Pileus Git - ~andy/linux/blob - 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
  35 #include <asm/reg.h>
  36 #include <asm/cputable.h>
  37 #include <asm/cacheflush.h>
  38 #include <asm/tlbflush.h>
  39 #include <asm/uaccess.h>
  40 #include <asm/io.h>
  41 #include <asm/kvm_ppc.h>
  42 #include <asm/kvm_book3s.h>
  43 #include <asm/mmu_context.h>
  44 #include <asm/lppaca.h>
  45 #include <asm/processor.h>
  46 #include <asm/cputhreads.h>
  47 #include <asm/page.h>
  48 #include <asm/hvcall.h>
  49 #include <asm/switch_to.h>
  50 #include <asm/smp.h>
  51 #include <linux/gfp.h>
  52 #include <linux/vmalloc.h>
  53 #include <linux/highmem.h>
  54 #include <linux/hugetlb.h>
  55
  56 /* #define EXIT_DEBUG */
  57 /* #define EXIT_DEBUG_SIMPLE */
  58 /* #define EXIT_DEBUG_INT */
  59
  60 /* Used to indicate that a guest page fault needs to be handled */
  61 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  62
  63 /* Used as a "null" value for timebase values */
  64 #define TB_NIL  (~(u64)0)
  65
  66 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
  67 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
  68
  69 void kvmppc_fast_vcpu_kick(struct kvm_vcpu *vcpu)
  70 {
  71         int me;
  72         int cpu = vcpu->cpu;
  73         wait_queue_head_t *wqp;
  74
  75         wqp = kvm_arch_vcpu_wq(vcpu);
  76         if (waitqueue_active(wqp)) {
  77                 wake_up_interruptible(wqp);
  78                 ++vcpu->stat.halt_wakeup;
  79         }
  80
  81         me = get_cpu();
  82
  83         /* CPU points to the first thread of the core */
  84         if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
  85                 int real_cpu = cpu + vcpu->arch.ptid;
  86                 if (paca[real_cpu].kvm_hstate.xics_phys)
  87                         xics_wake_cpu(real_cpu);
  88                 else if (cpu_online(cpu))
  89                         smp_send_reschedule(cpu);
  90         }
  91         put_cpu();
  92 }
  93
  94 /*
  95  * We use the vcpu_load/put functions to measure stolen time.
  96  * Stolen time is counted as time when either the vcpu is able to
  97  * run as part of a virtual core, but the task running the vcore
  98  * is preempted or sleeping, or when the vcpu needs something done
  99  * in the kernel by the task running the vcpu, but that task is
 100  * preempted or sleeping.  Those two things have to be counted
 101  * separately, since one of the vcpu tasks will take on the job
 102  * of running the core, and the other vcpu tasks in the vcore will
 103  * sleep waiting for it to do that, but that sleep shouldn't count
 104  * as stolen time.
 105  *
 106  * Hence we accumulate stolen time when the vcpu can run as part of
 107  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 108  * needs its task to do other things in the kernel (for example,
 109  * service a page fault) in busy_stolen.  We don't accumulate
 110  * stolen time for a vcore when it is inactive, or for a vcpu
 111  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 112  * a misnomer; it means that the vcpu task is not executing in
 113  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 114  * the kernel.  We don't have any way of dividing up that time
 115  * between time that the vcpu is genuinely stopped, time that
 116  * the task is actively working on behalf of the vcpu, and time
 117  * that the task is preempted, so we don't count any of it as
 118  * stolen.
 119  *
 120  * Updates to busy_stolen are protected by arch.tbacct_lock;
 121  * updates to vc->stolen_tb are protected by the arch.tbacct_lock
 122  * of the vcpu that has taken responsibility for running the vcore
 123  * (i.e. vc->runner).  The stolen times are measured in units of
 124  * timebase ticks.  (Note that the != TB_NIL checks below are
 125  * purely defensive; they should never fail.)
 126  */
 127
 128 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 129 {
 130         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 131
 132         spin_lock(&vcpu->arch.tbacct_lock);
 133         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
 134             vc->preempt_tb != TB_NIL) {
 135                 vc->stolen_tb += mftb() - vc->preempt_tb;
 136                 vc->preempt_tb = TB_NIL;
 137         }
 138         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 139             vcpu->arch.busy_preempt != TB_NIL) {
 140                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 141                 vcpu->arch.busy_preempt = TB_NIL;
 142         }
 143         spin_unlock(&vcpu->arch.tbacct_lock);
 144 }
 145
 146 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
 147 {
 148         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 149
 150         spin_lock(&vcpu->arch.tbacct_lock);
 151         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
 152                 vc->preempt_tb = mftb();
 153         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 154                 vcpu->arch.busy_preempt = mftb();
 155         spin_unlock(&vcpu->arch.tbacct_lock);
 156 }
 157
 158 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
 159 {
 160         vcpu->arch.shregs.msr = msr;
 161         kvmppc_end_cede(vcpu);
 162 }
 163
 164 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
 165 {
 166         vcpu->arch.pvr = pvr;
 167 }
 168
 169 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 170 {
 171         int r;
 172
 173         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 174         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 175                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
 176         for (r = 0; r < 16; ++r)
 177                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 178                        r, kvmppc_get_gpr(vcpu, r),
 179                        r+16, kvmppc_get_gpr(vcpu, r+16));
 180         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 181                vcpu->arch.ctr, vcpu->arch.lr);
 182         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 183                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 184         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 185                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 186         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 187                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 188         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 189                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 190         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 191         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 192                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 193         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 194         for (r = 0; r < vcpu->arch.slb_max; ++r)
 195                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 196                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 197         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 198                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 199                vcpu->arch.last_inst);
 200 }
 201
 202 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 203 {
 204         int r;
 205         struct kvm_vcpu *v, *ret = NULL;
 206
 207         mutex_lock(&kvm->lock);
 208         kvm_for_each_vcpu(r, v, kvm) {
 209                 if (v->vcpu_id == id) {
 210                         ret = v;
 211                         break;
 212                 }
 213         }
 214         mutex_unlock(&kvm->lock);
 215         return ret;
 216 }
 217
 218 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 219 {
 220         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 221         vpa->yield_count = 1;
 222 }
 223
 224 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 225                    unsigned long addr, unsigned long len)
 226 {
 227         /* check address is cacheline aligned */
 228         if (addr & (L1_CACHE_BYTES - 1))
 229                 return -EINVAL;
 230         spin_lock(&vcpu->arch.vpa_update_lock);
 231         if (v->next_gpa != addr || v->len != len) {
 232                 v->next_gpa = addr;
 233                 v->len = addr ? len : 0;
 234                 v->update_pending = 1;
 235         }
 236         spin_unlock(&vcpu->arch.vpa_update_lock);
 237         return 0;
 238 }
 239
 240 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 241 struct reg_vpa {
 242         u32 dummy;
 243         union {
 244                 u16 hword;
 245                 u32 word;
 246         } length;
 247 };
 248
 249 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 250 {
 251         if (vpap->update_pending)
 252                 return vpap->next_gpa != 0;
 253         return vpap->pinned_addr != NULL;
 254 }
 255
 256 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 257                                        unsigned long flags,
 258                                        unsigned long vcpuid, unsigned long vpa)
 259 {
 260         struct kvm *kvm = vcpu->kvm;
 261         unsigned long len, nb;
 262         void *va;
 263         struct kvm_vcpu *tvcpu;
 264         int err;
 265         int subfunc;
 266         struct kvmppc_vpa *vpap;
 267
 268         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 269         if (!tvcpu)
 270                 return H_PARAMETER;
 271
 272         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 273         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 274             subfunc == H_VPA_REG_SLB) {
 275                 /* Registering new area - address must be cache-line aligned */
 276                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 277                         return H_PARAMETER;
 278
 279                 /* convert logical addr to kernel addr and read length */
 280                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 281                 if (va == NULL)
 282                         return H_PARAMETER;
 283                 if (subfunc == H_VPA_REG_VPA)
 284                         len = ((struct reg_vpa *)va)->length.hword;
 285                 else
 286                         len = ((struct reg_vpa *)va)->length.word;
 287                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 288
 289                 /* Check length */
 290                 if (len > nb || len < sizeof(struct reg_vpa))
 291                         return H_PARAMETER;
 292         } else {
 293                 vpa = 0;
 294                 len = 0;
 295         }
 296
 297         err = H_PARAMETER;
 298         vpap = NULL;
 299         spin_lock(&tvcpu->arch.vpa_update_lock);
 300
 301         switch (subfunc) {
 302         case H_VPA_REG_VPA:             /* register VPA */
 303                 if (len < sizeof(struct lppaca))
 304                         break;
 305                 vpap = &tvcpu->arch.vpa;
 306                 err = 0;
 307                 break;
 308
 309         case H_VPA_REG_DTL:             /* register DTL */
 310                 if (len < sizeof(struct dtl_entry))
 311                         break;
 312                 len -= len % sizeof(struct dtl_entry);
 313
 314                 /* Check that they have previously registered a VPA */
 315                 err = H_RESOURCE;
 316                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 317                         break;
 318
 319                 vpap = &tvcpu->arch.dtl;
 320                 err = 0;
 321                 break;
 322
 323         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 324                 /* Check that they have previously registered a VPA */
 325                 err = H_RESOURCE;
 326                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 327                         break;
 328
 329                 vpap = &tvcpu->arch.slb_shadow;
 330                 err = 0;
 331                 break;
 332
 333         case H_VPA_DEREG_VPA:           /* deregister VPA */
 334                 /* Check they don't still have a DTL or SLB buf registered */
 335                 err = H_RESOURCE;
 336                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 337                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 338                         break;
 339
 340                 vpap = &tvcpu->arch.vpa;
 341                 err = 0;
 342                 break;
 343
 344         case H_VPA_DEREG_DTL:           /* deregister DTL */
 345                 vpap = &tvcpu->arch.dtl;
 346                 err = 0;
 347                 break;
 348
 349         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 350                 vpap = &tvcpu->arch.slb_shadow;
 351                 err = 0;
 352                 break;
 353         }
 354
 355         if (vpap) {
 356                 vpap->next_gpa = vpa;
 357                 vpap->len = len;
 358                 vpap->update_pending = 1;
 359         }
 360
 361         spin_unlock(&tvcpu->arch.vpa_update_lock);
 362
 363         return err;
 364 }
 365
 366 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 367 {
 368         struct kvm *kvm = vcpu->kvm;
 369         void *va;
 370         unsigned long nb;
 371         unsigned long gpa;
 372
 373         /*
 374          * We need to pin the page pointed to by vpap->next_gpa,
 375          * but we can't call kvmppc_pin_guest_page under the lock
 376          * as it does get_user_pages() and down_read().  So we
 377          * have to drop the lock, pin the page, then get the lock
 378          * again and check that a new area didn't get registered
 379          * in the meantime.
