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 #ifdef CONFIG_KVM_XICS
  85 static struct kernel_param_ops module_param_ops = {
  86         .set = param_set_int,
  87         .get = param_get_int,
  88 };
  89
  90 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect,
  91                                                         S_IRUGO | S_IWUSR);
  92 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
  93 #endif
  94
  95 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
  96 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
  97
  98 static bool kvmppc_ipi_thread(int cpu)
  99 {
 100         /* On POWER8 for IPIs to threads in the same core, use msgsnd */
 101         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
 102                 preempt_disable();
 103                 if (cpu_first_thread_sibling(cpu) ==
 104                     cpu_first_thread_sibling(smp_processor_id())) {
 105                         unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
 106                         msg |= cpu_thread_in_core(cpu);
 107                         smp_mb();
 108                         __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 109                         preempt_enable();
 110                         return true;
 111                 }
 112                 preempt_enable();
 113         }
 114
 115 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
 116         if (cpu >= 0 && cpu < nr_cpu_ids && paca[cpu].kvm_hstate.xics_phys) {
 117                 xics_wake_cpu(cpu);
 118                 return true;
 119         }
 120 #endif
 121
 122         return false;
 123 }
 124
 125 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
 126 {
 127         int cpu;
 128         struct swait_queue_head *wqp;
 129
 130         wqp = kvm_arch_vcpu_wq(vcpu);
 131         if (swait_active(wqp)) {
 132                 swake_up(wqp);
 133                 ++vcpu->stat.halt_wakeup;
 134         }
 135
 136         if (kvmppc_ipi_thread(vcpu->arch.thread_cpu))
 137                 return;
 138
 139         /* CPU points to the first thread of the core */
 140         cpu = vcpu->cpu;
 141         if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
 142                 smp_send_reschedule(cpu);
 143 }
 144
 145 /*
 146  * We use the vcpu_load/put functions to measure stolen time.
 147  * Stolen time is counted as time when either the vcpu is able to
 148  * run as part of a virtual core, but the task running the vcore
 149  * is preempted or sleeping, or when the vcpu needs something done
 150  * in the kernel by the task running the vcpu, but that task is
 151  * preempted or sleeping.  Those two things have to be counted
 152  * separately, since one of the vcpu tasks will take on the job
 153  * of running the core, and the other vcpu tasks in the vcore will
 154  * sleep waiting for it to do that, but that sleep shouldn't count
 155  * as stolen time.
 156  *
 157  * Hence we accumulate stolen time when the vcpu can run as part of
 158  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 159  * needs its task to do other things in the kernel (for example,
 160  * service a page fault) in busy_stolen.  We don't accumulate
 161  * stolen time for a vcore when it is inactive, or for a vcpu
 162  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 163  * a misnomer; it means that the vcpu task is not executing in
 164  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 165  * the kernel.  We don't have any way of dividing up that time
 166  * between time that the vcpu is genuinely stopped, time that
 167  * the task is actively working on behalf of the vcpu, and time
 168  * that the task is preempted, so we don't count any of it as
 169  * stolen.
 170  *
 171  * Updates to busy_stolen are protected by arch.tbacct_lock;
 172  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
 173  * lock.  The stolen times are measured in units of timebase ticks.
 174  * (Note that the != TB_NIL checks below are purely defensive;
 175  * they should never fail.)
 176  */
 177
 178 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
 179 {
 180         unsigned long flags;
 181
 182         spin_lock_irqsave(&vc->stoltb_lock, flags);
 183         vc->preempt_tb = mftb();
 184         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 185 }
 186
 187 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
 188 {
 189         unsigned long flags;
 190
 191         spin_lock_irqsave(&vc->stoltb_lock, flags);
 192         if (vc->preempt_tb != TB_NIL) {
 193                 vc->stolen_tb += mftb() - vc->preempt_tb;
 194                 vc->preempt_tb = TB_NIL;
 195         }
 196         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 197 }
 198
 199 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 200 {
 201         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 202         unsigned long flags;
 203
 204         /*
 205          * We can test vc->runner without taking the vcore lock,
 206          * because only this task ever sets vc->runner to this
 207          * vcpu, and once it is set to this vcpu, only this task
 208          * ever sets it to NULL.
 209          */
 210         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 211                 kvmppc_core_end_stolen(vc);
 212
 213         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 214         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 215             vcpu->arch.busy_preempt != TB_NIL) {
 216                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 217                 vcpu->arch.busy_preempt = TB_NIL;
 218         }
 219         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 220 }
 221
 222 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 223 {
 224         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 225         unsigned long flags;
 226
 227         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 228                 kvmppc_core_start_stolen(vc);
 229
 230         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 231         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 232                 vcpu->arch.busy_preempt = mftb();
 233         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 234 }
 235
 236 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
 237 {
 238         /*
 239          * Check for illegal transactional state bit combination
 240          * and if we find it, force the TS field to a safe state.
 241          */
 242         if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
 243                 msr &= ~MSR_TS_MASK;
 244         vcpu->arch.shregs.msr = msr;
 245         kvmppc_end_cede(vcpu);
 246 }
 247
 248 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 249 {
 250         vcpu->arch.pvr = pvr;
 251 }
 252
 253 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 254 {
 255         unsigned long pcr = 0;
 256         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 257
 258         if (arch_compat) {
 259                 switch (arch_compat) {
 260                 case PVR_ARCH_205:
 261                         /*
 262                          * If an arch bit is set in PCR, all the defined
 263                          * higher-order arch bits also have to be set.
 264                          */
 265                         pcr = PCR_ARCH_206 | PCR_ARCH_205;
 266                         break;
 267                 case PVR_ARCH_206:
 268                 case PVR_ARCH_206p:
 269                         pcr = PCR_ARCH_206;
 270                         break;
 271                 case PVR_ARCH_207:
 272                         break;
 273                 default:
 274                         return -EINVAL;
 275                 }
 276
 277                 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
 278                         /* POWER7 can't emulate POWER8 */
 279                         if (!(pcr & PCR_ARCH_206))
 280                                 return -EINVAL;
 281                         pcr &= ~PCR_ARCH_206;
 282                 }
 283         }
 284
 285         spin_lock(&vc->lock);
 286         vc->arch_compat = arch_compat;
 287         vc->pcr = pcr;
 288         spin_unlock(&vc->lock);
 289
 290         return 0;
 291 }
 292
 293 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 294 {
 295         int r;
 296
 297         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 298         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 299                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
 300         for (r = 0; r < 16; ++r)
 301                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 302                        r, kvmppc_get_gpr(vcpu, r),
 303                        r+16, kvmppc_get_gpr(vcpu, r+16));
 304         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 305                vcpu->arch.ctr, vcpu->arch.lr);
 306         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 307                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 308         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 309                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 310         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 311                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 312         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 313                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 314         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 315         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 316                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 317         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 318         for (r = 0; r < vcpu->arch.slb_max; ++r)
 319                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 320                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 321         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 322                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 323                vcpu->arch.last_inst);
 324 }
 325
 326 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 327 {
 328         struct kvm_vcpu *ret;
 329
 330         mutex_lock(&kvm->lock);
 331         ret = kvm_get_vcpu_by_id(kvm, id);
 332         mutex_unlock(&kvm->lock);
 333         return ret;
 334 }
 335
 336 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 337 {
 338         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 339         vpa->yield_count = cpu_to_be32(1);
 340 }
 341
 342 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 343                    unsigned long addr, unsigned long len)
 344 {
 345         /* check address is cacheline aligned */
 346         if (addr & (L1_CACHE_BYTES - 1))
 347                 return -EINVAL;
 348         spin_lock(&vcpu->arch.vpa_update_lock);
 349         if (v->next_gpa != addr || v->len != len) {
 350                 v->next_gpa = addr;
 351                 v->len = addr ? len : 0;
 352                 v->update_pending = 1;
 353         }
 354         spin_unlock(&vcpu->arch.vpa_update_lock);
 355         return 0;
 356 }
 357
 358 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 359 struct reg_vpa {
 360         u32 dummy;
 361         union {
 362                 __be16 hword;
 363                 __be32 word;
 364         } length;
 365 };
 366
 367 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 368 {
 369         if (vpap->update_pending)
 370                 return vpap->next_gpa != 0;
 371         return vpap->pinned_addr != NULL;
 372 }
 373
 374 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 375                                        unsigned long flags,
 376                                        unsigned long vcpuid, unsigned long vpa)
 377 {
 378         struct kvm *kvm = vcpu->kvm;
 379         unsigned long len, nb;
 380         void *va;
 381         struct kvm_vcpu *tvcpu;
 382         int err;
 383         int subfunc;
 384         struct kvmppc_vpa *vpap;
 385
 386         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 387         if (!tvcpu)
 388                 return H_PARAMETER;
 389
 390         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 391         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 392             subfunc == H_VPA_REG_SLB) {
 393                 /* Registering new area - address must be cache-line aligned */
 394                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 395                         return H_PARAMETER;
 396
 397                 /* convert logical addr to kernel addr and read length */
 398                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 399                 if (va == NULL)
 400                         return H_PARAMETER;
 401                 if (subfunc == H_VPA_REG_VPA)
 402                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 403                 else
 404                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 405                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 406
 407                 /* Check length */
 408                 if (len > nb || len < sizeof(struct reg_vpa))
 409                         return H_PARAMETER;
 410         } else {
 411                 vpa = 0;
 412                 len = 0;
 413         }
 414
 415         err = H_PARAMETER;
 416         vpap = NULL;
 417         spin_lock(&tvcpu->arch.vpa_update_lock);
 418
 419         switch (subfunc) {
 420         case H_VPA_REG_VPA:             /* register VPA */
 421                 if (len < sizeof(struct lppaca))
 422                         break;
 423                 vpap = &tvcpu->arch.vpa;
 424                 err = 0;
 425                 break;
 426
 427         case H_VPA_REG_DTL:             /* register DTL */
 428                 if (len < sizeof(struct dtl_entry))
 429                         break;
 430                 len -= len % sizeof(struct dtl_entry);
 431
 432                 /* Check that they have previously registered a VPA */
 433                 err = H_RESOURCE;
 434                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 435                         break;
 436
 437                 vpap = &tvcpu->arch.dtl;
 438                 err = 0;
 439                 break;
 440
 441         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 442                 /* Check that they have previously registered a VPA */
 443                 err = H_RESOURCE;
 444                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 445                         break;
 446
 447                 vpap = &tvcpu->arch.slb_shadow;
 448                 err = 0;
 449                 break;
 450
 451         case H_VPA_DEREG_VPA:           /* deregister VPA */
 452                 /* Check they don't still have a DTL or SLB buf registered */
 453                 err = H_RESOURCE;
 454                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 455                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 456                         break;
 457
 458                 vpap = &tvcpu->arch.vpa;
 459                 err = 0;
 460                 break;
 461
 462         case H_VPA_DEREG_DTL:           /* deregister DTL */
 463                 vpap = &tvcpu->arch.dtl;
 464                 err = 0;
 465                 break;
 466
 467         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 468                 vpap = &tvcpu->arch.slb_shadow;
 469                 err = 0;
 470                 break;
 471         }
 472
 473         if (vpap) {
 474                 vpap->next_gpa = vpa;
 475                 vpap->len = len;
 476                 vpap->update_pending = 1;
 477         }
 478
 479         spin_unlock(&tvcpu->arch.vpa_update_lock);
 480
 481         return err;
 482 }
 483
 484 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 485 {
 486         struct kvm *kvm = vcpu->kvm;
 487         void *va;
 488         unsigned long nb;
 489         unsigned long gpa;
 490
 491         /*
 492          * We need to pin the page pointed to by vpap->next_gpa,
 493          * but we can't call kvmppc_pin_guest_page under the lock
 494          * as it does get_user_pages() and down_read().  So we
 495          * have to drop the lock, pin the page, then get the lock
 496          * again and check that a new area didn't get registered
 497          * in the meantime.
 498          */
 499         for (;;) {
 500                 gpa = vpap->next_gpa;
 501                 spin_unlock(&vcpu->arch.vpa_update_lock);
 502                 va = NULL;
 503                 nb = 0;
 504                 if (gpa)
 505                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 506                 spin_lock(&vcpu->arch.vpa_update_lock);
 507                 if (gpa == vpap->next_gpa)
 508                         break;
 509                 /* sigh... unpin that one and try again */
 510                 if (va)
 511                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 512         }
 513
 514         vpap->update_pending = 0;
 515         if (va && nb < vpap->len) {
 516                 /*
 517                  * If it's now too short, it must be that userspace
 518                  * has changed the mappings underlying guest memory,
 519                  * so unregister the region.
 520                  */
 521                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 522                 va = NULL;
 523         }
 524         if (vpap->pinned_addr)
 525                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 526                                         vpap->dirty);
 527         vpap->gpa = gpa;
 528         vpap->pinned_addr = va;
 529         vpap->dirty = false;
 530         if (va)
 531                 vpap->pinned_end = va + vpap->len;
 532 }
 533
 534 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 535 {
 536         if (!(vcpu->arch.vpa.update_pending ||
 537               vcpu->arch.slb_shadow.update_pending ||
 538               vcpu->arch.dtl.update_pending))
 539                 return;
 540
 541         spin_lock(&vcpu->arch.vpa_update_lock);
 542         if (vcpu->arch.vpa.update_pending) {
 543                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 544                 if (vcpu->arch.vpa.pinned_addr)
 545                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 546         }
 547         if (vcpu->arch.dtl.update_pending) {
 548                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 549                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 550                 vcpu->arch.dtl_index = 0;
 551         }
 552         if (vcpu->arch.slb_shadow.update_pending)
 553                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 554         spin_unlock(&vcpu->arch.vpa_update_lock);
 555 }
 556
 557 /*
 558  * Return the accumulated stolen time for the vcore up until `now'.
