arch/powerpc/kvm/book3s_hv.c

   1 /*
   2  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
   3  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
   4  *
   5  * Authors:
   6  *    Paul Mackerras <paulus@au1.ibm.com>
   7  *    Alexander Graf <agraf@suse.de>
   8  *    Kevin Wolf <mail@kevin-wolf.de>
   9  *
  10  * Description: KVM functions specific to running on Book 3S
  11  * processors in hypervisor mode (specifically POWER7 and later).
  12  *
  13  * This file is derived from arch/powerpc/kvm/book3s.c,
  14  * by Alexander Graf <agraf@suse.de>.
  15  *
  16  * This program is free software; you can redistribute it and/or modify
  17  * it under the terms of the GNU General Public License, version 2, as
  18  * published by the Free Software Foundation.
  19  */
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/err.h>
  23 #include <linux/slab.h>
  24 #include <linux/preempt.h>
  25 #include <linux/sched.h>
  26 #include <linux/delay.h>
  27 #include <linux/export.h>
  28 #include <linux/fs.h>
  29 #include <linux/anon_inodes.h>
  30 #include <linux/cpumask.h>
  31 #include <linux/spinlock.h>
  32 #include <linux/page-flags.h>
  33 #include <linux/srcu.h>
  34 #include <linux/miscdevice.h>
  35
  36 #include <asm/reg.h>
  37 #include <asm/cputable.h>
  38 #include <asm/cache.h>
  39 #include <asm/cacheflush.h>
  40 #include <asm/tlbflush.h>
  41 #include <asm/uaccess.h>
  42 #include <asm/io.h>
  43 #include <asm/kvm_ppc.h>
  44 #include <asm/kvm_book3s.h>
  45 #include <asm/mmu_context.h>
  46 #include <asm/lppaca.h>
  47 #include <asm/processor.h>
  48 #include <asm/cputhreads.h>
  49 #include <asm/page.h>
  50 #include <asm/hvcall.h>
  51 #include <asm/switch_to.h>
  52 #include <asm/smp.h>
  53 #include <linux/gfp.h>
  54 #include <linux/vmalloc.h>
  55 #include <linux/highmem.h>
  56 #include <linux/hugetlb.h>
  57 #include <linux/module.h>
  58
  59 #include "book3s.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include "trace_hv.h"
  63
  64 /* #define EXIT_DEBUG */
  65 /* #define EXIT_DEBUG_SIMPLE */
  66 /* #define EXIT_DEBUG_INT */
  67
  68 /* Used to indicate that a guest page fault needs to be handled */
  69 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  70
  71 /* Used as a "null" value for timebase values */
  72 #define TB_NIL  (~(u64)0)
  73
  74 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
  75
  76 #if defined(CONFIG_PPC_64K_PAGES)
  77 #define MPP_BUFFER_ORDER        0
  78 #elif defined(CONFIG_PPC_4K_PAGES)
  79 #define MPP_BUFFER_ORDER        3
  80 #endif
  81
  82
  83 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
  84 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
  85
  86 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
  87 {
  88         int me;
  89         int cpu = vcpu->cpu;
  90         wait_queue_head_t *wqp;
  91
  92         wqp = kvm_arch_vcpu_wq(vcpu);
  93         if (waitqueue_active(wqp)) {
  94                 wake_up_interruptible(wqp);
  95                 ++vcpu->stat.halt_wakeup;
  96         }
  97
  98         me = get_cpu();
  99
 100         /* CPU points to the first thread of the core */
 101         if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
 102 #ifdef CONFIG_PPC_ICP_NATIVE
 103                 int real_cpu = cpu + vcpu->arch.ptid;
 104                 if (paca[real_cpu].kvm_hstate.xics_phys)
 105                         xics_wake_cpu(real_cpu);
 106                 else
 107 #endif
 108                 if (cpu_online(cpu))
 109                         smp_send_reschedule(cpu);
 110         }
 111         put_cpu();
 112 }
 113
 114 /*
 115  * We use the vcpu_load/put functions to measure stolen time.
 116  * Stolen time is counted as time when either the vcpu is able to
 117  * run as part of a virtual core, but the task running the vcore
 118  * is preempted or sleeping, or when the vcpu needs something done
 119  * in the kernel by the task running the vcpu, but that task is
 120  * preempted or sleeping.  Those two things have to be counted
 121  * separately, since one of the vcpu tasks will take on the job
 122  * of running the core, and the other vcpu tasks in the vcore will
 123  * sleep waiting for it to do that, but that sleep shouldn't count
 124  * as stolen time.
 125  *
 126  * Hence we accumulate stolen time when the vcpu can run as part of
 127  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 128  * needs its task to do other things in the kernel (for example,
 129  * service a page fault) in busy_stolen.  We don't accumulate
 130  * stolen time for a vcore when it is inactive, or for a vcpu
 131  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 132  * a misnomer; it means that the vcpu task is not executing in
 133  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 134  * the kernel.  We don't have any way of dividing up that time
 135  * between time that the vcpu is genuinely stopped, time that
 136  * the task is actively working on behalf of the vcpu, and time
 137  * that the task is preempted, so we don't count any of it as
 138  * stolen.
 139  *
 140  * Updates to busy_stolen are protected by arch.tbacct_lock;
 141  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
 142  * lock.  The stolen times are measured in units of timebase ticks.
 143  * (Note that the != TB_NIL checks below are purely defensive;
 144  * they should never fail.)
 145  */
 146
 147 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 148 {
 149         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 150         unsigned long flags;
 151
 152         /*
 153          * We can test vc->runner without taking the vcore lock,
 154          * because only this task ever sets vc->runner to this
 155          * vcpu, and once it is set to this vcpu, only this task
 156          * ever sets it to NULL.
 157          */
 158         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE) {
 159                 spin_lock_irqsave(&vc->stoltb_lock, flags);
 160                 if (vc->preempt_tb != TB_NIL) {
 161                         vc->stolen_tb += mftb() - vc->preempt_tb;
 162                         vc->preempt_tb = TB_NIL;
 163                 }
 164                 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 165         }
 166         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 167         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 168             vcpu->arch.busy_preempt != TB_NIL) {
 169                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 170                 vcpu->arch.busy_preempt = TB_NIL;
 171         }
 172         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 173 }
 174
 175 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 176 {
 177         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 178         unsigned long flags;
 179
 180         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE) {
 181                 spin_lock_irqsave(&vc->stoltb_lock, flags);
 182                 vc->preempt_tb = mftb();
 183                 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 184         }
 185         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 186         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 187                 vcpu->arch.busy_preempt = mftb();
 188         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 189 }
 190
 191 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
 192 {
 193         vcpu->arch.shregs.msr = msr;
 194         kvmppc_end_cede(vcpu);
 195 }
 196
 197 void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 198 {
 199         vcpu->arch.pvr = pvr;
 200 }
 201
 202 int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 203 {
 204         unsigned long pcr = 0;
 205         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 206
 207         if (arch_compat) {
 208                 switch (arch_compat) {
 209                 case PVR_ARCH_205:
 210                         /*
 211                          * If an arch bit is set in PCR, all the defined
 212                          * higher-order arch bits also have to be set.
 213                          */
 214                         pcr = PCR_ARCH_206 | PCR_ARCH_205;
 215                         break;
 216                 case PVR_ARCH_206:
 217                 case PVR_ARCH_206p:
 218                         pcr = PCR_ARCH_206;
 219                         break;
 220                 case PVR_ARCH_207:
 221                         break;
 222                 default:
 223                         return -EINVAL;
 224                 }
 225
 226                 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
 227                         /* POWER7 can't emulate POWER8 */
 228                         if (!(pcr & PCR_ARCH_206))
 229                                 return -EINVAL;
 230                         pcr &= ~PCR_ARCH_206;
 231                 }
 232         }
 233
 234         spin_lock(&vc->lock);
 235         vc->arch_compat = arch_compat;
 236         vc->pcr = pcr;
 237         spin_unlock(&vc->lock);
 238
 239         return 0;
 240 }
 241
 242 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 243 {
 244         int r;
 245
 246         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 247         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 248                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
 249         for (r = 0; r < 16; ++r)
 250                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 251                        r, kvmppc_get_gpr(vcpu, r),
 252                        r+16, kvmppc_get_gpr(vcpu, r+16));
 253         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 254                vcpu->arch.ctr, vcpu->arch.lr);
 255         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 256                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 257         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 258                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 259         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 260                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 261         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 262                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 263         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 264         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 265                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 266         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 267         for (r = 0; r < vcpu->arch.slb_max; ++r)
 268                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 269                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 270         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 271                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 272                vcpu->arch.last_inst);
 273 }
 274
 275 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 276 {
 277         int r;
 278         struct kvm_vcpu *v, *ret = NULL;
 279
 280         mutex_lock(&kvm->lock);
 281         kvm_for_each_vcpu(r, v, kvm) {
 282                 if (v->vcpu_id == id) {
 283                         ret = v;
 284                         break;
 285                 }
 286         }
 287         mutex_unlock(&kvm->lock);
 288         return ret;
 289 }
 290
 291 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 292 {
 293         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 294         vpa->yield_count = cpu_to_be32(1);
 295 }
 296
 297 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 298                    unsigned long addr, unsigned long len)
 299 {
 300         /* check address is cacheline aligned */
 301         if (addr & (L1_CACHE_BYTES - 1))
 302                 return -EINVAL;
 303         spin_lock(&vcpu->arch.vpa_update_lock);
 304         if (v->next_gpa != addr || v->len != len) {
 305                 v->next_gpa = addr;
 306                 v->len = addr ? len : 0;
 307                 v->update_pending = 1;
 308         }
 309         spin_unlock(&vcpu->arch.vpa_update_lock);
