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 #include <linux/module.h>
  56
  57 #include "book3s.h"
  58
  59 /* #define EXIT_DEBUG */
  60 /* #define EXIT_DEBUG_SIMPLE */
  61 /* #define EXIT_DEBUG_INT */
  62
  63 /* Used to indicate that a guest page fault needs to be handled */
  64 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  65
  66 /* Used as a "null" value for timebase values */
  67 #define TB_NIL  (~(u64)0)
  68
  69 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
  70 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
  71
  72 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
  73 {
  74         int me;
  75         int cpu = vcpu->cpu;
  76         wait_queue_head_t *wqp;
  77
  78         wqp = kvm_arch_vcpu_wq(vcpu);
  79         if (waitqueue_active(wqp)) {
  80                 wake_up_interruptible(wqp);
  81                 ++vcpu->stat.halt_wakeup;
  82         }
  83
  84         me = get_cpu();
  85
  86         /* CPU points to the first thread of the core */
  87         if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
  88                 int real_cpu = cpu + vcpu->arch.ptid;
  89                 if (paca[real_cpu].kvm_hstate.xics_phys)
  90                         xics_wake_cpu(real_cpu);
  91                 else if (cpu_online(cpu))
  92                         smp_send_reschedule(cpu);
  93         }
  94         put_cpu();
  95 }
  96
  97 /*
  98  * We use the vcpu_load/put functions to measure stolen time.
  99  * Stolen time is counted as time when either the vcpu is able to
 100  * run as part of a virtual core, but the task running the vcore
 101  * is preempted or sleeping, or when the vcpu needs something done
 102  * in the kernel by the task running the vcpu, but that task is
 103  * preempted or sleeping.  Those two things have to be counted
 104  * separately, since one of the vcpu tasks will take on the job
 105  * of running the core, and the other vcpu tasks in the vcore will
 106  * sleep waiting for it to do that, but that sleep shouldn't count
 107  * as stolen time.
 108  *
 109  * Hence we accumulate stolen time when the vcpu can run as part of
 110  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 111  * needs its task to do other things in the kernel (for example,
 112  * service a page fault) in busy_stolen.  We don't accumulate
 113  * stolen time for a vcore when it is inactive, or for a vcpu
 114  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 115  * a misnomer; it means that the vcpu task is not executing in
 116  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 117  * the kernel.  We don't have any way of dividing up that time
 118  * between time that the vcpu is genuinely stopped, time that
 119  * the task is actively working on behalf of the vcpu, and time
 120  * that the task is preempted, so we don't count any of it as
 121  * stolen.
 122  *
 123  * Updates to busy_stolen are protected by arch.tbacct_lock;
 124  * updates to vc->stolen_tb are protected by the arch.tbacct_lock
 125  * of the vcpu that has taken responsibility for running the vcore
 126  * (i.e. vc->runner).  The stolen times are measured in units of
 127  * timebase ticks.  (Note that the != TB_NIL checks below are
 128  * purely defensive; they should never fail.)
 129  */
 130
 131 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 132 {
 133         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 134         unsigned long flags;
 135
 136         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 137         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
 138             vc->preempt_tb != TB_NIL) {
 139                 vc->stolen_tb += mftb() - vc->preempt_tb;
 140                 vc->preempt_tb = TB_NIL;
 141         }
 142         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 143             vcpu->arch.busy_preempt != TB_NIL) {
 144                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 145                 vcpu->arch.busy_preempt = TB_NIL;
 146         }
 147         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 148 }
 149
 150 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 151 {
 152         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 153         unsigned long flags;
 154
 155         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 156         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
 157                 vc->preempt_tb = mftb();
 158         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 159                 vcpu->arch.busy_preempt = mftb();
 160         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 161 }
 162
 163 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
 164 {
 165         vcpu->arch.shregs.msr = msr;
 166         kvmppc_end_cede(vcpu);
 167 }
 168
 169 void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 170 {
 171         vcpu->arch.pvr = pvr;
 172 }
 173
 174 int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 175 {
 176         unsigned long pcr = 0;
 177         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 178
 179         if (arch_compat) {
 180                 if (!cpu_has_feature(CPU_FTR_ARCH_206))
 181                         return -EINVAL; /* 970 has no compat mode support */
 182
 183                 switch (arch_compat) {
 184                 case PVR_ARCH_205:
 185                         pcr = PCR_ARCH_205;
 186                         break;
 187                 case PVR_ARCH_206:
 188                 case PVR_ARCH_206p:
 189                         break;
 190                 default:
 191                         return -EINVAL;
 192                 }
 193         }
 194
 195         spin_lock(&vc->lock);
 196         vc->arch_compat = arch_compat;
