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 EXIT_DEBUG */
  62 /* #define EXIT_DEBUG_SIMPLE */
  63 /* #define EXIT_DEBUG_INT */
  64
  65 /* Used to indicate that a guest page fault needs to be handled */
  66 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  67
  68 /* Used as a "null" value for timebase values */
  69 #define TB_NIL  (~(u64)0)
  70
  71 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
  72
  73 #if defined(CONFIG_PPC_64K_PAGES)
  74 #define MPP_BUFFER_ORDER        0
  75 #elif defined(CONFIG_PPC_4K_PAGES)
  76 #define MPP_BUFFER_ORDER        3
  77 #endif
  78
  79
  80 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
  81 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
  82
  83 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
  84 {
  85         int me;
  86         int cpu = vcpu->cpu;
  87         wait_queue_head_t *wqp;
  88
  89         wqp = kvm_arch_vcpu_wq(vcpu);
  90         if (waitqueue_active(wqp)) {
  91                 wake_up_interruptible(wqp);
  92                 ++vcpu->stat.halt_wakeup;
  93         }
  94
  95         me = get_cpu();
  96
  97         /* CPU points to the first thread of the core */
  98         if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
  99 #ifdef CONFIG_PPC_ICP_NATIVE
 100                 int real_cpu = cpu + vcpu->arch.ptid;
 101                 if (paca[real_cpu].kvm_hstate.xics_phys)
 102                         xics_wake_cpu(real_cpu);
 103                 else
 104 #endif
 105                 if (cpu_online(cpu))
 106                         smp_send_reschedule(cpu);
 107         }
 108         put_cpu();
 109 }
 110
 111 /*
 112  * We use the vcpu_load/put functions to measure stolen time.
 113  * Stolen time is counted as time when either the vcpu is able to
 114  * run as part of a virtual core, but the task running the vcore
 115  * is preempted or sleeping, or when the vcpu needs something done
 116  * in the kernel by the task running the vcpu, but that task is
 117  * preempted or sleeping.  Those two things have to be counted
 118  * separately, since one of the vcpu tasks will take on the job
 119  * of running the core, and the other vcpu tasks in the vcore will
 120  * sleep waiting for it to do that, but that sleep shouldn't count
 121  * as stolen time.
 122  *
 123  * Hence we accumulate stolen time when the vcpu can run as part of
 124  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 125  * needs its task to do other things in the kernel (for example,
 126  * service a page fault) in busy_stolen.  We don't accumulate
 127  * stolen time for a vcore when it is inactive, or for a vcpu
 128  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 129  * a misnomer; it means that the vcpu task is not executing in
 130  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 131  * the kernel.  We don't have any way of dividing up that time
 132  * between time that the vcpu is genuinely stopped, time that
 133  * the task is actively working on behalf of the vcpu, and time
 134  * that the task is preempted, so we don't count any of it as
 135  * stolen.
 136  *
 137  * Updates to busy_stolen are protected by arch.tbacct_lock;
 138  * updates to vc->stolen_tb are protected by the arch.tbacct_lock
 139  * of the vcpu that has taken responsibility for running the vcore
 140  * (i.e. vc->runner).  The stolen times are measured in units of
 141  * timebase ticks.  (Note that the != TB_NIL checks below are
 142  * purely defensive; they should never fail.)
 143  */
 144
 145 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 146 {
 147         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 148         unsigned long flags;
 149
 150         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 151         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
 152             vc->preempt_tb != TB_NIL) {
 153                 vc->stolen_tb += mftb() - vc->preempt_tb;
 154                 vc->preempt_tb = TB_NIL;
 155         }
 156         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 157             vcpu->arch.busy_preempt != TB_NIL) {
 158                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 159                 vcpu->arch.busy_preempt = TB_NIL;
 160         }
 161         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 162 }
 163
 164 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 165 {
 166         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 167         unsigned long flags;
 168
 169         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 170         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
 171                 vc->preempt_tb = mftb();
 172         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 173                 vcpu->arch.busy_preempt = mftb();
 174         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 175 }
 176
 177 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
 178 {
 179         vcpu->arch.shregs.msr = msr;
 180         kvmppc_end_cede(vcpu);
 181 }
 182
 183 void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 184 {
 185         vcpu->arch.pvr = pvr;
 186 }
 187
 188 int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 189 {
 190         unsigned long pcr = 0;
 191         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 192
 193         if (arch_compat) {
 194                 if (!cpu_has_feature(CPU_FTR_ARCH_206))
 195                         return -EINVAL; /* 970 has no compat mode support */
 196
 197                 switch (arch_compat) {
 198                 case PVR_ARCH_205:
 199                         /*
 200                          * If an arch bit is set in PCR, all the defined
 201                          * higher-order arch bits also have to be set.
 202                          */
 203                         pcr = PCR_ARCH_206 | PCR_ARCH_205;
 204                         break;
 205                 case PVR_ARCH_206:
 206                 case PVR_ARCH_206p:
 207                         pcr = PCR_ARCH_206;
 208                         break;
 209                 case PVR_ARCH_207:
 210                         break;
 211                 default:
 212                         return -EINVAL;
 213                 }
 214
 215                 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
 216                         /* POWER7 can't emulate POWER8 */
 217                         if (!(pcr & PCR_ARCH_206))
 218                                 return -EINVAL;
 219                         pcr &= ~PCR_ARCH_206;
 220                 }
 221         }
 222
 223         spin_lock(&vc->lock);
 224         vc->arch_compat = arch_compat;
 225         vc->pcr = pcr;
 226         spin_unlock(&vc->lock);
 227
 228         return 0;
 229 }
 230
 231 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 232 {
 233         int r;
 234
 235         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 236         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 237                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
 238         for (r = 0; r < 16; ++r)
 239                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 240                        r, kvmppc_get_gpr(vcpu, r),
 241                        r+16, kvmppc_get_gpr(vcpu, r+16));
 242         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 243                vcpu->arch.ctr, vcpu->arch.lr);
 244         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 245                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 246         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 247                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 248         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 249                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 250         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 251                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 252         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 253         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 254                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 255         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 256         for (r = 0; r < vcpu->arch.slb_max; ++r)
 257                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 258                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 259         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 260                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 261                vcpu->arch.last_inst);
 262 }
 263
 264 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 265 {
 266         int r;
 267         struct kvm_vcpu *v, *ret = NULL;
 268
 269         mutex_lock(&kvm->lock);
 270         kvm_for_each_vcpu(r, v, kvm) {
 271                 if (v->vcpu_id == id) {
 272                         ret = v;
 273                         break;
 274                 }
 275         }
 276         mutex_unlock(&kvm->lock);
 277         return ret;
 278 }
 279
 280 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 281 {
 282         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 283         vpa->yield_count = cpu_to_be32(1);
 284 }
 285
 286 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 287                    unsigned long addr, unsigned long len)
 288 {
 289         /* check address is cacheline aligned */
 290         if (addr & (L1_CACHE_BYTES - 1))
 291                 return -EINVAL;
 292         spin_lock(&vcpu->arch.vpa_update_lock);
 293         if (v->next_gpa != addr || v->len != len) {
 294                 v->next_gpa = addr;
 295                 v->len = addr ? len : 0;
 296                 v->update_pending = 1;
 297         }
 298         spin_unlock(&vcpu->arch.vpa_update_lock);
 299         return 0;
 300 }
 301
 302 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 303 struct reg_vpa {
 304         u32 dummy;
 305         union {
 306                 __be16 hword;
 307                 __be32 word;
 308         } length;
 309 };
 310
 311 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 312 {
 313         if (vpap->update_pending)
 314                 return vpap->next_gpa != 0;
 315         return vpap->pinned_addr != NULL;
 316 }
 317
 318 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 319                                        unsigned long flags,
 320                                        unsigned long vcpuid, unsigned long vpa)
 321 {
 322         struct kvm *kvm = vcpu->kvm;
 323         unsigned long len, nb;
 324         void *va;
 325         struct kvm_vcpu *tvcpu;
 326         int err;
 327         int subfunc;
 328         struct kvmppc_vpa *vpap;
 329
 330         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 331         if (!tvcpu)
 332                 return H_PARAMETER;
 333
 334         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 335         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 336             subfunc == H_VPA_REG_SLB) {
 337                 /* Registering new area - address must be cache-line aligned */
 338                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 339                         return H_PARAMETER;
 340
 341                 /* convert logical addr to kernel addr and read length */
 342                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 343                 if (va == NULL)
 344                         return H_PARAMETER;
 345                 if (subfunc == H_VPA_REG_VPA)
 346                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 347                 else
 348                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 349                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 350
 351                 /* Check length */
 352                 if (len > nb || len < sizeof(struct reg_vpa))
 353                         return H_PARAMETER;
 354         } else {
 355                 vpa = 0;
 356                 len = 0;
 357         }
 358
 359         err = H_PARAMETER;
 360         vpap = NULL;
 361         spin_lock(&tvcpu->arch.vpa_update_lock);
 362
 363         switch (subfunc) {
 364         case H_VPA_REG_VPA:             /* register VPA */
 365                 if (len < sizeof(struct lppaca))
 366                         break;
 367                 vpap = &tvcpu->arch.vpa;
 368                 err = 0;
 369                 break;
 370
 371         case H_VPA_REG_DTL:             /* register DTL */
 372                 if (len < sizeof(struct dtl_entry))
 373                         break;
 374                 len -= len % sizeof(struct dtl_entry);
 375
 376                 /* Check that they have previously registered a VPA */
 377                 err = H_RESOURCE;
 378                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 379                         break;
 380
 381                 vpap = &tvcpu->arch.dtl;
 382                 err = 0;
 383                 break;
 384
 385         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 386                 /* Check that they have previously registered a VPA */
 387                 err = H_RESOURCE;
 388                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 389                         break;
 390
 391                 vpap = &tvcpu->arch.slb_shadow;
 392                 err = 0;
 393                 break;
 394
 395         case H_VPA_DEREG_VPA:           /* deregister VPA */
 396                 /* Check they don't still have a DTL or SLB buf registered */
 397                 err = H_RESOURCE;
 398                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 399                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 400                         break;
 401
 402                 vpap = &tvcpu->arch.vpa;
 403                 err = 0;
 404                 break;
 405
 406         case H_VPA_DEREG_DTL:           /* deregister DTL */
 407                 vpap = &tvcpu->arch.dtl;
 408                 err = 0;
 409                 break;
 410
 411         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 412                 vpap = &tvcpu->arch.slb_shadow;
 413                 err = 0;
 414                 break;
 415         }
 416
 417         if (vpap) {
 418                 vpap->next_gpa = vpa;
 419                 vpap->len = len;
 420                 vpap->update_pending = 1;
 421         }
 422
 423         spin_unlock(&tvcpu->arch.vpa_update_lock);
 424
 425         return err;
 426 }
 427
 428 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 429 {
 430         struct kvm *kvm = vcpu->kvm;
 431         void *va;
 432         unsigned long nb;
 433         unsigned long gpa;
 434
 435         /*
 436          * We need to pin the page pointed to by vpap->next_gpa,
 437          * but we can't call kvmppc_pin_guest_page under the lock
 438          * as it does get_user_pages() and down_read().  So we
 439          * have to drop the lock, pin the page, then get the lock
 440          * again and check that a new area didn't get registered
 441          * in the meantime.