 310         return 0;
 311 }
 312
 313 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 314 struct reg_vpa {
 315         u32 dummy;
 316         union {
 317                 __be16 hword;
 318                 __be32 word;
 319         } length;
 320 };
 321
 322 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 323 {
 324         if (vpap->update_pending)
 325                 return vpap->next_gpa != 0;
 326         return vpap->pinned_addr != NULL;
 327 }
 328
 329 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 330                                        unsigned long flags,
 331                                        unsigned long vcpuid, unsigned long vpa)
 332 {
 333         struct kvm *kvm = vcpu->kvm;
 334         unsigned long len, nb;
 335         void *va;
 336         struct kvm_vcpu *tvcpu;
 337         int err;
 338         int subfunc;
 339         struct kvmppc_vpa *vpap;
 340
 341         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 342         if (!tvcpu)
 343                 return H_PARAMETER;
 344
 345         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 346         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 347             subfunc == H_VPA_REG_SLB) {
 348                 /* Registering new area - address must be cache-line aligned */
 349                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 350                         return H_PARAMETER;
 351
 352                 /* convert logical addr to kernel addr and read length */
 353                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 354                 if (va == NULL)
 355                         return H_PARAMETER;
 356                 if (subfunc == H_VPA_REG_VPA)
 357                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 358                 else
 359                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 360                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 361
 362                 /* Check length */
 363                 if (len > nb || len < sizeof(struct reg_vpa))
 364                         return H_PARAMETER;
 365         } else {
 366                 vpa = 0;
 367                 len = 0;
 368         }
 369
 370         err = H_PARAMETER;
 371         vpap = NULL;
 372         spin_lock(&tvcpu->arch.vpa_update_lock);
 373
 374         switch (subfunc) {
 375         case H_VPA_REG_VPA:             /* register VPA */
 376                 if (len < sizeof(struct lppaca))
 377                         break;
 378                 vpap = &tvcpu->arch.vpa;
 379                 err = 0;
 380                 break;
 381
 382         case H_VPA_REG_DTL:             /* register DTL */
 383                 if (len < sizeof(struct dtl_entry))
 384                         break;
 385                 len -= len % sizeof(struct dtl_entry);
 386
 387                 /* Check that they have previously registered a VPA */
 388                 err = H_RESOURCE;
 389                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 390                         break;
 391
 392                 vpap = &tvcpu->arch.dtl;
 393                 err = 0;
 394                 break;
 395
 396         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 397                 /* Check that they have previously registered a VPA */
 398                 err = H_RESOURCE;
 399                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 400                         break;
 401
 402                 vpap = &tvcpu->arch.slb_shadow;
 403                 err = 0;
 404                 break;
 405
 406         case H_VPA_DEREG_VPA:           /* deregister VPA */
 407                 /* Check they don't still have a DTL or SLB buf registered */
 408                 err = H_RESOURCE;
 409                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 410                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 411                         break;
 412
 413                 vpap = &tvcpu->arch.vpa;
 414                 err = 0;
 415                 break;
 416
 417         case H_VPA_DEREG_DTL:           /* deregister DTL */
 418                 vpap = &tvcpu->arch.dtl;
 419                 err = 0;
 420                 break;
 421
 422         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 423                 vpap = &tvcpu->arch.slb_shadow;
 424                 err = 0;
 425                 break;
 426         }
 427
 428         if (vpap) {
 429                 vpap->next_gpa = vpa;
 430                 vpap->len = len;
 431                 vpap->update_pending = 1;
 432         }
 433
 434         spin_unlock(&tvcpu->arch.vpa_update_lock);
 435
 436         return err;
 437 }
 438
 439 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 440 {
 441         struct kvm *kvm = vcpu->kvm;
 442         void *va;
 443         unsigned long nb;
 444         unsigned long gpa;
 445
 446         /*
 447          * We need to pin the page pointed to by vpap->next_gpa,
 448          * but we can't call kvmppc_pin_guest_page under the lock
 449          * as it does get_user_pages() and down_read().  So we
 450          * have to drop the lock, pin the page, then get the lock
 451          * again and check that a new area didn't get registered
 452          * in the meantime.
 453          */
 454         for (;;) {
 455                 gpa = vpap->next_gpa;
 456                 spin_unlock(&vcpu->arch.vpa_update_lock);
 457                 va = NULL;
 458                 nb = 0;
 459                 if (gpa)
 460                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 461                 spin_lock(&vcpu->arch.vpa_update_lock);
 462                 if (gpa == vpap->next_gpa)
 463                         break;
 464                 /* sigh... unpin that one and try again */
 465                 if (va)
 466                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 467         }
 468
 469         vpap->update_pending = 0;
 470         if (va && nb < vpap->len) {
 471                 /*
 472                  * If it's now too short, it must be that userspace
 473                  * has changed the mappings underlying guest memory,
 474                  * so unregister the region.
 475                  */
 476                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 477                 va = NULL;
 478         }
 479         if (vpap->pinned_addr)
 480                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 481                                         vpap->dirty);
 482         vpap->gpa = gpa;
 483         vpap->pinned_addr = va;
 484         vpap->dirty = false;
 485         if (va)
 486                 vpap->pinned_end = va + vpap->len;
 487 }
 488
 489 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 490 {
 491         if (!(vcpu->arch.vpa.update_pending ||
 492               vcpu->arch.slb_shadow.update_pending ||
 493               vcpu->arch.dtl.update_pending))
 494                 return;
 495
 496         spin_lock(&vcpu->arch.vpa_update_lock);
 497         if (vcpu->arch.vpa.update_pending) {
 498                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 499                 if (vcpu->arch.vpa.pinned_addr)
 500                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 501         }
 502         if (vcpu->arch.dtl.update_pending) {
 503                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 504                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 505                 vcpu->arch.dtl_index = 0;
 506         }
 507         if (vcpu->arch.slb_shadow.update_pending)
 508                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 509         spin_unlock(&vcpu->arch.vpa_update_lock);
 510 }
 511
 512 /*
 513  * Return the accumulated stolen time for the vcore up until `now'.
 514  * The caller should hold the vcore lock.
 515  */
 516 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 517 {
 518         u64 p;
 519         unsigned long flags;
 520
 521         spin_lock_irqsave(&vc->stoltb_lock, flags);
 522         p = vc->stolen_tb;
 523         if (vc->vcore_state != VCORE_INACTIVE &&
 524             vc->preempt_tb != TB_NIL)
 525                 p += now - vc->preempt_tb;
 526         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 527         return p;
 528 }
 529
 530 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 531                                     struct kvmppc_vcore *vc)
 532 {
 533         struct dtl_entry *dt;
 534         struct lppaca *vpa;
 535         unsigned long stolen;
 536         unsigned long core_stolen;
 537         u64 now;
 538
 539         dt = vcpu->arch.dtl_ptr;
 540         vpa = vcpu->arch.vpa.pinned_addr;
 541         now = mftb();
 542         core_stolen = vcore_stolen_time(vc, now);
 543         stolen = core_stolen - vcpu->arch.stolen_logged;
 544         vcpu->arch.stolen_logged = core_stolen;
 545         spin_lock_irq(&vcpu->arch.tbacct_lock);
 546         stolen += vcpu->arch.busy_stolen;
 547         vcpu->arch.busy_stolen = 0;
 548         spin_unlock_irq(&vcpu->arch.tbacct_lock);
 549         if (!dt || !vpa)
 550                 return;
 551         memset(dt, 0, sizeof(struct dtl_entry));
 552         dt->dispatch_reason = 7;
 553         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 554         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 555         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 556         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 557         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 558         ++dt;
 559         if (dt == vcpu->arch.dtl.pinned_end)
 560                 dt = vcpu->arch.dtl.pinned_addr;
 561         vcpu->arch.dtl_ptr = dt;
 562         /* order writing *dt vs. writing vpa->dtl_idx */
 563         smp_wmb();
 564         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 565         vcpu->arch.dtl.dirty = true;
 566 }
 567
 568 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 569 {
 570         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 571                 return true;
 572         if ((!vcpu->arch.vcore->arch_compat) &&
 573             cpu_has_feature(CPU_FTR_ARCH_207S))
 574                 return true;
 575         return false;
 576 }
 577
 578 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 579                              unsigned long resource, unsigned long value1,
 580                              unsigned long value2)
 581 {
 582         switch (resource) {
 583         case H_SET_MODE_RESOURCE_SET_CIABR:
 584                 if (!kvmppc_power8_compatible(vcpu))
 585                         return H_P2;
 586                 if (value2)
 587                         return H_P4;
 588                 if (mflags)
 589                         return H_UNSUPPORTED_FLAG_START;
 590                 /* Guests can't breakpoint the hypervisor */
 591                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 592                         return H_P3;
 593                 vcpu->arch.ciabr  = value1;
 594                 return H_SUCCESS;
 595         case H_SET_MODE_RESOURCE_SET_DAWR:
 596                 if (!kvmppc_power8_compatible(vcpu))
 597                         return H_P2;
 598                 if (mflags)
 599                         return H_UNSUPPORTED_FLAG_START;
 600                 if (value2 & DABRX_HYP)
 601                         return H_P4;
 602                 vcpu->arch.dawr  = value1;
 603                 vcpu->arch.dawrx = value2;
 604                 return H_SUCCESS;
 605         default:
 606                 return H_TOO_HARD;
 607         }
 608 }
 609
 610 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 611 {
 612         struct kvmppc_vcore *vcore = target->arch.vcore;
 613
 614         /*
 615          * We expect to have been called by the real mode handler
 616          * (kvmppc_rm_h_confer()) which would have directly returned
 617          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 618          * have useful work to do and should not confer) so we don't
 619          * recheck that here.
