;
2280                 } else {
2281                         kvmppc_remove_runnable(vc, vcpu);
2282                         wake_up(&vcpu->arch.cpu_run);
2283                 }
2284         }
2285         list_del_init(&vc->preempt_list);
2286         if (!is_master) {
2287                 if (still_running > 0) {
2288                         kvmppc_vcore_preempt(vc);
2289                 } else if (vc->runner) {
2290                         vc->vcore_state = VCORE_PREEMPT;
2291                         kvmppc_core_start_stolen(vc);
2292                 } else {
2293                         vc->vcore_state = VCORE_INACTIVE;
2294                 }
2295                 if (vc->n_runnable > 0 && vc->runner == NULL) {
2296                         /* make sure there's a candidate runner awake */
2297                         vcpu = list_first_entry(&vc->runnable_threads,
2298                                                 struct kvm_vcpu, arch.run_list);
2299                         wake_up(&vcpu->arch.cpu_run);
2300                 }
2301         }
2302         spin_unlock(&vc->lock);
2303 }
2304
2305 /*
2306  * Run a set of guest threads on a physical core.
2307  * Called with vc->lock held.
2308  */
2309 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2310 {
2311         struct kvm_vcpu *vcpu, *vnext;
2312         int i;
2313         int srcu_idx;
2314         struct core_info core_info;
2315         struct kvmppc_vcore *pvc, *vcnext;
2316         struct kvm_split_mode split_info, *sip;
2317         int split, subcore_size, active;
2318         int sub;
2319         bool thr0_done;
2320         unsigned long cmd_bit, stat_bit;
2321         int pcpu, thr;
2322         int target_threads;
2323
2324         /*
2325          * Remove from the list any threads that have a signal pending
2326          * or need a VPA update done
2327          */
2328         prepare_threads(vc);
2329
2330         /* if the runner is no longer runnable, let the caller pick a new one */
2331         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2332                 return;
2333
2334         /*
2335          * Initialize *vc.
2336          */
2337         init_master_vcore(vc);
2338         vc->preempt_tb = TB_NIL;
2339
2340         /*
2341          * Make sure we are running on primary threads, and that secondary
2342          * threads are offline.  Also check if the number of threads in this
2343          * guest are greater than the current system threads per guest.
2344          */
2345         if ((threads_per_core > 1) &&
2346             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2347                 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2348                                          arch.run_list) {
2349                         vcpu->arch.ret = -EBUSY;
2350                         kvmppc_remove_runnable(vc, vcpu);
2351                         wake_up(&vcpu->arch.cpu_run);
2352                 }
2353                 goto out;
2354         }
2355
2356         /*
2357          * See if we could run any other vcores on the physical core
2358          * along with this one.
2359          */
2360         init_core_info(&core_info, vc);
2361         pcpu = smp_processor_id();
2362         target_threads = threads_per_subcore;
2363         if (target_smt_mode && target_smt_mode < target_threads)
2364                 target_threads = target_smt_mode;
2365         if (vc->num_threads < target_threads)
2366                 collect_piggybacks(&core_info, target_threads);
2367
2368         /* Decide on micro-threading (split-core) mode */
2369         subcore_size = threads_per_subcore;
2370         cmd_bit = stat_bit = 0;
2371         split = core_info.n_subcores;
2372         sip = NULL;
2373         if (split > 1) {
2374                 /* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2375                 if (split == 2 && (dynamic_mt_modes & 2)) {
2376                         cmd_bit = HID0_POWER8_1TO2LPAR;
2377                         stat_bit = HID0_POWER8_2LPARMODE;
2378                 } else {
2379                         split = 4;
2380                         cmd_bit = HID0_POWER8_1TO4LPAR;
2381                         stat_bit = HID0_POWER8_4LPARMODE;
2382                 }
2383                 subcore_size = MAX_SMT_THREADS / split;
2384                 sip = &split_info;
2385                 memset(&split_info, 0, sizeof(split_info));
2386                 split_info.rpr = mfspr(SPRN_RPR);
2387                 split_info.pmmar = mfspr(SPRN_PMMAR);
2388                 split_info.ldbar = mfspr(SPRN_LDBAR);
2389                 split_info.subcore_size = subcore_size;
2390                 for (sub = 0; sub < core_info.n_subcores; ++sub)
2391                         split_info.master_vcs[sub] =
2392                                 list_first_entry(&core_info.vcs[sub],
2393                                         struct kvmppc_vcore, preempt_list);
2394                 /* order writes to split_info before kvm_split_mode pointer */
2395                 smp_wmb();
2396         }
2397         pcpu = smp_processor_id();
2398         for (thr = 0; thr < threads_per_subcore; ++thr)
2399                 paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2400
2401         /* Initiate micro-threading (split-core) if required */
2402         if (cmd_bit) {