 380          */
 381         for (;;) {
 382                 gpa = vpap->next_gpa;
 383                 spin_unlock(&vcpu->arch.vpa_update_lock);
 384                 va = NULL;
 385                 nb = 0;
 386                 if (gpa)
 387                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 388                 spin_lock(&vcpu->arch.vpa_update_lock);
 389                 if (gpa == vpap->next_gpa)
 390                         break;
 391                 /* sigh... unpin that one and try again */
 392                 if (va)
 393                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 394         }
 395
 396         vpap->update_pending = 0;
 397         if (va && nb < vpap->len) {
 398                 /*
 399                  * If it's now too short, it must be that userspace
 400                  * has changed the mappings underlying guest memory,
 401                  * so unregister the region.
 402                  */
 403                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 404                 va = NULL;
 405         }
 406         if (vpap->pinned_addr)
 407                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 408                                         vpap->dirty);
 409         vpap->gpa = gpa;
 410         vpap->pinned_addr = va;
 411         vpap->dirty = false;
 412         if (va)
 413                 vpap->pinned_end = va + vpap->len;
 414 }
 415
 416 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 417 {
 418         if (!(vcpu->arch.vpa.update_pending ||
 419               vcpu->arch.slb_shadow.update_pending ||
 420               vcpu->arch.dtl.update_pending))
 421                 return;
 422
 423         spin_lock(&vcpu->arch.vpa_update_lock);
 424         if (vcpu->arch.vpa.update_pending) {
 425                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 426                 if (vcpu->arch.vpa.pinned_addr)
 427                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 428         }
 429         if (vcpu->arch.dtl.update_pending) {
 430                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 431                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 432                 vcpu->arch.dtl_index = 0;
 433         }
 434         if (vcpu->arch.slb_shadow.update_pending)
 435                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 436         spin_unlock(&vcpu->arch.vpa_update_lock);
 437 }
 438
 439 /*
 440  * Return the accumulated stolen time for the vcore up until `now'.
 441  * The caller should hold the vcore lock.
 442  */
 443 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 444 {
 445         u64 p;
 446
 447         /*
 448          * If we are the task running the vcore, then since we hold
 449          * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
 450          * can't be updated, so we don't need the tbacct_lock.
 451          * If the vcore is inactive, it can't become active (since we
 452          * hold the vcore lock), so the vcpu load/put functions won't
 453          * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
 454          */
 455         if (vc->vcore_state != VCORE_INACTIVE &&
 456             vc->runner->arch.run_task != current) {
 457                 spin_lock(&vc->runner->arch.tbacct_lock);
 458                 p = vc->stolen_tb;
 459                 if (vc->preempt_tb != TB_NIL)
 460                         p += now - vc->preempt_tb;
 461                 spin_unlock(&vc->runner->arch.tbacct_lock);
 462         } else {
 463                 p = vc->stolen_tb;
 464         }
 465         return p;
 466 }
 467
 468 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 469                                     struct kvmppc_vcore *vc)
 470 {
 471         struct dtl_entry *dt;
 472         struct lppaca *vpa;
 473         unsigned long stolen;
 474         unsigned long core_stolen;
 475         u64 now;
 476
 477         dt = vcpu->arch.dtl_ptr;
 478         vpa = vcpu->arch.vpa.pinned_addr;
 479         now = mftb();
 480         core_stolen = vcore_stolen_time(vc, now);
 481         stolen = core_stolen - vcpu->arch.stolen_logged;
 482         vcpu->arch.stolen_logged = core_stolen;
 483         spin_lock(&vcpu->arch.tbacct_lock);
 484         stolen += vcpu->arch.busy_stolen;
 485         vcpu->arch.busy_stolen = 0;
 486         spin_unlock(&vcpu->arch.tbacct_lock);
 487         if (!dt || !vpa)
 488                 return;
 489         memset(dt, 0, sizeof(struct dtl_entry));
 490         dt->dispatch_reason = 7;
 491         dt->processor_id = vc->pcpu + vcpu->arch.ptid;
 492         dt->timebase = now + vc->tb_offset;
 493         dt->enqueue_to_dispatch_time = stolen;
 494         dt->srr0 = kvmppc_get_pc(vcpu);
 495         dt->srr1 = vcpu->arch.shregs.msr;
 496         ++dt;
 497         if (dt == vcpu->arch.dtl.pinned_end)
 498                 dt = vcpu->arch.dtl.pinned_addr;
 499         vcpu->arch.dtl_ptr = dt;
 500         /* order writing *dt vs. writing vpa->dtl_idx */
 501         smp_wmb();
 502         vpa->dtl_idx = ++vcpu->arch.dtl_index;
 503         vcpu->arch.dtl.dirty = true;
 504 }
 505
 506 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 507 {
 508         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 509         unsigned long target, ret = H_SUCCESS;
 510         struct kvm_vcpu *tvcpu;
 511         int idx, rc;
 512
 513         switch (req) {
 514         case H_ENTER:
 515                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 516                 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
 517                                               kvmppc_get_gpr(vcpu, 5),
 518                                               kvmppc_get_gpr(vcpu, 6),
 519                                               kvmppc_get_gpr(vcpu, 7));
 520                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 521                 break;
 522         case H_CEDE:
 523                 break;
 524         case H_PROD:
 525                 target = kvmppc_get_gpr(vcpu, 4);
 526                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 527                 if (!tvcpu) {
 528                         ret = H_PARAMETER;
 529                         break;
 530                 }
 531                 tvcpu->arch.prodded = 1;
 532                 smp_mb();
 533                 if (vcpu->arch.ceded) {
 534                         if (waitqueue_active(&vcpu->wq)) {
 535                                 wake_up_interruptible(&vcpu->wq);
 536                                 vcpu->stat.halt_wakeup++;
 537                         }
 538                 }
 539                 break;
 540         case H_CONFER:
 541                 target = kvmppc_get_gpr(vcpu, 4);
 542                 if (target == -1)
 543                         break;
 544                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 545                 if (!tvcpu) {
 546                         ret = H_PARAMETER;
 547                         break;
 548                 }
 549                 kvm_vcpu_yield_to(tvcpu);
 550                 break;
 551         case H_REGISTER_VPA:
 552                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 553                                         kvmppc_get_gpr(vcpu, 5),
 554                                         kvmppc_get_gpr(vcpu, 6));
 555                 break;
 556         case H_RTAS:
 557                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 558                         return RESUME_HOST;
 559
 560                 rc = kvmppc_rtas_hcall(vcpu);
 561
 562                 if (rc == -ENOENT)
 563                         return RESUME_HOST;
 564                 else if (rc == 0)
 565                         break;
 566
 567                 /* Send the error out to userspace via KVM_RUN */
 568                 return rc;
 569
 570         case H_XIRR:
 571         case H_CPPR:
 572         case H_EOI:
 573         case H_IPI:
 574         case H_IPOLL:
 575         case H_XIRR_X:
 576                 if (kvmppc_xics_enabled(vcpu)) {
 577                         ret = kvmppc_xics_hcall(vcpu, req);
 578                         break;
 579                 } /* fallthrough */
 580         default:
 581                 return RESUME_HOST;
 582         }
 583         kvmppc_set_gpr(vcpu, 3, ret);
 584         vcpu->arch.hcall_needed = 0;
 585         return RESUME_GUEST;
 586 }
 587
 588 static int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 589                               struct task_struct *tsk)
 590 {
 591         int r = RESUME_HOST;
 592
 593         vcpu->stat.sum_exits++;
 594
 595         run->exit_reason = KVM_EXIT_UNKNOWN;
 596         run->ready_for_interrupt_injection = 1;
 597         switch (vcpu->arch.trap) {
 598         /* We're good on these - the host merely wanted to get our attention */
 599         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 600                 vcpu->stat.dec_exits++;
 601                 r = RESUME_GUEST;
 602                 break;
 603         case BOOK3S_INTERRUPT_EXTERNAL:
 604                 vcpu->stat.ext_intr_exits++;
 605                 r = RESUME_GUEST;
 606                 break;
 607         case BOOK3S_INTERRUPT_PERFMON:
 608                 r = RESUME_GUEST;
 609                 break;
 610         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 611                 /*
 612                  * Deliver a machine check interrupt to the guest.