 559  * The caller should hold the vcore lock.
 560  */
 561 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 562 {
 563         u64 p;
 564         unsigned long flags;
 565
 566         spin_lock_irqsave(&vc->stoltb_lock, flags);
 567         p = vc->stolen_tb;
 568         if (vc->vcore_state != VCORE_INACTIVE &&
 569             vc->preempt_tb != TB_NIL)
 570                 p += now - vc->preempt_tb;
 571         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 572         return p;
 573 }
 574
 575 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 576                                     struct kvmppc_vcore *vc)
 577 {
 578         struct dtl_entry *dt;
 579         struct lppaca *vpa;
 580         unsigned long stolen;
 581         unsigned long core_stolen;
 582         u64 now;
 583
 584         dt = vcpu->arch.dtl_ptr;
 585         vpa = vcpu->arch.vpa.pinned_addr;
 586         now = mftb();
 587         core_stolen = vcore_stolen_time(vc, now);
 588         stolen = core_stolen - vcpu->arch.stolen_logged;
 589         vcpu->arch.stolen_logged = core_stolen;
 590         spin_lock_irq(&vcpu->arch.tbacct_lock);
 591         stolen += vcpu->arch.busy_stolen;
 592         vcpu->arch.busy_stolen = 0;
 593         spin_unlock_irq(&vcpu->arch.tbacct_lock);
 594         if (!dt || !vpa)
 595                 return;
 596         memset(dt, 0, sizeof(struct dtl_entry));
 597         dt->dispatch_reason = 7;
 598         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 599         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 600         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 601         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 602         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 603         ++dt;
 604         if (dt == vcpu->arch.dtl.pinned_end)
 605                 dt = vcpu->arch.dtl.pinned_addr;
 606         vcpu->arch.dtl_ptr = dt;
 607         /* order writing *dt vs. writing vpa->dtl_idx */
 608         smp_wmb();
 609         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 610         vcpu->arch.dtl.dirty = true;
 611 }
 612
 613 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 614 {
 615         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 616                 return true;
 617         if ((!vcpu->arch.vcore->arch_compat) &&
 618             cpu_has_feature(CPU_FTR_ARCH_207S))
 619                 return true;
 620         return false;
 621 }
 622
 623 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 624                              unsigned long resource, unsigned long value1,
 625                              unsigned long value2)
 626 {
 627         switch (resource) {
 628         case H_SET_MODE_RESOURCE_SET_CIABR:
 629                 if (!kvmppc_power8_compatible(vcpu))
 630                         return H_P2;
 631                 if (value2)
 632                         return H_P4;
 633                 if (mflags)
 634                         return H_UNSUPPORTED_FLAG_START;
 635                 /* Guests can't breakpoint the hypervisor */
 636                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 637                         return H_P3;
 638                 vcpu->arch.ciabr  = value1;
 639                 return H_SUCCESS;
 640         case H_SET_MODE_RESOURCE_SET_DAWR:
 641                 if (!kvmppc_power8_compatible(vcpu))
 642                         return H_P2;
 643                 if (mflags)
 644                         return H_UNSUPPORTED_FLAG_START;
 645                 if (value2 & DABRX_HYP)
 646                         return H_P4;
 647                 vcpu->arch.dawr  = value1;
 648                 vcpu->arch.dawrx = value2;
 649                 return H_SUCCESS;
 650         default:
 651                 return H_TOO_HARD;
 652         }
 653 }
 654
 655 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 656 {
 657         struct kvmppc_vcore *vcore = target->arch.vcore;
 658
 659         /*
 660          * We expect to have been called by the real mode handler
 661          * (kvmppc_rm_h_confer()) which would have directly returned
 662          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 663          * have useful work to do and should not confer) so we don't
 664          * recheck that here.
 665          */
 666
 667         spin_lock(&vcore->lock);
 668         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 669             vcore->vcore_state != VCORE_INACTIVE &&
 670             vcore->runner)
 671                 target = vcore->runner;
 672         spin_unlock(&vcore->lock);
 673
 674         return kvm_vcpu_yield_to(target);
 675 }
 676
 677 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 678 {
 679         int yield_count = 0;
 680         struct lppaca *lppaca;
 681
 682         spin_lock(&vcpu->arch.vpa_update_lock);
 683         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 684         if (lppaca)
 685                 yield_count = be32_to_cpu(lppaca->yield_count);
 686         spin_unlock(&vcpu->arch.vpa_update_lock);
 687         return yield_count;
 688 }
 689
 690 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 691 {
 692         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 693         unsigned long target, ret = H_SUCCESS;
 694         int yield_count;
 695         struct kvm_vcpu *tvcpu;
 696         int idx, rc;
 697
 698         if (req <= MAX_HCALL_OPCODE &&
 699             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 700                 return RESUME_HOST;
 701
 702         switch (req) {
 703         case H_CEDE:
 704                 break;
 705         case H_PROD:
 706                 target = kvmppc_get_gpr(vcpu, 4);
 707                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 708                 if (!tvcpu) {
 709                         ret = H_PARAMETER;
 710                         break;
 711                 }
 712                 tvcpu->arch.prodded = 1;
 713                 smp_mb();
 714                 if (vcpu->arch.ceded) {
 715                         if (swait_active(&vcpu->wq)) {
 716                                 swake_up(&vcpu->wq);
 717                                 vcpu->stat.halt_wakeup++;
 718                         }
 719                 }
 720                 break;
 721         case H_CONFER:
 722                 target = kvmppc_get_gpr(vcpu, 4);
 723                 if (target == -1)
 724                         break;
 725                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 726                 if (!tvcpu) {
 727                         ret = H_PARAMETER;
 728                         break;
 729                 }
 730                 yield_count = kvmppc_get_gpr(vcpu, 5);
 731                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 732                         break;
 733                 kvm_arch_vcpu_yield_to(tvcpu);
 734                 break;
 735         case H_REGISTER_VPA:
 736                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 737                                         kvmppc_get_gpr(vcpu, 5),
 738                                         kvmppc_get_gpr(vcpu, 6));
 739                 break;
 740         case H_RTAS:
 741                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 742                         return RESUME_HOST;
 743
 744                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 745                 rc = kvmppc_rtas_hcall(vcpu);
 746                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 747
 748                 if (rc == -ENOENT)
 749                         return RESUME_HOST;
 750                 else if (rc == 0)
 751                         break;
 752
 753                 /* Send the error out to userspace via KVM_RUN */
 754                 return rc;
 755         case H_LOGICAL_CI_LOAD:
 756                 ret = kvmppc_h_logical_ci_load(vcpu);
 757                 if (ret == H_TOO_HARD)
 758                         return RESUME_HOST;
 759                 break;
 760         case H_LOGICAL_CI_STORE:
 761                 ret = kvmppc_h_logical_ci_store(vcpu);
 762                 if (ret == H_TOO_HARD)
 763                         return RESUME_HOST;
 764                 break;
 765         case H_SET_MODE:
 766                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 767                                         kvmppc_get_gpr(vcpu, 5),
 768                                         kvmppc_get_gpr(vcpu, 6),
 769                                         kvmppc_get_gpr(vcpu, 7));
 770                 if (ret == H_TOO_HARD)
 771                         return RESUME_HOST;
 772                 break;
 773         case H_XIRR:
 774         case H_CPPR:
 775         case H_EOI:
 776         case H_IPI:
 777         case H_IPOLL:
 778         case H_XIRR_X:
 779                 if (kvmppc_xics_enabled(vcpu)) {
 780                         ret = kvmppc_xics_hcall(vcpu, req);
 781                         break;
 782                 }
 783                 return RESUME_HOST;
 784         case H_PUT_TCE:
 785                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 786                                                 kvmppc_get_gpr(vcpu, 5),
 787                                                 kvmppc_get_gpr(vcpu, 6));
 788                 if (ret == H_TOO_HARD)
 789                         return RESUME_HOST;
 790                 break;
 791         case H_PUT_TCE_INDIRECT:
 792                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
 793                                                 kvmppc_get_gpr(vcpu, 5),
 794                                                 kvmppc_get_gpr(vcpu, 6),
 795                                                 kvmppc_get_gpr(vcpu, 7));
 796                 if (ret == H_TOO_HARD)
 797                         return RESUME_HOST;
 798                 break;
 799         case H_STUFF_TCE:
 800                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 801                                                 kvmppc_get_gpr(vcpu, 5),
 802                                                 kvmppc_get_gpr(vcpu, 6),
 803                                                 kvmppc_get_gpr(vcpu, 7));
 804                 if (ret == H_TOO_HARD)
 805                         return RESUME_HOST;
 806                 break;
 807         default:
 808                 return RESUME_HOST;
 809         }
 810         kvmppc_set_gpr(vcpu, 3, ret);
 811         vcpu->arch.hcall_needed = 0;
 812         return RESUME_GUEST;
 813 }
 814
 815 static int kvmppc_hcall_impl_hv(unsigned long cmd)
 816 {
 817         switch (cmd) {
 818         case H_CEDE:
 819         case H_PROD:
 820         case H_CONFER:
 821         case H_REGISTER_VPA:
 822         case H_SET_MODE:
 823         case H_LOGICAL_CI_LOAD:
 824         case H_LOGICAL_CI_STORE:
 825 #ifdef CONFIG_KVM_XICS
 826         case H_XIRR:
 827         case H_CPPR:
 828         case H_EOI:
 829         case H_IPI:
 830         case H_IPOLL:
 831         case H_XIRR_X:
 832 #endif
 833                 return 1;
 834         }
 835
 836         /* See if it's in the real-mode table */
 837         return kvmppc_hcall_impl_hv_realmode(cmd);
 838 }
 839
 840 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
 841                                         struct kvm_vcpu *vcpu)
 842 {
 843         u32 last_inst;
 844
 845         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
 846                                         EMULATE_DONE) {
 847                 /*
 848                  * Fetch failed, so return to guest and
 849                  * try executing it again.
 850                  */
 851                 return RESUME_GUEST;
 852         }
 853
 854         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
 855                 run->exit_reason = KVM_EXIT_DEBUG;
 856                 run->debug.arch.address = kvmppc_get_pc(vcpu);
 857                 return RESUME_HOST;
 858         } else {
 859                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 860                 return RESUME_GUEST;
 861         }
 862 }
 863
 864 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
 865                                  struct task_struct *tsk)
 866 {
 867         int r = RESUME_HOST;
 868
 869         vcpu->stat.sum_exits++;
 870
 871         /*
 872          * This can happen if an interrupt occurs in the last stages
 873          * of guest entry or the first stages of guest exit (i.e. after
 874          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
 875          * and before setting it to KVM_GUEST_MODE_HOST_HV).
 876          * That can happen due to a bug, or due to a machine check
 877          * occurring at just the wrong time.
 878          */
 879         if (vcpu->arch.shregs.msr & MSR_HV) {
 880                 printk(KERN_EMERG "KVM trap in HV mode!\n");
 881                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 882                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 883                         vcpu->arch.shregs.msr);
 884                 kvmppc_dump_regs(vcpu);
 885                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
 886                 run->hw.hardware_exit_reason = vcpu->arch.trap;
 887                 return RESUME_HOST;
 888         }
 889         run->exit_reason = KVM_EXIT_UNKNOWN;
 890         run->ready_for_interrupt_injection = 1;
 891         switch (vcpu->arch.trap) {
 892         /* We're good on these - the host merely wanted to get our attention */
 893         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 894                 vcpu->stat.dec_exits++;
 895                 r = RESUME_GUEST;
 896                 break;
 897         case BOOK3S_INTERRUPT_EXTERNAL:
 898         case BOOK3S_INTERRUPT_H_DOORBELL:
 899                 vcpu->stat.ext_intr_exits++;
 900                 r = RESUME_GUEST;
 901                 break;
 902         /* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
 903         case BOOK3S_INTERRUPT_HMI:
 904         case BOOK3S_INTERRUPT_PERFMON:
 905                 r = RESUME_GUEST;
 906                 break;
 907         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 908                 /*
 909                  * Deliver a machine check interrupt to the guest.
 910                  * We have to do this, even if the host has handled the
 911                  * machine check, because machine checks use SRR0/1 and
 912                  * the interrupt might have trashed guest state in them.
 913                  */
 914                 kvmppc_book3s_queue_irqprio(vcpu,
 915                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 916                 r = RESUME_GUEST;
 917                 break;
 918         case BOOK3S_INTERRUPT_PROGRAM:
 919         {
 920                 ulong flags;
 921                 /*
 922                  * Normally program interrupts are delivered directly
 923                  * to the guest by the hardware, but we can get here
 924                  * as a result of a hypervisor emulation interrupt
 925                  * (e40) getting turned into a 700 by BML RTAS.
 926                  */
 927                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 928                 kvmppc_core_queue_program(vcpu, flags);
 929                 r = RESUME_GUEST;
 930                 break;
 931         }
 932         case BOOK3S_INTERRUPT_SYSCALL:
 933         {
 934                 /* hcall - punt to userspace */
 935                 int i;
 936
 937                 /* hypercall with MSR_PR has already been handled in rmode,
 938                  * and never reaches here.
 939                  */
 940
 941                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 942                 for (i = 0; i < 9; ++i)
 943                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 944                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 945                 vcpu->arch.hcall_needed = 1;
 946                 r = RESUME_HOST;
 947                 break;
 948         }
 949         /*
 950          * We get these next two if the guest accesses a page which it thinks
 951          * it has mapped but which is not actually present, either because
 952          * it is for an emulated I/O device or because the corresonding
 953          * host page has been paged out.  Any other HDSI/HISI interrupts
 954          * have been handled already.