2403                 unsigned long hid0 = mfspr(SPRN_HID0);
2404
2405                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2406                 mb();
2407                 mtspr(SPRN_HID0, hid0);
2408                 isync();
2409                 for (;;) {
2410                         hid0 = mfspr(SPRN_HID0);
2411                         if (hid0 & stat_bit)
2412                                 break;
2413                         cpu_relax();
2414                 }
2415         }
2416
2417         /* Start all the threads */
2418         active = 0;
2419         for (sub = 0; sub < core_info.n_subcores; ++sub) {
2420                 thr = subcore_thread_map[sub];
2421                 thr0_done = false;
2422                 active |= 1 << thr;
2423                 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2424                         pvc->pcpu = pcpu + thr;
2425                         list_for_each_entry(vcpu, &pvc->runnable_threads,
2426                                             arch.run_list) {
2427                                 kvmppc_start_thread(vcpu, pvc);
2428                                 kvmppc_create_dtl_entry(vcpu, pvc);
2429                                 trace_kvm_guest_enter(vcpu);
2430                                 if (!vcpu->arch.ptid)
2431                                         thr0_done = true;
2432                                 active |= 1 << (thr + vcpu->arch.ptid);
2433                         }
2434                         /*
2435                          * We need to start the first thread of each subcore
2436                          * even if it doesn't have a vcpu.
2437                          */
2438                         if (pvc->master_vcore == pvc && !thr0_done)
2439                                 kvmppc_start_thread(NULL, pvc);
2440                         thr += pvc->num_threads;
2441                 }
2442         }
2443
2444         /*
2445          * Ensure that split_info.do_nap is set after setting
2446          * the vcore pointer in the PACA of the secondaries.
2447          */
2448         smp_mb();
2449         if (cmd_bit)
2450                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
2451
2452         /*
2453          * When doing micro-threading, poke the inactive threads as well.
2454          * This gets them to the nap instruction after kvm_do_nap,
2455          * which reduces the time taken to unsplit later.
2456          */
2457         if (split > 1)
2458                 for (thr = 1; thr < threads_per_subcore; ++thr)
2459                         if (!(active & (1 << thr)))
2460                                 kvmppc_ipi_thread(pcpu + thr);
2461
2462         vc->vcore_state = VCORE_RUNNING;
2463         preempt_disable();
2464
2465         trace_kvmppc_run_core(vc, 0);
2466
2467         for (sub = 0; sub < core_info.n_subcores; ++sub)
2468                 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2469                         spin_unlock(&pvc->lock);
2470
2471         kvm_guest_enter();
2472
2473         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2474
2475         __kvmppc_vcore_entry();
2476
2477         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2478
2479         spin_lock(&vc->lock);
2480         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2481         vc->vcore_state = VCORE_EXITING;
2482
2483         /* wait for secondary threads to finish writing their state to memory */
2484         kvmppc_wait_for_nap();
2485
2486         /* Return to whole-core mode if we split the core earlier */
2487         if (split > 1) {
2488                 unsigned long hid0 = mfspr(SPRN_HID0);
2489                 unsigned long loops = 0;
2490
2491                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2492                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2493                 mb();
2494                 mtspr(SPRN_HID0, hid0);
2495                 isync();
2496                 for (;;) {
2497                         hid0 = mfspr(SPRN_HID0);
2498                         if (!(hid0 & stat_bit))
2499                                 break;
2500                         cpu_relax();
2501                         ++loops;
2502                 }
2503                 split_info.do_nap = 0;
2504         }
2505
2506         /* Let secondaries go back to the offline loop */
2507         for (i = 0; i < threads_per_subcore; ++i) {
2508                 kvmppc_release_hwthread(pcpu + i);
2509                 if (sip && sip->napped[i])
2510                         kvmppc_ipi_thread(pcpu + i);
2511         }
2512
2513         spin_unlock(&vc->lock);
2514
2515         /* make sure updates to secondary vcpu structs are visible now */
2516         smp_mb();
2517         kvm_guest_exit();
2518
2519         for (sub = 0; sub < core_info.n_subcores; ++sub)
2520                 list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2521                                          preempt_list)
2522                         post_guest_process(pvc, pvc == vc);
2523
2524         spin_lock(&vc->lock);
2525         preempt_enable();
2526
2527  out:
2528         vc->vcore_state = VCORE_INACTIVE;
2529         trace_kvmppc_run_core(vc, 1);
2530 }
2531
2532 /*
2533  * Wait for some other vcpu thread to execute us, and
2534  * wake us up when we need to handle something in the host.