 613                  * We have to do this, even if the host has handled the
 614                  * machine check, because machine checks use SRR0/1 and
 615                  * the interrupt might have trashed guest state in them.
 616                  */
 617                 kvmppc_book3s_queue_irqprio(vcpu,
 618                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 619                 r = RESUME_GUEST;
 620                 break;
 621         case BOOK3S_INTERRUPT_PROGRAM:
 622         {
 623                 ulong flags;
 624                 /*
 625                  * Normally program interrupts are delivered directly
 626                  * to the guest by the hardware, but we can get here
 627                  * as a result of a hypervisor emulation interrupt
 628                  * (e40) getting turned into a 700 by BML RTAS.
 629                  */
 630                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 631                 kvmppc_core_queue_program(vcpu, flags);
 632                 r = RESUME_GUEST;
 633                 break;
 634         }
 635         case BOOK3S_INTERRUPT_SYSCALL:
 636         {
 637                 /* hcall - punt to userspace */
 638                 int i;
 639
 640                 if (vcpu->arch.shregs.msr & MSR_PR) {
 641                         /* sc 1 from userspace - reflect to guest syscall */
 642                         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_SYSCALL);
 643                         r = RESUME_GUEST;
 644                         break;
 645                 }
 646                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 647                 for (i = 0; i < 9; ++i)
 648                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 649                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 650                 vcpu->arch.hcall_needed = 1;
 651                 r = RESUME_HOST;
 652                 break;
 653         }
 654         /*
 655          * We get these next two if the guest accesses a page which it thinks
 656          * it has mapped but which is not actually present, either because
 657          * it is for an emulated I/O device or because the corresonding
 658          * host page has been paged out.  Any other HDSI/HISI interrupts
 659          * have been handled already.
 660          */
 661         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 662                 r = RESUME_PAGE_FAULT;
 663                 break;
 664         case BOOK3S_INTERRUPT_H_INST_STORAGE:
 665                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
 666                 vcpu->arch.fault_dsisr = 0;
 667                 r = RESUME_PAGE_FAULT;
 668                 break;
 669         /*
 670          * This occurs if the guest executes an illegal instruction.
 671          * We just generate a program interrupt to the guest, since
 672          * we don't emulate any guest instructions at this stage.
 673          */
 674         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 675                 kvmppc_core_queue_program(vcpu, 0x80000);
 676                 r = RESUME_GUEST;
 677                 break;
 678         default:
 679                 kvmppc_dump_regs(vcpu);
 680                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 681                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 682                         vcpu->arch.shregs.msr);
 683                 r = RESUME_HOST;
 684                 BUG();
 685                 break;
 686         }
 687
 688         return r;
 689 }
 690
 691 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 692                                   struct kvm_sregs *sregs)
 693 {
 694         int i;
 695
 696         memset(sregs, 0, sizeof(struct kvm_sregs));
 697         sregs->pvr = vcpu->arch.pvr;
 698         for (i = 0; i < vcpu->arch.slb_max; i++) {
 699                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
 700                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 701         }
 702
 703         return 0;
 704 }
 705
 706 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 707                                   struct kvm_sregs *sregs)
 708 {
 709         int i, j;
 710
 711         kvmppc_set_pvr(vcpu, sregs->pvr);
 712
 713         j = 0;
 714         for (i = 0; i < vcpu->arch.slb_nr; i++) {
 715                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
 716                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
 717                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
 718                         ++j;
 719                 }
 720         }
 721         vcpu->arch.slb_max = j;
 722
 723         return 0;
 724 }
 725
 726 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr)
 727 {
 728         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 729         u64 mask;
 730
 731         spin_lock(&vc->lock);
 732         /*
 733          * Userspace can only modify DPFD (default prefetch depth),
 734          * ILE (interrupt little-endian) and TC (translation control).
 735          */
 736         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
 737         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
 738         spin_unlock(&vc->lock);
 739 }
 740
 741 int kvmppc_get_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
 742 {
 743         int r = 0;
 744         long int i;
 745
 746         switch (id) {
 747         case KVM_REG_PPC_HIOR:
 748                 *val = get_reg_val(id, 0);
 749                 break;
 750         case KVM_REG_PPC_DABR:
 751                 *val = get_reg_val(id, vcpu->arch.dabr);
 752                 break;
 753         case KVM_REG_PPC_DSCR:
 754                 *val = get_reg_val(id, vcpu->arch.dscr);
 755                 break;
 756         case KVM_REG_PPC_PURR:
 757                 *val = get_reg_val(id, vcpu->arch.purr);
 758                 break;
 759         case KVM_REG_PPC_SPURR:
 760                 *val = get_reg_val(id, vcpu->arch.spurr);
 761                 break;
 762         case KVM_REG_PPC_AMR:
 763                 *val = get_reg_val(id, vcpu->arch.amr);
 764                 break;
 765         case KVM_REG_PPC_UAMOR:
 766                 *val = get_reg_val(id, vcpu->arch.uamor);
 767                 break;
 768         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
 769                 i = id - KVM_REG_PPC_MMCR0;
 770                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
 771                 break;
 772         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
 773                 i = id - KVM_REG_PPC_PMC1;
 774                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
 775                 break;
 776         case KVM_REG_PPC_SIAR:
 777                 *val = get_reg_val(id, vcpu->arch.siar);
 778                 break;
 779         case KVM_REG_PPC_SDAR:
 780                 *val = get_reg_val(id, vcpu->arch.sdar);
 781                 break;
 782 #ifdef CONFIG_VSX
 783         case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
 784                 if (cpu_has_feature(CPU_FTR_VSX)) {
 785                         /* VSX => FP reg i is stored in arch.vsr[2*i] */
 786                         long int i = id - KVM_REG_PPC_FPR0;
 787                         *val = get_reg_val(id, vcpu->arch.vsr[2 * i]);
 788                 } else {
 789                         /* let generic code handle it */
 790                         r = -EINVAL;
 791                 }
 792                 break;
 793         case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
 794                 if (cpu_has_feature(CPU_FTR_VSX)) {
 795                         long int i = id - KVM_REG_PPC_VSR0;
 796                         val->vsxval[0] = vcpu->arch.vsr[2 * i];
 797                         val->vsxval[1] = vcpu->arch.vsr[2 * i + 1];
 798                 } else {
 799                         r = -ENXIO;
 800                 }
 801                 break;
 802 #endif /* CONFIG_VSX */
 803         case KVM_REG_PPC_VPA_ADDR:
 804                 spin_lock(&vcpu->arch.vpa_update_lock);
 805                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
 806                 spin_unlock(&vcpu->arch.vpa_update_lock);
 807                 break;
 808         case KVM_REG_PPC_VPA_SLB:
 809                 spin_lock(&vcpu->arch.vpa_update_lock);
 810                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
 811                 val->vpaval.length = vcpu->arch.slb_shadow.len;
 812                 spin_unlock(&vcpu->arch.vpa_update_lock);
 813                 break;
 814         case KVM_REG_PPC_VPA_DTL:
 815                 spin_lock(&vcpu->arch.vpa_update_lock);
 816                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
 817                 val->vpaval.length = vcpu->arch.dtl.len;
 818                 spin_unlock(&vcpu->arch.vpa_update_lock);
 819                 break;
 820         case KVM_REG_PPC_TB_OFFSET:
 821                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
 822                 break;
 823         case KVM_REG_PPC_LPCR:
 824                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
 825                 break;
 826         case KVM_REG_PPC_PPR:
 827                 *val = get_reg_val(id, vcpu->arch.ppr);
 828                 break;
 829         default:
 830                 r = -EINVAL;
 831                 break;
 832         }
 833
 834         return r;
 835 }
 836
 837 int kvmppc_set_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
 838 {
 839         int r = 0;
 840         long int i;
 841         unsigned long addr, len;
 842
 843         switch (id) {
 844         case KVM_REG_PPC_HIOR:
 845                 /* Only allow this to be set to zero */
 846                 if (set_reg_val(id, *val))
 847                         r = -EINVAL;
 848                 break;