 620          */
 621
 622         spin_lock(&vcore->lock);
 623         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 624             vcore->vcore_state != VCORE_INACTIVE)
 625                 target = vcore->runner;
 626         spin_unlock(&vcore->lock);
 627
 628         return kvm_vcpu_yield_to(target);
 629 }
 630
 631 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 632 {
 633         int yield_count = 0;
 634         struct lppaca *lppaca;
 635
 636         spin_lock(&vcpu->arch.vpa_update_lock);
 637         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 638         if (lppaca)
 639                 yield_count = lppaca->yield_count;
 640         spin_unlock(&vcpu->arch.vpa_update_lock);
 641         return yield_count;
 642 }
 643
 644 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 645 {
 646         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 647         unsigned long target, ret = H_SUCCESS;
 648         int yield_count;
 649         struct kvm_vcpu *tvcpu;
 650         int idx, rc;
 651
 652         if (req <= MAX_HCALL_OPCODE &&
 653             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 654                 return RESUME_HOST;
 655
 656         switch (req) {
 657         case H_CEDE:
 658                 break;
 659         case H_PROD:
 660                 target = kvmppc_get_gpr(vcpu, 4);
 661                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 662                 if (!tvcpu) {
 663                         ret = H_PARAMETER;
 664                         break;
 665                 }
 666                 tvcpu->arch.prodded = 1;
 667                 smp_mb();
 668                 if (vcpu->arch.ceded) {
 669                         if (waitqueue_active(&vcpu->wq)) {
 670                                 wake_up_interruptible(&vcpu->wq);
 671                                 vcpu->stat.halt_wakeup++;
 672                         }
 673                 }
 674                 break;
 675         case H_CONFER:
 676                 target = kvmppc_get_gpr(vcpu, 4);
 677                 if (target == -1)
 678                         break;
 679                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 680                 if (!tvcpu) {
 681                         ret = H_PARAMETER;
 682                         break;
 683                 }
 684                 yield_count = kvmppc_get_gpr(vcpu, 5);
 685                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 686                         break;
 687                 kvm_arch_vcpu_yield_to(tvcpu);
 688                 break;
 689         case H_REGISTER_VPA:
 690                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 691                                         kvmppc_get_gpr(vcpu, 5),
 692                                         kvmppc_get_gpr(vcpu, 6));
 693                 break;
 694         case H_RTAS:
 695                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 696                         return RESUME_HOST;
 697
 698                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 699                 rc = kvmppc_rtas_hcall(vcpu);
 700                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 701
 702                 if (rc == -ENOENT)
 703                         return RESUME_HOST;
 704                 else if (rc == 0)
 705                         break;
 706
 707                 /* Send the error out to userspace via KVM_RUN */
 708                 return rc;
 709         case H_SET_MODE:
 710                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 711                                         kvmppc_get_gpr(vcpu, 5),
 712                                         kvmppc_get_gpr(vcpu, 6),
 713                                         kvmppc_get_gpr(vcpu, 7));
 714                 if (ret == H_TOO_HARD)
 715                         return RESUME_HOST;
 716                 break;
 717         case H_XIRR:
 718         case H_CPPR:
 719         case H_EOI:
 720         case H_IPI:
 721         case H_IPOLL:
 722         case H_XIRR_X:
 723                 if (kvmppc_xics_enabled(vcpu)) {
 724                         ret = kvmppc_xics_hcall(vcpu, req);
 725                         break;
 726                 } /* fallthrough */
 727         default:
 728                 return RESUME_HOST;
 729         }
 730         kvmppc_set_gpr(vcpu, 3, ret);
 731         vcpu->arch.hcall_needed = 0;
 732         return RESUME_GUEST;
 733 }
 734
 735 static int kvmppc_hcall_impl_hv(unsigned long cmd)
 736 {
 737         switch (cmd) {
 738         case H_CEDE:
 739         case H_PROD:
 740         case H_CONFER:
 741         case H_REGISTER_VPA:
 742         case H_SET_MODE:
 743 #ifdef CONFIG_KVM_XICS
 744         case H_XIRR:
 745         case H_CPPR:
 746         case H_EOI:
 747         case H_IPI:
 748         case H_IPOLL:
 749         case H_XIRR_X:
 750 #endif
 751                 return 1;
 752         }
 753
 754         /* See if it's in the real-mode table */
 755         return kvmppc_hcall_impl_hv_realmode(cmd);
 756 }
 757
 758 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
 759                                         struct kvm_vcpu *vcpu)
 760 {
 761         u32 last_inst;
 762
 763         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
 764                                         EMULATE_DONE) {
 765                 /*
 766                  * Fetch failed, so return to guest and
 767                  * try executing it again.
 768                  */
 769                 return RESUME_GUEST;
 770         }
 771
 772         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
 773                 run->exit_reason = KVM_EXIT_DEBUG;
 774                 run->debug.arch.address = kvmppc_get_pc(vcpu);
 775                 return RESUME_HOST;
 776         } else {
 777                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 778                 return RESUME_GUEST;
 779         }
 780 }
 781
 782 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
 783                                  struct task_struct *tsk)
 784 {
 785         int r = RESUME_HOST;
 786
 787         vcpu->stat.sum_exits++;
 788
 789         run->exit_reason = KVM_EXIT_UNKNOWN;
 790         run->ready_for_interrupt_injection = 1;
 791         switch (vcpu->arch.trap) {
 792         /* We're good on these - the host merely wanted to get our attention */
 793         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 794                 vcpu->stat.dec_exits++;
 795                 r = RESUME_GUEST;
 796                 break;
 797         case BOOK3S_INTERRUPT_EXTERNAL:
 798         case BOOK3S_INTERRUPT_H_DOORBELL:
 799                 vcpu->stat.ext_intr_exits++;
 800                 r = RESUME_GUEST;
 801                 break;
 802         /* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
 803         case BOOK3S_INTERRUPT_HMI:
 804         case BOOK3S_INTERRUPT_PERFMON:
 805                 r = RESUME_GUEST;
 806                 break;
 807         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 808                 /*
 809                  * Deliver a machine check interrupt to the guest.
 810                  * We have to do this, even if the host has handled the
 811                  * machine check, because machine checks use SRR0/1 and
 812                  * the interrupt might have trashed guest state in them.
 813                  */
 814                 kvmppc_book3s_queue_irqprio(vcpu,
 815                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 816                 r = RESUME_GUEST;
 817                 break;
 818         case BOOK3S_INTERRUPT_PROGRAM:
 819         {
 820                 ulong flags;
 821                 /*
 822                  * Normally program interrupts are delivered directly
 823                  * to the guest by the hardware, but we can get here
 824                  * as a result of a hypervisor emulation interrupt
 825                  * (e40) getting turned into a 700 by BML RTAS.
 826                  */
 827                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 828                 kvmppc_core_queue_program(vcpu, flags);
 829                 r = RESUME_GUEST;
 830                 break;
 831         }
 832         case BOOK3S_INTERRUPT_SYSCALL:
 833         {
 834                 /* hcall - punt to userspace */
 835                 int i;
 836
 837                 /* hypercall with MSR_PR has already been handled in rmode,
 838                  * and never reaches here.
 839                  */
 840
 841                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 842                 for (i = 0; i < 9; ++i)
 843                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 844                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 845                 vcpu->arch.hcall_needed = 1;
 846                 r = RESUME_HOST;
 847                 break;
 848         }
 849         /*
 850          * We get these next two if the guest accesses a page which it thinks
 851          * it has mapped but which is not actually present, either because
 852          * it is for an emulated I/O device or because the corresonding
 853          * host page has been paged out.  Any other HDSI/HISI interrupts
 854          * have been handled already.
 855          */
 856         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 857                 r = RESUME_PAGE_FAULT;
 858                 break;
 859         case BOOK3S_INTERRUPT_H_INST_STORAGE:
 860                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
 861                 vcpu->arch.fault_dsisr = 0;
 862                 r = RESUME_PAGE_FAULT;
 863                 break;
 864         /*
 865          * This occurs if the guest executes an illegal instruction.
 866          * If the guest debug is disabled, generate a program interrupt
 867          * to the guest. If guest debug is enabled, we need to check
 868          * whether the instruction is a software breakpoint instruction.
 869          * Accordingly return to Guest or Host.
 870          */
 871         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 872                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
 873                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
 874                                 swab32(vcpu->arch.emul_inst) :
 875                                 vcpu->arch.emul_inst;
 876                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
 877                         r = kvmppc_emulate_debug_inst(run, vcpu);
 878                 } else {
 879                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 880                         r = RESUME_GUEST;
 881                 }
 882                 break;
 883         /*
 884          * This occurs if the guest (kernel or userspace), does something that
 885          * is prohibited by HFSCR.  We just generate a program interrupt to
 886          * the guest.
 887          */
 888         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
 889                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 890                 r = RESUME_GUEST;
 891                 break;
 892         default:
 893                 kvmppc_dump_regs(vcpu);
 894                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 895                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 896                         vcpu->arch.shregs.msr);
 897                 run->hw.hardware_exit_reason = vcpu->arch.trap;
 898                 r = RESUME_HOST;
 899                 break;
 900         }
 901
 902         return r;
 903 }
 904
 905 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
 906                                             struct kvm_sregs *sregs)
 907 {
 908         int i;
 909
 910         memset(sregs, 0, sizeof(struct kvm_sregs));
 911         sregs->pvr = vcpu->arch.pvr;
 912         for (i = 0; i < vcpu->arch.slb_max; i++) {
 913                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
 914                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 915         }
 916
 917         return 0;
 918 }
 919
 920 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
 921                                             struct kvm_sregs *sregs)
 922 {
 923         int i, j;
 924
 925         /* Only accept the same PVR as the host's, since we can't spoof it */
 926         if (sregs->pvr != vcpu->arch.pvr)
 927                 return -EINVAL;
 928
 929         j = 0;
 930         for (i = 0; i < vcpu->arch.slb_nr; i++) {
 931                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
 932                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
 933                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
 934                         ++j;
 935                 }
 936         }
 937         vcpu->arch.slb_max = j;
 938
 939         return 0;
 940 }
 941
 942 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
 943                 bool preserve_top32)
 944 {
 945         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 946         u64 mask;
 947
 948         spin_lock(&vc->lock);
 949         /*
 950          * If ILE (interrupt little-endian) has changed, update the
 951          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
 952          */
 953         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
 954                 struct kvm *kvm = vcpu->kvm;
 955                 struct kvm_vcpu *vcpu;
 956                 int i;
 957
 958                 mutex_lock(&kvm->lock);
 959                 kvm_for_each_vcpu(i, vcpu, kvm) {
 960                         if (vcpu->arch.vcore != vc)
 961                                 continue;
 962                         if (new_lpcr & LPCR_ILE)
 963                                 vcpu->arch.intr_msr |= MSR_LE;
 964                         else
 965                                 vcpu->arch.intr_msr &= ~MSR_LE;
 966                 }
 967                 mutex_unlock(&kvm->lock);
 968         }
 969
 970         /*
 971          * Userspace can only modify DPFD (default prefetch depth),
 972          * ILE (interrupt little-endian) and TC (translation control).