2535  */
2536 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2537                                  struct kvm_vcpu *vcpu, int wait_state)
2538 {
2539         DEFINE_WAIT(wait);
2540
2541         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2542         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2543                 spin_unlock(&vc->lock);
2544                 schedule();
2545                 spin_lock(&vc->lock);
2546         }
2547         finish_wait(&vcpu->arch.cpu_run, &wait);
2548 }
2549
2550 /*
2551  * All the vcpus in this vcore are idle, so wait for a decrementer
2552  * or external interrupt to one of the vcpus.  vc->lock is held.
2553  */
2554 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2555 {
2556         struct kvm_vcpu *vcpu;
2557         int do_sleep = 1;
2558
2559         DEFINE_WAIT(wait);
2560
2561         prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2562
2563         /*
2564          * Check one last time for pending exceptions and ceded state after
2565          * we put ourselves on the wait queue
2566          */
2567         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
2568                 if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded) {
2569                         do_sleep = 0;
2570                         break;
2571                 }
2572         }
2573
2574         if (!do_sleep) {
2575                 finish_wait(&vc->wq, &wait);
2576                 return;
2577         }
2578
2579         vc->vcore_state = VCORE_SLEEPING;
2580         trace_kvmppc_vcore_blocked(vc, 0);
2581         spin_unlock(&vc->lock);
2582         schedule();
2583         finish_wait(&vc->wq, &wait);
2584         spin_lock(&vc->lock);
2585         vc->vcore_state = VCORE_INACTIVE;
2586         trace_kvmppc_vcore_blocked(vc, 1);
2587 }
2588
2589 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2590 {
2591         int n_ceded;
2592         struct kvmppc_vcore *vc;
2593         struct kvm_vcpu *v, *vn;
2594
2595         trace_kvmppc_run_vcpu_enter(vcpu);
2596
2597         kvm_run->exit_reason = 0;
2598         vcpu->arch.ret = RESUME_GUEST;
2599         vcpu->arch.trap = 0;
2600         kvmppc_update_vpas(vcpu);
2601
2602         /*
2603          * Synchronize with other threads in this virtual core
2604          */
2605         vc = vcpu->arch.vcore;
2606         spin_lock(&vc->lock);
2607         vcpu->arch.ceded = 0;
2608         vcpu->arch.run_task = current;
2609         vcpu->arch.kvm_run = kvm_run;
2610         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2611         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2612         vcpu->arch.busy_preempt = TB_NIL;
2613         list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
2614         ++vc->n_runnable;
2615
2616         /*
2617          * This happens the first time this is called for a vcpu.
2618          * If the vcore is already running, we may be able to start
2619          * this thread straight away and have it join in.