 442          */
 443         for (;;) {
 444                 gpa = vpap->next_gpa;
 445                 spin_unlock(&vcpu->arch.vpa_update_lock);
 446                 va = NULL;
 447                 nb = 0;
 448                 if (gpa)
 449                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 450                 spin_lock(&vcpu->arch.vpa_update_lock);
 451                 if (gpa == vpap->next_gpa)
 452                         break;
 453                 /* sigh... unpin that one and try again */
 454                 if (va)
 455                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 456         }
 457
 458         vpap->update_pending = 0;
 459         if (va && nb < vpap->len) {
 460                 /*
 461                  * If it's now too short, it must be that userspace
 462                  * has changed the mappings underlying guest memory,
 463                  * so unregister the region.
 464                  */
 465                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 466                 va = NULL;
 467         }
 468         if (vpap->pinned_addr)
 469                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 470                                         vpap->dirty);
 471         vpap->gpa = gpa;
 472         vpap->pinned_addr = va;
 473         vpap->dirty = false;
 474         if (va)
 475                 vpap->pinned_end = va + vpap->len;
 476 }
 477
 478 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 479 {
 480         if (!(vcpu->arch.vpa.update_pending ||
 481               vcpu->arch.slb_shadow.update_pending ||
 482               vcpu->arch.dtl.update_pending))
 483                 return;
 484
 485         spin_lock(&vcpu->arch.vpa_update_lock);
 486         if (vcpu->arch.vpa.update_pending) {
 487                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 488                 if (vcpu->arch.vpa.pinned_addr)
 489                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 490         }
 491         if (vcpu->arch.dtl.update_pending) {
 492                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 493                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 494                 vcpu->arch.dtl_index = 0;
 495         }
 496         if (vcpu->arch.slb_shadow.update_pending)
 497                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 498         spin_unlock(&vcpu->arch.vpa_update_lock);
 499 }
 500
 501 /*
 502  * Return the accumulated stolen time for the vcore up until `now'.
 503  * The caller should hold the vcore lock.
 504  */
 505 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 506 {
 507         u64 p;
 508
 509         /*
 510          * If we are the task running the vcore, then since we hold
 511          * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
 512          * can't be updated, so we don't need the tbacct_lock.
 513          * If the vcore is inactive, it can't become active (since we
 514          * hold the vcore lock), so the vcpu load/put functions won't
 515          * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
 516          */
 517         if (vc->vcore_state != VCORE_INACTIVE &&
 518             vc->runner->arch.run_task != current) {
 519                 spin_lock_irq(&vc->runner->arch.tbacct_lock);
 520                 p = vc->stolen_tb;
 521                 if (vc->preempt_tb != TB_NIL)
 522                         p += now - vc->preempt_tb;
 523                 spin_unlock_irq(&vc->runner->arch.tbacct_lock);
 524         } else {
 525                 p = vc->stolen_tb;
 526         }
 527         return p;
 528 }
 529
 530 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 531                                     struct kvmppc_vcore *vc)
 532 {
 533         struct dtl_entry *dt;
 534         struct lppaca *vpa;
 535         unsigned long stolen;
 536         unsigned long core_stolen;
 537         u64 now;
 538
 539         dt = vcpu->arch.dtl_ptr;
 540         vpa = vcpu->arch.vpa.pinned_addr;
 541         now = mftb();
 542         core_stolen = vcore_stolen_time(vc, now);
 543         stolen = core_stolen - vcpu->arch.stolen_logged;
 544         vcpu->arch.stolen_logged = core_stolen;
 545         spin_lock_irq(&vcpu->arch.tbacct_lock);
 546         stolen += vcpu->arch.busy_stolen;
 547         vcpu->arch.busy_stolen = 0;
 548         spin_unlock_irq(&vcpu->arch.tbacct_lock);
 549         if (!dt || !vpa)
 550                 return;
 551         memset(dt, 0, sizeof(struct dtl_entry));
 552         dt->dispatch_reason = 7;
 553         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 554         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 555         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 556         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 557         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 558         ++dt;
 559         if (dt == vcpu->arch.dtl.pinned_end)
 560                 dt = vcpu->arch.dtl.pinned_addr;
 561         vcpu->arch.dtl_ptr = dt;
 562         /* order writing *dt vs. writing vpa->dtl_idx */
 563         smp_wmb();
 564         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 565         vcpu->arch.dtl.dirty = true;
 566 }
 567
 568 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 569 {
 570         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 571                 return true;
 572         if ((!vcpu->arch.vcore->arch_compat) &&
 573             cpu_has_feature(CPU_FTR_ARCH_207S))
 574                 return true;
 575         return false;
 576 }
 577
 578 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 579                              unsigned long resource, unsigned long value1,
 580                              unsigned long value2)
 581 {
 582         switch (resource) {
 583         case H_SET_MODE_RESOURCE_SET_CIABR:
 584                 if (!kvmppc_power8_compatible(vcpu))
 585                         return H_P2;
 586                 if (value2)
 587                         return H_P4;
 588                 if (mflags)
 589                         return H_UNSUPPORTED_FLAG_START;
 590                 /* Guests can't breakpoint the hypervisor */
 591                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 592                         return H_P3;
 593                 vcpu->arch.ciabr  = value1;
 594                 return H_SUCCESS;
 595         case H_SET_MODE_RESOURCE_SET_DAWR:
 596                 if (!kvmppc_power8_compatible(vcpu))
 597                         return H_P2;
 598                 if (mflags)
 599                         return H_UNSUPPORTED_FLAG_START;
 600                 if (value2 & DABRX_HYP)
 601                         return H_P4;
 602                 vcpu->arch.dawr  = value1;
 603                 vcpu->arch.dawrx = value2;
 604                 return H_SUCCESS;
 605         default:
 606                 return H_TOO_HARD;
 607         }
 608 }
 609
 610 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 611 {
 612         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 613         unsigned long target, ret = H_SUCCESS;
 614         struct kvm_vcpu *tvcpu;
 615         int idx, rc;
 616
 617         if (req <= MAX_HCALL_OPCODE &&
 618             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 619                 return RESUME_HOST;
 620
 621         switch (req) {
 622         case H_ENTER:
 623                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 624                 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
 625                                               kvmppc_get_gpr(vcpu, 5),
 626                                               kvmppc_get_gpr(vcpu, 6),
 627                                               kvmppc_get_gpr(vcpu, 7));
 628                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 629                 break;
 630         case H_CEDE:
 631                 break;
 632         case H_PROD:
 633                 target = kvmppc_get_gpr(vcpu, 4);
 634                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 635                 if (!tvcpu) {
 636                         ret = H_PARAMETER;
 637                         break;
 638                 }
 639                 tvcpu->arch.prodded = 1;
 640                 smp_mb();
 641                 if (vcpu->arch.ceded) {
 642                         if (waitqueue_active(&vcpu->wq)) {
 643                                 wake_up_interruptible(&vcpu->wq);
 644                                 vcpu->stat.halt_wakeup++;
 645                         }
 646                 }
 647                 break;
 648         case H_CONFER:
 649                 target = kvmppc_get_gpr(vcpu, 4);
 650                 if (target == -1)
 651                         break;
 652                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 653                 if (!tvcpu) {
 654                         ret = H_PARAMETER;
 655                         break;
 656                 }
 657                 kvm_vcpu_yield_to(tvcpu);
 658                 break;
 659         case H_REGISTER_VPA:
 660                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 661                                         kvmppc_get_gpr(vcpu, 5),
 662                                         kvmppc_get_gpr(vcpu, 6));
 663                 break;
 664         case H_RTAS:
 665                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 666                         return RESUME_HOST;
 667
 668                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 669                 rc = kvmppc_rtas_hcall(vcpu);
 670                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 671
 672                 if (rc == -ENOENT)
 673                         return RESUME_HOST;
 674                 else if (rc == 0)
 675                         break;
 676
 677                 /* Send the error out to userspace via KVM_RUN */
 678                 return rc;
 679         case H_SET_MODE:
 680                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 681                                         kvmppc_get_gpr(vcpu, 5),
 682                                         kvmppc_get_gpr(vcpu, 6),
 683                                         kvmppc_get_gpr(vcpu, 7));
 684                 if (ret == H_TOO_HARD)
 685                         return RESUME_HOST;
 686                 break;
 687         case H_XIRR:
 688         case H_CPPR:
 689         case H_EOI:
 690         case H_IPI:
 691         case H_IPOLL:
 692         case H_XIRR_X:
 693                 if (kvmppc_xics_enabled(vcpu)) {
 694                         ret = kvmppc_xics_hcall(vcpu, req);
 695                         break;
 696                 } /* fallthrough */
 697         default:
 698                 return RESUME_HOST;
 699         }
 700         kvmppc_set_gpr(vcpu, 3, ret);
 701         vcpu->arch.hcall_needed = 0;
 702         return RESUME_GUEST;
 703 }
 704
 705 static int kvmppc_hcall_impl_hv(unsigned long cmd)
 706 {
 707         switch (cmd) {
 708         case H_CEDE:
 709         case H_PROD:
 710         case H_CONFER:
 711         case H_REGISTER_VPA:
 712         case H_SET_MODE:
 713 #ifdef CONFIG_KVM_XICS
 714         case H_XIRR:
 715         case H_CPPR:
 716         case H_EOI:
 717         case H_IPI:
 718         case H_IPOLL:
 719         case H_XIRR_X:
 720 #endif
 721                 return 1;
 722         }
 723
 724         /* See if it's in the real-mode table */
 725         return kvmppc_hcall_impl_hv_realmode(cmd);
 726 }
 727
 728 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
 729                                         struct kvm_vcpu *vcpu)
 730 {
 731         u32 last_inst;
 732
 733         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
 734                                         EMULATE_DONE) {
 735                 /*
 736                  * Fetch failed, so return to guest and
 737                  * try executing it again.