 973          * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
 974          */
 975         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
 976         if (cpu_has_feature(CPU_FTR_ARCH_207S))
 977                 mask |= LPCR_AIL;
 978
 979         /* Broken 32-bit version of LPCR must not clear top bits */
 980         if (preserve_top32)
 981                 mask &= 0xFFFFFFFF;
 982         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
 983         spin_unlock(&vc->lock);
 984 }
 985
 986 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
 987                                  union kvmppc_one_reg *val)
 988 {
 989         int r = 0;
 990         long int i;
 991
 992         switch (id) {
 993         case KVM_REG_PPC_DEBUG_INST:
 994                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
 995                 break;
 996         case KVM_REG_PPC_HIOR:
 997                 *val = get_reg_val(id, 0);
 998                 break;
 999         case KVM_REG_PPC_DABR:
1000                 *val = get_reg_val(id, vcpu->arch.dabr);
1001                 break;
1002         case KVM_REG_PPC_DABRX:
1003                 *val = get_reg_val(id, vcpu->arch.dabrx);
1004                 break;
1005         case KVM_REG_PPC_DSCR:
1006                 *val = get_reg_val(id, vcpu->arch.dscr);
1007                 break;
1008         case KVM_REG_PPC_PURR:
1009                 *val = get_reg_val(id, vcpu->arch.purr);
1010                 break;
1011         case KVM_REG_PPC_SPURR:
1012                 *val = get_reg_val(id, vcpu->arch.spurr);
1013                 break;
1014         case KVM_REG_PPC_AMR:
1015                 *val = get_reg_val(id, vcpu->arch.amr);
1016                 break;
1017         case KVM_REG_PPC_UAMOR:
1018                 *val = get_reg_val(id, vcpu->arch.uamor);
1019                 break;
1020         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1021                 i = id - KVM_REG_PPC_MMCR0;
1022                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1023                 break;
1024         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1025                 i = id - KVM_REG_PPC_PMC1;
1026                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1027                 break;
1028         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1029                 i = id - KVM_REG_PPC_SPMC1;
1030                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1031                 break;
1032         case KVM_REG_PPC_SIAR:
1033                 *val = get_reg_val(id, vcpu->arch.siar);
1034                 break;
1035         case KVM_REG_PPC_SDAR:
1036                 *val = get_reg_val(id, vcpu->arch.sdar);
1037                 break;
1038         case KVM_REG_PPC_SIER:
1039                 *val = get_reg_val(id, vcpu->arch.sier);
1040                 break;
1041         case KVM_REG_PPC_IAMR:
1042                 *val = get_reg_val(id, vcpu->arch.iamr);
1043                 break;
1044         case KVM_REG_PPC_PSPB:
1045                 *val = get_reg_val(id, vcpu->arch.pspb);
1046                 break;
1047         case KVM_REG_PPC_DPDES:
1048                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1049                 break;
1050         case KVM_REG_PPC_DAWR:
1051                 *val = get_reg_val(id, vcpu->arch.dawr);
1052                 break;
1053         case KVM_REG_PPC_DAWRX:
1054                 *val = get_reg_val(id, vcpu->arch.dawrx);
1055                 break;
1056         case KVM_REG_PPC_CIABR:
1057                 *val = get_reg_val(id, vcpu->arch.ciabr);
1058                 break;
1059         case KVM_REG_PPC_CSIGR:
1060                 *val = get_reg_val(id, vcpu->arch.csigr);
1061                 break;
1062         case KVM_REG_PPC_TACR:
1063                 *val = get_reg_val(id, vcpu->arch.tacr);
1064                 break;
1065         case KVM_REG_PPC_TCSCR:
1066                 *val = get_reg_val(id, vcpu->arch.tcscr);
1067                 break;
1068         case KVM_REG_PPC_PID:
1069                 *val = get_reg_val(id, vcpu->arch.pid);
1070                 break;
1071         case KVM_REG_PPC_ACOP:
1072                 *val = get_reg_val(id, vcpu->arch.acop);
1073                 break;
1074         case KVM_REG_PPC_WORT:
1075                 *val = get_reg_val(id, vcpu->arch.wort);
1076                 break;
1077         case KVM_REG_PPC_VPA_ADDR:
1078                 spin_lock(&vcpu->arch.vpa_update_lock);
1079                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1080                 spin_unlock(&vcpu->arch.vpa_update_lock);
1081                 break;
1082         case KVM_REG_PPC_VPA_SLB:
1083                 spin_lock(&vcpu->arch.vpa_update_lock);
1084                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1085                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1086                 spin_unlock(&vcpu->arch.vpa_update_lock);
1087                 break;
1088         case KVM_REG_PPC_VPA_DTL:
1089                 spin_lock(&vcpu->arch.vpa_update_lock);
1090                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1091                 val->vpaval.length = vcpu->arch.dtl.len;
1092                 spin_unlock(&vcpu->arch.vpa_update_lock);
1093                 break;
1094         case KVM_REG_PPC_TB_OFFSET:
1095                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1096                 break;
1097         case KVM_REG_PPC_LPCR:
1098         case KVM_REG_PPC_LPCR_64:
1099                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1100                 break;
1101         case KVM_REG_PPC_PPR:
1102                 *val = get_reg_val(id, vcpu->arch.ppr);
1103                 break;
1104 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1105         case KVM_REG_PPC_TFHAR:
1106                 *val = get_reg_val(id, vcpu->arch.tfhar);
1107                 break;
1108         case KVM_REG_PPC_TFIAR:
1109                 *val = get_reg_val(id, vcpu->arch.tfiar);
1110                 break;
1111         case KVM_REG_PPC_TEXASR:
1112                 *val = get_reg_val(id, vcpu->arch.texasr);
1113                 break;
1114         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1115                 i = id - KVM_REG_PPC_TM_GPR0;
1116                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1117                 break;
1118         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1119         {
1120                 int j;
1121                 i = id - KVM_REG_PPC_TM_VSR0;
1122                 if (i < 32)
1123                         for (j = 0; j < TS_FPRWIDTH; j++)
1124                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1125                 else {
1126                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1127                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1128                         else
1129                                 r = -ENXIO;
1130                 }
1131                 break;
1132         }
1133         case KVM_REG_PPC_TM_CR:
1134                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1135                 break;
1136         case KVM_REG_PPC_TM_LR:
1137                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1138                 break;
1139         case KVM_REG_PPC_TM_CTR:
1140                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1141                 break;
1142         case KVM_REG_PPC_TM_FPSCR:
1143                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1144                 break;
1145         case KVM_REG_PPC_TM_AMR:
1146                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1147                 break;
1148         case KVM_REG_PPC_TM_PPR:
1149                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1150                 break;
1151         case KVM_REG_PPC_TM_VRSAVE:
1152                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1153                 break;
1154         case KVM_REG_PPC_TM_VSCR:
1155                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1156                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1157                 else
1158                         r = -ENXIO;
1159                 break;
1160         case KVM_REG_PPC_TM_DSCR:
1161                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1162                 break;
1163         case KVM_REG_PPC_TM_TAR:
1164                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1165                 break;
1166 #endif
1167         case KVM_REG_PPC_ARCH_COMPAT:
1168                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1169                 break;
1170         default:
1171                 r = -EINVAL;
1172                 break;
1173         }
1174
1175         return r;
1176 }
1177
1178 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1179                                  union kvmppc_one_reg *val)
1180 {
1181         int r = 0;
1182         long int i;
1183         unsigned long addr, len;
1184
1185         switch (id) {
1186         case KVM_REG_PPC_HIOR:
1187                 /* Only allow this to be set to zero */
1188                 if (set_reg_val(id, *val))
1189                         r = -EINVAL;
1190                 break;
1191         case KVM_REG_PPC_DABR:
1192                 vcpu->arch.dabr = set_reg_val(id, *val);
1193                 break;
1194         case KVM_REG_PPC_DABRX:
1195                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1196                 break;
1197         case KVM_REG_PPC_DSCR:
1198                 vcpu->arch.dscr = set_reg_val(id, *val);
1199                 break;
1200         case KVM_REG_PPC_PURR:
1201                 vcpu->arch.purr = set_reg_val(id, *val);
1202                 break;
1203         case KVM_REG_PPC_SPURR:
1204                 vcpu->arch.spurr = set_reg_val(id, *val);
1205                 break;
1206         case KVM_REG_PPC_AMR:
1207                 vcpu->arch.amr = set_reg_val(id, *val);
1208                 break;
1209         case KVM_REG_PPC_UAMOR:
1210                 vcpu->arch.uamor = set_reg_val(id, *val);
1211                 break;
1212         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1213                 i = id - KVM_REG_PPC_MMCR0;
1214                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1215                 break;
1216         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1217                 i = id - KVM_REG_PPC_PMC1;
1218                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1219                 break;
1220         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1221                 i = id - KVM_REG_PPC_SPMC1;
1222                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1223                 break;
1224         case KVM_REG_PPC_SIAR:
1225                 vcpu->arch.siar = set_reg_val(id, *val);
1226                 break;
1227         case KVM_REG_PPC_SDAR:
1228                 vcpu->arch.sdar = set_reg_val(id, *val);
1229                 break;
1230         case KVM_REG_PPC_SIER:
1231                 vcpu->arch.sier = set_reg_val(id, *val);
1232                 break;
1233         case KVM_REG_PPC_IAMR:
1234                 vcpu->arch.iamr = set_reg_val(id, *val);
1235                 break;
1236         case KVM_REG_PPC_PSPB:
1237                 vcpu->arch.pspb = set_reg_val(id, *val);
1238                 break;
1239         case KVM_REG_PPC_DPDES:
1240                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1241                 break;
1242         case KVM_REG_PPC_DAWR:
1243                 vcpu->arch.dawr = set_reg_val(id, *val);
1244                 break;
1245         case KVM_REG_PPC_DAWRX:
1246                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1247                 break;
1248         case KVM_REG_PPC_CIABR:
1249                 vcpu->arch.ciabr = set_reg_val(id, *val);
1250                 /* Don't allow setting breakpoints in hypervisor code */
1251                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1252                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1253                 break;
1254         case KVM_REG_PPC_CSIGR:
1255                 vcpu->arch.csigr = set_reg_val(id, *val);
1256                 break;
1257         case KVM_REG_PPC_TACR:
1258                 vcpu->arch.tacr = set_reg_val(id, *val);