2620          */
2621         if (!signal_pending(current)) {
2622                 if (vc->vcore_state == VCORE_PIGGYBACK) {
2623                         struct kvmppc_vcore *mvc = vc->master_vcore;
2624                         if (spin_trylock(&mvc->lock)) {
2625                                 if (mvc->vcore_state == VCORE_RUNNING &&
2626                                     !VCORE_IS_EXITING(mvc)) {
2627                                         kvmppc_create_dtl_entry(vcpu, vc);
2628                                         kvmppc_start_thread(vcpu, vc);
2629                                         trace_kvm_guest_enter(vcpu);
2630                                 }
2631                                 spin_unlock(&mvc->lock);
2632                         }
2633                 } else if (vc->vcore_state == VCORE_RUNNING &&
2634                            !VCORE_IS_EXITING(vc)) {
2635                         kvmppc_create_dtl_entry(vcpu, vc);
2636                         kvmppc_start_thread(vcpu, vc);
2637                         trace_kvm_guest_enter(vcpu);
2638                 } else if (vc->vcore_state == VCORE_SLEEPING) {
2639                         wake_up(&vc->wq);
2640                 }
2641
2642         }
2643
2644         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2645                !signal_pending(current)) {
2646                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2647                         kvmppc_vcore_end_preempt(vc);
2648
2649                 if (vc->vcore_state != VCORE_INACTIVE) {
2650                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
2651                         continue;
2652                 }
2653                 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
2654                                          arch.run_list) {
2655                         kvmppc_core_prepare_to_enter(v);
2656                         if (signal_pending(v->arch.run_task)) {
2657                                 kvmppc_remove_runnable(vc, v);
2658                                 v->stat.signal_exits++;
2659                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2660                                 v->arch.ret = -EINTR;
2661                                 wake_up(&v->arch.cpu_run);
2662                         }
2663                 }
2664                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2665                         break;
2666                 n_ceded = 0;
2667                 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
2668                         if (!v->arch.pending_exceptions)
2669                                 n_ceded += v->arch.ceded;
2670                         else
2671                                 v->arch.ceded = 0;
2672                 }
2673                 vc->runner = vcpu;
2674                 if (n_ceded == vc->n_runnable) {
2675                         kvmppc_vcore_blocked(vc);
2676                 } else if (need_resched()) {
2677                         kvmppc_vcore_preempt(vc);
2678                         /* Let something else run */
2679                         cond_resched_lock(&vc->lock);
2680                         if (vc->vcore_state == VCORE_PREEMPT)
2681                                 kvmppc_vcore_end_preempt(vc);
2682                 } else {
2683                         kvmppc_run_core(vc);
2684                 }
2685                 vc->runner = NULL;
2686         }
2687
2688         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2689                (vc->vcore_state == VCORE_RUNNING ||
2690                 vc->vcore_state == VCORE_EXITING ||
2691                 vc->vcore_state == VCORE_PIGGYBACK))
2692                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
2693
2694         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2695                 kvmppc_vcore_end_preempt(vc);
2696
2697         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2698                 kvmppc_remove_runnable(vc, vcpu);
2699                 vcpu->stat.signal_exits++;
2700                 kvm_run->exit_reason = KVM_EXIT_INTR;
2701                 vcpu->arch.ret = -EINTR;
2702         }
2703
2704         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2705                 /* Wake up some vcpu to run the core */
2706                 v = list_first_entry(&vc->runnable_threads,
2707                                      struct kvm_vcpu, arch.run_list);
2708                 wake_up(&v->arch.cpu_run);
2709         }
2710
2711         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2712         spin_unlock(&vc->lock);
2713         return vcpu->arch.ret;
2714 }
2715
2716 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2717 {
2718         int r;
2719         int srcu_idx;
2720
2721         if (!vcpu->arch.sane) {
2722                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2723                 return -EINVAL;
2724         }
2725
2726         kvmppc_core_prepare_to_enter(vcpu);
2727
2728         /* No need to go into the guest when all we'll do is come back out */
2729         if (signal_pending(current)) {
2730                 run->exit_reason = KVM_EXIT_INTR;
2731                 return -EINTR;
2732         }
2733
2734         atomic_inc(&vcpu->kvm->arch.vcpus_running);
2735         /* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2736         smp_mb();
2737
2738         /* On the first time here, set up HTAB and VRMA */
2739         if (!vcpu->kvm->arch.hpte_setup_done) {
2740                 r = kvmppc_hv_setup_htab_rma(vcpu);
2741                 if (r)