 849         case KVM_REG_PPC_DABR:
 850                 vcpu->arch.dabr = set_reg_val(id, *val);
 851                 break;
 852         case KVM_REG_PPC_DSCR:
 853                 vcpu->arch.dscr = set_reg_val(id, *val);
 854                 break;
 855         case KVM_REG_PPC_PURR:
 856                 vcpu->arch.purr = set_reg_val(id, *val);
 857                 break;
 858         case KVM_REG_PPC_SPURR:
 859                 vcpu->arch.spurr = set_reg_val(id, *val);
 860                 break;
 861         case KVM_REG_PPC_AMR:
 862                 vcpu->arch.amr = set_reg_val(id, *val);
 863                 break;
 864         case KVM_REG_PPC_UAMOR:
 865                 vcpu->arch.uamor = set_reg_val(id, *val);
 866                 break;
 867         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
 868                 i = id - KVM_REG_PPC_MMCR0;
 869                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
 870                 break;
 871         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
 872                 i = id - KVM_REG_PPC_PMC1;
 873                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
 874                 break;
 875         case KVM_REG_PPC_SIAR:
 876                 vcpu->arch.siar = set_reg_val(id, *val);
 877                 break;
 878         case KVM_REG_PPC_SDAR:
 879                 vcpu->arch.sdar = set_reg_val(id, *val);
 880                 break;
 881 #ifdef CONFIG_VSX
 882         case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
 883                 if (cpu_has_feature(CPU_FTR_VSX)) {
 884                         /* VSX => FP reg i is stored in arch.vsr[2*i] */
 885                         long int i = id - KVM_REG_PPC_FPR0;
 886                         vcpu->arch.vsr[2 * i] = set_reg_val(id, *val);
 887                 } else {
 888                         /* let generic code handle it */
 889                         r = -EINVAL;
 890                 }
 891                 break;
 892         case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
 893                 if (cpu_has_feature(CPU_FTR_VSX)) {
 894                         long int i = id - KVM_REG_PPC_VSR0;
 895                         vcpu->arch.vsr[2 * i] = val->vsxval[0];
 896                         vcpu->arch.vsr[2 * i + 1] = val->vsxval[1];
 897                 } else {
 898                         r = -ENXIO;
 899                 }
 900                 break;
 901 #endif /* CONFIG_VSX */
 902         case KVM_REG_PPC_VPA_ADDR:
 903                 addr = set_reg_val(id, *val);
 904                 r = -EINVAL;
 905                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
 906                               vcpu->arch.dtl.next_gpa))
 907                         break;
 908                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
 909                 break;
 910         case KVM_REG_PPC_VPA_SLB:
 911                 addr = val->vpaval.addr;
 912                 len = val->vpaval.length;
 913                 r = -EINVAL;
 914                 if (addr && !vcpu->arch.vpa.next_gpa)
 915                         break;
 916                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
 917                 break;
 918         case KVM_REG_PPC_VPA_DTL:
 919                 addr = val->vpaval.addr;
 920                 len = val->vpaval.length;
 921                 r = -EINVAL;
 922                 if (addr && (len < sizeof(struct dtl_entry) ||
 923                              !vcpu->arch.vpa.next_gpa))
 924                         break;
 925                 len -= len % sizeof(struct dtl_entry);
 926                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
 927                 break;
 928         case KVM_REG_PPC_TB_OFFSET:
 929                 /* round up to multiple of 2^24 */
 930                 vcpu->arch.vcore->tb_offset =
 931                         ALIGN(set_reg_val(id, *val), 1UL << 24);
 932                 break;
 933         case KVM_REG_PPC_LPCR:
 934                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val));
 935                 break;
 936         case KVM_REG_PPC_PPR:
 937                 vcpu->arch.ppr = set_reg_val(id, *val);
 938                 break;
 939         default:
 940                 r = -EINVAL;
 941                 break;
 942         }
 943
 944         return r;
 945 }
 946
 947 int kvmppc_core_check_processor_compat(void)
 948 {
 949         if (cpu_has_feature(CPU_FTR_HVMODE))
 950                 return 0;
 951         return -EIO;
 952 }
 953
 954 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
 955 {
 956         struct kvm_vcpu *vcpu;
 957         int err = -EINVAL;
 958         int core;
 959         struct kvmppc_vcore *vcore;
 960
 961         core = id / threads_per_core;
 962         if (core >= KVM_MAX_VCORES)
 963                 goto out;
 964
 965         err = -ENOMEM;
 966         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 967         if (!vcpu)
 968                 goto out;
 969
 970         err = kvm_vcpu_init(vcpu, kvm, id);
 971         if (err)
 972                 goto free_vcpu;
 973
 974         vcpu->arch.shared = &vcpu->arch.shregs;
 975         vcpu->arch.mmcr[0] = MMCR0_FC;
 976         vcpu->arch.ctrl = CTRL_RUNLATCH;
 977         /* default to host PVR, since we can't spoof it */
 978         vcpu->arch.pvr = mfspr(SPRN_PVR);
 979         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
 980         spin_lock_init(&vcpu->arch.vpa_update_lock);
 981         spin_lock_init(&vcpu->arch.tbacct_lock);
 982         vcpu->arch.busy_preempt = TB_NIL;
 983
 984         kvmppc_mmu_book3s_hv_init(vcpu);
 985
 986         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
 987
 988         init_waitqueue_head(&vcpu->arch.cpu_run);
 989
 990         mutex_lock(&kvm->lock);
 991         vcore = kvm->arch.vcores[core];
 992         if (!vcore) {
 993                 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
 994                 if (vcore) {
 995                         INIT_LIST_HEAD(&vcore->runnable_threads);
 996                         spin_lock_init(&vcore->lock);
 997                         init_waitqueue_head(&vcore->wq);
 998                         vcore->preempt_tb = TB_NIL;
 999                         vcore->lpcr = kvm->arch.lpcr;
1000                 }
1001                 kvm->arch.vcores[core] = vcore;
1002                 kvm->arch.online_vcores++;
1003         }
1004         mutex_unlock(&kvm->lock);
1005
1006         if (!vcore)
1007                 goto free_vcpu;
1008
1009         spin_lock(&vcore->lock);
1010         ++vcore->num_threads;
1011         spin_unlock(&vcore->lock);
1012         vcpu->arch.vcore = vcore;
1013
1014         vcpu->arch.cpu_type = KVM_CPU_3S_64;
1015         kvmppc_sanity_check(vcpu);
1016
1017         return vcpu;
1018
1019 free_vcpu:
1020         kmem_cache_free(kvm_vcpu_cache, vcpu);
1021 out:
1022         return ERR_PTR(err);
1023 }
1024
1025 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1026 {
1027         if (vpa->pinned_addr)
1028                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1029                                         vpa->dirty);
1030 }
1031
1032 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
1033 {
1034         spin_lock(&vcpu->arch.vpa_update_lock);
1035         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1036         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1037         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1038         spin_unlock(&vcpu->arch.vpa_update_lock);
1039         kvm_vcpu_uninit(vcpu);
1040         kmem_cache_free(kvm_vcpu_cache, vcpu);
1041 }
1042
1043 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1044 {
1045         unsigned long dec_nsec, now;
1046
1047         now = get_tb();
1048         if (now > vcpu->arch.dec_expires) {
1049                 /* decrementer has already gone negative */
1050                 kvmppc_core_queue_dec(vcpu);
1051                 kvmppc_core_prepare_to_enter(vcpu);
1052                 return;
1053         }
1054         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1055                    / tb_ticks_per_sec;
1056         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1057                       HRTIMER_MODE_REL);
1058         vcpu->arch.timer_running = 1;
1059 }
1060
1061 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1062 {
1063         vcpu->arch.ceded = 0;
1064         if (vcpu->arch.timer_running) {
1065                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1066                 vcpu->arch.timer_running = 0;
1067         }
1068 }
1069
1070 extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
1071
1072 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1073                                    struct kvm_vcpu *vcpu)
1074 {
1075         u64 now;
1076
1077         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1078                 return;
1079         spin_lock(&vcpu->arch.tbacct_lock);
1080         now = mftb();
1081         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1082                 vcpu->arch.stolen_logged;
1083         vcpu->arch.busy_preempt = now;
1084         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1085         spin_unlock(&vcpu->arch.tbacct_lock);
1086         --vc->n_runnable;
1087         list_del(&vcpu->arch.run_list);
1088 }
1089
1090 static int kvmppc_grab_hwthread(int cpu)
1091 {
1092         struct paca_struct *tpaca;
1093         long timeout = 1000;
1094
1095         tpaca = &paca[cpu];
1096
1097         /* Ensure the thread won't go into the kernel if it wakes */
1098         tpaca->kvm_hstate.hwthread_req = 1;
1099         tpaca->kvm_hstate.kvm_vcpu = NULL;
1100
1101         /*
1102          * If the thread is already executing in the kernel (e.g. handling
1103          * a stray interrupt), wait for it to get back to nap mode.