1259                 break;
1260         case KVM_REG_PPC_TCSCR:
1261                 vcpu->arch.tcscr = set_reg_val(id, *val);
1262                 break;
1263         case KVM_REG_PPC_PID:
1264                 vcpu->arch.pid = set_reg_val(id, *val);
1265                 break;
1266         case KVM_REG_PPC_ACOP:
1267                 vcpu->arch.acop = set_reg_val(id, *val);
1268                 break;
1269         case KVM_REG_PPC_WORT:
1270                 vcpu->arch.wort = set_reg_val(id, *val);
1271                 break;
1272         case KVM_REG_PPC_VPA_ADDR:
1273                 addr = set_reg_val(id, *val);
1274                 r = -EINVAL;
1275                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1276                               vcpu->arch.dtl.next_gpa))
1277                         break;
1278                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1279                 break;
1280         case KVM_REG_PPC_VPA_SLB:
1281                 addr = val->vpaval.addr;
1282                 len = val->vpaval.length;
1283                 r = -EINVAL;
1284                 if (addr && !vcpu->arch.vpa.next_gpa)
1285                         break;
1286                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1287                 break;
1288         case KVM_REG_PPC_VPA_DTL:
1289                 addr = val->vpaval.addr;
1290                 len = val->vpaval.length;
1291                 r = -EINVAL;
1292                 if (addr && (len < sizeof(struct dtl_entry) ||
1293                              !vcpu->arch.vpa.next_gpa))
1294                         break;
1295                 len -= len % sizeof(struct dtl_entry);
1296                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1297                 break;
1298         case KVM_REG_PPC_TB_OFFSET:
1299                 /* round up to multiple of 2^24 */
1300                 vcpu->arch.vcore->tb_offset =
1301                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1302                 break;
1303         case KVM_REG_PPC_LPCR:
1304                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1305                 break;
1306         case KVM_REG_PPC_LPCR_64:
1307                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1308                 break;
1309         case KVM_REG_PPC_PPR:
1310                 vcpu->arch.ppr = set_reg_val(id, *val);
1311                 break;
1312 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1313         case KVM_REG_PPC_TFHAR:
1314                 vcpu->arch.tfhar = set_reg_val(id, *val);
1315                 break;
1316         case KVM_REG_PPC_TFIAR:
1317                 vcpu->arch.tfiar = set_reg_val(id, *val);
1318                 break;
1319         case KVM_REG_PPC_TEXASR:
1320                 vcpu->arch.texasr = set_reg_val(id, *val);
1321                 break;
1322         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1323                 i = id - KVM_REG_PPC_TM_GPR0;
1324                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1325                 break;
1326         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1327         {
1328                 int j;
1329                 i = id - KVM_REG_PPC_TM_VSR0;
1330                 if (i < 32)
1331                         for (j = 0; j < TS_FPRWIDTH; j++)
1332                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1333                 else
1334                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1335                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1336                         else
1337                                 r = -ENXIO;
1338                 break;
1339         }
1340         case KVM_REG_PPC_TM_CR:
1341                 vcpu->arch.cr_tm = set_reg_val(id, *val);
1342                 break;
1343         case KVM_REG_PPC_TM_LR:
1344                 vcpu->arch.lr_tm = set_reg_val(id, *val);
1345                 break;
1346         case KVM_REG_PPC_TM_CTR:
1347                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1348                 break;
1349         case KVM_REG_PPC_TM_FPSCR:
1350                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1351                 break;
1352         case KVM_REG_PPC_TM_AMR:
1353                 vcpu->arch.amr_tm = set_reg_val(id, *val);
1354                 break;
1355         case KVM_REG_PPC_TM_PPR:
1356                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1357                 break;
1358         case KVM_REG_PPC_TM_VRSAVE:
1359                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1360                 break;
1361         case KVM_REG_PPC_TM_VSCR:
1362                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1363                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1364                 else
1365                         r = - ENXIO;
1366                 break;
1367         case KVM_REG_PPC_TM_DSCR:
1368                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1369                 break;
1370         case KVM_REG_PPC_TM_TAR:
1371                 vcpu->arch.tar_tm = set_reg_val(id, *val);
1372                 break;
1373 #endif
1374         case KVM_REG_PPC_ARCH_COMPAT:
1375                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1376                 break;
1377         default:
1378                 r = -EINVAL;
1379                 break;
1380         }
1381
1382         return r;
1383 }
1384
1385 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1386 {
1387         struct kvmppc_vcore *vcore;
1388
1389         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1390
1391         if (vcore == NULL)
1392                 return NULL;
1393
1394         INIT_LIST_HEAD(&vcore->runnable_threads);
1395         spin_lock_init(&vcore->lock);
1396         spin_lock_init(&vcore->stoltb_lock);
1397         init_waitqueue_head(&vcore->wq);
1398         vcore->preempt_tb = TB_NIL;
1399         vcore->lpcr = kvm->arch.lpcr;
1400         vcore->first_vcpuid = core * threads_per_subcore;
1401         vcore->kvm = kvm;
1402
1403         vcore->mpp_buffer_is_valid = false;
1404
1405         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1406                 vcore->mpp_buffer = (void *)__get_free_pages(
1407                         GFP_KERNEL|__GFP_ZERO,
1408                         MPP_BUFFER_ORDER);
1409
1410         return vcore;
1411 }
1412
1413 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1414                                                    unsigned int id)
1415 {
1416         struct kvm_vcpu *vcpu;
1417         int err = -EINVAL;
1418         int core;
1419         struct kvmppc_vcore *vcore;
1420
1421         core = id / threads_per_subcore;
1422         if (core >= KVM_MAX_VCORES)
1423                 goto out;
1424
1425         err = -ENOMEM;
1426         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1427         if (!vcpu)
1428                 goto out;
1429
1430         err = kvm_vcpu_init(vcpu, kvm, id);
1431         if (err)
1432                 goto free_vcpu;
1433
1434         vcpu->arch.shared = &vcpu->arch.shregs;
1435 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1436         /*
1437          * The shared struct is never shared on HV,
1438          * so we can always use host endianness
1439          */
1440 #ifdef __BIG_ENDIAN__
1441         vcpu->arch.shared_big_endian = true;
1442 #else
1443         vcpu->arch.shared_big_endian = false;
1444 #endif
1445 #endif
1446         vcpu->arch.mmcr[0] = MMCR0_FC;
1447         vcpu->arch.ctrl = CTRL_RUNLATCH;
1448         /* default to host PVR, since we can't spoof it */
1449         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1450         spin_lock_init(&vcpu->arch.vpa_update_lock);
1451         spin_lock_init(&vcpu->arch.tbacct_lock);
1452         vcpu->arch.busy_preempt = TB_NIL;
1453         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1454
1455         kvmppc_mmu_book3s_hv_init(vcpu);
1456
1457         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1458
1459         init_waitqueue_head(&vcpu->arch.cpu_run);
1460
1461         mutex_lock(&kvm->lock);
1462         vcore = kvm->arch.vcores[core];
1463         if (!vcore) {
1464                 vcore = kvmppc_vcore_create(kvm, core);
1465                 kvm->arch.vcores[core] = vcore;
1466                 kvm->arch.online_vcores++;
1467         }
1468         mutex_unlock(&kvm->lock);
1469
1470         if (!vcore)
1471                 goto free_vcpu;
1472
1473         spin_lock(&vcore->lock);
1474         ++vcore->num_threads;
1475         spin_unlock(&vcore->lock);
1476         vcpu->arch.vcore = vcore;
1477         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1478
1479         vcpu->arch.cpu_type = KVM_CPU_3S_64;
1480         kvmppc_sanity_check(vcpu);
1481
1482         return vcpu;
1483
1484 free_vcpu:
1485         kmem_cache_free(kvm_vcpu_cache, vcpu);
1486 out:
1487         return ERR_PTR(err);
1488 }
1489
1490 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1491 {
1492         if (vpa->pinned_addr)
1493                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1494                                         vpa->dirty);
1495 }
1496
1497 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1498 {
1499         spin_lock(&vcpu->arch.vpa_update_lock);
1500         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1501         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1502         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1503         spin_unlock(&vcpu->arch.vpa_update_lock);
1504         kvm_vcpu_uninit(vcpu);
1505         kmem_cache_free(kvm_vcpu_cache, vcpu);
1506 }
1507
1508 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1509 {
1510         /* Indicate we want to get back into the guest */
1511         return 1;
1512 }
1513
1514 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1515 {
1516         unsigned long dec_nsec, now;
1517
1518         now = get_tb();
1519         if (now > vcpu->arch.dec_expires) {
1520                 /* decrementer has already gone negative */
1521                 kvmppc_core_queue_dec(vcpu);
1522                 kvmppc_core_prepare_to_enter(vcpu);
1523                 return;
1524         }
1525         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1526                    / tb_ticks_per_sec;
1527         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1528                       HRTIMER_MODE_REL);
1529         vcpu->arch.timer_running = 1;
1530 }
1531
1532 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1533 {
1534         vcpu->arch.ceded = 0;
1535         if (vcpu->arch.timer_running) {
1536                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1537                 vcpu->arch.timer_running = 0;
1538         }
1539 }
1540
1541 extern void __kvmppc_vcore_entry(void);
1542
1543 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1544                                    struct kvm_vcpu *vcpu)
1545 {
1546         u64 now;
1547
1548         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1549                 return;
1550         spin_lock_irq(&vcpu->arch.tbacct_lock);
1551         now = mftb();
1552         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1553                 vcpu->arch.stolen_logged;
1554         vcpu->arch.busy_preempt = now;
1555         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1556         spin_unlock_irq(&vcpu->arch.tbacct_lock);
1557         --vc->n_runnable;
1558         list_del(&vcpu->arch.run_list);
1559 }
1560
1561 static int kvmppc_grab_hwthread(int cpu)
1562 {
1563         struct paca_struct *tpaca;
1564         long timeout = 10000;
1565
1566         tpaca = &paca[cpu];
1567
1568         /* Ensure the thread won't go into the kernel if it wakes */
1569         tpaca->kvm_hstate.hwthread_req = 1;
1570         tpaca->kvm_hstate.kvm_vcpu = NULL;
1571
1572         /*
1573          * If the thread is already executing in the kernel (e.g. handling
1574          * a stray interrupt), wait for it to get back to nap mode.
1575          * The smp_mb() is to ensure that our setting of hwthread_req
1576          * is visible before we look at hwthread_state, so if this
1577          * races with the code at system_reset_pSeries and the thread
1578          * misses our setting of hwthread_req, we are sure to see its
1579          * setting of hwthread_state, and vice versa.