 955          */
 956         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 957                 r = RESUME_PAGE_FAULT;
 958                 break;
 959         case BOOK3S_INTERRUPT_H_INST_STORAGE:
 960                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
 961                 vcpu->arch.fault_dsisr = 0;
 962                 r = RESUME_PAGE_FAULT;
 963                 break;
 964         /*
 965          * This occurs if the guest executes an illegal instruction.
 966          * If the guest debug is disabled, generate a program interrupt
 967          * to the guest. If guest debug is enabled, we need to check
 968          * whether the instruction is a software breakpoint instruction.
 969          * Accordingly return to Guest or Host.
 970          */
 971         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 972                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
 973                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
 974                                 swab32(vcpu->arch.emul_inst) :
 975                                 vcpu->arch.emul_inst;
 976                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
 977                         r = kvmppc_emulate_debug_inst(run, vcpu);
 978                 } else {
 979                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 980                         r = RESUME_GUEST;
 981                 }
 982                 break;
 983         /*
 984          * This occurs if the guest (kernel or userspace), does something that
 985          * is prohibited by HFSCR.  We just generate a program interrupt to
 986          * the guest.
 987          */
 988         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
 989                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 990                 r = RESUME_GUEST;
 991                 break;
 992         default:
 993                 kvmppc_dump_regs(vcpu);
 994                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 995                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 996                         vcpu->arch.shregs.msr);
 997                 run->hw.hardware_exit_reason = vcpu->arch.trap;
 998                 r = RESUME_HOST;
 999                 break;
1000         }
1001
1002         return r;
1003 }
1004
1005 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1006                                             struct kvm_sregs *sregs)
1007 {
1008         int i;
1009
1010         memset(sregs, 0, sizeof(struct kvm_sregs));
1011         sregs->pvr = vcpu->arch.pvr;
1012         for (i = 0; i < vcpu->arch.slb_max; i++) {
1013                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1014                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1015         }
1016
1017         return 0;
1018 }
1019
1020 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1021                                             struct kvm_sregs *sregs)
1022 {
1023         int i, j;
1024
1025         /* Only accept the same PVR as the host's, since we can't spoof it */
1026         if (sregs->pvr != vcpu->arch.pvr)
1027                 return -EINVAL;
1028
1029         j = 0;
1030         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1031                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1032                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1033                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1034                         ++j;
1035                 }
1036         }
1037         vcpu->arch.slb_max = j;
1038
1039         return 0;
1040 }
1041
1042 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1043                 bool preserve_top32)
1044 {
1045         struct kvm *kvm = vcpu->kvm;
1046         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1047         u64 mask;
1048
1049         mutex_lock(&kvm->lock);
1050         spin_lock(&vc->lock);
1051         /*
1052          * If ILE (interrupt little-endian) has changed, update the
1053          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1054          */
1055         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1056                 struct kvm_vcpu *vcpu;
1057                 int i;
1058
1059                 kvm_for_each_vcpu(i, vcpu, kvm) {
1060                         if (vcpu->arch.vcore != vc)
1061                                 continue;
1062                         if (new_lpcr & LPCR_ILE)
1063                                 vcpu->arch.intr_msr |= MSR_LE;
1064                         else
1065                                 vcpu->arch.intr_msr &= ~MSR_LE;
1066                 }
1067         }
1068
1069         /*
1070          * Userspace can only modify DPFD (default prefetch depth),
1071          * ILE (interrupt little-endian) and TC (translation control).
1072          * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
1073          */
1074         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1075         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1076                 mask |= LPCR_AIL;
1077
1078         /* Broken 32-bit version of LPCR must not clear top bits */
1079         if (preserve_top32)
1080                 mask &= 0xFFFFFFFF;
1081         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1082         spin_unlock(&vc->lock);
1083         mutex_unlock(&kvm->lock);
1084 }
1085
1086 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1087                                  union kvmppc_one_reg *val)
1088 {
1089         int r = 0;
1090         long int i;
1091
1092         switch (id) {
1093         case KVM_REG_PPC_DEBUG_INST:
1094                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1095                 break;
1096         case KVM_REG_PPC_HIOR:
1097                 *val = get_reg_val(id, 0);
1098                 break;
1099         case KVM_REG_PPC_DABR:
1100                 *val = get_reg_val(id, vcpu->arch.dabr);
1101                 break;
1102         case KVM_REG_PPC_DABRX:
1103                 *val = get_reg_val(id, vcpu->arch.dabrx);
1104                 break;
1105         case KVM_REG_PPC_DSCR:
1106                 *val = get_reg_val(id, vcpu->arch.dscr);
1107                 break;
1108         case KVM_REG_PPC_PURR:
1109                 *val = get_reg_val(id, vcpu->arch.purr);
1110                 break;
1111         case KVM_REG_PPC_SPURR:
1112                 *val = get_reg_val(id, vcpu->arch.spurr);
1113                 break;
1114         case KVM_REG_PPC_AMR:
1115                 *val = get_reg_val(id, vcpu->arch.amr);
1116                 break;
1117         case KVM_REG_PPC_UAMOR:
1118                 *val = get_reg_val(id, vcpu->arch.uamor);
1119                 break;
1120         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1121                 i = id - KVM_REG_PPC_MMCR0;
1122                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1123                 break;
1124         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1125                 i = id - KVM_REG_PPC_PMC1;
1126                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1127                 break;
1128         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1129                 i = id - KVM_REG_PPC_SPMC1;
1130                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1131                 break;
1132         case KVM_REG_PPC_SIAR:
1133                 *val = get_reg_val(id, vcpu->arch.siar);
1134                 break;
1135         case KVM_REG_PPC_SDAR:
1136                 *val = get_reg_val(id, vcpu->arch.sdar);
1137                 break;
1138         case KVM_REG_PPC_SIER:
1139                 *val = get_reg_val(id, vcpu->arch.sier);
1140                 break;
1141         case KVM_REG_PPC_IAMR:
1142                 *val = get_reg_val(id, vcpu->arch.iamr);
1143                 break;
1144         case KVM_REG_PPC_PSPB:
1145                 *val = get_reg_val(id, vcpu->arch.pspb);
1146                 break;
1147         case KVM_REG_PPC_DPDES:
1148                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1149                 break;
1150         case KVM_REG_PPC_DAWR:
1151                 *val = get_reg_val(id, vcpu->arch.dawr);
1152                 break;
1153         case KVM_REG_PPC_DAWRX:
1154                 *val = get_reg_val(id, vcpu->arch.dawrx);
1155                 break;
1156         case KVM_REG_PPC_CIABR:
1157                 *val = get_reg_val(id, vcpu->arch.ciabr);
1158                 break;
1159         case KVM_REG_PPC_CSIGR:
1160                 *val = get_reg_val(id, vcpu->arch.csigr);
1161                 break;
1162         case KVM_REG_PPC_TACR:
1163                 *val = get_reg_val(id, vcpu->arch.tacr);
1164                 break;
1165         case KVM_REG_PPC_TCSCR:
1166                 *val = get_reg_val(id, vcpu->arch.tcscr);
1167                 break;
1168         case KVM_REG_PPC_PID:
1169                 *val = get_reg_val(id, vcpu->arch.pid);
1170                 break;
1171         case KVM_REG_PPC_ACOP:
1172                 *val = get_reg_val(id, vcpu->arch.acop);
1173                 break;
1174         case KVM_REG_PPC_WORT:
1175                 *val = get_reg_val(id, vcpu->arch.wort);
1176                 break;
1177         case KVM_REG_PPC_VPA_ADDR:
1178                 spin_lock(&vcpu->arch.vpa_update_lock);
1179                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1180                 spin_unlock(&vcpu->arch.vpa_update_lock);
1181                 break;
1182         case KVM_REG_PPC_VPA_SLB:
1183                 spin_lock(&vcpu->arch.vpa_update_lock);
1184                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1185                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1186                 spin_unlock(&vcpu->arch.vpa_update_lock);
1187                 break;
1188         case KVM_REG_PPC_VPA_DTL:
1189                 spin_lock(&vcpu->arch.vpa_update_lock);
1190                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1191                 val->vpaval.length = vcpu->arch.dtl.len;
1192                 spin_unlock(&vcpu->arch.vpa_update_lock);
1193                 break;
1194         case KVM_REG_PPC_TB_OFFSET:
1195                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1196                 break;
1197         case KVM_REG_PPC_LPCR:
1198         case KVM_REG_PPC_LPCR_64:
1199                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1200                 break;
1201         case KVM_REG_PPC_PPR:
1202                 *val = get_reg_val(id, vcpu->arch.ppr);
1203                 break;
1204 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1205         case KVM_REG_PPC_TFHAR:
1206                 *val = get_reg_val(id, vcpu->arch.tfhar);
1207                 break;
1208         case KVM_REG_PPC_TFIAR:
1209                 *val = get_reg_val(id, vcpu->arch.tfiar);
1210                 break;
1211         case KVM_REG_PPC_TEXASR:
1212                 *val = get_reg_val(id, vcpu->arch.texasr);
1213                 break;
1214         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1215                 i = id - KVM_REG_PPC_TM_GPR0;
1216                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1217                 break;
1218         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1219         {
1220                 int j;
1221                 i = id - KVM_REG_PPC_TM_VSR0;
1222                 if (i < 32)
1223                         for (j = 0; j < TS_FPRWIDTH; j++)
1224                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1225                 else {
1226                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1227                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1228                         else
1229                                 r = -ENXIO;
1230                 }
1231                 break;
1232         }
1233         case KVM_REG_PPC_TM_CR:
1234                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1235                 break;
1236         case KVM_REG_PPC_TM_LR:
1237                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1238                 break;
1239         case KVM_REG_PPC_TM_CTR:
1240                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1241                 break;
1242         case KVM_REG_PPC_TM_FPSCR:
1243                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1244                 break;
1245         case KVM_REG_PPC_TM_AMR:
1246                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1247                 break;
1248         case KVM_REG_PPC_TM_PPR:
1249                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1250                 break;
1251         case KVM_REG_PPC_TM_VRSAVE:
1252                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1253                 break;
1254         case KVM_REG_PPC_TM_VSCR:
1255                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1256                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1257                 else
1258                         r = -ENXIO;
1259                 break;
1260         case KVM_REG_PPC_TM_DSCR:
1261                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1262                 break;
1263         case KVM_REG_PPC_TM_TAR:
1264                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1265                 break;
1266 #endif
1267         case KVM_REG_PPC_ARCH_COMPAT:
1268                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1269                 break;
1270         default:
1271                 r = -EINVAL;
1272                 break;
1273         }
1274
1275         return r;
1276 }
1277
1278 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1279                                  union kvmppc_one_reg *val)
1280 {
1281         int r = 0;
1282         long int i;
1283         unsigned long addr, len;
1284
1285         switch (id) {
1286         case KVM_REG_PPC_HIOR:
1287                 /* Only allow this to be set to zero */
1288                 if (set_reg_val(id, *val))
1289                         r = -EINVAL;
1290                 break;
1291         case KVM_REG_PPC_DABR:
1292                 vcpu->arch.dabr = set_reg_val(id, *val);
1293                 break;
1294         case KVM_REG_PPC_DABRX:
1295                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1296                 break;
1297         case KVM_REG_PPC_DSCR:
1298                 vcpu->arch.dscr = set_reg_val(id, *val);
1299                 break;
1300         case KVM_REG_PPC_PURR:
1301                 vcpu->arch.purr = set_reg_val(id, *val);
1302                 break;
1303         case KVM_REG_PPC_SPURR:
1304                 vcpu->arch.spurr = set_reg_val(id, *val);
1305                 break;
1306         case KVM_REG_PPC_AMR:
1307                 vcpu->arch.amr = set_reg_val(id, *val);
1308                 break;
1309         case KVM_REG_PPC_UAMOR:
1310                 vcpu->arch.uamor = set_reg_val(id, *val);
1311                 break;
1312         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1313                 i = id - KVM_REG_PPC_MMCR0;
1314                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1315                 break;
1316         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1317                 i = id - KVM_REG_PPC_PMC1;
1318                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1319                 break;
1320         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1321                 i = id - KVM_REG_PPC_SPMC1;
1322                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1323                 break;
1324         case KVM_REG_PPC_SIAR:
1325                 vcpu->arch.siar = set_reg_val(id, *val);
1326                 break;
1327         case KVM_REG_PPC_SDAR:
1328                 vcpu->arch.sdar = set_reg_val(id, *val);
1329                 break;
1330         case KVM_REG_PPC_SIER:
1331                 vcpu->arch.sier = set_reg_val(id, *val);
1332                 break;
1333         case KVM_REG_PPC_IAMR:
1334                 vcpu->arch.iamr = set_reg_val(id, *val);
1335                 break;
1336         case KVM_REG_PPC_PSPB:
1337                 vcpu->arch.pspb = set_reg_val(id, *val);
1338                 break;
1339         case KVM_REG_PPC_DPDES:
1340                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1341                 break;
1342         case KVM_REG_PPC_DAWR:
1343                 vcpu->arch.dawr = set_reg_val(id, *val);
1344                 break;
1345         case KVM_REG_PPC_DAWRX:
1346                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1347                 break;
1348         case KVM_REG_PPC_CIABR:
1349                 vcpu->arch.ciabr = set_reg_val(id, *val);
1350                 /* Don't allow setting breakpoints in hypervisor code */
1351                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1352                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1353                 break;
1354         case KVM_REG_PPC_CSIGR:
1355                 vcpu->arch.csigr = set_reg_val(id, *val);
1356                 break;
1357         case KVM_REG_PPC_TACR:
1358                 vcpu->arch.tacr = set_reg_val(id, *val);
1359                 break;
1360         case KVM_REG_PPC_TCSCR:
1361                 vcpu->arch.tcscr = set_reg_val(id, *val);
1362                 break;
1363         case KVM_REG_PPC_PID:
1364                 vcpu->arch.pid = set_reg_val(id, *val);
1365                 break;
1366         case KVM_REG_PPC_ACOP:
1367                 vcpu->arch.acop = set_reg_val(id, *val);
1368                 break;
1369         case KVM_REG_PPC_WORT:
1370                 vcpu->arch.wort = set_reg_val(id, *val);
1371                 break;
1372         case KVM_REG_PPC_VPA_ADDR:
1373                 addr = set_reg_val(id, *val);
1374                 r = -EINVAL;
1375                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1376                               vcpu->arch.dtl.next_gpa))
1377                         break;
1378                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1379                 break;
1380         case KVM_REG_PPC_VPA_SLB:
1381                 addr = val->vpaval.addr;
1382                 len = val->vpaval.length;
1383                 r = -EINVAL;
1384                 if (addr && !vcpu->arch.vpa.next_gpa)
1385                         break;
1386                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1387                 break;
1388         case KVM_REG_PPC_VPA_DTL:
1389                 addr = val->vpaval.addr;
1390                 len = val->vpaval.length;
1391                 r = -EINVAL;
1392                 if (addr && (len < sizeof(struct dtl_entry) ||
1393                              !vcpu->arch.vpa.next_gpa))
1394                         break;
1395                 len -= len % sizeof(struct dtl_entry);
1396                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1397                 break;
1398         case KVM_REG_PPC_TB_OFFSET:
1399                 /* round up to multiple of 2^24 */
1400                 vcpu->arch.vcore->tb_offset =
1401                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1402                 break;
1403         case KVM_REG_PPC_LPCR:
1404                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1405                 break;
1406         case KVM_REG_PPC_LPCR_64:
1407                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1408                 break;
1409         case KVM_REG_PPC_PPR:
1410                 vcpu->arch.ppr = set_reg_val(id, *val);
1411                 break;
1412 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1413         case KVM_REG_PPC_TFHAR:
1414                 vcpu->arch.tfhar = set_reg_val(id, *val);
1415                 break;
1416         case KVM_REG_PPC_TFIAR:
1417                 vcpu->arch.tfiar = set_reg_val(id, *val);
1418                 break;
1419         case KVM_REG_PPC_TEXASR:
1420                 vcpu->arch.texasr = set_reg_val(id, *val);
1421                 break;
1422         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1423                 i = id - KVM_REG_PPC_TM_GPR0;
1424                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1425                 break;
1426         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1427         {
1428                 int j;
1429                 i = id - KVM_REG_PPC_TM_VSR0;
1430                 if (i < 32)
1431                         for (j = 0; j < TS_FPRWIDTH; j++)
1432                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1433                 else
1434                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1435                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1436                         else
1437                                 r = -ENXIO;
1438                 break;
1439         }
1440         case KVM_REG_PPC_TM_CR:
1441                 vcpu->arch.cr_tm = set_reg_val(id, *val);
1442                 break;
1443         case KVM_REG_PPC_TM_LR:
1444                 vcpu->arch.lr_tm = set_reg_val(id, *val);
1445                 break;
1446         case KVM_REG_PPC_TM_CTR:
1447                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1448                 break;
1449         case KVM_REG_PPC_TM_FPSCR:
1450                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1451                 break;
1452         case KVM_REG_PPC_TM_AMR:
1453                 vcpu->arch.amr_tm = set_reg_val(id, *val);
1454                 break;
1455         case KVM_REG_PPC_TM_PPR:
1456                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1457                 break;
1458         case KVM_REG_PPC_TM_VRSAVE:
1459                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1460                 break;
1461         case KVM_REG_PPC_TM_VSCR:
1462                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1463                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1464                 else
1465                         r = - ENXIO;
1466                 break;
1467         case KVM_REG_PPC_TM_DSCR:
1468                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1469                 break;
1470         case KVM_REG_PPC_TM_TAR:
1471                 vcpu->arch.tar_tm = set_reg_val(id, *val);
1472                 break;
1473 #endif
1474         case KVM_REG_PPC_ARCH_COMPAT:
1475                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1476                 break;
1477         default:
1478                 r = -EINVAL;
1479                 break;
1480         }
1481
1482         return r;
1483 }
1484
1485 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1486 {
1487         struct kvmppc_vcore *vcore;
1488
1489         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1490
1491         if (vcore == NULL)
1492                 return NULL;
1493
1494         INIT_LIST_HEAD(&vcore->runnable_threads);
1495         spin_lock_init(&vcore->lock);
1496         spin_lock_init(&vcore->stoltb_lock);
1497         init_swait_queue_head(&vcore->wq);
1498         vcore->preempt_tb = TB_NIL;
1499         vcore->lpcr = kvm->arch.lpcr;
1500         vcore->first_vcpuid = core * threads_per_subcore;
1501         vcore->kvm = kvm;
1502         INIT_LIST_HEAD(&vcore->preempt_list);
1503
1504         return vcore;
1505 }
1506
1507 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1508 static struct debugfs_timings_element {
1509         const char *name;
1510         size_t offset;
1511 } timings[] = {
1512         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
1513         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
1514         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
1515         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
1516         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
1517 };
1518
1519 #define N_TIMINGS       (sizeof(timings) / sizeof(timings[0]))
1520
1521 struct debugfs_timings_state {
1522         struct kvm_vcpu *vcpu;
1523         unsigned int    buflen;
1524         char            buf[N_TIMINGS * 100];
1525 };
1526
1527 static int debugfs_timings_open(struct inode *inode, struct file *file)
1528 {
1529         struct kvm_vcpu *vcpu = inode->i_private;
1530         struct debugfs_timings_state *p;
1531
1532         p = kzalloc(sizeof(*p), GFP_KERNEL);
1533         if (!p)
1534                 return -ENOMEM;
1535
1536         kvm_get_kvm(vcpu->kvm);
1537         p->vcpu = vcpu;
1538         file->private_data = p;
1539
1540         return nonseekable_open(inode, file);
1541 }
1542
1543 static int debugfs_timings_release(struct inode *inode, struct file *file)
1544 {
1545         struct debugfs_timings_state *p = file->private_data;
1546
1547         kvm_put_kvm(p->vcpu->kvm);
1548         kfree(p);
1549         return 0;
1550 }
1551
1552 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1553                                     size_t len, loff_t *ppos)
1554 {
1555         struct debugfs_timings_state *p = file->private_data;
1556         struct kvm_vcpu *vcpu = p->vcpu;
1557         char *s, *buf_end;
1558         struct kvmhv_tb_accumulator tb;
1559         u64 count;
1560         loff_t pos;
1561         ssize_t n;
1562         int i, loops;
1563         bool ok;
1564
1565         if (!p->buflen) {
1566                 s = p->buf;
1567                 buf_end = s + sizeof(p->buf);
1568                 for (i = 0; i < N_TIMINGS; ++i) {
1569                         struct kvmhv_tb_accumulator *acc;
1570
1571                         acc = (struct kvmhv_tb_accumulator *)
1572                                 ((unsigned long)vcpu + timings[i].offset);
1573                         ok = false;
1574                         for (loops = 0; loops < 1000; ++loops) {
1575                                 count = acc->seqcount;
1576                                 if (!(count & 1)) {
1577                                         smp_rmb();
1578                                         tb = *acc;
1579                                         smp_rmb();
1580                                         if (count == acc->seqcount) {
1581                                                 ok = true;
1582                                                 break;
1583                                         }
1584                                 }
1585                                 udelay(1);
1586                         }
1587                         if (!ok)
1588                                 snprintf(s, buf_end - s, "%s: stuck\n",
1589                                         timings[i].name);
1590                         else
1591                                 snprintf(s, buf_end - s,
1592                                         "%s: %llu %llu %llu %llu\n",
1593                                         timings[i].name, count / 2,
1594                                         tb_to_ns(tb.tb_total),
1595                                         tb_to_ns(tb.tb_min),
1596                                         tb_to_ns(tb.tb_max));
1597                         s += strlen(s);
1598                 }
1599                 p->buflen = s - p->buf;
1600         }
1601
1602         pos = *ppos;
1603         if (pos >= p->buflen)
1604                 return 0;
1605         if (len > p->buflen - pos)
1606                 len = p->buflen - pos;
1607         n = copy_to_user(buf, p->buf + pos, len);
1608         if (n) {
1609                 if (n == len)
1610                         return -EFAULT;
1611                 len -= n;
1612         }
1613         *ppos = pos + len;
1614         return len;
1615 }
1616
1617 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1618                                      size_t len, loff_t *ppos)
1619 {
1620         return -EACCES;
1621 }
1622
1623 static const struct file_operations debugfs_timings_ops = {
1624         .owner   = THIS_MODULE,
1625         .open    = debugfs_timings_open,
1626         .release = debugfs_timings_release,
1627         .read    = debugfs_timings_read,
1628         .write   = debugfs_timings_write,
1629         .llseek  = generic_file_llseek,
1630 };
1631
1632 /* Create a debugfs directory for the vcpu */
1633 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1634 {
1635         char buf[16];
1636         struct kvm *kvm = vcpu->kvm;
1637
1638         snprintf(buf, sizeof(buf), "vcpu%u", id);
1639         if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1640                 return;
1641         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1642         if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1643                 return;
1644         vcpu->arch.debugfs_timings =
1645                 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1646                                     vcpu, &debugfs_timings_ops);
1647 }
1648
1649 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1650 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1651 {
1652 }
1653 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1654
1655 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1656                                                    unsigned int id)
1657 {
1658         struct kvm_vcpu *vcpu;
1659         int err = -EINVAL;
1660         int core;
1661         struct kvmppc_vcore *vcore;
1662
1663         core = id / threads_per_subcore;
1664         if (core >= KVM_MAX_VCORES)
1665                 goto out;
1666
1667         err = -ENOMEM;
1668         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1669         if (!vcpu)
1670                 goto out;
1671
1672         err = kvm_vcpu_init(vcpu, kvm, id);
1673         if (err)
1674                 goto free_vcpu;
1675
1676         vcpu->arch.shared = &vcpu->arch.shregs;
1677 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1678         /*
1679          * The shared struct is never shared on HV,
1680          * so we can always use host endianness
1681          */
1682 #ifdef __BIG_ENDIAN__
1683         vcpu->arch.shared_big_endian = true;
1684 #else
1685         vcpu->arch.shared_big_endian = false;
1686 #endif
1687 #endif
1688         vcpu->arch.mmcr[0] = MMCR0_FC;
1689         vcpu->arch.ctrl = CTRL_RUNLATCH;
1690         /* default to host PVR, since we can't spoof it */
1691         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1692         spin_lock_init(&vcpu->arch.vpa_update_lock);
1693         spin_lock_init(&vcpu->arch.tbacct_lock);
1694         vcpu->arch.busy_preempt = TB_NIL;
1695         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1696
1697         kvmppc_mmu_book3s_hv_init(vcpu);
1698
1699         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1700
1701         init_waitqueue_head(&vcpu->arch.cpu_run);
1702
1703         mutex_lock(&kvm->lock);
1704         vcore = kvm->arch.vcores[core];
1705         if (!vcore) {
1706                 vcore = kvmppc_vcore_create(kvm, core);
1707                 kvm->arch.vcores[core] = vcore;
1708                 kvm->arch.online_vcores++;
1709         }
1710         mutex_unlock(&kvm->lock);
1711
1712         if (!vcore)
1713                 goto free_vcpu;
1714
1715         spin_lock(&vcore->lock);
1716         ++vcore->num_threads;
1717         spin_unlock(&vcore->lock);
1718         vcpu->arch.vcore = vcore;
1719         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1720         vcpu->arch.thread_cpu = -1;
1721
1722         vcpu->arch.cpu_type = KVM_CPU_3S_64;
1723         kvmppc_sanity_check(vcpu);
1724
1725         debugfs_vcpu_init(vcpu, id);
1726
1727         return vcpu;
1728
1729 free_vcpu:
1730         kmem_cache_free(kvm_vcpu_cache, vcpu);
1731 out:
1732         return ERR_PTR(err);
1733 }
1734
1735 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1736 {
1737         if (vpa->pinned_addr)
1738                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1739                                         vpa->dirty);
1740 }
1741
1742 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1743 {
1744         spin_lock(&vcpu->arch.vpa_update_lock);
1745         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1746         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1747         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1748         spin_unlock(&vcpu->arch.vpa_update_lock);
1749         kvm_vcpu_uninit(vcpu);
1750         kmem_cache_free(kvm_vcpu_cache, vcpu);
1751 }
1752
1753 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1754 {
1755         /* Indicate we want to get back into the guest */
1756         return 1;
1757 }
1758
1759 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1760 {
1761         unsigned long dec_nsec, now;
1762
1763         now = get_tb();
1764         if (now > vcpu->arch.dec_expires) {
1765                 /* decrementer has already gone negative */
1766                 kvmppc_core_queue_dec(vcpu);
1767                 kvmppc_core_prepare_to_enter(vcpu);
1768                 return;
1769         }
1770         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1771                    / tb_ticks_per_sec;
1772         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1773                       HRTIMER_MODE_REL);
1774         vcpu->arch.timer_running = 1;
1775 }
1776
1777 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1778 {
1779         vcpu->arch.ceded = 0;
1780         if (vcpu->arch.timer_running) {
1781                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1782                 vcpu->arch.timer_running = 0;
1783         }
1784 }
1785
1786 extern void __kvmppc_vcore_entry(void);
1787
1788 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1789                                    struct kvm_vcpu *vcpu)
1790 {
1791         u64 now;
1792
1793         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1794                 return;
1795         spin_lock_irq(&vcpu->arch.tbacct_lock);
1796         now = mftb();
1797         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1798                 vcpu->arch.stolen_logged;
1799         vcpu->arch.busy_preempt = now;
1800         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1801         spin_unlock_irq(&vcpu->arch.tbacct_lock);
1802         --vc->n_runnable;
1803         list_del(&vcpu->arch.run_list);
1804 }
1805
1806 static int kvmppc_grab_hwthread(int cpu)
1807 {
1808         struct paca_struct *tpaca;
1809         long timeout = 10000;
1810
1811         tpaca = &paca[cpu];
1812
1813         /* Ensure the thread won't go into the kernel if it wakes */
1814         tpaca->kvm_hstate.kvm_vcpu = NULL;
1815         tpaca->kvm_hstate.kvm_vcore = NULL;
1816         tpaca->kvm_hstate.napping = 0;
1817         smp_wmb();
1818         tpaca->kvm_hstate.hwthread_req = 1;
1819
1820         /*
1821          * If the thread is already executing in the kernel (e.g. handling
1822          * a stray interrupt), wait for it to get back to nap mode.