 738                  */
 739                 return RESUME_GUEST;
 740         }
 741
 742         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
 743                 run->exit_reason = KVM_EXIT_DEBUG;
 744                 run->debug.arch.address = kvmppc_get_pc(vcpu);
 745                 return RESUME_HOST;
 746         } else {
 747                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 748                 return RESUME_GUEST;
 749         }
 750 }
 751
 752 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
 753                                  struct task_struct *tsk)
 754 {
 755         int r = RESUME_HOST;
 756
 757         vcpu->stat.sum_exits++;
 758
 759         run->exit_reason = KVM_EXIT_UNKNOWN;
 760         run->ready_for_interrupt_injection = 1;
 761         switch (vcpu->arch.trap) {
 762         /* We're good on these - the host merely wanted to get our attention */
 763         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 764                 vcpu->stat.dec_exits++;
 765                 r = RESUME_GUEST;
 766                 break;
 767         case BOOK3S_INTERRUPT_EXTERNAL:
 768         case BOOK3S_INTERRUPT_H_DOORBELL:
 769                 vcpu->stat.ext_intr_exits++;
 770                 r = RESUME_GUEST;
 771                 break;
 772         /* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
 773         case BOOK3S_INTERRUPT_HMI:
 774         case BOOK3S_INTERRUPT_PERFMON:
 775                 r = RESUME_GUEST;
 776                 break;
 777         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 778                 /*
 779                  * Deliver a machine check interrupt to the guest.
 780                  * We have to do this, even if the host has handled the
 781                  * machine check, because machine checks use SRR0/1 and
 782                  * the interrupt might have trashed guest state in them.
 783                  */
 784                 kvmppc_book3s_queue_irqprio(vcpu,
 785                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 786                 r = RESUME_GUEST;
 787                 break;
 788         case BOOK3S_INTERRUPT_PROGRAM:
 789         {
 790                 ulong flags;
 791                 /*
 792                  * Normally program interrupts are delivered directly
 793                  * to the guest by the hardware, but we can get here
 794                  * as a result of a hypervisor emulation interrupt
 795                  * (e40) getting turned into a 700 by BML RTAS.
 796                  */
 797                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 798                 kvmppc_core_queue_program(vcpu, flags);
 799                 r = RESUME_GUEST;
 800                 break;
 801         }
 802         case BOOK3S_INTERRUPT_SYSCALL:
 803         {
 804                 /* hcall - punt to userspace */
 805                 int i;
 806
 807                 /* hypercall with MSR_PR has already been handled in rmode,
 808                  * and never reaches here.
 809                  */
 810
 811                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 812                 for (i = 0; i < 9; ++i)
 813                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 814                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 815                 vcpu->arch.hcall_needed = 1;
 816                 r = RESUME_HOST;
 817                 break;
 818         }
 819         /*
 820          * We get these next two if the guest accesses a page which it thinks
 821          * it has mapped but which is not actually present, either because
 822          * it is for an emulated I/O device or because the corresonding
 823          * host page has been paged out.  Any other HDSI/HISI interrupts
 824          * have been handled already.
 825          */
 826         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 827                 r = RESUME_PAGE_FAULT;
 828                 break;
 829         case BOOK3S_INTERRUPT_H_INST_STORAGE:
 830                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
 831                 vcpu->arch.fault_dsisr = 0;
 832                 r = RESUME_PAGE_FAULT;
 833                 break;
 834         /*
 835          * This occurs if the guest executes an illegal instruction.
 836          * If the guest debug is disabled, generate a program interrupt
 837          * to the guest. If guest debug is enabled, we need to check
 838          * whether the instruction is a software breakpoint instruction.
 839          * Accordingly return to Guest or Host.
 840          */
 841         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 842                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
 843                         r = kvmppc_emulate_debug_inst(run, vcpu);
 844                 } else {
 845                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 846                         r = RESUME_GUEST;
 847                 }
 848                 break;
 849         /*
 850          * This occurs if the guest (kernel or userspace), does something that
 851          * is prohibited by HFSCR.  We just generate a program interrupt to
 852          * the guest.
 853          */
 854         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
 855                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 856                 r = RESUME_GUEST;
 857                 break;
 858         default:
 859                 kvmppc_dump_regs(vcpu);
 860                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 861                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 862                         vcpu->arch.shregs.msr);
 863                 run->hw.hardware_exit_reason = vcpu->arch.trap;
 864                 r = RESUME_HOST;
 865                 break;
 866         }
 867
 868         return r;
 869 }
 870
 871 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
 872                                             struct kvm_sregs *sregs)
 873 {
 874         int i;
 875
 876         memset(sregs, 0, sizeof(struct kvm_sregs));
 877         sregs->pvr = vcpu->arch.pvr;
 878         for (i = 0; i < vcpu->arch.slb_max; i++) {
 879                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
 880                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 881         }
 882
 883         return 0;
 884 }
 885
 886 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
 887                                             struct kvm_sregs *sregs)
 888 {
 889         int i, j;
 890
 891         /* Only accept the same PVR as the host's, since we can't spoof it */
 892         if (sregs->pvr != vcpu->arch.pvr)
 893                 return -EINVAL;
 894
 895         j = 0;
 896         for (i = 0; i < vcpu->arch.slb_nr; i++) {
 897                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
 898                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
 899                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
 900                         ++j;
 901                 }
 902         }
 903         vcpu->arch.slb_max = j;
 904
 905         return 0;
 906 }
 907
 908 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
 909                 bool preserve_top32)
 910 {
 911         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 912         u64 mask;
 913
 914         spin_lock(&vc->lock);
 915         /*
 916          * If ILE (interrupt little-endian) has changed, update the
 917          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
 918          */
 919         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
 920                 struct kvm *kvm = vcpu->kvm;
 921                 struct kvm_vcpu *vcpu;
 922                 int i;
 923
 924                 mutex_lock(&kvm->lock);
 925                 kvm_for_each_vcpu(i, vcpu, kvm) {
 926                         if (vcpu->arch.vcore != vc)
 927                                 continue;
 928                         if (new_lpcr & LPCR_ILE)
 929                                 vcpu->arch.intr_msr |= MSR_LE;
 930                         else
 931                                 vcpu->arch.intr_msr &= ~MSR_LE;
 932                 }
 933                 mutex_unlock(&kvm->lock);
 934         }
 935
 936         /*
 937          * Userspace can only modify DPFD (default prefetch depth),
 938          * ILE (interrupt little-endian) and TC (translation control).
 939          * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
 940          */
 941         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
 942         if (cpu_has_feature(CPU_FTR_ARCH_207S))
 943                 mask |= LPCR_AIL;
 944
 945         /* Broken 32-bit version of LPCR must not clear top bits */
 946         if (preserve_top32)
 947                 mask &= 0xFFFFFFFF;
 948         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
 949         spin_unlock(&vc->lock);
 950 }
 951
 952 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
 953                                  union kvmppc_one_reg *val)
 954 {
 955         int r = 0;
 956         long int i;
 957
 958         switch (id) {
 959         case KVM_REG_PPC_DEBUG_INST:
 960                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
 961                 break;
 962         case KVM_REG_PPC_HIOR:
 963                 *val = get_reg_val(id, 0);
 964                 break;
 965         case KVM_REG_PPC_DABR:
 966                 *val = get_reg_val(id, vcpu->arch.dabr);
 967                 break;
 968         case KVM_REG_PPC_DABRX:
 969                 *val = get_reg_val(id, vcpu->arch.dabrx);
 970                 break;
 971         case KVM_REG_PPC_DSCR:
 972                 *val = get_reg_val(id, vcpu->arch.dscr);
 973                 break;
 974         case KVM_REG_PPC_PURR:
 975                 *val = get_reg_val(id, vcpu->arch.purr);
 976                 break;
 977         case KVM_REG_PPC_SPURR:
 978                 *val = get_reg_val(id, vcpu->arch.spurr);
 979                 break;
 980         case KVM_REG_PPC_AMR:
 981                 *val = get_reg_val(id, vcpu->arch.amr);
 982                 break;
 983         case KVM_REG_PPC_UAMOR:
 984                 *val = get_reg_val(id, vcpu->arch.uamor);
 985                 break;
 986         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
 987                 i = id - KVM_REG_PPC_MMCR0;
 988                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
 989                 break;
 990         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
 991                 i = id - KVM_REG_PPC_PMC1;
 992                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
 993                 break;
 994         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
 995                 i = id - KVM_REG_PPC_SPMC1;
 996                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
 997                 break;
 998         case KVM_REG_PPC_SIAR:
 999                 *val = get_reg_val(id, vcpu->arch.siar);
1000                 break;
1001         case KVM_REG_PPC_SDAR:
1002                 *val = get_reg_val(id, vcpu->arch.sdar);
1003                 break;
1004         case KVM_REG_PPC_SIER:
1005                 *val = get_reg_val(id, vcpu->arch.sier);
1006                 break;
1007         case KVM_REG_PPC_IAMR:
1008                 *val = get_reg_val(id, vcpu->arch.iamr);
1009                 break;
1010         case KVM_REG_PPC_PSPB:
1011                 *val = get_reg_val(id, vcpu->arch.pspb);
1012                 break;
1013         case KVM_REG_PPC_DPDES:
1014                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1015                 break;
1016         case KVM_REG_PPC_DAWR:
1017                 *val = get_reg_val(id, vcpu->arch.dawr);
1018                 break;
1019         case KVM_REG_PPC_DAWRX:
1020                 *val = get_reg_val(id, vcpu->arch.dawrx);
1021                 break;
1022         case KVM_REG_PPC_CIABR:
1023                 *val = get_reg_val(id, vcpu->arch.ciabr);
1024                 break;
1025         case KVM_REG_PPC_CSIGR:
1026                 *val = get_reg_val(id, vcpu->arch.csigr);
1027                 break;
1028         case KVM_REG_PPC_TACR:
1029                 *val = get_reg_val(id, vcpu->arch.tacr);
1030                 break;
1031         case KVM_REG_PPC_TCSCR:
1032                 *val = get_reg_val(id, vcpu->arch.tcscr);
1033                 break;
1034         case KVM_REG_PPC_PID:
1035                 *val = get_reg_val(id, vcpu->arch.pid);
1036                 break;
1037         case KVM_REG_PPC_ACOP:
1038                 *val = get_reg_val(id, vcpu->arch.acop);
1039                 break;
1040         case KVM_REG_PPC_WORT:
1041                 *val = get_reg_val(id, vcpu->arch.wort);
1042                 break;
1043         case KVM_REG_PPC_VPA_ADDR:
1044                 spin_lock(&vcpu->arch.vpa_update_lock);