2742                         goto out;
2743         }
2744
2745         flush_all_to_thread(current);
2746
2747         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2748         vcpu->arch.pgdir = current->mm->pgd;
2749         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2750
2751         do {
2752                 r = kvmppc_run_vcpu(run, vcpu);
2753
2754                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2755                     !(vcpu->arch.shregs.msr & MSR_PR)) {
2756                         trace_kvm_hcall_enter(vcpu);
2757                         r = kvmppc_pseries_do_hcall(vcpu);
2758                         trace_kvm_hcall_exit(vcpu, r);
2759                         kvmppc_core_prepare_to_enter(vcpu);
2760                 } else if (r == RESUME_PAGE_FAULT) {
2761                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2762                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
2763                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2764                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2765                 }
2766         } while (is_kvmppc_resume_guest(r));
2767
2768  out:
2769         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2770         atomic_dec(&vcpu->kvm->arch.vcpus_running);
2771         return r;
2772 }
2773
2774 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2775                                      int linux_psize)
2776 {
2777         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2778
2779         if (!def->shift)
2780                 return;
2781         (*sps)->page_shift = def->shift;
2782         (*sps)->slb_enc = def->sllp;
2783         (*sps)->enc[0].page_shift = def->shift;
2784         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
2785         /*
2786          * Add 16MB MPSS support if host supports it
2787          */
2788         if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2789                 (*sps)->enc[1].page_shift = 24;
2790                 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2791         }
2792         (*sps)++;
2793 }
2794
2795 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2796                                          struct kvm_ppc_smmu_info *info)
2797 {
2798         struct kvm_ppc_one_seg_page_size *sps;
2799
2800         info->flags = KVM_PPC_PAGE_SIZES_REAL;
2801         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2802                 info->flags |= KVM_PPC_1T_SEGMENTS;
2803         info->slb_size = mmu_slb_size;
2804
2805         /* We only support these sizes for now, and no muti-size segments */
2806         sps = &info->sps[0];
2807         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2808         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2809         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2810
2811         return 0;
2812 }
2813
2814 /*
2815  * Get (and clear) the dirty memory log for a memory slot.
2816  */
2817 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2818                                          struct kvm_dirty_log *log)
2819 {
2820         struct kvm_memslots *slots;
2821         struct kvm_memory_slot *memslot;
2822         int r;
2823         unsigned long n;
2824
2825         mutex_lock(&kvm->slots_lock);
2826
2827         r = -EINVAL;
2828         if (log->slot >= KVM_USER_MEM_SLOTS)
2829                 goto out;
2830
2831         slots = kvm_memslots(kvm);
2832         memslot = id_to_memslot(slots, log->slot);
2833         r = -ENOENT;
2834         if (!memslot->dirty_bitmap)
2835                 goto out;
2836
2837         n = kvm_dirty_bitmap_bytes(memslot);
2838         memset(memslot->dirty_bitmap, 0, n);
2839
2840         r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2841         if (r)
2842                 goto out;
2843
2844         r = -EFAULT;
2845         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2846                 goto out;
2847
2848         r = 0;
2849 out:
2850         mutex_unlock(&kvm->slots_lock);
2851         return r;
2852 }
2853
2854 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2855                                         struct kvm_memory_slot *dont)
2856 {
2857         if (!dont || free->arch.rmap != dont->arch.rmap) {
2858                 vfree(free->arch.rmap);
2859                 free->arch.rmap = NULL;
2860         }
2861 }
2862
2863 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2864                                          unsigned long npages)
2865 {
2866         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2867         if (!slot->arch.rmap)
2868                 return -ENOMEM;
2869
2870         return 0;
2871 }
2872
2873 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2874                                         struct kvm_memory_slot *memslot,
2875                                         const struct kvm_userspace_memory_region *mem)
2876 {
2877         return 0;
2878 }
2879
2880 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2881                                 const struct kvm_userspace_memory_region *mem,
2882                                 const struct kvm_memory_slot *old,
2883                                 const struct kvm_memory_slot *new)
2884 {
2885         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2886         struct kvm_memslots *slots;
2887         struct kvm_memory_slot *memslot;
2888
2889         if (npages && old->npages) {
2890                 /*
2891                  * If modifying a memslot, reset all the rmap dirty bits.