1823          * The smp_mb() is to ensure that our setting of hwthread_req
1824          * is visible before we look at hwthread_state, so if this
1825          * races with the code at system_reset_pSeries and the thread
1826          * misses our setting of hwthread_req, we are sure to see its
1827          * setting of hwthread_state, and vice versa.
1828          */
1829         smp_mb();
1830         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1831                 if (--timeout <= 0) {
1832                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1833                         return -EBUSY;
1834                 }
1835                 udelay(1);
1836         }
1837         return 0;
1838 }
1839
1840 static void kvmppc_release_hwthread(int cpu)
1841 {
1842         struct paca_struct *tpaca;
1843
1844         tpaca = &paca[cpu];
1845         tpaca->kvm_hstate.hwthread_req = 0;
1846         tpaca->kvm_hstate.kvm_vcpu = NULL;
1847         tpaca->kvm_hstate.kvm_vcore = NULL;
1848         tpaca->kvm_hstate.kvm_split_mode = NULL;
1849 }
1850
1851 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
1852 {
1853         int cpu;
1854         struct paca_struct *tpaca;
1855         struct kvmppc_vcore *mvc = vc->master_vcore;
1856
1857         cpu = vc->pcpu;
1858         if (vcpu) {
1859                 if (vcpu->arch.timer_running) {
1860                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1861                         vcpu->arch.timer_running = 0;
1862                 }
1863                 cpu += vcpu->arch.ptid;
1864                 vcpu->cpu = mvc->pcpu;
1865                 vcpu->arch.thread_cpu = cpu;
1866         }
1867         tpaca = &paca[cpu];
1868         tpaca->kvm_hstate.kvm_vcpu = vcpu;
1869         tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
1870         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
1871         smp_wmb();
1872         tpaca->kvm_hstate.kvm_vcore = mvc;
1873         if (cpu != smp_processor_id())
1874                 kvmppc_ipi_thread(cpu);
1875 }
1876
1877 static void kvmppc_wait_for_nap(void)
1878 {
1879         int cpu = smp_processor_id();
1880         int i, loops;
1881
1882         for (loops = 0; loops < 1000000; ++loops) {
1883                 /*
1884                  * Check if all threads are finished.
1885                  * We set the vcore pointer when starting a thread
1886                  * and the thread clears it when finished, so we look
1887                  * for any threads that still have a non-NULL vcore ptr.
1888                  */
1889                 for (i = 1; i < threads_per_subcore; ++i)
1890                         if (paca[cpu + i].kvm_hstate.kvm_vcore)
1891                                 break;
1892                 if (i == threads_per_subcore) {
1893                         HMT_medium();
1894                         return;
1895                 }
1896                 HMT_low();
1897         }
1898         HMT_medium();
1899         for (i = 1; i < threads_per_subcore; ++i)
1900                 if (paca[cpu + i].kvm_hstate.kvm_vcore)
1901                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
1902 }
1903
1904 /*
1905  * Check that we are on thread 0 and that any other threads in
1906  * this core are off-line.  Then grab the threads so they can't
1907  * enter the kernel.
1908  */
1909 static int on_primary_thread(void)
1910 {
1911         int cpu = smp_processor_id();
1912         int thr;
1913
1914         /* Are we on a primary subcore? */
1915         if (cpu_thread_in_subcore(cpu))
1916                 return 0;
1917
1918         thr = 0;
1919         while (++thr < threads_per_subcore)
1920                 if (cpu_online(cpu + thr))
1921                         return 0;
1922
1923         /* Grab all hw threads so they can't go into the kernel */
1924         for (thr = 1; thr < threads_per_subcore; ++thr) {
1925                 if (kvmppc_grab_hwthread(cpu + thr)) {
1926                         /* Couldn't grab one; let the others go */
1927                         do {
1928                                 kvmppc_release_hwthread(cpu + thr);
1929                         } while (--thr > 0);
1930                         return 0;
1931                 }
1932         }
1933         return 1;
1934 }
1935
1936 /*
1937  * A list of virtual cores for each physical CPU.
1938  * These are vcores that could run but their runner VCPU tasks are
1939  * (or may be) preempted.
1940  */
1941 struct preempted_vcore_list {
1942         struct list_head        list;
1943         spinlock_t              lock;
1944 };
1945
1946 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
1947
1948 static void init_vcore_lists(void)
1949 {
1950         int cpu;
1951
1952         for_each_possible_cpu(cpu) {
1953                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
1954                 spin_lock_init(&lp->lock);
1955                 INIT_LIST_HEAD(&lp->list);
1956         }
1957 }
1958
1959 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
1960 {
1961         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
1962
1963         vc->vcore_state = VCORE_PREEMPT;
1964         vc->pcpu = smp_processor_id();
1965         if (vc->num_threads < threads_per_subcore) {
1966                 spin_lock(&lp->lock);
1967                 list_add_tail(&vc->preempt_list, &lp->list);
1968                 spin_unlock(&lp->lock);
1969         }
1970
1971         /* Start accumulating stolen time */
1972         kvmppc_core_start_stolen(vc);
1973 }
1974
1975 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
1976 {
1977         struct preempted_vcore_list *lp;
1978
1979         kvmppc_core_end_stolen(vc);
1980         if (!list_empty(&vc->preempt_list)) {
1981                 lp = &per_cpu(preempted_vcores, vc->pcpu);
1982                 spin_lock(&lp->lock);
1983                 list_del_init(&vc->preempt_list);
1984                 spin_unlock(&lp->lock);
1985         }
1986         vc->vcore_state = VCORE_INACTIVE;
1987 }
1988
1989 /*
1990  * This stores information about the virtual cores currently
1991  * assigned to a physical core.
1992  */
1993 struct core_info {
1994         int             n_subcores;
1995         int             max_subcore_threads;
1996         int             total_threads;
1997         int             subcore_threads[MAX_SUBCORES];
1998         struct kvm      *subcore_vm[MAX_SUBCORES];
1999         struct list_head vcs[MAX_SUBCORES];
2000 };
2001
2002 /*
2003  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2004  * respectively in 2-way micro-threading (split-core) mode.
2005  */
2006 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2007
2008 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2009 {
2010         int sub;
2011
2012         memset(cip, 0, sizeof(*cip));
2013         cip->n_subcores = 1;
2014         cip->max_subcore_threads = vc->num_threads;
2015         cip->total_threads = vc->num_threads;
2016         cip->subcore_threads[0] = vc->num_threads;
2017         cip->subcore_vm[0] = vc->kvm;
2018         for (sub = 0; sub < MAX_SUBCORES; ++sub)
2019                 INIT_LIST_HEAD(&cip->vcs[sub]);
2020         list_add_tail(&vc->preempt_list, &cip->vcs[0]);
2021 }
2022
2023 static bool subcore_config_ok(int n_subcores, int n_threads)
2024 {
2025         /* Can only dynamically split if unsplit to begin with */
2026         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2027                 return false;
2028         if (n_subcores > MAX_SUBCORES)
2029                 return false;
2030         if (n_subcores > 1) {
2031                 if (!(dynamic_mt_modes & 2))
2032                         n_subcores = 4;
2033                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2034                         return false;
2035         }
2036
2037         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2038 }
2039
2040 static void init_master_vcore(struct kvmppc_vcore *vc)
2041 {
2042         vc->master_vcore = vc;
2043         vc->entry_exit_map = 0;
2044         vc->in_guest = 0;
2045         vc->napping_threads = 0;
2046         vc->conferring_threads = 0;
2047 }
2048
2049 /*
2050  * See if the existing subcores can be split into 3 (or fewer) subcores
2051  * of at most two threads each, so we can fit in another vcore.  This
2052  * assumes there are at most two subcores and at most 6 threads in total.
2053  */
2054 static bool can_split_piggybacked_subcores(struct core_info *cip)
2055 {
2056         int sub, new_sub;
2057         int large_sub = -1;
2058         int thr;
2059         int n_subcores = cip->n_subcores;
2060         struct kvmppc_vcore *vc, *vcnext;
2061         struct kvmppc_vcore *master_vc = NULL;
2062
2063         for (sub = 0; sub < cip->n_subcores; ++sub) {
2064                 if (cip->subcore_threads[sub] <= 2)
2065                         continue;
2066                 if (large_sub >= 0)
2067                         return false;
2068                 large_sub = sub;
2069                 vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2070                                       preempt_list);
2071                 if (vc->num_threads > 2)
2072                         return false;
2073                 n_subcores += (cip->subcore_threads[sub] - 1) >> 1;
2074         }
2075         if (large_sub < 0 || !subcore_config_ok(n_subcores + 1, 2))
2076                 return false;
2077
2078         /*
2079          * Seems feasible, so go through and move vcores to new subcores.
2080          * Note that when we have two or more vcores in one subcore,
2081          * all those vcores must have only one thread each.
2082          */
2083         new_sub = cip->n_subcores;
2084         thr = 0;
2085         sub = large_sub;
2086         list_for_each_entry_safe(vc, vcnext, &cip->vcs[sub], preempt_list) {
2087                 if (thr >= 2) {
2088                         list_del(&vc->preempt_list);
2089                         list_add_tail(&vc->preempt_list, &cip->vcs[new_sub]);
2090                         /* vc->num_threads must be 1 */
2091                         if (++cip->subcore_threads[new_sub] == 1) {
2092                                 cip->subcore_vm[new_sub] = vc->kvm;
2093                                 init_master_vcore(vc);
2094                                 master_vc = vc;
2095                                 ++cip->n_subcores;
2096                         } else {
2097                                 vc->master_vcore = master_vc;
2098                                 ++new_sub;
2099                         }
2100                 }
2101                 thr += vc->num_threads;
2102         }
2103         cip->subcore_threads[large_sub] = 2;
2104         cip->max_subcore_threads = 2;
2105
2106         return true;
2107 }
2108
2109 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2110 {
2111         int n_threads = vc->num_threads;
2112         int sub;
2113
2114         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2115                 return false;
2116
2117         if (n_threads < cip->max_subcore_threads)
2118                 n_threads = cip->max_subcore_threads;
2119         if (subcore_config_ok(cip->n_subcores + 1, n_threads)) {
2120                 cip->max_subcore_threads = n_threads;
2121         } else if (cip->n_subcores <= 2 && cip->total_threads <= 6 &&
2122                    vc->num_threads <= 2) {
2123                 /*
2124                  * We may be able to fit another subcore in by
2125                  * splitting an existing subcore with 3 or 4
2126                  * threads into two 2-thread subcores, or one
2127                  * with 5 or 6 threads into three subcores.
2128                  * We can only do this if those subcores have
2129                  * piggybacked virtual cores.
2130                  */
2131                 if (!can_split_piggybacked_subcores(cip))
2132                         return false;
2133         } else {
2134                 return false;
2135         }
2136
2137         sub = cip->n_subcores;
2138         ++cip->n_subcores;
2139         cip->total_threads += vc->num_threads;
2140         cip->subcore_threads[sub] = vc->num_threads;
2141         cip->subcore_vm[sub] = vc->kvm;
2142         init_master_vcore(vc);
2143         list_del(&vc->preempt_list);
2144         list_add_tail(&vc->preempt_list, &cip->vcs[sub]);
2145
2146         return true;
2147 }
2148
2149 static bool can_piggyback_subcore(struct kvmppc_vcore *pvc,
2150                                   struct core_info *cip, int sub)
2151 {
2152         struct kvmppc_vcore *vc;
2153         int n_thr;
2154
2155         vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2156                               preempt_list);
2157
2158         /* require same VM and same per-core reg values */
2159         if (pvc->kvm != vc->kvm ||
2160             pvc->tb_offset != vc->tb_offset ||
2161             pvc->pcr != vc->pcr ||
2162             pvc->lpcr != vc->lpcr)
2163                 return false;
2164
2165         /* P8 guest with > 1 thread per core would see wrong TIR value */
2166         if (cpu_has_feature(CPU_FTR_ARCH_207S) &&
2167             (vc->num_threads > 1 || pvc->num_threads > 1))
2168                 return false;
2169
2170         n_thr = cip->subcore_threads[sub] + pvc->num_threads;
2171         if (n_thr > cip->max_subcore_threads) {
2172                 if (!subcore_config_ok(cip->n_subcores, n_thr))
2173                         return false;
2174                 cip->max_subcore_threads = n_thr;
2175         }
2176
2177         cip->total_threads += pvc->num_threads;
2178         cip->subcore_threads[sub] = n_thr;
2179         pvc->master_vcore = vc;
2180         list_del(&pvc->preempt_list);
2181         list_add_tail(&pvc->preempt_list, &cip->vcs[sub]);
2182
2183         return true;
2184 }
2185
2186 /*
2187  * Work out whether it is possible to piggyback the execution of
2188  * vcore *pvc onto the execution of the other vcores described in *cip.