1045                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1046                 spin_unlock(&vcpu->arch.vpa_update_lock);
1047                 break;
1048         case KVM_REG_PPC_VPA_SLB:
1049                 spin_lock(&vcpu->arch.vpa_update_lock);
1050                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1051                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1052                 spin_unlock(&vcpu->arch.vpa_update_lock);
1053                 break;
1054         case KVM_REG_PPC_VPA_DTL:
1055                 spin_lock(&vcpu->arch.vpa_update_lock);
1056                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1057                 val->vpaval.length = vcpu->arch.dtl.len;
1058                 spin_unlock(&vcpu->arch.vpa_update_lock);
1059                 break;
1060         case KVM_REG_PPC_TB_OFFSET:
1061                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1062                 break;
1063         case KVM_REG_PPC_LPCR:
1064         case KVM_REG_PPC_LPCR_64:
1065                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1066                 break;
1067         case KVM_REG_PPC_PPR:
1068                 *val = get_reg_val(id, vcpu->arch.ppr);
1069                 break;
1070 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1071         case KVM_REG_PPC_TFHAR:
1072                 *val = get_reg_val(id, vcpu->arch.tfhar);
1073                 break;
1074         case KVM_REG_PPC_TFIAR:
1075                 *val = get_reg_val(id, vcpu->arch.tfiar);
1076                 break;
1077         case KVM_REG_PPC_TEXASR:
1078                 *val = get_reg_val(id, vcpu->arch.texasr);
1079                 break;
1080         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1081                 i = id - KVM_REG_PPC_TM_GPR0;
1082                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1083                 break;
1084         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1085         {
1086                 int j;
1087                 i = id - KVM_REG_PPC_TM_VSR0;
1088                 if (i < 32)
1089                         for (j = 0; j < TS_FPRWIDTH; j++)
1090                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1091                 else {
1092                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1093                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1094                         else
1095                                 r = -ENXIO;
1096                 }
1097                 break;
1098         }
1099         case KVM_REG_PPC_TM_CR:
1100                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1101                 break;
1102         case KVM_REG_PPC_TM_LR:
1103                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1104                 break;
1105         case KVM_REG_PPC_TM_CTR:
1106                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1107                 break;
1108         case KVM_REG_PPC_TM_FPSCR:
1109                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1110                 break;
1111         case KVM_REG_PPC_TM_AMR:
1112                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1113                 break;
1114         case KVM_REG_PPC_TM_PPR:
1115                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1116                 break;
1117         case KVM_REG_PPC_TM_VRSAVE:
1118                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1119                 break;
1120         case KVM_REG_PPC_TM_VSCR:
1121                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1122                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1123                 else
1124                         r = -ENXIO;
1125                 break;
1126         case KVM_REG_PPC_TM_DSCR:
1127                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1128                 break;
1129         case KVM_REG_PPC_TM_TAR:
1130                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1131                 break;
1132 #endif
1133         case KVM_REG_PPC_ARCH_COMPAT:
1134                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1135                 break;
1136         default:
1137                 r = -EINVAL;
1138                 break;
1139         }
1140
1141         return r;
1142 }
1143
1144 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1145                                  union kvmppc_one_reg *val)
1146 {
1147         int r = 0;
1148         long int i;
1149         unsigned long addr, len;
1150
1151         switch (id) {
1152         case KVM_REG_PPC_HIOR:
1153                 /* Only allow this to be set to zero */
1154                 if (set_reg_val(id, *val))
1155                         r = -EINVAL;
1156                 break;
1157         case KVM_REG_PPC_DABR:
1158                 vcpu->arch.dabr = set_reg_val(id, *val);
1159                 break;
1160         case KVM_REG_PPC_DABRX:
1161                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1162                 break;
1163         case KVM_REG_PPC_DSCR:
1164                 vcpu->arch.dscr = set_reg_val(id, *val);
1165                 break;
1166         case KVM_REG_PPC_PURR:
1167                 vcpu->arch.purr = set_reg_val(id, *val);
1168                 break;
1169         case KVM_REG_PPC_SPURR:
1170                 vcpu->arch.spurr = set_reg_val(id, *val);
1171                 break;
1172         case KVM_REG_PPC_AMR:
1173                 vcpu->arch.amr = set_reg_val(id, *val);
1174                 break;
1175         case KVM_REG_PPC_UAMOR:
1176                 vcpu->arch.uamor = set_reg_val(id, *val);
1177                 break;
1178         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1179                 i = id - KVM_REG_PPC_MMCR0;
1180                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1181                 break;
1182         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1183                 i = id - KVM_REG_PPC_PMC1;
1184                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1185                 break;
1186         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1187                 i = id - KVM_REG_PPC_SPMC1;
1188                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1189                 break;
1190         case KVM_REG_PPC_SIAR:
1191                 vcpu->arch.siar = set_reg_val(id, *val);
1192                 break;
1193         case KVM_REG_PPC_SDAR:
1194                 vcpu->arch.sdar = set_reg_val(id, *val);
1195                 break;
1196         case KVM_REG_PPC_SIER:
1197                 vcpu->arch.sier = set_reg_val(id, *val);
1198                 break;
1199         case KVM_REG_PPC_IAMR:
1200                 vcpu->arch.iamr = set_reg_val(id, *val);
1201                 break;
1202         case KVM_REG_PPC_PSPB:
1203                 vcpu->arch.pspb = set_reg_val(id, *val);
1204                 break;
1205         case KVM_REG_PPC_DPDES:
1206                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1207                 break;
1208         case KVM_REG_PPC_DAWR:
1209                 vcpu->arch.dawr = set_reg_val(id, *val);
1210                 break;
1211         case KVM_REG_PPC_DAWRX:
1212                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1213                 break;
1214         case KVM_REG_PPC_CIABR:
1215                 vcpu->arch.ciabr = set_reg_val(id, *val);
1216                 /* Don't allow setting breakpoints in hypervisor code */
1217                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1218                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1219                 break;
1220         case KVM_REG_PPC_CSIGR:
1221                 vcpu->arch.csigr = set_reg_val(id, *val);
1222                 break;
1223         case KVM_REG_PPC_TACR:
1224                 vcpu->arch.tacr = set_reg_val(id, *val);
1225                 break;
1226         case KVM_REG_PPC_TCSCR:
1227                 vcpu->arch.tcscr = set_reg_val(id, *val);
1228                 break;
1229         case KVM_REG_PPC_PID:
1230                 vcpu->arch.pid = set_reg_val(id, *val);
1231                 break;
1232         case KVM_REG_PPC_ACOP:
1233                 vcpu->arch.acop = set_reg_val(id, *val);
1234                 break;
1235         case KVM_REG_PPC_WORT:
1236                 vcpu->arch.wort = set_reg_val(id, *val);
1237                 break;
1238         case KVM_REG_PPC_VPA_ADDR:
1239                 addr = set_reg_val(id, *val);
1240                 r = -EINVAL;
1241                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1242                               vcpu->arch.dtl.next_gpa))
1243                         break;
1244                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1245                 break;
1246         case KVM_REG_PPC_VPA_SLB:
1247                 addr = val->vpaval.addr;
1248                 len = val->vpaval.length;
1249                 r = -EINVAL;
1250                 if (addr && !vcpu->arch.vpa.next_gpa)
1251                         break;
1252                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1253                 break;
1254         case KVM_REG_PPC_VPA_DTL:
1255                 addr = val->vpaval.addr;
1256                 len = val->vpaval.length;
1257                 r = -EINVAL;
1258                 if (addr && (len < sizeof(struct dtl_entry) ||
1259                              !vcpu->arch.vpa.next_gpa))
1260                         break;
1261                 len -= len % sizeof(struct dtl_entry);
1262                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1263                 break;
1264         case KVM_REG_PPC_TB_OFFSET:
1265                 /* round up to multiple of 2^24 */
1266                 vcpu->arch.vcore->tb_offset =
1267                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1268                 break;
1269         case KVM_REG_PPC_LPCR:
1270                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1271                 break;
1272         case KVM_REG_PPC_LPCR_64:
1273                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1274                 break;
1275         case KVM_REG_PPC_PPR:
1276                 vcpu->arch.ppr = set_reg_val(id, *val);
1277                 break;
1278 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1279         case KVM_REG_PPC_TFHAR:
1280                 vcpu->arch.tfhar = set_reg_val(id, *val);
1281                 break;
1282         case KVM_REG_PPC_TFIAR:
1283                 vcpu->arch.tfiar = set_reg_val(id, *val);
1284                 break;
1285         case KVM_REG_PPC_TEXASR:
1286                 vcpu->arch.texasr = set_reg_val(id, *val);
1287                 break;
1288         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1289                 i = id - KVM_REG_PPC_TM_GPR0;
1290                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1291                 break;
1292         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1293         {
1294                 int j;
1295                 i = id - KVM_REG_PPC_TM_VSR0;
1296                 if (i < 32)
1297                         for (j = 0; j < TS_FPRWIDTH; j++)
1298                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1299                 else
1300                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1301                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1302                         else
1303                                 r = -ENXIO;
1304                 break;
1305         }
1306         case KVM_REG_PPC_TM_CR:
1307                 vcpu->arch.cr_tm = set_reg_val(id, *val);
1308                 break;
1309         case KVM_REG_PPC_TM_LR:
1310                 vcpu->arch.lr_tm = set_reg_val(id, *val);
1311                 break;
1312         case KVM_REG_PPC_TM_CTR:
1313                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1314                 break;
1315         case KVM_REG_PPC_TM_FPSCR:
1316                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1317                 break;
1318         case KVM_REG_PPC_TM_AMR:
1319                 vcpu->arch.amr_tm = set_reg_val(id, *val);
1320                 break;
1321         case KVM_REG_PPC_TM_PPR:
1322                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1323                 break;
1324         case KVM_REG_PPC_TM_VRSAVE:
1325                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1326                 break;
1327         case KVM_REG_PPC_TM_VSCR:
1328                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1329                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1330                 else
1331                         r = - ENXIO;
1332                 break;
1333         case KVM_REG_PPC_TM_DSCR:
1334                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1335                 break;