1104          * The smp_mb() is to ensure that our setting of hwthread_req
1105          * is visible before we look at hwthread_state, so if this
1106          * races with the code at system_reset_pSeries and the thread
1107          * misses our setting of hwthread_req, we are sure to see its
1108          * setting of hwthread_state, and vice versa.
1109          */
1110         smp_mb();
1111         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1112                 if (--timeout <= 0) {
1113                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1114                         return -EBUSY;
1115                 }
1116                 udelay(1);
1117         }
1118         return 0;
1119 }
1120
1121 static void kvmppc_release_hwthread(int cpu)
1122 {
1123         struct paca_struct *tpaca;
1124
1125         tpaca = &paca[cpu];
1126         tpaca->kvm_hstate.hwthread_req = 0;
1127         tpaca->kvm_hstate.kvm_vcpu = NULL;
1128 }
1129
1130 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1131 {
1132         int cpu;
1133         struct paca_struct *tpaca;
1134         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1135
1136         if (vcpu->arch.timer_running) {
1137                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1138                 vcpu->arch.timer_running = 0;
1139         }
1140         cpu = vc->pcpu + vcpu->arch.ptid;
1141         tpaca = &paca[cpu];
1142         tpaca->kvm_hstate.kvm_vcpu = vcpu;
1143         tpaca->kvm_hstate.kvm_vcore = vc;
1144         tpaca->kvm_hstate.napping = 0;
1145         vcpu->cpu = vc->pcpu;
1146         smp_wmb();
1147 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1148         if (vcpu->arch.ptid) {
1149                 xics_wake_cpu(cpu);
1150                 ++vc->n_woken;
1151         }
1152 #endif
1153 }
1154
1155 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1156 {
1157         int i;
1158
1159         HMT_low();
1160         i = 0;
1161         while (vc->nap_count < vc->n_woken) {
1162                 if (++i >= 1000000) {
1163                         pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1164                                vc->nap_count, vc->n_woken);
1165                         break;
1166                 }
1167                 cpu_relax();
1168         }
1169         HMT_medium();
1170 }
1171
1172 /*
1173  * Check that we are on thread 0 and that any other threads in
1174  * this core are off-line.  Then grab the threads so they can't
1175  * enter the kernel.
1176  */
1177 static int on_primary_thread(void)
1178 {
1179         int cpu = smp_processor_id();
1180         int thr = cpu_thread_in_core(cpu);
1181
1182         if (thr)
1183                 return 0;
1184         while (++thr < threads_per_core)
1185                 if (cpu_online(cpu + thr))
1186                         return 0;
1187
1188         /* Grab all hw threads so they can't go into the kernel */
1189         for (thr = 1; thr < threads_per_core; ++thr) {
1190                 if (kvmppc_grab_hwthread(cpu + thr)) {
1191                         /* Couldn't grab one; let the others go */
1192                         do {
1193                                 kvmppc_release_hwthread(cpu + thr);
1194                         } while (--thr > 0);
1195                         return 0;
1196                 }
1197         }
1198         return 1;
1199 }
1200
1201 /*
1202  * Run a set of guest threads on a physical core.
1203  * Called with vc->lock held.
1204  */
1205 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1206 {
1207         struct kvm_vcpu *vcpu, *vcpu0, *vnext;
1208         long ret;
1209         u64 now;
1210         int ptid, i, need_vpa_update;
1211         int srcu_idx;
1212         struct kvm_vcpu *vcpus_to_update[threads_per_core];
1213
1214         /* don't start if any threads have a signal pending */
1215         need_vpa_update = 0;
1216         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1217                 if (signal_pending(vcpu->arch.run_task))
1218                         return;
1219                 if (vcpu->arch.vpa.update_pending ||
1220                     vcpu->arch.slb_shadow.update_pending ||
1221                     vcpu->arch.dtl.update_pending)
1222                         vcpus_to_update[need_vpa_update++] = vcpu;
1223         }
1224
1225         /*
1226          * Initialize *vc, in particular vc->vcore_state, so we can
1227          * drop the vcore lock if necessary.
1228          */
1229         vc->n_woken = 0;
1230         vc->nap_count = 0;
1231         vc->entry_exit_count = 0;
1232         vc->vcore_state = VCORE_STARTING;
1233         vc->in_guest = 0;
1234         vc->napping_threads = 0;
1235
1236         /*
1237          * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1238          * which can't be called with any spinlocks held.
1239          */
1240         if (need_vpa_update) {
1241                 spin_unlock(&vc->lock);
1242                 for (i = 0; i < need_vpa_update; ++i)
1243                         kvmppc_update_vpas(vcpus_to_update[i]);
1244                 spin_lock(&vc->lock);
1245         }
1246
1247         /*
1248          * Assign physical thread IDs, first to non-ceded vcpus
1249          * and then to ceded ones.
1250          */
1251         ptid = 0;
1252         vcpu0 = NULL;
1253         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1254                 if (!vcpu->arch.ceded) {
1255                         if (!ptid)
1256                                 vcpu0 = vcpu;
1257                         vcpu->arch.ptid = ptid++;
1258                 }
1259         }
1260         if (!vcpu0)
1261                 goto out;       /* nothing to run; should never happen */
1262         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1263                 if (vcpu->arch.ceded)
1264                         vcpu->arch.ptid = ptid++;
1265
1266         /*
1267          * Make sure we are running on thread 0, and that
1268          * secondary threads are offline.