1580          */
1581         smp_mb();
1582         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1583                 if (--timeout <= 0) {
1584                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1585                         return -EBUSY;
1586                 }
1587                 udelay(1);
1588         }
1589         return 0;
1590 }
1591
1592 static void kvmppc_release_hwthread(int cpu)
1593 {
1594         struct paca_struct *tpaca;
1595
1596         tpaca = &paca[cpu];
1597         tpaca->kvm_hstate.hwthread_req = 0;
1598         tpaca->kvm_hstate.kvm_vcpu = NULL;
1599 }
1600
1601 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1602 {
1603         int cpu;
1604         struct paca_struct *tpaca;
1605         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1606
1607         if (vcpu->arch.timer_running) {
1608                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1609                 vcpu->arch.timer_running = 0;
1610         }
1611         cpu = vc->pcpu + vcpu->arch.ptid;
1612         tpaca = &paca[cpu];
1613         tpaca->kvm_hstate.kvm_vcpu = vcpu;
1614         tpaca->kvm_hstate.kvm_vcore = vc;
1615         tpaca->kvm_hstate.ptid = vcpu->arch.ptid;
1616         vcpu->cpu = vc->pcpu;
1617         smp_wmb();
1618 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1619         if (cpu != smp_processor_id()) {
1620                 xics_wake_cpu(cpu);
1621                 if (vcpu->arch.ptid)
1622                         ++vc->n_woken;
1623         }
1624 #endif
1625 }
1626
1627 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1628 {
1629         int i;
1630
1631         HMT_low();
1632         i = 0;
1633         while (vc->nap_count < vc->n_woken) {
1634                 if (++i >= 1000000) {
1635                         pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1636                                vc->nap_count, vc->n_woken);
1637                         break;
1638                 }
1639                 cpu_relax();
1640         }
1641         HMT_medium();
1642 }
1643
1644 /*
1645  * Check that we are on thread 0 and that any other threads in
1646  * this core are off-line.  Then grab the threads so they can't
1647  * enter the kernel.
1648  */
1649 static int on_primary_thread(void)
1650 {
1651         int cpu = smp_processor_id();
1652         int thr;
1653
1654         /* Are we on a primary subcore? */
1655         if (cpu_thread_in_subcore(cpu))
1656                 return 0;
1657
1658         thr = 0;
1659         while (++thr < threads_per_subcore)
1660                 if (cpu_online(cpu + thr))
1661                         return 0;
1662
1663         /* Grab all hw threads so they can't go into the kernel */
1664         for (thr = 1; thr < threads_per_subcore; ++thr) {
1665                 if (kvmppc_grab_hwthread(cpu + thr)) {
1666                         /* Couldn't grab one; let the others go */
1667                         do {
1668                                 kvmppc_release_hwthread(cpu + thr);
1669                         } while (--thr > 0);
1670                         return 0;
1671                 }
1672         }
1673         return 1;
1674 }
1675
1676 static void kvmppc_start_saving_l2_cache(struct kvmppc_vcore *vc)
1677 {
1678         phys_addr_t phy_addr, mpp_addr;
1679
1680         phy_addr = (phys_addr_t)virt_to_phys(vc->mpp_buffer);
1681         mpp_addr = phy_addr & PPC_MPPE_ADDRESS_MASK;
1682
1683         mtspr(SPRN_MPPR, mpp_addr | PPC_MPPR_FETCH_ABORT);
1684         logmpp(mpp_addr | PPC_LOGMPP_LOG_L2);
1685
1686         vc->mpp_buffer_is_valid = true;
1687 }
1688
1689 static void kvmppc_start_restoring_l2_cache(const struct kvmppc_vcore *vc)
1690 {
1691         phys_addr_t phy_addr, mpp_addr;
1692
1693         phy_addr = virt_to_phys(vc->mpp_buffer);
1694         mpp_addr = phy_addr & PPC_MPPE_ADDRESS_MASK;
1695
1696         /* We must abort any in-progress save operations to ensure
1697          * the table is valid so that prefetch engine knows when to
1698          * stop prefetching. */
1699         logmpp(mpp_addr | PPC_LOGMPP_LOG_ABORT);
1700         mtspr(SPRN_MPPR, mpp_addr | PPC_MPPR_FETCH_WHOLE_TABLE);
1701 }
1702
1703 /*
1704  * Run a set of guest threads on a physical core.
1705  * Called with vc->lock held.
1706  */
1707 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1708 {
1709         struct kvm_vcpu *vcpu, *vnext;
1710         long ret;
1711         u64 now;
1712         int i, need_vpa_update;
1713         int srcu_idx;
1714         struct kvm_vcpu *vcpus_to_update[threads_per_core];
1715
1716         /* don't start if any threads have a signal pending */
1717         need_vpa_update = 0;
1718         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1719                 if (signal_pending(vcpu->arch.run_task))
1720                         return;
1721                 if (vcpu->arch.vpa.update_pending ||
1722                     vcpu->arch.slb_shadow.update_pending ||
1723                     vcpu->arch.dtl.update_pending)
1724                         vcpus_to_update[need_vpa_update++] = vcpu;
1725         }
1726
1727         /*
1728          * Initialize *vc, in particular vc->vcore_state, so we can
1729          * drop the vcore lock if necessary.
1730          */
1731         vc->n_woken = 0;
1732         vc->nap_count = 0;
1733         vc->entry_exit_count = 0;
1734         vc->preempt_tb = TB_NIL;
1735         vc->vcore_state = VCORE_STARTING;
1736         vc->in_guest = 0;
1737         vc->napping_threads = 0;
1738         vc->conferring_threads = 0;
1739
1740         /*
1741          * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1742          * which can't be called with any spinlocks held.
1743          */
1744         if (need_vpa_update) {
1745                 spin_unlock(&vc->lock);
1746                 for (i = 0; i < need_vpa_update; ++i)
1747                         kvmppc_update_vpas(vcpus_to_update[i]);
1748                 spin_lock(&vc->lock);
1749         }
1750
1751         /*
1752          * Make sure we are running on primary threads, and that secondary
1753          * threads are offline.  Also check if the number of threads in this
1754          * guest are greater than the current system threads per guest.
1755          */
1756         if ((threads_per_core > 1) &&
1757             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
1758                 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1759                         vcpu->arch.ret = -EBUSY;
1760                 goto out;
1761         }
1762
1763
1764         vc->pcpu = smp_processor_id();
1765         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1766                 kvmppc_start_thread(vcpu);
1767                 kvmppc_create_dtl_entry(vcpu, vc);
1768                 trace_kvm_guest_enter(vcpu);
1769         }
1770
1771         /* Set this explicitly in case thread 0 doesn't have a vcpu */
1772         get_paca()->kvm_hstate.kvm_vcore = vc;
1773         get_paca()->kvm_hstate.ptid = 0;
1774
1775         vc->vcore_state = VCORE_RUNNING;
1776         preempt_disable();
1777
1778         trace_kvmppc_run_core(vc, 0);
1779
1780         spin_unlock(&vc->lock);
1781
1782         kvm_guest_enter();
1783
1784         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
1785
1786         if (vc->mpp_buffer_is_valid)
1787                 kvmppc_start_restoring_l2_cache(vc);
1788
1789         __kvmppc_vcore_entry();
1790
1791         spin_lock(&vc->lock);
1792
1793         if (vc->mpp_buffer)
1794                 kvmppc_start_saving_l2_cache(vc);
1795
1796         /* disable sending of IPIs on virtual external irqs */
1797         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1798                 vcpu->cpu = -1;
1799         /* wait for secondary threads to finish writing their state to memory */
1800         if (vc->nap_count < vc->n_woken)
1801                 kvmppc_wait_for_nap(vc);
1802         for (i = 0; i < threads_per_subcore; ++i)
1803                 kvmppc_release_hwthread(vc->pcpu + i);
1804         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
1805         vc->vcore_state = VCORE_EXITING;
1806         spin_unlock(&vc->lock);
1807
1808         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
1809
1810         /* make sure updates to secondary vcpu structs are visible now */
1811         smp_mb();
1812         kvm_guest_exit();
1813
1814         preempt_enable();
1815         cond_resched();
1816
1817         spin_lock(&vc->lock);
1818         now = get_tb();
1819         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1820                 /* cancel pending dec exception if dec is positive */
1821                 if (now < vcpu->arch.dec_expires &&
1822                     kvmppc_core_pending_dec(vcpu))
1823                         kvmppc_core_dequeue_dec(vcpu);
1824
1825                 trace_kvm_guest_exit(vcpu);
1826
1827                 ret = RESUME_GUEST;
1828                 if (vcpu->arch.trap)
1829                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1830                                                     vcpu->arch.run_task);
1831
1832                 vcpu->arch.ret = ret;
1833                 vcpu->arch.trap = 0;
1834
1835                 if (vcpu->arch.ceded) {
1836                         if (!is_kvmppc_resume_guest(ret))
1837                                 kvmppc_end_cede(vcpu);
1838                         else
1839                                 kvmppc_set_timer(vcpu);
1840                 }
1841         }
1842
1843  out:
1844         vc->vcore_state = VCORE_INACTIVE;
1845         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1846                                  arch.run_list) {
1847                 if (!is_kvmppc_resume_guest(vcpu->arch.ret)) {
1848                         kvmppc_remove_runnable(vc, vcpu);
1849                         wake_up(&vcpu->arch.cpu_run);
1850                 }
1851         }
1852
1853         trace_kvmppc_run_core(vc, 1);
1854 }
1855
1856 /*
1857  * Wait for some other vcpu thread to execute us, and
1858  * wake us up when we need to handle something in the host.
1859  */
1860 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1861 {
1862         DEFINE_WAIT(wait);
1863
1864         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1865         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1866                 schedule();
1867         finish_wait(&vcpu->arch.cpu_run, &wait);
1868 }
1869
1870 /*
1871  * All the vcpus in this vcore are idle, so wait for a decrementer
1872  * or external interrupt to one of the vcpus.  vc->lock is held.