1336         case KVM_REG_PPC_TM_TAR:
1337                 vcpu->arch.tar_tm = set_reg_val(id, *val);
1338                 break;
1339 #endif
1340         case KVM_REG_PPC_ARCH_COMPAT:
1341                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1342                 break;
1343         default:
1344                 r = -EINVAL;
1345                 break;
1346         }
1347
1348         return r;
1349 }
1350
1351 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1352 {
1353         struct kvmppc_vcore *vcore;
1354
1355         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1356
1357         if (vcore == NULL)
1358                 return NULL;
1359
1360         INIT_LIST_HEAD(&vcore->runnable_threads);
1361         spin_lock_init(&vcore->lock);
1362         init_waitqueue_head(&vcore->wq);
1363         vcore->preempt_tb = TB_NIL;
1364         vcore->lpcr = kvm->arch.lpcr;
1365         vcore->first_vcpuid = core * threads_per_subcore;
1366         vcore->kvm = kvm;
1367
1368         vcore->mpp_buffer_is_valid = false;
1369
1370         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1371                 vcore->mpp_buffer = (void *)__get_free_pages(
1372                         GFP_KERNEL|__GFP_ZERO,
1373                         MPP_BUFFER_ORDER);
1374
1375         return vcore;
1376 }
1377
1378 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1379                                                    unsigned int id)
1380 {
1381         struct kvm_vcpu *vcpu;
1382         int err = -EINVAL;
1383         int core;
1384         struct kvmppc_vcore *vcore;
1385
1386         core = id / threads_per_subcore;
1387         if (core >= KVM_MAX_VCORES)
1388                 goto out;
1389
1390         err = -ENOMEM;
1391         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1392         if (!vcpu)
1393                 goto out;
1394
1395         err = kvm_vcpu_init(vcpu, kvm, id);
1396         if (err)
1397                 goto free_vcpu;
1398
1399         vcpu->arch.shared = &vcpu->arch.shregs;
1400 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1401         /*
1402          * The shared struct is never shared on HV,
1403          * so we can always use host endianness
1404          */
1405 #ifdef __BIG_ENDIAN__
1406         vcpu->arch.shared_big_endian = true;
1407 #else
1408         vcpu->arch.shared_big_endian = false;
1409 #endif
1410 #endif
1411         vcpu->arch.mmcr[0] = MMCR0_FC;
1412         vcpu->arch.ctrl = CTRL_RUNLATCH;
1413         /* default to host PVR, since we can't spoof it */
1414         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1415         spin_lock_init(&vcpu->arch.vpa_update_lock);
1416         spin_lock_init(&vcpu->arch.tbacct_lock);
1417         vcpu->arch.busy_preempt = TB_NIL;
1418         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1419
1420         kvmppc_mmu_book3s_hv_init(vcpu);
1421
1422         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1423
1424         init_waitqueue_head(&vcpu->arch.cpu_run);
1425
1426         mutex_lock(&kvm->lock);
1427         vcore = kvm->arch.vcores[core];
1428         if (!vcore) {
1429                 vcore = kvmppc_vcore_create(kvm, core);
1430                 kvm->arch.vcores[core] = vcore;
1431                 kvm->arch.online_vcores++;
1432         }
1433         mutex_unlock(&kvm->lock);
1434
1435         if (!vcore)
1436                 goto free_vcpu;
1437
1438         spin_lock(&vcore->lock);
1439         ++vcore->num_threads;
1440         spin_unlock(&vcore->lock);
1441         vcpu->arch.vcore = vcore;
1442         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1443
1444         vcpu->arch.cpu_type = KVM_CPU_3S_64;
1445         kvmppc_sanity_check(vcpu);
1446
1447         return vcpu;
1448
1449 free_vcpu:
1450         kmem_cache_free(kvm_vcpu_cache, vcpu);
1451 out:
1452         return ERR_PTR(err);
1453 }
1454
1455 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1456 {
1457         if (vpa->pinned_addr)
1458                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1459                                         vpa->dirty);
1460 }
1461
1462 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1463 {
1464         spin_lock(&vcpu->arch.vpa_update_lock);
1465         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1466         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1467         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1468         spin_unlock(&vcpu->arch.vpa_update_lock);
1469         kvm_vcpu_uninit(vcpu);
1470         kmem_cache_free(kvm_vcpu_cache, vcpu);
1471 }
1472
1473 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1474 {
1475         /* Indicate we want to get back into the guest */
1476         return 1;
1477 }
1478
1479 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1480 {
1481         unsigned long dec_nsec, now;
1482
1483         now = get_tb();
1484         if (now > vcpu->arch.dec_expires) {
1485                 /* decrementer has already gone negative */
1486                 kvmppc_core_queue_dec(vcpu);
1487                 kvmppc_core_prepare_to_enter(vcpu);
1488                 return;
1489         }
1490         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1491                    / tb_ticks_per_sec;
1492         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1493                       HRTIMER_MODE_REL);
1494         vcpu->arch.timer_running = 1;
1495 }
1496
1497 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1498 {
1499         vcpu->arch.ceded = 0;
1500         if (vcpu->arch.timer_running) {
1501                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1502                 vcpu->arch.timer_running = 0;
1503         }
1504 }
1505
1506 extern void __kvmppc_vcore_entry(void);
1507
1508 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1509                                    struct kvm_vcpu *vcpu)
1510 {
1511         u64 now;
1512
1513         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1514                 return;
1515         spin_lock_irq(&vcpu->arch.tbacct_lock);
1516         now = mftb();
1517         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1518                 vcpu->arch.stolen_logged;
1519         vcpu->arch.busy_preempt = now;
1520         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1521         spin_unlock_irq(&vcpu->arch.tbacct_lock);
1522         --vc->n_runnable;
1523         list_del(&vcpu->arch.run_list);
1524 }
1525
1526 static int kvmppc_grab_hwthread(int cpu)
1527 {
1528         struct paca_struct *tpaca;
1529         long timeout = 10000;
1530
1531         tpaca = &paca[cpu];
1532
1533         /* Ensure the thread won't go into the kernel if it wakes */
1534         tpaca->kvm_hstate.hwthread_req = 1;
1535         tpaca->kvm_hstate.kvm_vcpu = NULL;
1536
1537         /*
1538          * If the thread is already executing in the kernel (e.g. handling
1539          * a stray interrupt), wait for it to get back to nap mode.
1540          * The smp_mb() is to ensure that our setting of hwthread_req
1541          * is visible before we look at hwthread_state, so if this
1542          * races with the code at system_reset_pSeries and the thread
1543          * misses our setting of hwthread_req, we are sure to see its
1544          * setting of hwthread_state, and vice versa.
1545          */
1546         smp_mb();
1547         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1548                 if (--timeout <= 0) {
1549                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1550                         return -EBUSY;
1551                 }
1552                 udelay(1);
1553         }
1554         return 0;
1555 }
1556
1557 static void kvmppc_release_hwthread(int cpu)
1558 {
1559         struct paca_struct *tpaca;
1560
1561         tpaca = &paca[cpu];
1562         tpaca->kvm_hstate.hwthread_req = 0;
1563         tpaca->kvm_hstate.kvm_vcpu = NULL;
1564 }
1565
1566 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1567 {
1568         int cpu;
1569         struct paca_struct *tpaca;
1570         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1571
1572         if (vcpu->arch.timer_running) {
1573                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1574                 vcpu->arch.timer_running = 0;
1575         }
1576         cpu = vc->pcpu + vcpu->arch.ptid;
1577         tpaca = &paca[cpu];
1578         tpaca->kvm_hstate.kvm_vcpu = vcpu;
1579         tpaca->kvm_hstate.kvm_vcore = vc;
1580         tpaca->kvm_hstate.ptid = vcpu->arch.ptid;
1581         vcpu->cpu = vc->pcpu;
1582         smp_wmb();
1583 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1584         if (cpu != smp_processor_id()) {
1585                 xics_wake_cpu(cpu);
1586                 if (vcpu->arch.ptid)
1587                         ++vc->n_woken;
1588         }
1589 #endif
1590 }
1591
1592 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1593 {
1594         int i;
1595
1596         HMT_low();
1597         i = 0;
1598         while (vc->nap_count < vc->n_woken) {
1599                 if (++i >= 1000000) {
1600                         pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1601                                vc->nap_count, vc->n_woken);
1602                         break;
1603                 }
1604                 cpu_relax();
1605         }
1606         HMT_medium();
1607 }
1608
1609 /*
1610  * Check that we are on thread 0 and that any other threads in
1611  * this core are off-line.  Then grab the threads so they can't
1612  * enter the kernel.
1613  */
1614 static int on_primary_thread(void)
1615 {
1616         int cpu = smp_processor_id();
1617         int thr;
1618
1619         /* Are we on a primary subcore? */
1620         if (cpu_thread_in_subcore(cpu))
1621                 return 0;
1622
1623         thr = 0;
1624         while (++thr < threads_per_subcore)
1625                 if (cpu_online(cpu + thr))
1626                         return 0;
1627
1628         /* Grab all hw threads so they can't go into the kernel */
1629         for (thr = 1; thr < threads_per_subcore; ++thr) {
1630                 if (kvmppc_grab_hwthread(cpu + thr)) {
1631                         /* Couldn't grab one; let the others go */
1632                         do {
1633                                 kvmppc_release_hwthread(cpu + thr);
1634                         } while (--thr > 0);
1635                         return 0;
1636                 }
1637         }
1638         return 1;
1639 }
1640
1641 static void kvmppc_start_saving_l2_cache(struct kvmppc_vcore *vc)
1642 {
1643         phys_addr_t phy_addr, mpp_addr;
1644
1645         phy_addr = (phys_addr_t)virt_to_phys(vc->mpp_buffer);
1646         mpp_addr = phy_addr & PPC_MPPE_ADDRESS_MASK;
1647
1648         mtspr(SPRN_MPPR, mpp_addr | PPC_MPPR_FETCH_ABORT);
1649         logmpp(mpp_addr | PPC_LOGMPP_LOG_L2);
1650
1651         vc->mpp_buffer_is_valid = true;
1652 }
1653
1654 static void kvmppc_start_restoring_l2_cache(const struct kvmppc_vcore *vc)
1655 {
1656         phys_addr_t phy_addr, mpp_addr;
1657
1658         phy_addr = virt_to_phys(vc->mpp_buffer);
1659         mpp_addr = phy_addr & PPC_MPPE_ADDRESS_MASK;
1660
1661         /* We must abort any in-progress save operations to ensure
1662          * the table is valid so that prefetch engine knows when to
1663          * stop prefetching. */
1664         logmpp(mpp_addr | PPC_LOGMPP_LOG_ABORT);
1665         mtspr(SPRN_MPPR, mpp_addr | PPC_MPPR_FETCH_WHOLE_TABLE);
1666 }
1667
1668 /*
1669  * Run a set of guest threads on a physical core.
1670  * Called with vc->lock held.
1671  */
1672 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1673 {
1674         struct kvm_vcpu *vcpu, *vnext;
1675         long ret;
1676         u64 now;
1677         int i, need_vpa_update;
1678         int srcu_idx;
1679         struct kvm_vcpu *vcpus_to_update[threads_per_core];
1680
1681         /* don't start if any threads have a signal pending */
1682         need_vpa_update = 0;
1683         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1684                 if (signal_pending(vcpu->arch.run_task))
1685                         return;
1686                 if (vcpu->arch.vpa.update_pending ||
1687                     vcpu->arch.slb_shadow.update_pending ||
1688                     vcpu->arch.dtl.update_pending)
1689                         vcpus_to_update[need_vpa_update++] = vcpu;
1690         }
1691
1692         /*
1693          * Initialize *vc, in particular vc->vcore_state, so we can
1694          * drop the vcore lock if necessary.
1695          */
1696         vc->n_woken = 0;
1697         vc->nap_count = 0;
1698         vc->entry_exit_count = 0;
1699         vc->vcore_state = VCORE_STARTING;
1700         vc->in_guest = 0;
1701         vc->napping_threads = 0;
1702
1703         /*
1704          * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1705          * which can't be called with any spinlocks held.