2892                  * If this is a new memslot, we don't need to do anything
2893                  * since the rmap array starts out as all zeroes,
2894                  * i.e. no pages are dirty.
2895                  */
2896                 slots = kvm_memslots(kvm);
2897                 memslot = id_to_memslot(slots, mem->slot);
2898                 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2899         }
2900 }
2901
2902 /*
2903  * Update LPCR values in kvm->arch and in vcores.
2904  * Caller must hold kvm->lock.
2905  */
2906 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2907 {
2908         long int i;
2909         u32 cores_done = 0;
2910
2911         if ((kvm->arch.lpcr & mask) == lpcr)
2912                 return;
2913
2914         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2915
2916         for (i = 0; i < KVM_MAX_VCORES; ++i) {
2917                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2918                 if (!vc)
2919                         continue;
2920                 spin_lock(&vc->lock);
2921                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2922                 spin_unlock(&vc->lock);
2923                 if (++cores_done >= kvm->arch.online_vcores)
2924                         break;
2925         }
2926 }
2927
2928 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2929 {
2930         return;
2931 }
2932
2933 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2934 {
2935         int err = 0;
2936         struct kvm *kvm = vcpu->kvm;
2937         unsigned long hva;
2938         struct kvm_memory_slot *memslot;
2939         struct vm_area_struct *vma;
2940         unsigned long lpcr = 0, senc;
2941         unsigned long psize, porder;
2942         int srcu_idx;
2943
2944         mutex_lock(&kvm->lock);
2945         if (kvm->arch.hpte_setup_done)
2946                 goto out;       /* another vcpu beat us to it */
2947
2948         /* Allocate hashed page table (if not done already) and reset it */
2949         if (!kvm->arch.hpt_virt) {
2950                 err = kvmppc_alloc_hpt(kvm, NULL);
2951                 if (err) {
2952                         pr_err("KVM: Couldn't alloc HPT\n");
2953                         goto out;
2954                 }
2955         }
2956
2957         /* Look up the memslot for guest physical address 0 */
2958         srcu_idx = srcu_read_lock(&kvm->srcu);
2959         memslot = gfn_to_memslot(kvm, 0);
2960
2961         /* We must have some memory at 0 by now */
2962         err = -EINVAL;
2963         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2964                 goto out_srcu;
2965
2966         /* Look up the VMA for the start of this memory slot */
2967         hva = memslot->userspace_addr;
2968         down_read(&current->mm->mmap_sem);
2969         vma = find_vma(current->mm, hva);
2970         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2971                 goto up_out;
2972
2973         psize = vma_kernel_pagesize(vma);
2974         porder = __ilog2(psize);
2975
2976         up_read(&current->mm->mmap_sem);
2977
2978         /* We can handle 4k, 64k or 16M pages in the VRMA */
2979         err = -EINVAL;
2980         if (!(psize == 0x1000 || psize == 0x10000 ||
2981               psize == 0x1000000))
2982                 goto out_srcu;
2983
2984         /* Update VRMASD field in the LPCR */
2985         senc = slb_pgsize_encoding(psize);
2986         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2987                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
2988         /* the -4 is to account for senc values starting at 0x10 */
2989         lpcr = senc << (LPCR_VRMASD_SH - 4);
2990
2991         /* Create HPTEs in the hash page table for the VRMA */
2992         kvmppc_map_vrma(vcpu, memslot, porder);
2993
2994         kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
2995
2996         /* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
2997         smp_wmb();
2998         kvm->arch.hpte_setup_done = 1;
2999         err = 0;
3000  out_srcu:
3001         srcu_read_unlock(&kvm->srcu, srcu_idx);
3002  out:
3003         mutex_unlock(&kvm->lock);
3004         return err;
3005
3006  up_out:
3007         up_read(&current->mm->mmap_sem);
3008         goto out_srcu;
3009 }
3010
3011 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3012 {
3013         unsigned long lpcr, lpid;
3014         char buf[32];
3015
3016         /* Allocate the guest's logical partition ID */
3017
3018         lpid = kvmppc_alloc_lpid();
3019         if ((long)lpid < 0)
3020                 return -ENOMEM;
3021         kvm->arch.lpid = lpid;
3022
3023         /*
3024          * Since we don't flush the TLB when tearing down a VM,
3025          * and this lpid might have previously been used,
3026          * make sure we flush on each core before running the new VM.