 197         vc->pcr = pcr;
 198         spin_unlock(&vc->lock);
 199
 200         return 0;
 201 }
 202
 203 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 204 {
 205         int r;
 206
 207         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 208         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 209                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
 210         for (r = 0; r < 16; ++r)
 211                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 212                        r, kvmppc_get_gpr(vcpu, r),
 213                        r+16, kvmppc_get_gpr(vcpu, r+16));
 214         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 215                vcpu->arch.ctr, vcpu->arch.lr);
 216         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 217                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 218         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 219                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 220         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 221                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 222         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 223                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 224         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 225         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 226                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 227         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 228         for (r = 0; r < vcpu->arch.slb_max; ++r)
 229                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 230                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 231         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 232                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 233                vcpu->arch.last_inst);
 234 }
 235
 236 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 237 {
 238         int r;
 239         struct kvm_vcpu *v, *ret = NULL;
 240
 241         mutex_lock(&kvm->lock);
 242         kvm_for_each_vcpu(r, v, kvm) {
 243                 if (v->vcpu_id == id) {
 244                         ret = v;
 245                         break;
 246                 }
 247         }
 248         mutex_unlock(&kvm->lock);
 249         return ret;
 250 }
 251
 252 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 253 {
 254         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 255         vpa->yield_count = 1;
 256 }
 257
 258 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 259                    unsigned long addr, unsigned long len)
 260 {
 261         /* check address is cacheline aligned */
 262         if (addr & (L1_CACHE_BYTES - 1))
 263                 return -EINVAL;
 264         spin_lock(&vcpu->arch.vpa_update_lock);
 265         if (v->next_gpa != addr || v->len != len) {
 266                 v->next_gpa = addr;
 267                 v->len = addr ? len : 0;
 268                 v->update_pending = 1;
 269         }
 270         spin_unlock(&vcpu->arch.vpa_update_lock);
 271         return 0;
 272 }
 273
 274 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 275 struct reg_vpa {
 276         u32 dummy;
 277         union {
 278                 u16 hword;
 279                 u32 word;
 280         } length;
 281 };
 282
 283 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 284 {
 285         if (vpap->update_pending)
 286                 return vpap->next_gpa != 0;
 287         return vpap->pinned_addr != NULL;
 288 }
 289
 290 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 291                                        unsigned long flags,
 292                                        unsigned long vcpuid, unsigned long vpa)
 293 {
 294         struct kvm *kvm = vcpu->kvm;
 295         unsigned long len, nb;
 296         void *va;
 297         struct kvm_vcpu *tvcpu;
 298         int err;
 299         int subfunc;
 300         struct kvmppc_vpa *vpap;
 301
 302         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 303         if (!tvcpu)
 304                 return H_PARAMETER;
 305
 306         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 307         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 308             subfunc == H_VPA_REG_SLB) {
 309                 /* Registering new area - address must be cache-line aligned */
 310                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 311                         return H_PARAMETER;
 312
 313                 /* convert logical addr to kernel addr and read length */
 314                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 315                 if (va == NULL)
 316                         return H_PARAMETER;
 317                 if (subfunc == H_VPA_REG_VPA)
 318                         len = ((struct reg_vpa *)va)->length.hword;
 319                 else
 320                         len = ((struct reg_vpa *)va)->length.word;
 321                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 322
 323                 /* Check length */
 324                 if (len > nb || len < sizeof(struct reg_vpa))
 325                         return H_PARAMETER;
 326         } else {
 327                 vpa = 0;
 328                 len = 0;
 329         }
 330
 331         err = H_PARAMETER;
 332         vpap = NULL;
 333         spin_lock(&tvcpu->arch.vpa_update_lock);
 334
 335         switch (subfunc) {
 336         case H_VPA_REG_VPA:             /* register VPA */
 337                 if (len < sizeof(struct lppaca))
 338                         break;
 339                 vpap = &tvcpu->arch.vpa;