1706          */
1707         if (need_vpa_update) {
1708                 spin_unlock(&vc->lock);
1709                 for (i = 0; i < need_vpa_update; ++i)
1710                         kvmppc_update_vpas(vcpus_to_update[i]);
1711                 spin_lock(&vc->lock);
1712         }
1713
1714         /*
1715          * Make sure we are running on primary threads, and that secondary
1716          * threads are offline.  Also check if the number of threads in this
1717          * guest are greater than the current system threads per guest.
1718          */
1719         if ((threads_per_core > 1) &&
1720             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
1721                 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1722                         vcpu->arch.ret = -EBUSY;
1723                 goto out;
1724         }
1725
1726
1727         vc->pcpu = smp_processor_id();
1728         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1729                 kvmppc_start_thread(vcpu);
1730                 kvmppc_create_dtl_entry(vcpu, vc);
1731         }
1732
1733         /* Set this explicitly in case thread 0 doesn't have a vcpu */
1734         get_paca()->kvm_hstate.kvm_vcore = vc;
1735         get_paca()->kvm_hstate.ptid = 0;
1736
1737         vc->vcore_state = VCORE_RUNNING;
1738         preempt_disable();
1739         spin_unlock(&vc->lock);
1740
1741         kvm_guest_enter();
1742
1743         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
1744
1745         if (vc->mpp_buffer_is_valid)
1746                 kvmppc_start_restoring_l2_cache(vc);
1747
1748         __kvmppc_vcore_entry();
1749
1750         spin_lock(&vc->lock);
1751
1752         if (vc->mpp_buffer)
1753                 kvmppc_start_saving_l2_cache(vc);
1754
1755         /* disable sending of IPIs on virtual external irqs */
1756         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1757                 vcpu->cpu = -1;
1758         /* wait for secondary threads to finish writing their state to memory */
1759         if (vc->nap_count < vc->n_woken)
1760                 kvmppc_wait_for_nap(vc);
1761         for (i = 0; i < threads_per_subcore; ++i)
1762                 kvmppc_release_hwthread(vc->pcpu + i);
1763         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
1764         vc->vcore_state = VCORE_EXITING;
1765         spin_unlock(&vc->lock);
1766
1767         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
1768
1769         /* make sure updates to secondary vcpu structs are visible now */
1770         smp_mb();
1771         kvm_guest_exit();
1772
1773         preempt_enable();
1774         cond_resched();
1775
1776         spin_lock(&vc->lock);
1777         now = get_tb();
1778         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1779                 /* cancel pending dec exception if dec is positive */
1780                 if (now < vcpu->arch.dec_expires &&
1781                     kvmppc_core_pending_dec(vcpu))
1782                         kvmppc_core_dequeue_dec(vcpu);
1783
1784                 ret = RESUME_GUEST;
1785                 if (vcpu->arch.trap)
1786                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1787                                                     vcpu->arch.run_task);
1788
1789                 vcpu->arch.ret = ret;
1790                 vcpu->arch.trap = 0;
1791
1792                 if (vcpu->arch.ceded) {
1793                         if (!is_kvmppc_resume_guest(ret))
1794                                 kvmppc_end_cede(vcpu);
1795                         else
1796                                 kvmppc_set_timer(vcpu);
1797                 }
1798         }
1799
1800  out:
1801         vc->vcore_state = VCORE_INACTIVE;
1802         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1803                                  arch.run_list) {
1804                 if (!is_kvmppc_resume_guest(vcpu->arch.ret)) {
1805                         kvmppc_remove_runnable(vc, vcpu);
1806                         wake_up(&vcpu->arch.cpu_run);
1807                 }
1808         }
1809 }
1810
1811 /*
1812  * Wait for some other vcpu thread to execute us, and
1813  * wake us up when we need to handle something in the host.
1814  */
1815 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1816 {
1817         DEFINE_WAIT(wait);
1818
1819         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1820         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1821                 schedule();
1822         finish_wait(&vcpu->arch.cpu_run, &wait);
1823 }
1824
1825 /*
1826  * All the vcpus in this vcore are idle, so wait for a decrementer
1827  * or external interrupt to one of the vcpus.  vc->lock is held.
1828  */
1829 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1830 {
1831         struct kvm_vcpu *vcpu;
1832         int do_sleep = 1;
1833
1834         DEFINE_WAIT(wait);
1835
1836         prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1837
1838         /*
1839          * Check one last time for pending exceptions and ceded state after
1840          * we put ourselves on the wait queue
1841          */
1842         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1843                 if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded) {
1844                         do_sleep = 0;
1845                         break;
1846                 }
1847         }
1848
1849         if (!do_sleep) {
1850                 finish_wait(&vc->wq, &wait);
1851                 return;
1852         }
1853
1854         vc->vcore_state = VCORE_SLEEPING;
1855         spin_unlock(&vc->lock);
1856         schedule();
1857         finish_wait(&vc->wq, &wait);
1858         spin_lock(&vc->lock);
1859         vc->vcore_state = VCORE_INACTIVE;
1860 }
1861
1862 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1863 {
1864         int n_ceded;
1865         struct kvmppc_vcore *vc;
1866         struct kvm_vcpu *v, *vn;
1867
1868         kvm_run->exit_reason = 0;
1869         vcpu->arch.ret = RESUME_GUEST;
1870         vcpu->arch.trap = 0;
1871         kvmppc_update_vpas(vcpu);
1872
1873         /*
1874          * Synchronize with other threads in this virtual core
1875          */
1876         vc = vcpu->arch.vcore;
1877         spin_lock(&vc->lock);
1878         vcpu->arch.ceded = 0;
1879         vcpu->arch.run_task = current;
1880         vcpu->arch.kvm_run = kvm_run;
1881         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
1882         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1883         vcpu->arch.busy_preempt = TB_NIL;
1884         list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1885         ++vc->n_runnable;
1886
1887         /*
1888          * This happens the first time this is called for a vcpu.
1889          * If the vcore is already running, we may be able to start
1890          * this thread straight away and have it join in.
1891          */
1892         if (!signal_pending(current)) {
1893                 if (vc->vcore_state == VCORE_RUNNING &&
1894                     VCORE_EXIT_COUNT(vc) == 0) {
1895                         kvmppc_create_dtl_entry(vcpu, vc);
1896                         kvmppc_start_thread(vcpu);
1897                 } else if (vc->vcore_state == VCORE_SLEEPING) {
1898                         wake_up(&vc->wq);
1899                 }
1900
1901         }
1902
1903         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1904                !signal_pending(current)) {
1905                 if (vc->vcore_state != VCORE_INACTIVE) {
1906                         spin_unlock(&vc->lock);
1907                         kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1908                         spin_lock(&vc->lock);
1909                         continue;
1910                 }
1911                 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1912                                          arch.run_list) {
1913                         kvmppc_core_prepare_to_enter(v);
1914                         if (signal_pending(v->arch.run_task)) {
1915                                 kvmppc_remove_runnable(vc, v);
1916                                 v->stat.signal_exits++;
1917                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1918                                 v->arch.ret = -EINTR;
1919                                 wake_up(&v->arch.cpu_run);
1920                         }
1921                 }
1922                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1923                         break;
1924                 vc->runner = vcpu;
1925                 n_ceded = 0;
1926                 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
1927                         if (!v->arch.pending_exceptions)
1928                                 n_ceded += v->arch.ceded;
1929                         else
1930                                 v->arch.ceded = 0;
1931                 }
1932                 if (n_ceded == vc->n_runnable)
1933                         kvmppc_vcore_blocked(vc);
1934                 else
1935                         kvmppc_run_core(vc);
1936                 vc->runner = NULL;
1937         }
1938
1939         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1940                (vc->vcore_state == VCORE_RUNNING ||
1941                 vc->vcore_state == VCORE_EXITING)) {
1942                 spin_unlock(&vc->lock);
1943                 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1944                 spin_lock(&vc->lock);
1945         }
1946
1947         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1948                 kvmppc_remove_runnable(vc, vcpu);
1949                 vcpu->stat.signal_exits++;
1950                 kvm_run->exit_reason = KVM_EXIT_INTR;
1951                 vcpu->arch.ret = -EINTR;
1952         }
1953
1954         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1955                 /* Wake up some vcpu to run the core */
1956                 v = list_first_entry(&vc->runnable_threads,
1957                                      struct kvm_vcpu, arch.run_list);
1958                 wake_up(&v->arch.cpu_run);
1959         }
1960
1961         spin_unlock(&vc->lock);
1962         return vcpu->arch.ret;
1963 }
1964
1965 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
1966 {
1967         int r;
1968         int srcu_idx;
1969
1970         if (!vcpu->arch.sane) {
1971                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1972                 return -EINVAL;
1973         }
1974
1975         kvmppc_core_prepare_to_enter(vcpu);
1976
1977         /* No need to go into the guest when all we'll do is come back out */
1978         if (signal_pending(current)) {
1979                 run->exit_reason = KVM_EXIT_INTR;
1980                 return -EINTR;
1981         }
1982
1983         atomic_inc(&vcpu->kvm->arch.vcpus_running);
1984         /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1985         smp_mb();
1986
1987         /* On the first time here, set up HTAB and VRMA or RMA */
1988         if (!vcpu->kvm->arch.rma_setup_done) {
1989                 r = kvmppc_hv_setup_htab_rma(vcpu);
1990                 if (r)
1991                         goto out;
1992         }
1993
1994         flush_fp_to_thread(current);
1995         flush_altivec_to_thread(current);
1996         flush_vsx_to_thread(current);
1997         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1998         vcpu->arch.pgdir = current->mm->pgd;
1999         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2000
2001         do {
2002                 r = kvmppc_run_vcpu(run, vcpu);
2003
2004                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2005                     !(vcpu->arch.shregs.msr & MSR_PR)) {
2006                         r = kvmppc_pseries_do_hcall(vcpu);
2007                         kvmppc_core_prepare_to_enter(vcpu);
2008                 } else if (r == RESUME_PAGE_FAULT) {
2009                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2010                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
2011                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2012                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2013                 }
2014         } while (is_kvmppc_resume_guest(r));
2015
2016  out:
2017         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2018         atomic_dec(&vcpu->kvm->arch.vcpus_running);
2019         return r;
2020 }
2021
2022
2023 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
2024    Assumes POWER7 or PPC970. */
2025 static inline int lpcr_rmls(unsigned long rma_size)
2026 {
2027         switch (rma_size) {
2028         case 32ul << 20:        /* 32 MB */
2029                 if (cpu_has_feature(CPU_FTR_ARCH_206))
2030                         return 8;       /* only supported on POWER7 */
2031                 return -1;
2032         case 64ul << 20:        /* 64 MB */
2033                 return 3;
2034         case 128ul << 20:       /* 128 MB */
2035                 return 7;
2036         case 256ul << 20:       /* 256 MB */
2037                 return 4;
2038         case 1ul << 30:         /* 1 GB */
2039                 return 2;
2040         case 16ul << 30:        /* 16 GB */
2041                 return 1;
2042         case 256ul << 30:       /* 256 GB */
2043                 return 0;
2044         default:
2045                 return -1;
2046         }
2047 }
2048
2049 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2050 {
2051         struct page *page;
2052         struct kvm_rma_info *ri = vma->vm_file->private_data;
2053
2054         if (vmf->pgoff >= kvm_rma_pages)
2055                 return VM_FAULT_SIGBUS;
2056
2057         page = pfn_to_page(ri->base_pfn + vmf->pgoff);
2058         get_page(page);
2059         vmf->page = page;
2060         return 0;
2061 }
2062
2063 static const struct vm_operations_struct kvm_rma_vm_ops = {
2064         .fault = kvm_rma_fault,
2065 };
2066
2067 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
2068 {
2069         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2070         vma->vm_ops = &kvm_rma_vm_ops;
2071         return 0;
2072 }
2073
2074 static int kvm_rma_release(struct inode *inode, struct file *filp)
2075 {
2076         struct kvm_rma_info *ri = filp->private_data;
2077
2078         kvm_release_rma(ri);
2079         return 0;
2080 }
2081
2082 static const struct file_operations kvm_rma_fops = {
2083         .mmap           = kvm_rma_mmap,
2084         .release        = kvm_rma_release,
2085 };
2086
2087 static long kvm_vm_ioctl_allocate_rma(struct kvm *kvm,
2088                                       struct kvm_allocate_rma *ret)
2089 {
2090         long fd;
2091         struct kvm_rma_info *ri;
2092         /*
2093          * Only do this on PPC970 in HV mode
2094          */
2095         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
2096             !cpu_has_feature(CPU_FTR_ARCH_201))
2097                 return -EINVAL;
2098
2099         if (!kvm_rma_pages)
2100                 return -EINVAL;
2101
2102         ri = kvm_alloc_rma();
2103         if (!ri)
2104                 return -ENOMEM;
2105
2106         fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC);
2107         if (fd < 0)
2108                 kvm_release_rma(ri);
2109
2110         ret->rma_size = kvm_rma_pages << PAGE_SHIFT;
2111         return fd;
2112 }
2113
2114 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2115                                      int linux_psize)
2116 {
2117         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2118
2119         if (!def->shift)
2120                 return;
2121         (*sps)->page_shift = def->shift;
2122         (*sps)->slb_enc = def->sllp;
2123         (*sps)->enc[0].page_shift = def->shift;
2124         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
2125         /*
2126          * Add 16MB MPSS support if host supports it
2127          */
2128         if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2129                 (*sps)->enc[1].page_shift = 24;
2130                 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2131         }
2132         (*sps)++;
2133 }
2134
2135 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2136                                          struct kvm_ppc_smmu_info *info)
2137 {
2138         struct kvm_ppc_one_seg_page_size *sps;
2139
2140         info->flags = KVM_PPC_PAGE_SIZES_REAL;
2141         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2142                 info->flags |= KVM_PPC_1T_SEGMENTS;
2143         info->slb_size = mmu_slb_size;
2144
2145         /* We only support these sizes for now, and no muti-size segments */
2146         sps = &info->sps[0];
2147         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2148         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2149         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2150
2151         return 0;
2152 }
2153
2154 /*
2155  * Get (and clear) the dirty memory log for a memory slot.