3027          */
3028         cpumask_setall(&kvm->arch.need_tlb_flush);
3029
3030         /* Start out with the default set of hcalls enabled */
3031         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3032                sizeof(kvm->arch.enabled_hcalls));
3033
3034         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3035
3036         /* Init LPCR for virtual RMA mode */
3037         kvm->arch.host_lpid = mfspr(SPRN_LPID);
3038         kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3039         lpcr &= LPCR_PECE | LPCR_LPES;
3040         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3041                 LPCR_VPM0 | LPCR_VPM1;
3042         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3043                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3044         /* On POWER8 turn on online bit to enable PURR/SPURR */
3045         if (cpu_has_feature(CPU_FTR_ARCH_207S))
3046                 lpcr |= LPCR_ONL;
3047         kvm->arch.lpcr = lpcr;
3048
3049         /*
3050          * Track that we now have a HV mode VM active. This blocks secondary
3051          * CPU threads from coming online.
3052          */
3053         kvm_hv_vm_activated();
3054
3055         /*
3056          * Create a debugfs directory for the VM
3057          */
3058         snprintf(buf, sizeof(buf), "vm%d", current->pid);
3059         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3060         if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3061                 kvmppc_mmu_debugfs_init(kvm);
3062
3063         return 0;
3064 }
3065
3066 static void kvmppc_free_vcores(struct kvm *kvm)
3067 {
3068         long int i;
3069
3070         for (i = 0; i < KVM_MAX_VCORES; ++i)
3071                 kfree(kvm->arch.vcores[i]);
3072         kvm->arch.online_vcores = 0;
3073 }
3074
3075 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3076 {
3077         debugfs_remove_recursive(kvm->arch.debugfs_dir);
3078
3079         kvm_hv_vm_deactivated();
3080
3081         kvmppc_free_vcores(kvm);
3082
3083         kvmppc_free_hpt(kvm);
3084 }
3085
3086 /* We don't need to emulate any privileged instructions or dcbz */
3087 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3088                                      unsigned int inst, int *advance)
3089 {
3090         return EMULATE_FAIL;
3091 }
3092
3093 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3094                                         ulong spr_val)
3095 {
3096         return EMULATE_FAIL;
3097 }
3098
3099 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3100                                         ulong *spr_val)
3101 {
3102         return EMULATE_FAIL;
3103 }
3104
3105 static int kvmppc_core_check_processor_compat_hv(void)
3106 {
3107         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3108             !cpu_has_feature(CPU_FTR_ARCH_206))
3109                 return -EIO;
3110         return 0;
3111 }
3112
3113 static long kvm_arch_vm_ioctl_hv(struct file *filp,
3114                                  unsigned int ioctl, unsigned long arg)
3115 {
3116         struct kvm *kvm __maybe_unused = filp->private_data;
3117         void __user *argp = (void __user *)arg;
3118         long r;
3119
3120         switch (ioctl) {
3121
3122         case KVM_PPC_ALLOCATE_HTAB: {
3123                 u32 htab_order;
3124
3125                 r = -EFAULT;
3126                 if (get_user(htab_order, (u32 __user *)argp))
3127                         break;
3128                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
3129                 if (r)
3130                         break;
3131                 r = -EFAULT;
3132                 if (put_user(htab_order, (u32 __user *)argp))
3133                         break;
3134                 r = 0;
3135                 break;
3136         }
3137
3138         case KVM_PPC_GET_HTAB_FD: {
3139                 struct kvm_get_htab_fd ghf;
3140
3141                 r = -EFAULT;
3142                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
3143                         break;
3144                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3145                 break;
3146         }
3147
3148         default:
3149                 r = -ENOTTY;
3150         }
3151
3152         return r;
3153 }
3154
3155 /*
3156  * List of hcall numbers to enable by default.