2189  */
2190 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2191                           int target_threads)
2192 {
2193         int sub;
2194
2195         if (cip->total_threads + pvc->num_threads > target_threads)
2196                 return false;
2197         for (sub = 0; sub < cip->n_subcores; ++sub)
2198                 if (cip->subcore_threads[sub] &&
2199                     can_piggyback_subcore(pvc, cip, sub))
2200                         return true;
2201
2202         if (can_dynamic_split(pvc, cip))
2203                 return true;
2204
2205         return false;
2206 }
2207
2208 static void prepare_threads(struct kvmppc_vcore *vc)
2209 {
2210         struct kvm_vcpu *vcpu, *vnext;
2211
2212         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2213                                  arch.run_list) {
2214                 if (signal_pending(vcpu->arch.run_task))
2215                         vcpu->arch.ret = -EINTR;
2216                 else if (vcpu->arch.vpa.update_pending ||
2217                          vcpu->arch.slb_shadow.update_pending ||
2218                          vcpu->arch.dtl.update_pending)
2219                         vcpu->arch.ret = RESUME_GUEST;
2220                 else
2221                         continue;
2222                 kvmppc_remove_runnable(vc, vcpu);
2223                 wake_up(&vcpu->arch.cpu_run);
2224         }
2225 }
2226
2227 static void collect_piggybacks(struct core_info *cip, int target_threads)
2228 {
2229         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2230         struct kvmppc_vcore *pvc, *vcnext;
2231
2232         spin_lock(&lp->lock);
2233         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2234                 if (!spin_trylock(&pvc->lock))
2235                         continue;
2236                 prepare_threads(pvc);
2237                 if (!pvc->n_runnable) {
2238                         list_del_init(&pvc->preempt_list);
2239                         if (pvc->runner == NULL) {
2240                                 pvc->vcore_state = VCORE_INACTIVE;
2241                                 kvmppc_core_end_stolen(pvc);
2242                         }
2243                         spin_unlock(&pvc->lock);
2244                         continue;
2245                 }
2246                 if (!can_piggyback(pvc, cip, target_threads)) {
2247                         spin_unlock(&pvc->lock);
2248                         continue;
2249                 }
2250                 kvmppc_core_end_stolen(pvc);
2251                 pvc->vcore_state = VCORE_PIGGYBACK;
2252                 if (cip->total_threads >= target_threads)
2253                         break;
2254         }
2255         spin_unlock(&lp->lock);
2256 }
2257
2258 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2259 {
2260         int still_running = 0;
2261         u64 now;
2262         long ret;
2263         struct kvm_vcpu *vcpu, *vnext;
2264
2265         spin_lock(&vc->lock);
2266         now = get_tb();
2267         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2268                                  arch.run_list) {
2269                 /* cancel pending dec exception if dec is positive */
2270                 if (now < vcpu->arch.dec_expires &&
2271                     kvmppc_core_pending_dec(vcpu))
2272                         kvmppc_core_dequeue_dec(vcpu);
2273
2274                 trace_kvm_guest_exit(vcpu);
2275
2276                 ret = RESUME_GUEST;
2277                 if (vcpu->arch.trap)
2278                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2279                                                     vcpu->arch.run_task);
2280
2281                 vcpu->arch.ret = ret;
2282                 vcpu->arch.trap = 0;
2283
2284                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2285                         if (vcpu->arch.pending_exceptions)
2286                                 kvmppc_core_prepare_to_enter(vcpu);
2287                         if (vcpu->arch.ceded)
2288                                 kvmppc_set_timer(vcpu);
2289                         else
2290                                 ++still_running;
2291                 } else {
2292                         kvmppc_remove_runnable(vc, vcpu);
2293                         wake_up(&vcpu->arch.cpu_run);
2294                 }
2295         }
2296         list_del_init(&vc->preempt_list);
2297         if (!is_master) {
2298                 if (still_running > 0) {
2299                         kvmppc_vcore_preempt(vc);
2300                 } else if (vc->runner) {
2301                         vc->vcore_state = VCORE_PREEMPT;
2302                         kvmppc_core_start_stolen(vc);
2303                 } else {
2304                         vc->vcore_state = VCORE_INACTIVE;
2305                 }
2306                 if (vc->n_runnable > 0 && vc->runner == NULL) {
2307                         /* make sure there's a candidate runner awake */
2308                         vcpu = list_first_entry(&vc->runnable_threads,
2309                                                 struct kvm_vcpu, arch.run_list);
2310                         wake_up(&vcpu->arch.cpu_run);
2311                 }
2312         }
2313         spin_unlock(&vc->lock);
2314 }
2315
2316 /*
2317  * Clear core from the list of active host cores as we are about to
2318  * enter the guest. Only do this if it is the primary thread of the
2319  * core (not if a subcore) that is entering the guest.
2320  */
2321 static inline void kvmppc_clear_host_core(int cpu)
2322 {
2323         int core;
2324
2325         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2326                 return;
2327         /*
2328          * Memory barrier can be omitted here as we will do a smp_wmb()
2329          * later in kvmppc_start_thread and we need ensure that state is
2330          * visible to other CPUs only after we enter guest.
2331          */
2332         core = cpu >> threads_shift;
2333         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2334 }
2335
2336 /*
2337  * Advertise this core as an active host core since we exited the guest
2338  * Only need to do this if it is the primary thread of the core that is
2339  * exiting.
2340  */
2341 static inline void kvmppc_set_host_core(int cpu)
2342 {
2343         int core;
2344
2345         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2346                 return;
2347
2348         /*
2349          * Memory barrier can be omitted here because we do a spin_unlock
2350          * immediately after this which provides the memory barrier.
2351          */
2352         core = cpu >> threads_shift;
2353         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2354 }
2355
2356 /*
2357  * Run a set of guest threads on a physical core.
2358  * Called with vc->lock held.
2359  */
2360 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2361 {
2362         struct kvm_vcpu *vcpu, *vnext;
2363         int i;
2364         int srcu_idx;
2365         struct core_info core_info;
2366         struct kvmppc_vcore *pvc, *vcnext;
2367         struct kvm_split_mode split_info, *sip;
2368         int split, subcore_size, active;
2369         int sub;
2370         bool thr0_done;
2371         unsigned long cmd_bit, stat_bit;
2372         int pcpu, thr;
2373         int target_threads;
2374
2375         /*
2376          * Remove from the list any threads that have a signal pending
2377          * or need a VPA update done
2378          */
2379         prepare_threads(vc);
2380
2381         /* if the runner is no longer runnable, let the caller pick a new one */
2382         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2383                 return;
2384
2385         /*
2386          * Initialize *vc.
2387          */
2388         init_master_vcore(vc);
2389         vc->preempt_tb = TB_NIL;
2390
2391         /*
2392          * Make sure we are running on primary threads, and that secondary
2393          * threads are offline.  Also check if the number of threads in this
2394          * guest are greater than the current system threads per guest.
2395          */
2396         if ((threads_per_core > 1) &&
2397             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2398                 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2399                                          arch.run_list) {
2400                         vcpu->arch.ret = -EBUSY;
2401                         kvmppc_remove_runnable(vc, vcpu);
2402                         wake_up(&vcpu->arch.cpu_run);
2403                 }
2404                 goto out;
2405         }
2406
2407         /*
2408          * See if we could run any other vcores on the physical core
2409          * along with this one.
2410          */
2411         init_core_info(&core_info, vc);
2412         pcpu = smp_processor_id();
2413         target_threads = threads_per_subcore;
2414         if (target_smt_mode && target_smt_mode < target_threads)
2415                 target_threads = target_smt_mode;
2416         if (vc->num_threads < target_threads)
2417                 collect_piggybacks(&core_info, target_threads);
2418
2419         /* Decide on micro-threading (split-core) mode */
2420         subcore_size = threads_per_subcore;
2421         cmd_bit = stat_bit = 0;
2422         split = core_info.n_subcores;
2423         sip = NULL;
2424         if (split > 1) {
2425                 /* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2426                 if (split == 2 && (dynamic_mt_modes & 2)) {
2427                         cmd_bit = HID0_POWER8_1TO2LPAR;
2428                         stat_bit = HID0_POWER8_2LPARMODE;
2429                 } else {
2430                         split = 4;
2431                         cmd_bit = HID0_POWER8_1TO4LPAR;
2432                         stat_bit = HID0_POWER8_4LPARMODE;
2433                 }
2434                 subcore_size = MAX_SMT_THREADS / split;
2435                 sip = &split_info;
2436                 memset(&split_info, 0, sizeof(split_info));
2437                 split_info.rpr = mfspr(SPRN_RPR);
2438                 split_info.pmmar = mfspr(SPRN_PMMAR);
2439                 split_info.ldbar = mfspr(SPRN_LDBAR);
2440                 split_info.subcore_size = subcore_size;
2441                 for (sub = 0; sub < core_info.n_subcores; ++sub)
2442                         split_info.master_vcs[sub] =
2443                                 list_first_entry(&core_info.vcs[sub],
2444                                         struct kvmppc_vcore, preempt_list);
2445                 /* order writes to split_info before kvm_split_mode pointer */
2446                 smp_wmb();
2447         }
2448         pcpu = smp_processor_id();
2449         for (thr = 0; thr < threads_per_subcore; ++thr)
2450                 paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2451
2452         /* Initiate micro-threading (split-core) if required */
2453         if (cmd_bit) {
2454                 unsigned long hid0 = mfspr(SPRN_HID0);
2455
2456                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2457                 mb();
2458                 mtspr(SPRN_HID0, hid0);
2459                 isync();
2460                 for (;;) {
2461                         hid0 = mfspr(SPRN_HID0);
2462                         if (hid0 & stat_bit)
2463                                 break;
2464                         cpu_relax();
2465                 }
2466         }
2467
2468         kvmppc_clear_host_core(pcpu);
2469
2470         /* Start all the threads */
2471         active = 0;
2472         for (sub = 0; sub < core_info.n_subcores; ++sub) {
2473                 thr = subcore_thread_map[sub];
2474                 thr0_done = false;
2475                 active |= 1 << thr;
2476                 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2477                         pvc->pcpu = pcpu + thr;
2478                         list_for_each_entry(vcpu, &pvc->runnable_threads,
2479                                             arch.run_list) {
2480                                 kvmppc_start_thread(vcpu, pvc);
2481                                 kvmppc_create_dtl_entry(vcpu, pvc);
2482                                 trace_kvm_guest_enter(vcpu);
2483                                 if (!vcpu->arch.ptid)
2484                                         thr0_done = true;
2485                                 active |= 1 << (thr + vcpu->arch.ptid);
2486                         }
2487                         /*
2488                          * We need to start the first thread of each subcore
2489                          * even if it doesn't have a vcpu.
2490                          */
2491                         if (pvc->master_vcore == pvc && !thr0_done)
2492                                 kvmppc_start_thread(NULL, pvc);
2493                         thr += pvc->num_threads;
2494                 }
2495         }
2496
2497         /*
2498          * Ensure that split_info.do_nap is set after setting
2499          * the vcore pointer in the PACA of the secondaries.
2500          */
2501         smp_mb();
2502         if (cmd_bit)
2503                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
2504
2505         /*
2506          * When doing micro-threading, poke the inactive threads as well.
2507          * This gets them to the nap instruction after kvm_do_nap,
2508          * which reduces the time taken to unsplit later.
2509          */
2510         if (split > 1)
2511                 for (thr = 1; thr < threads_per_subcore; ++thr)
2512                         if (!(active & (1 << thr)))
2513                                 kvmppc_ipi_thread(pcpu + thr);
2514
2515         vc->vcore_state = VCORE_RUNNING;
2516         preempt_disable();
2517
2518         trace_kvmppc_run_core(vc, 0);
2519
2520         for (sub = 0; sub < core_info.n_subcores; ++sub)
2521                 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2522                         spin_unlock(&pvc->lock);
2523
2524         kvm_guest_enter();
2525
2526         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2527
2528         __kvmppc_vcore_entry();
2529
2530         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2531
2532         spin_lock(&vc->lock);
2533         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2534         vc->vcore_state = VCORE_EXITING;
2535
2536         /* wait for secondary threads to finish writing their state to memory */
2537         kvmppc_wait_for_nap();
2538
2539         /* Return to whole-core mode if we split the core earlier */
2540         if (split > 1) {
2541                 unsigned long hid0 = mfspr(SPRN_HID0);
2542                 unsigned long loops = 0;
2543
2544                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2545                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2546                 mb();
2547                 mtspr(SPRN_HID0, hid0);
2548                 isync();
2549                 for (;;) {
2550                         hid0 = mfspr(SPRN_HID0);
2551                         if (!(hid0 & stat_bit))
2552                                 break;
2553                         cpu_relax();
2554                         ++loops;
2555                 }
2556                 split_info.do_nap = 0;
2557         }
2558
2559         /* Let secondaries go back to the offline loop */
2560         for (i = 0; i < threads_per_subcore; ++i) {
2561                 kvmppc_release_hwthread(pcpu + i);
2562                 if (sip && sip->napped[i])
2563                         kvmppc_ipi_thread(pcpu + i);
2564         }
2565
2566         kvmppc_set_host_core(pcpu);
2567
2568         spin_unlock(&vc->lock);
2569
2570         /* make sure updates to secondary vcpu structs are visible now */
2571         smp_mb();
2572         kvm_guest_exit();
2573
2574         for (sub = 0; sub < core_info.n_subcores; ++sub)
2575                 list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2576                                          preempt_list)
2577                         post_guest_process(pvc, pvc == vc);
2578
2579         spin_lock(&vc->lock);
2580         preempt_enable();
2581
2582  out:
2583         vc->vcore_state = VCORE_INACTIVE;
2584         trace_kvmppc_run_core(vc, 1);
2585 }
2586
2587 /*
2588  * Wait for some other vcpu thread to execute us, and
2589  * wake us up when we need to handle something in the host.