2156  */
2157 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2158                                          struct kvm_dirty_log *log)
2159 {
2160         struct kvm_memory_slot *memslot;
2161         int r;
2162         unsigned long n;
2163
2164         mutex_lock(&kvm->slots_lock);
2165
2166         r = -EINVAL;
2167         if (log->slot >= KVM_USER_MEM_SLOTS)
2168                 goto out;
2169
2170         memslot = id_to_memslot(kvm->memslots, log->slot);
2171         r = -ENOENT;
2172         if (!memslot->dirty_bitmap)
2173                 goto out;
2174
2175         n = kvm_dirty_bitmap_bytes(memslot);
2176         memset(memslot->dirty_bitmap, 0, n);
2177
2178         r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2179         if (r)
2180                 goto out;
2181
2182         r = -EFAULT;
2183         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2184                 goto out;
2185
2186         r = 0;
2187 out:
2188         mutex_unlock(&kvm->slots_lock);
2189         return r;
2190 }
2191
2192 static void unpin_slot(struct kvm_memory_slot *memslot)
2193 {
2194         unsigned long *physp;
2195         unsigned long j, npages, pfn;
2196         struct page *page;
2197
2198         physp = memslot->arch.slot_phys;
2199         npages = memslot->npages;
2200         if (!physp)
2201                 return;
2202         for (j = 0; j < npages; j++) {
2203                 if (!(physp[j] & KVMPPC_GOT_PAGE))
2204                         continue;
2205                 pfn = physp[j] >> PAGE_SHIFT;
2206                 page = pfn_to_page(pfn);
2207                 SetPageDirty(page);
2208                 put_page(page);
2209         }
2210 }
2211
2212 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2213                                         struct kvm_memory_slot *dont)
2214 {
2215         if (!dont || free->arch.rmap != dont->arch.rmap) {
2216                 vfree(free->arch.rmap);
2217                 free->arch.rmap = NULL;
2218         }
2219         if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
2220                 unpin_slot(free);
2221                 vfree(free->arch.slot_phys);
2222                 free->arch.slot_phys = NULL;
2223         }
2224 }
2225
2226 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2227                                          unsigned long npages)
2228 {
2229         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2230         if (!slot->arch.rmap)
2231                 return -ENOMEM;
2232         slot->arch.slot_phys = NULL;
2233
2234         return 0;
2235 }
2236
2237 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2238                                         struct kvm_memory_slot *memslot,
2239                                         struct kvm_userspace_memory_region *mem)
2240 {
2241         unsigned long *phys;
2242
2243         /* Allocate a slot_phys array if needed */
2244         phys = memslot->arch.slot_phys;
2245         if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
2246                 phys = vzalloc(memslot->npages * sizeof(unsigned long));
2247                 if (!phys)
2248                         return -ENOMEM;
2249                 memslot->arch.slot_phys = phys;
2250         }
2251
2252         return 0;
2253 }
2254
2255 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2256                                 struct kvm_userspace_memory_region *mem,
2257                                 const struct kvm_memory_slot *old)
2258 {
2259         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2260         struct kvm_memory_slot *memslot;
2261
2262         if (npages && old->npages) {
2263                 /*
2264                  * If modifying a memslot, reset all the rmap dirty bits.
2265                  * If this is a new memslot, we don't need to do anything
2266                  * since the rmap array starts out as all zeroes,
2267                  * i.e. no pages are dirty.
2268                  */
2269                 memslot = id_to_memslot(kvm->memslots, mem->slot);
2270                 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2271         }
2272 }
2273
2274 /*
2275  * Update LPCR values in kvm->arch and in vcores.
2276  * Caller must hold kvm->lock.
2277  */
2278 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2279 {
2280         long int i;
2281         u32 cores_done = 0;
2282
2283         if ((kvm->arch.lpcr & mask) == lpcr)
2284                 return;
2285
2286         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2287
2288         for (i = 0; i < KVM_MAX_VCORES; ++i) {
2289                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2290                 if (!vc)
2291                         continue;
2292                 spin_lock(&vc->lock);
2293                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2294                 spin_unlock(&vc->lock);
2295                 if (++cores_done >= kvm->arch.online_vcores)
2296                         break;
2297         }
2298 }
2299
2300 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2301 {
2302         return;
2303 }
2304
2305 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2306 {
2307         int err = 0;
2308         struct kvm *kvm = vcpu->kvm;
2309         struct kvm_rma_info *ri = NULL;
2310         unsigned long hva;
2311         struct kvm_memory_slot *memslot;
2312         struct vm_area_struct *vma;
2313         unsigned long lpcr = 0, senc;
2314         unsigned long lpcr_mask = 0;
2315         unsigned long psize, porder;
2316         unsigned long rma_size;
2317         unsigned long rmls;
2318         unsigned long *physp;
2319         unsigned long i, npages;
2320         int srcu_idx;
2321
2322         mutex_lock(&kvm->lock);
2323         if (kvm->arch.rma_setup_done)
2324                 goto out;       /* another vcpu beat us to it */
2325
2326         /* Allocate hashed page table (if not done already) and reset it */
2327         if (!kvm->arch.hpt_virt) {
2328                 err = kvmppc_alloc_hpt(kvm, NULL);
2329                 if (err) {
2330                         pr_err("KVM: Couldn't alloc HPT\n");
2331                         goto out;
2332                 }
2333         }
2334
2335         /* Look up the memslot for guest physical address 0 */
2336         srcu_idx = srcu_read_lock(&kvm->srcu);
2337         memslot = gfn_to_memslot(kvm, 0);
2338
2339         /* We must have some memory at 0 by now */
2340         err = -EINVAL;
2341         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2342                 goto out_srcu;
2343
2344         /* Look up the VMA for the start of this memory slot */
2345         hva = memslot->userspace_addr;
2346         down_read(&current->mm->mmap_sem);
2347         vma = find_vma(current->mm, hva);
2348         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2349                 goto up_out;
2350
2351         psize = vma_kernel_pagesize(vma);
2352         porder = __ilog2(psize);
2353
2354         /* Is this one of our preallocated RMAs? */
2355         if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
2356             hva == vma->vm_start)
2357                 ri = vma->vm_file->private_data;
2358
2359         up_read(&current->mm->mmap_sem);
2360
2361         if (!ri) {
2362                 /* On POWER7, use VRMA; on PPC970, give up */
2363                 err = -EPERM;
2364                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2365                         pr_err("KVM: CPU requires an RMO\n");
2366                         goto out_srcu;
2367                 }
2368
2369                 /* We can handle 4k, 64k or 16M pages in the VRMA */
2370                 err = -EINVAL;
2371                 if (!(psize == 0x1000 || psize == 0x10000 ||
2372                       psize == 0x1000000))
2373                         goto out_srcu;
2374
2375                 /* Update VRMASD field in the LPCR */
2376                 senc = slb_pgsize_encoding(psize);
2377                 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2378                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
2379                 lpcr_mask = LPCR_VRMASD;
2380                 /* the -4 is to account for senc values starting at 0x10 */
2381                 lpcr = senc << (LPCR_VRMASD_SH - 4);
2382
2383                 /* Create HPTEs in the hash page table for the VRMA */
2384                 kvmppc_map_vrma(vcpu, memslot, porder);
2385
2386         } else {
2387                 /* Set up to use an RMO region */
2388                 rma_size = kvm_rma_pages;
2389                 if (rma_size > memslot->npages)
2390                         rma_size = memslot->npages;
2391                 rma_size <<= PAGE_SHIFT;
2392                 rmls = lpcr_rmls(rma_size);
2393                 err = -EINVAL;
2394                 if ((long)rmls < 0) {
2395                         pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
2396                         goto out_srcu;
2397                 }
2398                 atomic_inc(&ri->use_count);
2399                 kvm->arch.rma = ri;
2400
2401                 /* Update LPCR and RMOR */
2402                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2403                         /* PPC970; insert RMLS value (split field) in HID4 */
2404                         lpcr_mask = (1ul << HID4_RMLS0_SH) |
2405                                 (3ul << HID4_RMLS2_SH) | HID4_RMOR;
2406                         lpcr = ((rmls >> 2) << HID4_RMLS0_SH) |
2407                                 ((rmls & 3) << HID4_RMLS2_SH);
2408                         /* RMOR is also in HID4 */
2409                         lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
2410                                 << HID4_RMOR_SH;
2411                 } else {
2412                         /* POWER7 */
2413                         lpcr_mask = LPCR_VPM0 | LPCR_VRMA_L | LPCR_RMLS;
2414                         lpcr = rmls << LPCR_RMLS_SH;
2415                         kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT;
2416                 }
2417                 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
2418                         ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
2419
2420                 /* Initialize phys addrs of pages in RMO */
2421                 npages = kvm_rma_pages;
2422                 porder = __ilog2(npages);
2423                 physp = memslot->arch.slot_phys;
2424                 if (physp) {
2425                         if (npages > memslot->npages)
2426                                 npages = memslot->npages;
2427                         spin_lock(&kvm->arch.slot_phys_lock);
2428                         for (i = 0; i < npages; ++i)
2429                                 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
2430                                         porder;
2431                         spin_unlock(&kvm->arch.slot_phys_lock);
2432                 }
2433         }
2434
2435         kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
2436
2437         /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
2438         smp_wmb();
2439         kvm->arch.rma_setup_done = 1;
2440         err = 0;
2441  out_srcu:
2442         srcu_read_unlock(&kvm->srcu, srcu_idx);
2443  out:
2444         mutex_unlock(&kvm->lock);
2445         return err;
2446
2447  up_out:
2448         up_read(&current->mm->mmap_sem);
2449         goto out_srcu;
2450 }
2451
2452 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2453 {
2454         unsigned long lpcr, lpid;
2455
2456         /* Allocate the guest's logical partition ID */
2457
2458         lpid = kvmppc_alloc_lpid();
2459         if ((long)lpid < 0)
2460                 return -ENOMEM;
2461         kvm->arch.lpid = lpid;
2462
2463         /*
2464          * Since we don't flush the TLB when tearing down a VM,
2465          * and this lpid might have previously been used,
2466          * make sure we flush on each core before running the new VM.