1269          */
1270         if (threads_per_core > 1 && !on_primary_thread()) {
1271                 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1272                         vcpu->arch.ret = -EBUSY;
1273                 goto out;
1274         }
1275
1276         vc->pcpu = smp_processor_id();
1277         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1278                 kvmppc_start_thread(vcpu);
1279                 kvmppc_create_dtl_entry(vcpu, vc);
1280         }
1281
1282         vc->vcore_state = VCORE_RUNNING;
1283         preempt_disable();
1284         spin_unlock(&vc->lock);
1285
1286         kvm_guest_enter();
1287
1288         srcu_idx = srcu_read_lock(&vcpu0->kvm->srcu);
1289
1290         __kvmppc_vcore_entry(NULL, vcpu0);
1291
1292         spin_lock(&vc->lock);
1293         /* disable sending of IPIs on virtual external irqs */
1294         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1295                 vcpu->cpu = -1;
1296         /* wait for secondary threads to finish writing their state to memory */
1297         if (vc->nap_count < vc->n_woken)
1298                 kvmppc_wait_for_nap(vc);
1299         for (i = 0; i < threads_per_core; ++i)
1300                 kvmppc_release_hwthread(vc->pcpu + i);
1301         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
1302         vc->vcore_state = VCORE_EXITING;
1303         spin_unlock(&vc->lock);
1304
1305         srcu_read_unlock(&vcpu0->kvm->srcu, srcu_idx);
1306
1307         /* make sure updates to secondary vcpu structs are visible now */
1308         smp_mb();
1309         kvm_guest_exit();
1310
1311         preempt_enable();
1312         kvm_resched(vcpu);
1313
1314         spin_lock(&vc->lock);
1315         now = get_tb();
1316         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1317                 /* cancel pending dec exception if dec is positive */
1318                 if (now < vcpu->arch.dec_expires &&
1319                     kvmppc_core_pending_dec(vcpu))
1320                         kvmppc_core_dequeue_dec(vcpu);
1321
1322                 ret = RESUME_GUEST;
1323                 if (vcpu->arch.trap)
1324                         ret = kvmppc_handle_exit(vcpu->arch.kvm_run, vcpu,
1325                                                  vcpu->arch.run_task);
1326
1327                 vcpu->arch.ret = ret;
1328                 vcpu->arch.trap = 0;
1329
1330                 if (vcpu->arch.ceded) {
1331                         if (ret != RESUME_GUEST)
1332                                 kvmppc_end_cede(vcpu);
1333                         else
1334                                 kvmppc_set_timer(vcpu);
1335                 }
1336         }
1337
1338  out:
1339         vc->vcore_state = VCORE_INACTIVE;
1340         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1341                                  arch.run_list) {
1342                 if (vcpu->arch.ret != RESUME_GUEST) {
1343                         kvmppc_remove_runnable(vc, vcpu);
1344                         wake_up(&vcpu->arch.cpu_run);
1345                 }
1346         }
1347 }
1348
1349 /*
1350  * Wait for some other vcpu thread to execute us, and
1351  * wake us up when we need to handle something in the host.
1352  */
1353 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1354 {
1355         DEFINE_WAIT(wait);
1356
1357         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1358         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1359                 schedule();
1360         finish_wait(&vcpu->arch.cpu_run, &wait);
1361 }
1362
1363 /*
1364  * All the vcpus in this vcore are idle, so wait for a decrementer
1365  * or external interrupt to one of the vcpus.  vc->lock is held.
1366  */
1367 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1368 {
1369         DEFINE_WAIT(wait);
1370
1371         prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1372         vc->vcore_state = VCORE_SLEEPING;
1373         spin_unlock(&vc->lock);
1374         schedule();
1375         finish_wait(&vc->wq, &wait);
1376         spin_lock(&vc->lock);
1377         vc->vcore_state = VCORE_INACTIVE;
1378 }
1379
1380 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1381 {
1382         int n_ceded;
1383         struct kvmppc_vcore *vc;
1384         struct kvm_vcpu *v, *vn;
1385
1386         kvm_run->exit_reason = 0;
1387         vcpu->arch.ret = RESUME_GUEST;
1388         vcpu->arch.trap = 0;
1389         kvmppc_update_vpas(vcpu);
1390
1391         /*
1392          * Synchronize with other threads in this virtual core
1393          */
1394         vc = vcpu->arch.vcore;
1395         spin_lock(&vc->lock);
1396         vcpu->arch.ceded = 0;
1397         vcpu->arch.run_task = current;
1398         vcpu->arch.kvm_run = kvm_run;
1399         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
1400         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1401         vcpu->arch.busy_preempt = TB_NIL;
1402         list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1403         ++vc->n_runnable;
1404
1405         /*
1406          * This happens the first time this is called for a vcpu.
1407          * If the vcore is already running, we may be able to start
1408          * this thread straight away and have it join in.
1409          */
1410         if (!signal_pending(current)) {
1411                 if (vc->vcore_state == VCORE_RUNNING &&
1412                     VCORE_EXIT_COUNT(vc) == 0) {
1413                         vcpu->arch.ptid = vc->n_runnable - 1;
1414                         kvmppc_create_dtl_entry(vcpu, vc);
1415                         kvmppc_start_thread(vcpu);
1416                 } else if (vc->vcore_state == VCORE_SLEEPING) {
1417                         wake_up(&vc->wq);
1418                 }
1419
1420         }
1421
1422         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1423                !signal_pending(current)) {
1424                 if (vc->vcore_state != VCORE_INACTIVE) {
1425                         spin_unlock(&vc->lock);
1426                         kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1427                         spin_lock(&vc->lock);
1428                         continue;
1429                 }
1430                 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1431                                          arch.run_list) {
1432                         kvmppc_core_prepare_to_enter(v);
1433                         if (signal_pending(v->arch.run_task)) {
1434                                 kvmppc_remove_runnable(vc, v);
1435                                 v->stat.signal_exits++;
1436                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1437                                 v->arch.ret = -EINTR;
1438                                 wake_up(&v->arch.cpu_run);
1439                         }
1440                 }
1441                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1442                         break;
1443                 vc->runner = vcpu;
1444                 n_ceded = 0;
1445                 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
1446                         if (!v->arch.pending_exceptions)
1447                                 n_ceded += v->arch.ceded;
1448                         else
1449                                 v->arch.ceded = 0;
1450                 }
1451                 if (n_ceded == vc->n_runnable)
1452                         kvmppc_vcore_blocked(vc);
1453                 else
1454                         kvmppc_run_core(vc);
1455                 vc->runner = NULL;
1456         }
1457
1458         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1459                (vc->vcore_state == VCORE_RUNNING ||
1460                 vc->vcore_state == VCORE_EXITING)) {
1461                 spin_unlock(&vc->lock);
1462                 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1463                 spin_lock(&vc->lock);
1464         }
1465
1466         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1467                 kvmppc_remove_runnable(vc, vcpu);
1468                 vcpu->stat.signal_exits++;
1469                 kvm_run->exit_reason = KVM_EXIT_INTR;
1470                 vcpu->arch.ret = -EINTR;
1471         }
1472
1473         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1474                 /* Wake up some vcpu to run the core */
1475                 v = list_first_entry(&vc->runnable_threads,
1476                                      struct kvm_vcpu, arch.run_list);
1477                 wake_up(&v->arch.cpu_run);
1478         }
1479
1480         spin_unlock(&vc->lock);
1481         return vcpu->arch.ret;
1482 }
1483
1484 int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
1485 {
1486         int r;
1487         int srcu_idx;
1488
1489         if (!vcpu->arch.sane) {
1490                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1491                 return -EINVAL;
1492         }
1493
1494         kvmppc_core_prepare_to_enter(vcpu);
1495
1496         /* No need to go into the guest when all we'll do is come back out */
1497         if (signal_pending(current)) {
1498                 run->exit_reason = KVM_EXIT_INTR;
1499                 return -EINTR;
1500         }
1501
1502         atomic_inc(&vcpu->kvm->arch.vcpus_running);
1503         /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1504         smp_mb();
1505
1506         /* On the first time here, set up HTAB and VRMA or RMA */
1507         if (!vcpu->kvm->arch.rma_setup_done) {
1508                 r = kvmppc_hv_setup_htab_rma(vcpu);
1509                 if (r)
1510                         goto out;
1511         }
1512
1513         flush_fp_to_thread(current);
1514         flush_altivec_to_thread(current);
1515         flush_vsx_to_thread(current);