2590  */
2591 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2592                                  struct kvm_vcpu *vcpu, int wait_state)
2593 {
2594         DEFINE_WAIT(wait);
2595
2596         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2597         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2598                 spin_unlock(&vc->lock);
2599                 schedule();
2600                 spin_lock(&vc->lock);
2601         }
2602         finish_wait(&vcpu->arch.cpu_run, &wait);
2603 }
2604
2605 /*
2606  * All the vcpus in this vcore are idle, so wait for a decrementer
2607  * or external interrupt to one of the vcpus.  vc->lock is held.
2608  */
2609 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2610 {
2611         struct kvm_vcpu *vcpu;
2612         int do_sleep = 1;
2613         DECLARE_SWAITQUEUE(wait);
2614
2615         prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2616
2617         /*
2618          * Check one last time for pending exceptions and ceded state after
2619          * we put ourselves on the wait queue
2620          */
2621         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
2622                 if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded) {
2623                         do_sleep = 0;
2624                         break;
2625                 }
2626         }
2627
2628         if (!do_sleep) {
2629                 finish_swait(&vc->wq, &wait);
2630                 return;
2631         }
2632
2633         vc->vcore_state = VCORE_SLEEPING;
2634         trace_kvmppc_vcore_blocked(vc, 0);
2635         spin_unlock(&vc->lock);
2636         schedule();
2637         finish_swait(&vc->wq, &wait);
2638         spin_lock(&vc->lock);
2639         vc->vcore_state = VCORE_INACTIVE;
2640         trace_kvmppc_vcore_blocked(vc, 1);
2641 }
2642
2643 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2644 {
2645         int n_ceded;
2646         struct kvmppc_vcore *vc;
2647         struct kvm_vcpu *v, *vn;
2648
2649         trace_kvmppc_run_vcpu_enter(vcpu);
2650
2651         kvm_run->exit_reason = 0;
2652         vcpu->arch.ret = RESUME_GUEST;
2653         vcpu->arch.trap = 0;
2654         kvmppc_update_vpas(vcpu);
2655
2656         /*
2657          * Synchronize with other threads in this virtual core
2658          */
2659         vc = vcpu->arch.vcore;
2660         spin_lock(&vc->lock);
2661         vcpu->arch.ceded = 0;
2662         vcpu->arch.run_task = current;
2663         vcpu->arch.kvm_run = kvm_run;
2664         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2665         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2666         vcpu->arch.busy_preempt = TB_NIL;
2667         list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
2668         ++vc->n_runnable;
2669
2670         /*
2671          * This happens the first time this is called for a vcpu.
2672          * If the vcore is already running, we may be able to start
2673          * this thread straight away and have it join in.
2674          */
2675         if (!signal_pending(current)) {
2676                 if (vc->vcore_state == VCORE_PIGGYBACK) {
2677                         struct kvmppc_vcore *mvc = vc->master_vcore;
2678                         if (spin_trylock(&mvc->lock)) {
2679                                 if (mvc->vcore_state == VCORE_RUNNING &&
2680                                     !VCORE_IS_EXITING(mvc)) {
2681                                         kvmppc_create_dtl_entry(vcpu, vc);
2682                                         kvmppc_start_thread(vcpu, vc);
2683                                         trace_kvm_guest_enter(vcpu);
2684                                 }
2685                                 spin_unlock(&mvc->lock);
2686                         }
2687                 } else if (vc->vcore_state == VCORE_RUNNING &&
2688                            !VCORE_IS_EXITING(vc)) {
2689                         kvmppc_create_dtl_entry(vcpu, vc);
2690                         kvmppc_start_thread(vcpu, vc);
2691                         trace_kvm_guest_enter(vcpu);
2692                 } else if (vc->vcore_state == VCORE_SLEEPING) {
2693                         swake_up(&vc->wq);
2694                 }
2695
2696         }
2697
2698         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2699                !signal_pending(current)) {
2700                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2701                         kvmppc_vcore_end_preempt(vc);
2702
2703                 if (vc->vcore_state != VCORE_INACTIVE) {
2704                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
2705                         continue;
2706                 }
2707                 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
2708                                          arch.run_list) {
2709                         kvmppc_core_prepare_to_enter(v);
2710                         if (signal_pending(v->arch.run_task)) {
2711                                 kvmppc_remove_runnable(vc, v);
2712                                 v->stat.signal_exits++;
2713                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2714                                 v->arch.ret = -EINTR;
2715                                 wake_up(&v->arch.cpu_run);
2716                         }
2717                 }
2718                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2719                         break;
2720                 n_ceded = 0;
2721                 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
2722                         if (!v->arch.pending_exceptions)
2723                                 n_ceded += v->arch.ceded;
2724                         else
2725                                 v->arch.ceded = 0;
2726                 }
2727                 vc->runner = vcpu;
2728                 if (n_ceded == vc->n_runnable) {
2729                         kvmppc_vcore_blocked(vc);
2730                 } else if (need_resched()) {
2731                         kvmppc_vcore_preempt(vc);
2732                         /* Let something else run */
2733                         cond_resched_lock(&vc->lock);
2734                         if (vc->vcore_state == VCORE_PREEMPT)
2735                                 kvmppc_vcore_end_preempt(vc);
2736                 } else {
2737                         kvmppc_run_core(vc);
2738                 }
2739                 vc->runner = NULL;
2740         }
2741
2742         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2743                (vc->vcore_state == VCORE_RUNNING ||
2744                 vc->vcore_state == VCORE_EXITING ||
2745                 vc->vcore_state == VCORE_PIGGYBACK))
2746                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
2747
2748         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2749                 kvmppc_vcore_end_preempt(vc);
2750
2751         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2752                 kvmppc_remove_runnable(vc, vcpu);
2753                 vcpu->stat.signal_exits++;
2754                 kvm_run->exit_reason = KVM_EXIT_INTR;
2755                 vcpu->arch.ret = -EINTR;
2756         }
2757
2758         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2759                 /* Wake up some vcpu to run the core */
2760                 v = list_first_entry(&vc->runnable_threads,
2761                                      struct kvm_vcpu, arch.run_list);
2762                 wake_up(&v->arch.cpu_run);
2763         }
2764
2765         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2766         spin_unlock(&vc->lock);
2767         return vcpu->arch.ret;
2768 }
2769
2770 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2771 {
2772         int r;
2773         int srcu_idx;
2774
2775         if (!vcpu->arch.sane) {
2776                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2777                 return -EINVAL;
2778         }
2779
2780         kvmppc_core_prepare_to_enter(vcpu);
2781
2782         /* No need to go into the guest when all we'll do is come back out */
2783         if (signal_pending(current)) {
2784                 run->exit_reason = KVM_EXIT_INTR;
2785                 return -EINTR;
2786         }
2787
2788         atomic_inc(&vcpu->kvm->arch.vcpus_running);
2789         /* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2790         smp_mb();
2791
2792         /* On the first time here, set up HTAB and VRMA */
2793         if (!vcpu->kvm->arch.hpte_setup_done) {
2794                 r = kvmppc_hv_setup_htab_rma(vcpu);
2795                 if (r)
2796                         goto out;
2797         }
2798
2799         flush_all_to_thread(current);
2800
2801         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2802         vcpu->arch.pgdir = current->mm->pgd;
2803         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2804
2805         do {
2806                 r = kvmppc_run_vcpu(run, vcpu);
2807
2808                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2809                     !(vcpu->arch.shregs.msr & MSR_PR)) {
2810                         trace_kvm_hcall_enter(vcpu);
2811                         r = kvmppc_pseries_do_hcall(vcpu);
2812                         trace_kvm_hcall_exit(vcpu, r);
2813                         kvmppc_core_prepare_to_enter(vcpu);
2814                 } else if (r == RESUME_PAGE_FAULT) {
2815                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2816                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
2817                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2818                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2819                 }
2820         } while (is_kvmppc_resume_guest(r));
2821
2822  out:
2823         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2824         atomic_dec(&vcpu->kvm->arch.vcpus_running);
2825         return r;
2826 }
2827
2828 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2829                                      int linux_psize)
2830 {
2831         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2832
2833         if (!def->shift)
2834                 return;
2835         (*sps)->page_shift = def->shift;
2836         (*sps)->slb_enc = def->sllp;
2837         (*sps)->enc[0].page_shift = def->shift;
2838         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
2839         /*
2840          * Add 16MB MPSS support if host supports it
2841          */
2842         if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2843                 (*sps)->enc[1].page_shift = 24;
2844                 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2845         }
2846         (*sps)++;
2847 }
2848
2849 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2850                                          struct kvm_ppc_smmu_info *info)
2851 {
2852         struct kvm_ppc_one_seg_page_size *sps;
2853
2854         info->flags = KVM_PPC_PAGE_SIZES_REAL;
2855         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2856                 info->flags |= KVM_PPC_1T_SEGMENTS;
2857         info->slb_size = mmu_slb_size;
2858
2859         /* We only support these sizes for now, and no muti-size segments */
2860         sps = &info->sps[0];
2861         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2862         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2863         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2864
2865         return 0;
2866 }
2867
2868 /*
2869  * Get (and clear) the dirty memory log for a memory slot.
2870  */
2871 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2872                                          struct kvm_dirty_log *log)
2873 {
2874         struct kvm_memslots *slots;
2875         struct kvm_memory_slot *memslot;
2876         int r;
2877         unsigned long n;
2878
2879         mutex_lock(&kvm->slots_lock);
2880
2881         r = -EINVAL;
2882         if (log->slot >= KVM_USER_MEM_SLOTS)
2883                 goto out;
2884
2885         slots = kvm_memslots(kvm);
2886         memslot = id_to_memslot(slots, log->slot);
2887         r = -ENOENT;
2888         if (!memslot->dirty_bitmap)
2889                 goto out;
2890
2891         n = kvm_dirty_bitmap_bytes(memslot);
2892         memset(memslot->dirty_bitmap, 0, n);
2893
2894         r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2895         if (r)
2896                 goto out;
2897
2898         r = -EFAULT;
2899         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2900                 goto out;
2901
2902         r = 0;
2903 out:
2904         mutex_unlock(&kvm->slots_lock);
2905         return r;
2906 }
2907
2908 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2909                                         struct kvm_memory_slot *dont)
2910 {
2911         if (!dont || free->arch.rmap != dont->arch.rmap) {
2912                 vfree(free->arch.rmap);
2913                 free->arch.rmap = NULL;
2914         }
2915 }
2916
2917 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2918                                          unsigned long npages)
2919 {
2920         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2921         if (!slot->arch.rmap)
2922                 return -ENOMEM;
2923
2924         return 0;
2925 }
2926
2927 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2928                                         struct kvm_memory_slot *memslot,
2929                                         const struct kvm_userspace_memory_region *mem)
2930 {
2931         return 0;
2932 }
2933
2934 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2935                                 const struct kvm_userspace_memory_region *mem,
2936                                 const struct kvm_memory_slot *old,
2937                                 const struct kvm_memory_slot *new)
2938 {
2939         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2940         struct kvm_memslots *slots;
2941         struct kvm_memory_slot *memslot;
2942
2943         if (npages && old->npages) {
2944                 /*
2945                  * If modifying a memslot, reset all the rmap dirty bits.
2946                  * If this is a new memslot, we don't need to do anything
2947                  * since the rmap array starts out as all zeroes,
2948                  * i.e. no pages are dirty.
2949                  */
2950                 slots = kvm_memslots(kvm);
2951                 memslot = id_to_memslot(slots, mem->slot);
2952                 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2953         }
2954 }
2955
2956 /*
2957  * Update LPCR values in kvm->arch and in vcores.
2958  * Caller must hold kvm->lock.