2467          */
2468         cpumask_setall(&kvm->arch.need_tlb_flush);
2469
2470         /* Start out with the default set of hcalls enabled */
2471         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
2472                sizeof(kvm->arch.enabled_hcalls));
2473
2474         kvm->arch.rma = NULL;
2475
2476         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
2477
2478         if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2479                 /* PPC970; HID4 is effectively the LPCR */
2480                 kvm->arch.host_lpid = 0;
2481                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
2482                 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
2483                 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
2484                         ((lpid & 0xf) << HID4_LPID5_SH);
2485         } else {
2486                 /* POWER7; init LPCR for virtual RMA mode */
2487                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
2488                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2489                 lpcr &= LPCR_PECE | LPCR_LPES;
2490                 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
2491                         LPCR_VPM0 | LPCR_VPM1;
2492                 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2493                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
2494                 /* On POWER8 turn on online bit to enable PURR/SPURR */
2495                 if (cpu_has_feature(CPU_FTR_ARCH_207S))
2496                         lpcr |= LPCR_ONL;
2497         }
2498         kvm->arch.lpcr = lpcr;
2499
2500         kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
2501         spin_lock_init(&kvm->arch.slot_phys_lock);
2502
2503         /*
2504          * Track that we now have a HV mode VM active. This blocks secondary
2505          * CPU threads from coming online.
2506          */
2507         kvm_hv_vm_activated();
2508
2509         return 0;
2510 }
2511
2512 static void kvmppc_free_vcores(struct kvm *kvm)
2513 {
2514         long int i;
2515
2516         for (i = 0; i < KVM_MAX_VCORES; ++i) {
2517                 if (kvm->arch.vcores[i] && kvm->arch.vcores[i]->mpp_buffer) {
2518                         struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2519                         free_pages((unsigned long)vc->mpp_buffer,
2520                                    MPP_BUFFER_ORDER);
2521                 }
2522                 kfree(kvm->arch.vcores[i]);
2523         }
2524         kvm->arch.online_vcores = 0;
2525 }
2526
2527 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
2528 {
2529         kvm_hv_vm_deactivated();
2530
2531         kvmppc_free_vcores(kvm);
2532         if (kvm->arch.rma) {
2533                 kvm_release_rma(kvm->arch.rma);
2534                 kvm->arch.rma = NULL;
2535         }
2536
2537         kvmppc_free_hpt(kvm);
2538 }
2539
2540 /* We don't need to emulate any privileged instructions or dcbz */
2541 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2542                                      unsigned int inst, int *advance)
2543 {
2544         return EMULATE_FAIL;
2545 }
2546
2547 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2548                                         ulong spr_val)
2549 {
2550         return EMULATE_FAIL;
2551 }
2552
2553 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2554                                         ulong *spr_val)
2555 {
2556         return EMULATE_FAIL;
2557 }
2558
2559 static int kvmppc_core_check_processor_compat_hv(void)
2560 {
2561         if (!cpu_has_feature(CPU_FTR_HVMODE))
2562                 return -EIO;
2563         return 0;
2564 }
2565
2566 static long kvm_arch_vm_ioctl_hv(struct file *filp,
2567                                  unsigned int ioctl, unsigned long arg)
2568 {
2569         struct kvm *kvm __maybe_unused = filp->private_data;
2570         void __user *argp = (void __user *)arg;
2571         long r;
2572
2573         switch (ioctl) {
2574
2575         case KVM_ALLOCATE_RMA: {
2576                 struct kvm_allocate_rma rma;
2577                 struct kvm *kvm = filp->private_data;
2578
2579                 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
2580                 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
2581                         r = -EFAULT;
2582                 break;
2583         }
2584
2585         case KVM_PPC_ALLOCATE_HTAB: {
2586                 u32 htab_order;
2587
2588                 r = -EFAULT;
2589                 if (get_user(htab_order, (u32 __user *)argp))
2590                         break;
2591                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2592                 if (r)
2593                         break;
2594                 r = -EFAULT;
2595                 if (put_user(htab_order, (u32 __user *)argp))
2596                         break;
2597                 r = 0;
2598                 break;
2599         }
2600
2601         case KVM_PPC_GET_HTAB_FD: {
2602                 struct kvm_get_htab_fd ghf;
2603
2604                 r = -EFAULT;
2605                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
2606                         break;
2607                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2608                 break;
2609         }
2610
2611         default:
2612                 r = -ENOTTY;
2613         }
2614
2615         return r;
2616 }
2617
2618 /*
2619  * List of hcall numbers to enable by default.
2620  * For compatibility with old userspace, we enable by default
2621  * all hcalls that were implemented before the hcall-enabling
2622  * facility was added.  Note this list should not include H_RTAS.
2623  */
2624 static unsigned int default_hcall_list[] = {
2625         H_REMOVE,
2626         H_ENTER,
2627         H_READ,
2628         H_PROTECT,
2629         H_BULK_REMOVE,
2630         H_GET_TCE,
2631         H_PUT_TCE,
2632         H_SET_DABR,
2633         H_SET_XDABR,
2634         H_CEDE,
2635         H_PROD,
2636         H_CONFER,
2637         H_REGISTER_VPA,
2638 #ifdef CONFIG_KVM_XICS
2639         H_EOI,
2640         H_CPPR,
2641         H_IPI,
2642         H_IPOLL,
2643         H_XIRR,
2644         H_XIRR_X,
2645 #endif
2646         0
2647 };
2648
2649 static void init_default_hcalls(void)
2650 {
2651         int i;
2652         unsigned int hcall;
2653
2654         for (i = 0; default_hcall_list[i]; ++i) {
2655                 hcall = default_hcall_list[i];
2656                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
2657                 __set_bit(hcall / 4, default_enabled_hcalls);
2658         }
2659 }
2660
2661 static struct kvmppc_ops kvm_ops_hv = {
2662         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2663         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2664         .get_one_reg = kvmppc_get_one_reg_hv,
2665         .set_one_reg = kvmppc_set_one_reg_hv,
2666         .vcpu_load   = kvmppc_core_vcpu_load_hv,
2667         .vcpu_put    = kvmppc_core_vcpu_put_hv,
2668         .set_msr     = kvmppc_set_msr_hv,
2669         .vcpu_run    = kvmppc_vcpu_run_hv,
2670         .vcpu_create = kvmppc_core_vcpu_create_hv,
2671         .vcpu_free   = kvmppc_core_vcpu_free_hv,
2672         .check_requests = kvmppc_core_check_requests_hv,
2673         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
2674         .flush_memslot  = kvmppc_core_flush_memslot_hv,
2675         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2676         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
2677         .unmap_hva = kvm_unmap_hva_hv,
2678         .unmap_hva_range = kvm_unmap_hva_range_hv,
2679         .age_hva  = kvm_age_hva_hv,
2680         .test_age_hva = kvm_test_age_hva_hv,
2681         .set_spte_hva = kvm_set_spte_hva_hv,
2682         .mmu_destroy  = kvmppc_mmu_destroy_hv,
2683         .free_memslot = kvmppc_core_free_memslot_hv,
2684         .create_memslot = kvmppc_core_create_memslot_hv,
2685         .init_vm =  kvmppc_core_init_vm_hv,
2686         .destroy_vm = kvmppc_core_destroy_vm_hv,
2687         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2688         .emulate_op = kvmppc_core_emulate_op_hv,
2689         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2690         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2691         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2692         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
2693         .hcall_implemented = kvmppc_hcall_impl_hv,
2694 };
2695
2696 static int kvmppc_book3s_init_hv(void)
2697 {
2698         int r;
2699         /*
2700          * FIXME!! Do we need to check on all cpus ?
2701          */
2702         r = kvmppc_core_check_processor_compat_hv();
2703         if (r < 0)
2704                 return -ENODEV;
2705
2706         kvm_ops_hv.owner = THIS_MODULE;
2707         kvmppc_hv_ops = &kvm_ops_hv;
2708
2709         init_default_hcalls();
2710
2711         r = kvmppc_mmu_hv_init();
2712         return r;
2713 }
2714
2715 static void kvmppc_book3s_exit_hv(void)
2716 {
2717         kvmppc_hv_ops = NULL;
2718 }
2719
2720 module_init(kvmppc_book3s_init_hv);
2721 module_exit(kvmppc_book3s_exit_hv);
2722 MODULE_LICENSE("GPL");
2723 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2724 MODULE_ALIAS("devname:kvm");