1516         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1517         vcpu->arch.pgdir = current->mm->pgd;
1518         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1519
1520         do {
1521                 r = kvmppc_run_vcpu(run, vcpu);
1522
1523                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1524                     !(vcpu->arch.shregs.msr & MSR_PR)) {
1525                         r = kvmppc_pseries_do_hcall(vcpu);
1526                         kvmppc_core_prepare_to_enter(vcpu);
1527                 } else if (r == RESUME_PAGE_FAULT) {
1528                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1529                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
1530                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1531                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1532                 }
1533         } while (r == RESUME_GUEST);
1534
1535  out:
1536         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1537         atomic_dec(&vcpu->kvm->arch.vcpus_running);
1538         return r;
1539 }
1540
1541
1542 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1543    Assumes POWER7 or PPC970. */
1544 static inline int lpcr_rmls(unsigned long rma_size)
1545 {
1546         switch (rma_size) {
1547         case 32ul << 20:        /* 32 MB */
1548                 if (cpu_has_feature(CPU_FTR_ARCH_206))
1549                         return 8;       /* only supported on POWER7 */
1550                 return -1;
1551         case 64ul << 20:        /* 64 MB */
1552                 return 3;
1553         case 128ul << 20:       /* 128 MB */
1554                 return 7;
1555         case 256ul << 20:       /* 256 MB */
1556                 return 4;
1557         case 1ul << 30:         /* 1 GB */
1558                 return 2;
1559         case 16ul << 30:        /* 16 GB */
1560                 return 1;
1561         case 256ul << 30:       /* 256 GB */
1562                 return 0;
1563         default:
1564                 return -1;
1565         }
1566 }
1567
1568 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1569 {
1570         struct page *page;
1571         struct kvm_rma_info *ri = vma->vm_file->private_data;
1572
1573         if (vmf->pgoff >= kvm_rma_pages)
1574                 return VM_FAULT_SIGBUS;
1575
1576         page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1577         get_page(page);
1578         vmf->page = page;
1579         return 0;
1580 }
1581
1582 static const struct vm_operations_struct kvm_rma_vm_ops = {
1583         .fault = kvm_rma_fault,
1584 };
1585
1586 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1587 {
1588         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1589         vma->vm_ops = &kvm_rma_vm_ops;
1590         return 0;
1591 }
1592
1593 static int kvm_rma_release(struct inode *inode, struct file *filp)
1594 {
1595         struct kvm_rma_info *ri = filp->private_data;
1596
1597         kvm_release_rma(ri);
1598         return 0;
1599 }
1600
1601 static const struct file_operations kvm_rma_fops = {
1602         .mmap           = kvm_rma_mmap,
1603         .release        = kvm_rma_release,
1604 };
1605
1606 long kvm_vm_ioctl_allocate_rma(struct kvm *kvm, struct kvm_allocate_rma *ret)
1607 {
1608         long fd;
1609         struct kvm_rma_info *ri;
1610         /*
1611          * Only do this on PPC970 in HV mode
1612          */
1613         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
1614             !cpu_has_feature(CPU_FTR_ARCH_201))
1615                 return -EINVAL;
1616
1617         if (!kvm_rma_pages)
1618                 return -EINVAL;
1619
1620         ri = kvm_alloc_rma();
1621         if (!ri)
1622                 return -ENOMEM;
1623
1624         fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC);
1625         if (fd < 0)
1626                 kvm_release_rma(ri);
1627
1628         ret->rma_size = kvm_rma_pages << PAGE_SHIFT;
1629         return fd;
1630 }
1631
1632 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1633                                      int linux_psize)
1634 {
1635         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1636
1637         if (!def->shift)
1638                 return;
1639         (*sps)->page_shift = def->shift;
1640         (*sps)->slb_enc = def->sllp;
1641         (*sps)->enc[0].page_shift = def->shift;
1642         /*
1643          * Only return base page encoding. We don't want to return
1644          * all the supporting pte_enc, because our H_ENTER doesn't
1645          * support MPSS yet. Once they do, we can start passing all
1646          * support pte_enc here
1647          */
1648         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
1649         (*sps)++;
1650 }
1651
1652 int kvm_vm_ioctl_get_smmu_info(struct kvm *kvm, struct kvm_ppc_smmu_info *info)
1653 {
1654         struct kvm_ppc_one_seg_page_size *sps;
1655
1656         info->flags = KVM_PPC_PAGE_SIZES_REAL;
1657         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1658                 info->flags |= KVM_PPC_1T_SEGMENTS;
1659         info->slb_size = mmu_slb_size;
1660
1661         /* We only support these sizes for now, and no muti-size segments */
1662         sps = &info->sps[0];
1663         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1664         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1665         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1666
1667         return 0;
1668 }
1669
1670 /*
1671  * Get (and clear) the dirty memory log for a memory slot.
1672  */
1673 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1674 {
1675         struct kvm_memory_slot *memslot;
1676         int r;
1677         unsigned long n;
1678
1679         mutex_lock(&kvm->slots_lock);
1680
1681         r = -EINVAL;
1682         if (log->slot >= KVM_USER_MEM_SLOTS)
1683                 goto out;
1684
1685         memslot = id_to_memslot(kvm->memslots, log->slot);
1686         r = -ENOENT;
1687         if (!memslot->dirty_bitmap)
1688                 goto out;
1689
1690         n = kvm_dirty_bitmap_bytes(memslot);
1691         memset(memslot->dirty_bitmap, 0, n);
1692
1693         r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
1694         if (r)
1695                 goto out;
1696
1697         r = -EFAULT;
1698         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1699                 goto out;
1700
1701         r = 0;
1702 out:
1703         mutex_unlock(&kvm->slots_lock);
1704         return r;
1705 }
1706
1707 static void unpin_slot(struct kvm_memory_slot *memslot)
1708 {
1709         unsigned long *physp;
1710         unsigned long j, npages, pfn;
1711         struct page *page;
1712
1713         physp = memslot->arch.slot_phys;
1714         npages = memslot->npages;
1715         if (!physp)
1716                 return;
1717         for (j = 0; j < npages; j++) {
1718                 if (!(physp[j] & KVMPPC_GOT_PAGE))
1719                         continue;
1720                 pfn = physp[j] >> PAGE_SHIFT;
1721                 page = pfn_to_page(pfn);
1722                 SetPageDirty(page);
1723                 put_page(page);
1724         }
1725 }
1726
1727 void kvmppc_core_free_memslot(struct kvm_memory_slot *free,
1728                               struct kvm_memory_slot *dont)
1729 {
1730         if (!dont || free->arch.rmap != dont->arch.rmap) {
1731                 vfree(free->arch.rmap);
1732                 free->arch.rmap = NULL;
1733         }
1734         if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
1735                 unpin_slot(free);
1736                 vfree(free->arch.slot_phys);
1737                 free->arch.slot_phys = NULL;
1738         }
1739 }
1740
1741 int kvmppc_core_create_memslot(struct kvm_memory_slot *slot,
1742                                unsigned long npages)
1743 {
1744         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
1745         if (!slot->arch.rmap)
1746                 return -ENOMEM;
1747         slot->arch.slot_phys = NULL;
1748
1749         return 0;
1750 }
1751
1752 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1753                                       struct kvm_memory_slot *memslot,
1754                                       struct kvm_userspace_memory_region *mem)
1755 {
1756         unsigned long *phys;
1757
1758         /* Allocate a slot_phys array if needed */
1759         phys = memslot->arch.slot_phys;
1760         if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
1761                 phys = vzalloc(memslot->npages * sizeof(unsigned long));
1762                 if (!phys)
1763                         return -ENOMEM;
1764                 memslot->arch.slot_phys = phys;
1765         }
1766
1767         return 0;
1768 }
1769
1770 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1771                                       struct kvm_userspace_memory_region *mem,
1772                                       const struct kvm_memory_slot *old)
1773 {
1774         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
1775         struct kvm_memory_slot *memslot;
1776
1777         if (npages && old->npages) {
1778                 /*
1779                  * If modifying a memslot, reset all the rmap dirty bits.
1780                  * If this is a new memslot, we don't need to do anything
1781                  * since the rmap array starts out as all zeroes,
1782                  * i.e. no pages are dirty.
1783                  */
1784                 memslot = id_to_memslot(kvm->memslots, mem->slot);
1785                 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
1786         }
1787 }
1788
1789 /*
1790  * Update LPCR values in kvm->arch and in vcores.
1791  * Caller must hold kvm->lock.