2959  */
2960 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2961 {
2962         long int i;
2963         u32 cores_done = 0;
2964
2965         if ((kvm->arch.lpcr & mask) == lpcr)
2966                 return;
2967
2968         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2969
2970         for (i = 0; i < KVM_MAX_VCORES; ++i) {
2971                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2972                 if (!vc)
2973                         continue;
2974                 spin_lock(&vc->lock);
2975                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2976                 spin_unlock(&vc->lock);
2977                 if (++cores_done >= kvm->arch.online_vcores)
2978                         break;
2979         }
2980 }
2981
2982 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2983 {
2984         return;
2985 }
2986
2987 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2988 {
2989         int err = 0;
2990         struct kvm *kvm = vcpu->kvm;
2991         unsigned long hva;
2992         struct kvm_memory_slot *memslot;
2993         struct vm_area_struct *vma;
2994         unsigned long lpcr = 0, senc;
2995         unsigned long psize, porder;
2996         int srcu_idx;
2997
2998         mutex_lock(&kvm->lock);
2999         if (kvm->arch.hpte_setup_done)
3000                 goto out;       /* another vcpu beat us to it */
3001
3002         /* Allocate hashed page table (if not done already) and reset it */
3003         if (!kvm->arch.hpt_virt) {
3004                 err = kvmppc_alloc_hpt(kvm, NULL);
3005                 if (err) {
3006                         pr_err("KVM: Couldn't alloc HPT\n");
3007                         goto out;
3008                 }
3009         }
3010
3011         /* Look up the memslot for guest physical address 0 */
3012         srcu_idx = srcu_read_lock(&kvm->srcu);
3013         memslot = gfn_to_memslot(kvm, 0);
3014
3015         /* We must have some memory at 0 by now */
3016         err = -EINVAL;
3017         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3018                 goto out_srcu;
3019
3020         /* Look up the VMA for the start of this memory slot */
3021         hva = memslot->userspace_addr;
3022         down_read(&current->mm->mmap_sem);
3023         vma = find_vma(current->mm, hva);
3024         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3025                 goto up_out;
3026
3027         psize = vma_kernel_pagesize(vma);
3028         porder = __ilog2(psize);
3029
3030         up_read(&current->mm->mmap_sem);
3031
3032         /* We can handle 4k, 64k or 16M pages in the VRMA */
3033         err = -EINVAL;
3034         if (!(psize == 0x1000 || psize == 0x10000 ||
3035               psize == 0x1000000))
3036                 goto out_srcu;
3037
3038         /* Update VRMASD field in the LPCR */
3039         senc = slb_pgsize_encoding(psize);
3040         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3041                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3042         /* the -4 is to account for senc values starting at 0x10 */
3043         lpcr = senc << (LPCR_VRMASD_SH - 4);
3044
3045         /* Create HPTEs in the hash page table for the VRMA */
3046         kvmppc_map_vrma(vcpu, memslot, porder);
3047
3048         kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3049
3050         /* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
3051         smp_wmb();
3052         kvm->arch.hpte_setup_done = 1;
3053         err = 0;
3054  out_srcu:
3055         srcu_read_unlock(&kvm->srcu, srcu_idx);
3056  out:
3057         mutex_unlock(&kvm->lock);
3058         return err;
3059
3060  up_out:
3061         up_read(&current->mm->mmap_sem);
3062         goto out_srcu;
3063 }
3064
3065 #ifdef CONFIG_KVM_XICS
3066 static int kvmppc_cpu_notify(struct notifier_block *self, unsigned long action,
3067                         void *hcpu)
3068 {
3069         unsigned long cpu = (long)hcpu;
3070
3071         switch (action) {
3072         case CPU_UP_PREPARE:
3073         case CPU_UP_PREPARE_FROZEN:
3074                 kvmppc_set_host_core(cpu);
3075                 break;
3076
3077 #ifdef CONFIG_HOTPLUG_CPU
3078         case CPU_DEAD:
3079         case CPU_DEAD_FROZEN:
3080         case CPU_UP_CANCELED:
3081         case CPU_UP_CANCELED_FROZEN:
3082                 kvmppc_clear_host_core(cpu);
3083                 break;
3084 #endif
3085         default:
3086                 break;
3087         }
3088
3089         return NOTIFY_OK;
3090 }
3091
3092 static struct notifier_block kvmppc_cpu_notifier = {
3093             .notifier_call = kvmppc_cpu_notify,
3094 };
3095
3096 /*
3097  * Allocate a per-core structure for managing state about which cores are
3098  * running in the host versus the guest and for exchanging data between
3099  * real mode KVM and CPU running in the host.
3100  * This is only done for the first VM.
3101  * The allocated structure stays even if all VMs have stopped.
3102  * It is only freed when the kvm-hv module is unloaded.
3103  * It's OK for this routine to fail, we just don't support host
3104  * core operations like redirecting H_IPI wakeups.
3105  */
3106 void kvmppc_alloc_host_rm_ops(void)
3107 {
3108         struct kvmppc_host_rm_ops *ops;
3109         unsigned long l_ops;
3110         int cpu, core;
3111         int size;
3112
3113         /* Not the first time here ? */
3114         if (kvmppc_host_rm_ops_hv != NULL)
3115                 return;
3116
3117         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3118         if (!ops)
3119                 return;
3120
3121         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3122         ops->rm_core = kzalloc(size, GFP_KERNEL);
3123
3124         if (!ops->rm_core) {
3125                 kfree(ops);
3126                 return;
3127         }
3128
3129         get_online_cpus();
3130
3131         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3132                 if (!cpu_online(cpu))
3133                         continue;
3134
3135                 core = cpu >> threads_shift;
3136                 ops->rm_core[core].rm_state.in_host = 1;
3137         }
3138
3139         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3140
3141         /*
3142          * Make the contents of the kvmppc_host_rm_ops structure visible
3143          * to other CPUs before we assign it to the global variable.
3144          * Do an atomic assignment (no locks used here), but if someone
3145          * beats us to it, just free our copy and return.
3146          */
3147         smp_wmb();
3148         l_ops = (unsigned long) ops;
3149
3150         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3151                 put_online_cpus();
3152                 kfree(ops->rm_core);
3153                 kfree(ops);
3154                 return;
3155         }
3156
3157         register_cpu_notifier(&kvmppc_cpu_notifier);
3158
3159         put_online_cpus();
3160 }
3161
3162 void kvmppc_free_host_rm_ops(void)
3163 {
3164         if (kvmppc_host_rm_ops_hv) {
3165                 unregister_cpu_notifier(&kvmppc_cpu_notifier);
3166                 kfree(kvmppc_host_rm_ops_hv->rm_core);
3167                 kfree(kvmppc_host_rm_ops_hv);
3168                 kvmppc_host_rm_ops_hv = NULL;
3169         }
3170 }
3171 #endif
3172
3173 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3174 {
3175         unsigned long lpcr, lpid;
3176         char buf[32];
3177
3178         /* Allocate the guest's logical partition ID */
3179
3180         lpid = kvmppc_alloc_lpid();
3181         if ((long)lpid < 0)
3182                 return -ENOMEM;
3183         kvm->arch.lpid = lpid;
3184
3185         kvmppc_alloc_host_rm_ops();
3186
3187         /*
3188          * Since we don't flush the TLB when tearing down a VM,
3189          * and this lpid might have previously been used,
3190          * make sure we flush on each core before running the new VM.
3191          */
3192         cpumask_setall(&kvm->arch.need_tlb_flush);
3193
3194         /* Start out with the default set of hcalls enabled */
3195         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3196                sizeof(kvm->arch.enabled_hcalls));
3197
3198         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3199
3200         /* Init LPCR for virtual RMA mode */
3201         kvm->arch.host_lpid = mfspr(SPRN_LPID);
3202         kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3203         lpcr &= LPCR_PECE | LPCR_LPES;
3204         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3205                 LPCR_VPM0 | LPCR_VPM1;
3206         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3207                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3208         /* On POWER8 turn on online bit to enable PURR/SPURR */
3209         if (cpu_has_feature(CPU_FTR_ARCH_207S))
3210                 lpcr |= LPCR_ONL;
3211         kvm->arch.lpcr = lpcr;
3212
3213         /*
3214          * Track that we now have a HV mode VM active. This blocks secondary
3215          * CPU threads from coming online.
3216          */
3217         kvm_hv_vm_activated();
3218
3219         /*
3220          * Create a debugfs directory for the VM
3221          */
3222         snprintf(buf, sizeof(buf), "vm%d", current->pid);
3223         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3224         if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3225                 kvmppc_mmu_debugfs_init(kvm);
3226
3227         return 0;
3228 }
3229
3230 static void kvmppc_free_vcores(struct kvm *kvm)
3231 {
3232         long int i;
3233
3234         for (i = 0; i < KVM_MAX_VCORES; ++i)
3235                 kfree(kvm->arch.vcores[i]);
3236         kvm->arch.online_vcores = 0;
3237 }
3238
3239 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3240 {
3241         debugfs_remove_recursive(kvm->arch.debugfs_dir);
3242
3243         kvm_hv_vm_deactivated();
3244
3245         kvmppc_free_vcores(kvm);
3246
3247         kvmppc_free_hpt(kvm);
3248 }
3249
3250 /* We don't need to emulate any privileged instructions or dcbz */
3251 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3252                                      unsigned int inst, int *advance)
3253 {
3254         return EMULATE_FAIL;
3255 }
3256
3257 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3258                                         ulong spr_val)
3259 {
3260         return EMULATE_FAIL;
3261 }
3262
3263 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3264                                         ulong *spr_val)
3265 {
3266         return EMULATE_FAIL;
3267 }
3268
3269 static int kvmppc_core_check_processor_compat_hv(void)
3270 {
3271         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3272             !cpu_has_feature(CPU_FTR_ARCH_206))
3273                 return -EIO;
3274         return 0;
3275 }
3276
3277 static long kvm_arch_vm_ioctl_hv(struct file *filp,
3278                                  unsigned int ioctl, unsigned long arg)
3279 {
3280         struct kvm *kvm __maybe_unused = filp->private_data;
3281         void __user *argp = (void __user *)arg;
3282         long r;
3283
3284         switch (ioctl) {
3285
3286         case KVM_PPC_ALLOCATE_HTAB: {
3287                 u32 htab_order;
3288
3289                 r = -EFAULT;
3290                 if (get_user(htab_order, (u32 __user *)argp))
3291                         break;
3292                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
3293                 if (r)
3294                         break;
3295                 r = -EFAULT;
3296                 if (put_user(htab_order, (u32 __user *)argp))
3297                         break;
3298                 r = 0;
3299                 break;
3300         }
3301
3302         case KVM_PPC_GET_HTAB_FD: {
3303                 struct kvm_get_htab_fd ghf;
3304
3305                 r = -EFAULT;
3306                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
3307                         break;
3308                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3309                 break;
3310         }
3311
3312         default:
3313                 r = -ENOTTY;
3314         }
3315
3316         return r;
3317 }
3318
3319 /*
3320  * List of hcall numbers to enable by default.
3321  * For compatibility with old userspace, we enable by default
3322  * all hcalls that were implemented before the hcall-enabling
3323  * facility was added.  Note this list should not include H_RTAS.
3324  */
3325 static unsigned int default_hcall_list[] = {
3326         H_REMOVE,
3327         H_ENTER,
3328         H_READ,
3329         H_PROTECT,
3330         H_BULK_REMOVE,
3331         H_GET_TCE,
3332         H_PUT_TCE,
3333         H_SET_DABR,
3334         H_SET_XDABR,
3335         H_CEDE,
3336         H_PROD,
3337         H_CONFER,
3338         H_REGISTER_VPA,
3339 #ifdef CONFIG_KVM_XICS
3340         H_EOI,
3341         H_CPPR,
3342         H_IPI,
3343         H_IPOLL,
3344         H_XIRR,
3345         H_XIRR_X,
3346 #endif
3347         0
3348 };
3349
3350 static void init_default_hcalls(void)
3351 {
3352         int i;
3353         unsigned int hcall;
3354
3355         for (i = 0; default_hcall_list[i]; ++i) {
3356                 hcall = default_hcall_list[i];
3357                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3358                 __set_bit(hcall / 4, default_enabled_hcalls);
3359         }
3360 }
3361
3362 static struct kvmppc_ops kvm_ops_hv = {
3363         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3364         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3365         .get_one_reg = kvmppc_get_one_reg_hv,
3366         .set_one_reg = kvmppc_set_one_reg_hv,
3367         .vcpu_load   = kvmppc_core_vcpu_load_hv,
3368         .vcpu_put    = kvmppc_core_vcpu_put_hv,
3369         .set_msr     = kvmppc_set_msr_hv,
3370         .vcpu_run    = kvmppc_vcpu_run_hv,
3371         .vcpu_create = kvmppc_core_vcpu_create_hv,
3372         .vcpu_free   = kvmppc_core_vcpu_free_hv,
3373         .check_requests = kvmppc_core_check_requests_hv,
3374         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
3375         .flush_memslot  = kvmppc_core_flush_memslot_hv,
3376         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3377         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
3378         .unmap_hva = kvm_unmap_hva_hv,
3379         .unmap_hva_range = kvm_unmap_hva_range_hv,
3380         .age_hva  = kvm_age_hva_hv,
3381         .test_age_hva = kvm_test_age_hva_hv,
3382         .set_spte_hva = kvm_set_spte_hva_hv,
3383         .mmu_destroy  = kvmppc_mmu_destroy_hv,
3384         .free_memslot = kvmppc_core_free_memslot_hv,
3385         .create_memslot = kvmppc_core_create_memslot_hv,
3386         .init_vm =  kvmppc_core_init_vm_hv,
3387         .destroy_vm = kvmppc_core_destroy_vm_hv,
3388         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3389         .emulate_op = kvmppc_core_emulate_op_hv,
3390         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3391         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3392         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3393         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
3394         .hcall_implemented = kvmppc_hcall_impl_hv,
3395 };
3396
3397 static int kvmppc_book3s_init_hv(void)
3398 {
3399         int r;
3400         /*
3401          * FIXME!! Do we need to check on all cpus ?
3402          */
3403         r = kvmppc_core_check_processor_compat_hv();
3404         if (r < 0)
3405                 return -ENODEV;
3406
3407         kvm_ops_hv.owner = THIS_MODULE;
3408         kvmppc_hv_ops = &kvm_ops_hv;
3409
3410         init_default_hcalls();
3411
3412         init_vcore_lists();
3413
3414         r = kvmppc_mmu_hv_init();
3415         return r;
3416 }
3417
3418 static void kvmppc_book3s_exit_hv(void)
3419 {
3420         kvmppc_free_host_rm_ops();
3421         kvmppc_hv_ops = NULL;
3422 }
3423
3424 module_init(kvmppc_book3s_init_hv);
3425 module_exit(kvmppc_book3s_exit_hv);
3426 MODULE_LICENSE("GPL");
3427 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3428 MODULE_ALIAS("devname:kvm");