3157  * For compatibility with old userspace, we enable by default
3158  * all hcalls that were implemented before the hcall-enabling
3159  * facility was added.  Note this list should not include H_RTAS.
3160  */
3161 static unsigned int default_hcall_list[] = {
3162         H_REMOVE,
3163         H_ENTER,
3164         H_READ,
3165         H_PROTECT,
3166         H_BULK_REMOVE,
3167         H_GET_TCE,
3168         H_PUT_TCE,
3169         H_SET_DABR,
3170         H_SET_XDABR,
3171         H_CEDE,
3172         H_PROD,
3173         H_CONFER,
3174         H_REGISTER_VPA,
3175 #ifdef CONFIG_KVM_XICS
3176         H_EOI,
3177         H_CPPR,
3178         H_IPI,
3179         H_IPOLL,
3180         H_XIRR,
3181         H_XIRR_X,
3182 #endif
3183         0
3184 };
3185
3186 static void init_default_hcalls(void)
3187 {
3188         int i;
3189         unsigned int hcall;
3190
3191         for (i = 0; default_hcall_list[i]; ++i) {
3192                 hcall = default_hcall_list[i];
3193                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3194                 __set_bit(hcall / 4, default_enabled_hcalls);
3195         }
3196 }
3197
3198 static struct kvmppc_ops kvm_ops_hv = {
3199         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3200         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3201         .get_one_reg = kvmppc_get_one_reg_hv,
3202         .set_one_reg = kvmppc_set_one_reg_hv,
3203         .vcpu_load   = kvmppc_core_vcpu_load_hv,
3204         .vcpu_put    = kvmppc_core_vcpu_put_hv,
3205         .set_msr     = kvmppc_set_msr_hv,
3206         .vcpu_run    = kvmppc_vcpu_run_hv,
3207         .vcpu_create = kvmppc_core_vcpu_create_hv,
3208         .vcpu_free   = kvmppc_core_vcpu_free_hv,
3209         .check_requests = kvmppc_core_check_requests_hv,
3210         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
3211         .flush_memslot  = kvmppc_core_flush_memslot_hv,
3212         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3213         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
3214         .unmap_hva = kvm_unmap_hva_hv,
3215         .unmap_hva_range = kvm_unmap_hva_range_hv,
3216         .age_hva  = kvm_age_hva_hv,
3217         .test_age_hva = kvm_test_age_hva_hv,
3218         .set_spte_hva = kvm_set_spte_hva_hv,
3219         .mmu_destroy  = kvmppc_mmu_destroy_hv,
3220         .free_memslot = kvmppc_core_free_memslot_hv,
3221         .create_memslot = kvmppc_core_create_memslot_hv,
3222         .init_vm =  kvmppc_core_init_vm_hv,
3223         .destroy_vm = kvmppc_core_destroy_vm_hv,
3224         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3225         .emulate_op = kvmppc_core_emulate_op_hv,
3226         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3227         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3228         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3229         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
3230         .hcall_implemented = kvmppc_hcall_impl_hv,
3231 };
3232
3233 static int kvmppc_book3s_init_hv(void)
3234 {
3235         int r;
3236         /*
3237          * FIXME!! Do we need to check on all cpus ?
3238          */
3239         r = kvmppc_core_check_processor_compat_hv();
3240         if (r < 0)
3241                 return -ENODEV;
3242
3243         kvm_ops_hv.owner = THIS_MODULE;
3244         kvmppc_hv_ops = &kvm_ops_hv;
3245
3246         init_default_hcalls();
3247
3248         init_vcore_lists();
3249
3250         r = kvmppc_mmu_hv_init();
3251         return r;
3252 }
3253
3254 static void kvmppc_book3s_exit_hv(void)
3255 {
3256         kvmppc_hv_ops = NULL;
3257 }
3258
3259 module_init(kvmppc_book3s_init_hv);
3260 module_exit(kvmppc_book3s_exit_hv);
3261 MODULE_LICENSE("GPL");
3262 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3263 MODULE_ALIAS("devname:kvm");