1792  */
1793 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
1794 {
1795         long int i;
1796         u32 cores_done = 0;
1797
1798         if ((kvm->arch.lpcr & mask) == lpcr)
1799                 return;
1800
1801         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
1802
1803         for (i = 0; i < KVM_MAX_VCORES; ++i) {
1804                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
1805                 if (!vc)
1806                         continue;
1807                 spin_lock(&vc->lock);
1808                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
1809                 spin_unlock(&vc->lock);
1810                 if (++cores_done >= kvm->arch.online_vcores)
1811                         break;
1812         }
1813 }
1814
1815 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
1816 {
1817         int err = 0;
1818         struct kvm *kvm = vcpu->kvm;
1819         struct kvm_rma_info *ri = NULL;
1820         unsigned long hva;
1821         struct kvm_memory_slot *memslot;
1822         struct vm_area_struct *vma;
1823         unsigned long lpcr = 0, senc;
1824         unsigned long lpcr_mask = 0;
1825         unsigned long psize, porder;
1826         unsigned long rma_size;
1827         unsigned long rmls;
1828         unsigned long *physp;
1829         unsigned long i, npages;
1830         int srcu_idx;
1831
1832         mutex_lock(&kvm->lock);
1833         if (kvm->arch.rma_setup_done)
1834                 goto out;       /* another vcpu beat us to it */
1835
1836         /* Allocate hashed page table (if not done already) and reset it */
1837         if (!kvm->arch.hpt_virt) {
1838                 err = kvmppc_alloc_hpt(kvm, NULL);
1839                 if (err) {
1840                         pr_err("KVM: Couldn't alloc HPT\n");
1841                         goto out;
1842                 }
1843         }
1844
1845         /* Look up the memslot for guest physical address 0 */
1846         srcu_idx = srcu_read_lock(&kvm->srcu);
1847         memslot = gfn_to_memslot(kvm, 0);
1848
1849         /* We must have some memory at 0 by now */
1850         err = -EINVAL;
1851         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1852                 goto out_srcu;
1853
1854         /* Look up the VMA for the start of this memory slot */
1855         hva = memslot->userspace_addr;
1856         down_read(&current->mm->mmap_sem);
1857         vma = find_vma(current->mm, hva);
1858         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
1859                 goto up_out;
1860
1861         psize = vma_kernel_pagesize(vma);
1862         porder = __ilog2(psize);
1863
1864         /* Is this one of our preallocated RMAs? */
1865         if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
1866             hva == vma->vm_start)
1867                 ri = vma->vm_file->private_data;
1868
1869         up_read(&current->mm->mmap_sem);
1870
1871         if (!ri) {
1872                 /* On POWER7, use VRMA; on PPC970, give up */
1873                 err = -EPERM;
1874                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1875                         pr_err("KVM: CPU requires an RMO\n");
1876                         goto out_srcu;
1877                 }
1878
1879                 /* We can handle 4k, 64k or 16M pages in the VRMA */
1880                 err = -EINVAL;
1881                 if (!(psize == 0x1000 || psize == 0x10000 ||
1882                       psize == 0x1000000))
1883                         goto out_srcu;
1884
1885                 /* Update VRMASD field in the LPCR */
1886                 senc = slb_pgsize_encoding(psize);
1887                 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1888                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
1889                 lpcr_mask = LPCR_VRMASD;
1890                 /* the -4 is to account for senc values starting at 0x10 */
1891                 lpcr = senc << (LPCR_VRMASD_SH - 4);
1892
1893                 /* Create HPTEs in the hash page table for the VRMA */
1894                 kvmppc_map_vrma(vcpu, memslot, porder);
1895
1896         } else {
1897                 /* Set up to use an RMO region */
1898                 rma_size = kvm_rma_pages;
1899                 if (rma_size > memslot->npages)
1900                         rma_size = memslot->npages;
1901                 rma_size <<= PAGE_SHIFT;
1902                 rmls = lpcr_rmls(rma_size);
1903                 err = -EINVAL;
1904                 if ((long)rmls < 0) {
1905                         pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
1906                         goto out_srcu;
1907                 }
1908                 atomic_inc(&ri->use_count);
1909                 kvm->arch.rma = ri;
1910
1911                 /* Update LPCR and RMOR */
1912                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1913                         /* PPC970; insert RMLS value (split field) in HID4 */
1914                         lpcr_mask = (1ul << HID4_RMLS0_SH) |
1915                                 (3ul << HID4_RMLS2_SH) | HID4_RMOR;
1916                         lpcr = ((rmls >> 2) << HID4_RMLS0_SH) |
1917                                 ((rmls & 3) << HID4_RMLS2_SH);
1918                         /* RMOR is also in HID4 */
1919                         lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
1920                                 << HID4_RMOR_SH;
1921                 } else {
1922                         /* POWER7 */
1923                         lpcr_mask = LPCR_VPM0 | LPCR_VRMA_L | LPCR_RMLS;
1924                         lpcr = rmls << LPCR_RMLS_SH;
1925                         kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT;
1926                 }
1927                 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
1928                         ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
1929
1930                 /* Initialize phys addrs of pages in RMO */
1931                 npages = kvm_rma_pages;
1932                 porder = __ilog2(npages);
1933                 physp = memslot->arch.slot_phys;
1934                 if (physp) {
1935                         if (npages > memslot->npages)
1936                                 npages = memslot->npages;
1937                         spin_lock(&kvm->arch.slot_phys_lock);
1938                         for (i = 0; i < npages; ++i)
1939                                 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
1940                                         porder;
1941                         spin_unlock(&kvm->arch.slot_phys_lock);
1942                 }
1943         }
1944
1945         kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
1946
1947         /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
1948         smp_wmb();
1949         kvm->arch.rma_setup_done = 1;
1950         err = 0;
1951  out_srcu:
1952         srcu_read_unlock(&kvm->srcu, srcu_idx);
1953  out:
1954         mutex_unlock(&kvm->lock);
1955         return err;
1956
1957  up_out:
1958         up_read(&current->mm->mmap_sem);
1959         goto out_srcu;
1960 }
1961
1962 int kvmppc_core_init_vm(struct kvm *kvm)
1963 {
1964         unsigned long lpcr, lpid;
1965
1966         /* Allocate the guest's logical partition ID */
1967
1968         lpid = kvmppc_alloc_lpid();
1969         if ((long)lpid < 0)
1970                 return -ENOMEM;
1971         kvm->arch.lpid = lpid;
1972
1973         /*
1974          * Since we don't flush the TLB when tearing down a VM,
1975          * and this lpid might have previously been used,
1976          * make sure we flush on each core before running the new VM.
1977          */
1978         cpumask_setall(&kvm->arch.need_tlb_flush);
1979
1980         INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
1981         INIT_LIST_HEAD(&kvm->arch.rtas_tokens);
1982
1983         kvm->arch.rma = NULL;
1984
1985         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
1986
1987         if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1988                 /* PPC970; HID4 is effectively the LPCR */
1989                 kvm->arch.host_lpid = 0;
1990                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
1991                 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
1992                 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
1993                         ((lpid & 0xf) << HID4_LPID5_SH);
1994         } else {
1995                 /* POWER7; init LPCR for virtual RMA mode */
1996                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
1997                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
1998                 lpcr &= LPCR_PECE | LPCR_LPES;
1999                 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
2000                         LPCR_VPM0 | LPCR_VPM1;
2001                 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2002                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
2003         }
2004         kvm->arch.lpcr = lpcr;
2005
2006         kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
2007         spin_lock_init(&kvm->arch.slot_phys_lock);
2008
2009         /*
2010          * Don't allow secondary CPU threads to come online
2011          * while any KVM VMs exist.
2012          */
2013         inhibit_secondary_onlining();
2014
2015         return 0;
2016 }
2017
2018 void kvmppc_core_destroy_vm(struct kvm *kvm)
2019 {
2020         uninhibit_secondary_onlining();
2021
2022         if (kvm->arch.rma) {
2023                 kvm_release_rma(kvm->arch.rma);
2024                 kvm->arch.rma = NULL;
2025         }
2026
2027         kvmppc_rtas_tokens_free(kvm);
2028
2029         kvmppc_free_hpt(kvm);
2030         WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
2031 }
2032
2033 /* These are stubs for now */
2034 void kvmppc_mmu_pte_pflush(struct kvm_vcpu *vcpu, ulong pa_start, ulong pa_end)
2035 {
2036 }
2037
2038 /* We don't need to emulate any privileged instructions or dcbz */
2039 int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu,
2040                            unsigned int inst, int *advance)
2041 {
2042         return EMULATE_FAIL;
2043 }
2044
2045 int kvmppc_core_emulate_mtspr(struct kvm_vcpu *vcpu, int sprn, ulong spr_val)
2046 {
2047         return EMULATE_FAIL;
2048 }
2049
2050 int kvmppc_core_emulate_mfspr(struct kvm_vcpu *vcpu, int sprn, ulong *spr_val)
2051 {
2052         return EMULATE_FAIL;
2053 }
2054
2055 static int kvmppc_book3s_hv_init(void)
2056 {
2057         int r;
2058
2059         r = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2060
2061         if (r)
2062                 return r;
2063
2064         r = kvmppc_mmu_hv_init();
2065
2066         return r;
2067 }
2068
2069 static void kvmppc_book3s_hv_exit(void)
2070 {
2071         kvm_exit();
2072 }
2073
2074 module_init(kvmppc_book3s_hv_init);
2075 module_exit(kvmppc_book3s_hv_exit);