1873  */
1874 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1875 {
1876         struct kvm_vcpu *vcpu;
1877         int do_sleep = 1;
1878
1879         DEFINE_WAIT(wait);
1880
1881         prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1882
1883         /*
1884          * Check one last time for pending exceptions and ceded state after
1885          * we put ourselves on the wait queue
1886          */
1887         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1888                 if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded) {
1889                         do_sleep = 0;
1890                         break;
1891                 }
1892         }
1893
1894         if (!do_sleep) {
1895                 finish_wait(&vc->wq, &wait);
1896                 return;
1897         }
1898
1899         vc->vcore_state = VCORE_SLEEPING;
1900         trace_kvmppc_vcore_blocked(vc, 0);
1901         spin_unlock(&vc->lock);
1902         schedule();
1903         finish_wait(&vc->wq, &wait);
1904         spin_lock(&vc->lock);
1905         vc->vcore_state = VCORE_INACTIVE;
1906         trace_kvmppc_vcore_blocked(vc, 1);
1907 }
1908
1909 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1910 {
1911         int n_ceded;
1912         struct kvmppc_vcore *vc;
1913         struct kvm_vcpu *v, *vn;
1914
1915         trace_kvmppc_run_vcpu_enter(vcpu);
1916
1917         kvm_run->exit_reason = 0;
1918         vcpu->arch.ret = RESUME_GUEST;
1919         vcpu->arch.trap = 0;
1920         kvmppc_update_vpas(vcpu);
1921
1922         /*
1923          * Synchronize with other threads in this virtual core
1924          */
1925         vc = vcpu->arch.vcore;
1926         spin_lock(&vc->lock);
1927         vcpu->arch.ceded = 0;
1928         vcpu->arch.run_task = current;
1929         vcpu->arch.kvm_run = kvm_run;
1930         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
1931         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1932         vcpu->arch.busy_preempt = TB_NIL;
1933         list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1934         ++vc->n_runnable;
1935
1936         /*
1937          * This happens the first time this is called for a vcpu.
1938          * If the vcore is already running, we may be able to start
1939          * this thread straight away and have it join in.
1940          */
1941         if (!signal_pending(current)) {
1942                 if (vc->vcore_state == VCORE_RUNNING &&
1943                     VCORE_EXIT_COUNT(vc) == 0) {
1944                         kvmppc_create_dtl_entry(vcpu, vc);
1945                         kvmppc_start_thread(vcpu);
1946                         trace_kvm_guest_enter(vcpu);
1947                 } else if (vc->vcore_state == VCORE_SLEEPING) {
1948                         wake_up(&vc->wq);
1949                 }
1950
1951         }
1952
1953         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1954                !signal_pending(current)) {
1955                 if (vc->vcore_state != VCORE_INACTIVE) {
1956                         spin_unlock(&vc->lock);
1957                         kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1958                         spin_lock(&vc->lock);
1959                         continue;
1960                 }
1961                 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1962                                          arch.run_list) {
1963                         kvmppc_core_prepare_to_enter(v);
1964                         if (signal_pending(v->arch.run_task)) {
1965                                 kvmppc_remove_runnable(vc, v);
1966                                 v->stat.signal_exits++;
1967                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1968                                 v->arch.ret = -EINTR;
1969                                 wake_up(&v->arch.cpu_run);
1970                         }
1971                 }
1972                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1973                         break;
1974                 vc->runner = vcpu;
1975                 n_ceded = 0;
1976                 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
1977                         if (!v->arch.pending_exceptions)
1978                                 n_ceded += v->arch.ceded;
1979                         else
1980                                 v->arch.ceded = 0;
1981                 }
1982                 if (n_ceded == vc->n_runnable)
1983                         kvmppc_vcore_blocked(vc);
1984                 else
1985                         kvmppc_run_core(vc);
1986                 vc->runner = NULL;
1987         }
1988
1989         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1990                (vc->vcore_state == VCORE_RUNNING ||
1991                 vc->vcore_state == VCORE_EXITING)) {
1992                 spin_unlock(&vc->lock);
1993                 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1994                 spin_lock(&vc->lock);
1995         }
1996
1997         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1998                 kvmppc_remove_runnable(vc, vcpu);
1999                 vcpu->stat.signal_exits++;
2000                 kvm_run->exit_reason = KVM_EXIT_INTR;
2001                 vcpu->arch.ret = -EINTR;
2002         }
2003
2004         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2005                 /* Wake up some vcpu to run the core */
2006                 v = list_first_entry(&vc->runnable_threads,
2007                                      struct kvm_vcpu, arch.run_list);
2008                 wake_up(&v->arch.cpu_run);
2009         }
2010
2011         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2012         spin_unlock(&vc->lock);
2013         return vcpu->arch.ret;
2014 }
2015
2016 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2017 {
2018         int r;
2019         int srcu_idx;
2020
2021         if (!vcpu->arch.sane) {
2022                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2023                 return -EINVAL;
2024         }
2025
2026         kvmppc_core_prepare_to_enter(vcpu);
2027
2028         /* No need to go into the guest when all we'll do is come back out */
2029         if (signal_pending(current)) {
2030                 run->exit_reason = KVM_EXIT_INTR;
2031                 return -EINTR;
2032         }
2033
2034         atomic_inc(&vcpu->kvm->arch.vcpus_running);
2035         /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
2036         smp_mb();
2037
2038         /* On the first time here, set up HTAB and VRMA */
2039         if (!vcpu->kvm->arch.rma_setup_done) {
2040                 r = kvmppc_hv_setup_htab_rma(vcpu);
2041                 if (r)
2042                         goto out;
2043         }
2044
2045         flush_fp_to_thread(current);
2046         flush_altivec_to_thread(current);
2047         flush_vsx_to_thread(current);
2048         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2049         vcpu->arch.pgdir = current->mm->pgd;
2050         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2051
2052         do {
2053                 r = kvmppc_run_vcpu(run, vcpu);
2054
2055                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2056                     !(vcpu->arch.shregs.msr & MSR_PR)) {
2057                         trace_kvm_hcall_enter(vcpu);
2058                         r = kvmppc_pseries_do_hcall(vcpu);
2059                         trace_kvm_hcall_exit(vcpu, r);
2060                         kvmppc_core_prepare_to_enter(vcpu);
2061                 } else if (r == RESUME_PAGE_FAULT) {
2062                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2063                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
2064                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2065                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2066                 }
2067         } while (is_kvmppc_resume_guest(r));
2068
2069  out:
2070         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2071         atomic_dec(&vcpu->kvm->arch.vcpus_running);
2072         return r;
2073 }
2074
2075 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2076                                      int linux_psize)
2077 {
2078         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2079
2080         if (!def->shift)
2081                 return;
2082         (*sps)->page_shift = def->shift;
2083         (*sps)->slb_enc = def->sllp;
2084         (*sps)->enc[0].page_shift = def->shift;
2085         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
2086         /*
2087          * Add 16MB MPSS support if host supports it
2088          */
2089         if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2090                 (*sps)->enc[1].page_shift = 24;
2091                 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2092         }
2093         (*sps)++;
2094 }
2095
2096 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2097                                          struct kvm_ppc_smmu_info *info)
2098 {
2099         struct kvm_ppc_one_seg_page_size *sps;
2100
2101         info->flags = KVM_PPC_PAGE_SIZES_REAL;
2102         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2103                 info->flags |= KVM_PPC_1T_SEGMENTS;
2104         info->slb_size = mmu_slb_size;
2105
2106         /* We only support these sizes for now, and no muti-size segments */
2107         sps = &info->sps[0];
2108         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2109         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2110         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2111
2112         return 0;
2113 }
2114
2115 /*
2116  * Get (and clear) the dirty memory log for a memory slot.
2117  */
2118 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2119                                          struct kvm_dirty_log *log)
2120 {
2121         struct kvm_memory_slot *memslot;
2122         int r;
2123         unsigned long n;
2124
2125         mutex_lock(&kvm->slots_lock);
2126
2127         r = -EINVAL;
2128         if (log->slot >= KVM_USER_MEM_SLOTS)
2129                 goto out;
2130
2131         memslot = id_to_memslot(kvm->memslots, log->slot);
2132         r = -ENOENT;
2133         if (!memslot->dirty_bitmap)
2134                 goto out;
2135
2136         n = kvm_dirty_bitmap_bytes(memslot);
2137         memset(memslot->dirty_bitmap, 0, n);
2138
2139         r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2140         if (r)
2141                 goto out;
2142
2143         r = -EFAULT;
2144         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2145                 goto out;
2146
2147         r = 0;
2148 out:
2149         mutex_unlock(&kvm->slots_lock);
2150         return r;
2151 }
2152
2153 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2154                                         struct kvm_memory_slot *dont)
2155 {
2156         if (!dont || free->arch.rmap != dont->arch.rmap) {
2157                 vfree(free->arch.rmap);
2158                 free->arch.rmap = NULL;
2159         }
2160 }
2161
2162 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2163                                          unsigned long npages)
2164 {
2165         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2166         if (!slot->arch.rmap)
2167                 return -ENOMEM;
2168
2169         return 0;
2170 }
2171
2172 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2173                                         struct kvm_memory_slot *memslot,
2174                                         struct kvm_userspace_memory_region *mem)
2175 {
2176         return 0;
2177 }
2178
2179 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2180                                 struct kvm_userspace_memory_region *mem,
2181                                 const struct kvm_memory_slot *old)
2182 {
2183         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2184         struct kvm_memory_slot *memslot;
2185
2186         if (npages && old->npages) {
2187                 /*
2188                  * If modifying a memslot, reset all the rmap dirty bits.
2189                  * If this is a new memslot, we don't need to do anything
2190                  * since the rmap array starts out as all zeroes,
2191                  * i.e. no pages are dirty.
2192                  */
2193                 memslot = id_to_memslot(kvm->memslots, mem->slot);
2194                 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2195         }
2196 }
2197
2198 /*
2199  * Update LPCR values in kvm->arch and in vcores.
2200  * Caller must hold kvm->lock.
2201  */
2202 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2203 {
2204         long int i;
2205         u32 cores_done = 0;
2206
2207         if ((kvm->arch.lpcr & mask) == lpcr)
2208                 return;
2209
2210         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2211
2212         for (i = 0; i < KVM_MAX_VCORES; ++i) {
2213                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2214                 if (!vc)
2215                         continue;
2216                 spin_lock(&vc->lock);
2217                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2218                 spin_unlock(&vc->lock);
2219                 if (++cores_done >= kvm->arch.online_vcores)
2220                         break;
2221         }
2222 }
2223
2224 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2225 {
2226         return;
2227 }
2228
2229 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2230 {
2231         int err = 0;
2232         struct kvm *kvm = vcpu->kvm;
2233         unsigned long hva;
2234         struct kvm_memory_slot *memslot;
2235         struct vm_area_struct *vma;
2236         unsigned long lpcr = 0, senc;
2237         unsigned long psize, porder;
2238         int srcu_idx;
2239
2240         mutex_lock(&kvm->lock);
2241         if (kvm->arch.rma_setup_done)
2242                 goto out;       /* another vcpu beat us to it */
2243
2244         /* Allocate hashed page table (if not done already) and reset it */
2245         if (!kvm->arch.hpt_virt) {
2246                 err = kvmppc_alloc_hpt(kvm, NULL);
2247                 if (err) {
2248                         pr_err("KVM: Couldn't alloc HPT\n");
2249                         goto out;
2250                 }
2251         }
2252
2253         /* Look up the memslot for guest physical address 0 */
2254         srcu_idx = srcu_read_lock(&kvm->srcu);
2255         memslot = gfn_to_memslot(kvm, 0);
2256
2257         /* We must have some memory at 0 by now */
2258         err = -EINVAL;
2259         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2260                 goto out_srcu;
2261
2262         /* Look up the VMA for the start of this memory slot */
2263         hva = memslot->userspace_addr;
2264         down_read(&current->mm->mmap_sem);
2265         vma = find_vma(current->mm, hva);
2266         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2267                 goto up_out;
2268
2269         psize = vma_kernel_pagesize(vma);
2270         porder = __ilog2(psize);
2271
2272         up_read(&current->mm->mmap_sem);
2273
2274         /* We can handle 4k, 64k or 16M pages in the VRMA */
2275         err = -EINVAL;
2276         if (!(psize == 0x1000 || psize == 0x10000 ||
2277               psize == 0x1000000))
2278                 goto out_srcu;
2279
2280         /* Update VRMASD field in the LPCR */
2281         senc = slb_pgsize_encoding(psize);
2282         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2283                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
2284         /* the -4 is to account for senc values starting at 0x10 */
2285         lpcr = senc << (LPCR_VRMASD_SH - 4);
2286
2287         /* Create HPTEs in the hash page table for the VRMA */
2288         kvmppc_map_vrma(vcpu, memslot, porder);
2289
2290         kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
2291
2292         /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
2293         smp_wmb();
2294         kvm->arch.rma_setup_done = 1;
2295         err = 0;
2296  out_srcu:
2297         srcu_read_unlock(&kvm->srcu, srcu_idx);
2298  out:
2299         mutex_unlock(&kvm->lock);
2300         return err;
2301
2302  up_out:
2303         up_read(&current->mm->mmap_sem);
2304         goto out_srcu;
2305 }
2306
2307 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2308 {
2309         unsigned long lpcr, lpid;
2310
2311         /* Allocate the guest's logical partition ID */
2312
2313         lpid = kvmppc_alloc_lpid();
2314         if ((long)lpid < 0)
2315                 return -ENOMEM;
2316         kvm->arch.lpid = lpid;
2317
2318         /*
2319          * Since we don't flush the TLB when tearing down a VM,
2320          * and this lpid might have previously been used,
2321          * make sure we flush on each core before running the new VM.
2322          */
2323         cpumask_setall(&kvm->arch.need_tlb_flush);
2324
2325         /* Start out with the default set of hcalls enabled */
2326         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
2327                sizeof(kvm->arch.enabled_hcalls));
2328
2329         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
2330
2331         /* Init LPCR for virtual RMA mode */
2332         kvm->arch.host_lpid = mfspr(SPRN_LPID);
2333         kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2334         lpcr &= LPCR_PECE | LPCR_LPES;
2335         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
2336                 LPCR_VPM0 | LPCR_VPM1;
2337         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2338                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
2339         /* On POWER8 turn on online bit to enable PURR/SPURR */
2340         if (cpu_has_feature(CPU_FTR_ARCH_207S))
2341                 lpcr |= LPCR_ONL;
2342         kvm->arch.lpcr = lpcr;
2343
2344         /*
2345          * Track that we now have a HV mode VM active. This blocks secondary
2346          * CPU threads from coming online.
2347          */
2348         kvm_hv_vm_activated();
2349
2350         return 0;
2351 }
2352
2353 static void kvmppc_free_vcores(struct kvm *kvm)
2354 {
2355         long int i;
2356
2357         for (i = 0; i < KVM_MAX_VCORES; ++i) {
2358                 if (kvm->arch.vcores[i] && kvm->arch.vcores[i]->mpp_buffer) {
2359                         struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2360                         free_pages((unsigned long)vc->mpp_buffer,
2361                                    MPP_BUFFER_ORDER);
2362                 }
2363                 kfree(kvm->arch.vcores[i]);
2364         }
2365         kvm->arch.online_vcores = 0;
2366 }
2367
2368 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
2369 {
2370         kvm_hv_vm_deactivated();
2371
2372         kvmppc_free_vcores(kvm);
2373
2374         kvmppc_free_hpt(kvm);
2375 }
2376
2377 /* We don't need to emulate any privileged instructions or dcbz */
2378 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2379                                      unsigned int inst, int *advance)
2380 {
2381         return EMULATE_FAIL;
2382 }
2383
2384 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2385                                         ulong spr_val)
2386 {
2387         return EMULATE_FAIL;
2388 }
2389
2390 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2391                                         ulong *spr_val)
2392 {
2393         return EMULATE_FAIL;
2394 }
2395
2396 static int kvmppc_core_check_processor_compat_hv(void)
2397 {
2398         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
2399             !cpu_has_feature(CPU_FTR_ARCH_206))
2400                 return -EIO;
2401         return 0;
2402 }
2403
2404 static long kvm_arch_vm_ioctl_hv(struct file *filp,
2405                                  unsigned int ioctl, unsigned long arg)
2406 {
2407         struct kvm *kvm __maybe_unused = filp->private_data;
2408         void __user *argp = (void __user *)arg;
2409         long r;
2410
2411         switch (ioctl) {
2412
2413         case KVM_PPC_ALLOCATE_HTAB: {
2414                 u32 htab_order;
2415
2416                 r = -EFAULT;
2417                 if (get_user(htab_order, (u32 __user *)argp))
2418                         break;
2419                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2420                 if (r)
2421                         break;
2422                 r = -EFAULT;
2423                 if (put_user(htab_order, (u32 __user *)argp))
2424                         break;
2425                 r = 0;
2426                 break;
2427         }
2428
2429         case KVM_PPC_GET_HTAB_FD: {
2430                 struct kvm_get_htab_fd ghf;
2431
2432                 r = -EFAULT;
2433                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
2434                         break;
2435                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2436                 break;
2437         }
2438
2439         default:
2440                 r = -ENOTTY;
2441         }
2442
2443         return r;
2444 }
2445
2446 /*
2447  * List of hcall numbers to enable by default.
2448  * For compatibility with old userspace, we enable by default
2449  * all hcalls that were implemented before the hcall-enabling
2450  * facility was added.  Note this list should not include H_RTAS.
2451  */
2452 static unsigned int default_hcall_list[] = {
2453         H_REMOVE,
2454         H_ENTER,
2455         H_READ,
2456         H_PROTECT,
2457         H_BULK_REMOVE,
2458         H_GET_TCE,
2459         H_PUT_TCE,
2460         H_SET_DABR,
2461         H_SET_XDABR,
2462         H_CEDE,
2463         H_PROD,
2464         H_CONFER,
2465         H_REGISTER_VPA,
2466 #ifdef CONFIG_KVM_XICS
2467         H_EOI,
2468         H_CPPR,
2469         H_IPI,
2470         H_IPOLL,
2471         H_XIRR,
2472         H_XIRR_X,
2473 #endif
2474         0
2475 };
2476
2477 static void init_default_hcalls(void)
2478 {
2479         int i;
2480         unsigned int hcall;
2481
2482         for (i = 0; default_hcall_list[i]; ++i) {
2483                 hcall = default_hcall_list[i];
2484                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
2485                 __set_bit(hcall / 4, default_enabled_hcalls);
2486         }
2487 }
2488
2489 static struct kvmppc_ops kvm_ops_hv = {
2490         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2491         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2492         .get_one_reg = kvmppc_get_one_reg_hv,
2493         .set_one_reg = kvmppc_set_one_reg_hv,
2494         .vcpu_load   = kvmppc_core_vcpu_load_hv,
2495         .vcpu_put    = kvmppc_core_vcpu_put_hv,
2496         .set_msr     = kvmppc_set_msr_hv,
2497         .vcpu_run    = kvmppc_vcpu_run_hv,
2498         .vcpu_create = kvmppc_core_vcpu_create_hv,
2499         .vcpu_free   = kvmppc_core_vcpu_free_hv,
2500         .check_requests = kvmppc_core_check_requests_hv,
2501         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
2502         .flush_memslot  = kvmppc_core_flush_memslot_hv,
2503         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2504         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
2505         .unmap_hva = kvm_unmap_hva_hv,
2506         .unmap_hva_range = kvm_unmap_hva_range_hv,
2507         .age_hva  = kvm_age_hva_hv,
2508         .test_age_hva = kvm_test_age_hva_hv,
2509         .set_spte_hva = kvm_set_spte_hva_hv,
2510         .mmu_destroy  = kvmppc_mmu_destroy_hv,
2511         .free_memslot = kvmppc_core_free_memslot_hv,
2512         .create_memslot = kvmppc_core_create_memslot_hv,
2513         .init_vm =  kvmppc_core_init_vm_hv,
2514         .destroy_vm = kvmppc_core_destroy_vm_hv,
2515         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2516         .emulate_op = kvmppc_core_emulate_op_hv,
2517         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2518         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2519         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2520         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
2521         .hcall_implemented = kvmppc_hcall_impl_hv,
2522 };
2523
2524 static int kvmppc_book3s_init_hv(void)
2525 {
2526         int r;
2527         /*
2528          * FIXME!! Do we need to check on all cpus ?
2529          */
2530         r = kvmppc_core_check_processor_compat_hv();
2531         if (r < 0)
2532                 return -ENODEV;
2533
2534         kvm_ops_hv.owner = THIS_MODULE;
2535         kvmppc_hv_ops = &kvm_ops_hv;
2536
2537         init_default_hcalls();
2538
2539         r = kvmppc_mmu_hv_init();
2540         return r;
2541 }
2542
2543 static void kvmppc_book3s_exit_hv(void)
2544 {
2545         kvmppc_hv_ops = NULL;
2546 }
2547
2548 module_init(kvmppc_book3s_init_hv);
2549 module_exit(kvmppc_book3s_exit_hv);
2550 MODULE_LICENSE("GPL");
2551 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2552 MODULE_ALIAS("devname:kvm");