 340                 err = 0;
 341                 break;
 342
 343         case H_VPA_REG_DTL:             /* register DTL */
 344                 if (len < sizeof(struct dtl_entry))
 345                         break;
 346                 len -= len % sizeof(struct dtl_entry);
 347
 348                 /* Check that they have previously registered a VPA */
 349                 err = H_RESOURCE;
 350                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 351                         break;
 352
 353                 vpap = &tvcpu->arch.dtl;
 354                 err = 0;
 355                 break;
 356
 357         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 358                 /* Check that they have previously registered a VPA */
 359                 err = H_RESOURCE;
 360                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 361                         break;
 362
 363                 vpap = &tvcpu->arch.slb_shadow;
 364                 err = 0;
 365                 break;
 366
 367         case H_VPA_DEREG_VPA:           /* deregister VPA */
 368                 /* Check they don't still have a DTL or SLB buf registered */
 369                 err = H_RESOURCE;
 370                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 371                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 372                         break;
 373
 374                 vpap = &tvcpu->arch.vpa;
 375                 err = 0;
 376                 break;
 377
 378         case H_VPA_DEREG_DTL:           /* deregister DTL */
 379                 vpap = &tvcpu->arch.dtl;
 380                 err = 0;
 381                 break;
 382
 383         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 384                 vpap = &tvcpu->arch.slb_shadow;
 385                 err = 0;
 386                 break;
 387         }
 388
 389         if (vpap) {
 390                 vpap->next_gpa = vpa;
 391                 vpap->len = len;
 392                 vpap->update_pending = 1;
 393         }
 394
 395         spin_unlock(&tvcpu->arch.vpa_update_lock);
 396
 397         return err;
 398 }
 399
 400 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 401 {
 402         struct kvm *kvm = vcpu->kvm;
 403         void *va;
 404         unsigned long nb;
 405         unsigned long gpa;
 406
 407         /*
 408          * We need to pin the page pointed to by vpap->next_gpa,
 409          * but we can't call kvmppc_pin_guest_page under the lock
 410          * as it does get_user_pages() and down_read().  So we
 411          * have to drop the lock, pin the page, then get the lock
 412          * again and check that a new area didn't get registered
 413          * in the meantime.
 414          */
 415         for (;;) {
 416                 gpa = vpap->next_gpa;
 417                 spin_unlock(&vcpu->arch.vpa_update_lock);
 418                 va = NULL;
 419                 nb = 0;
 420                 if (gpa)
 421                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 422                 spin_lock(&vcpu->arch.vpa_update_lock);
 423                 if (gpa == vpap->next_gpa)
 424                         break;
 425                 /* sigh... unpin that one and try again */
 426                 if (va)
 427                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 428         }
 429
 430         vpap->update_pending = 0;
 431         if (va && nb < vpap->len) {
 432                 /*
 433                  * If it's now too short, it must be that userspace
 434                  * has changed the mappings underlying guest memory,
 435                  * so unregister the region.
 436                  */
 437                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 438                 va = NULL;
 439         }
 440         if (vpap->pinned_addr)
 441                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 442                                         vpap->dirty);
 443         vpap->gpa = gpa;
 444         vpap->pinned_addr = va;
 445         vpap->dirty = false;
 446         if (va)
 447                 vpap->pinned_end = va + vpap->len;
 448 }
 449
 450 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 451 {
 452         if (!(vcpu->arch.vpa.update_pending ||
 453               vcpu->arch.slb_shadow.update_pending ||
 454               vcpu->arch.dtl.update_pending))
 455                 return;
 456
 457         spin_lock(&vcpu->arch.vpa_update_lock);
 458         if (vcpu->arch.vpa.update_pending) {
 459                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 460                 if (vcpu->arch.vpa.pinned_addr)
 461                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 462         }
 463         if (vcpu->arch.dtl.update_pending) {
 464                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 465                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 466                 vcpu->arch.dtl_index = 0;
 467         }
 468         if (vcpu->arch.slb_shadow.update_pending)
 469                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 470         spin_unlock(&vcpu->arch.vpa_update_lock);
 471 }
 472
 473 /*
 474  * Return the accumulated stolen time for the vcore up until `now'.
 475  * The caller should hold the vcore lock.
 476  */
 477 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 478 {
 479         u64 p;
 480
 481         /*
 482          * If we are the task running the vcore, then since we hold
 483          * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
 484          * can't be updated, so we don't need the tbacct_lock.
 485          * If the vcore is inactive, it can't become active (since we
 486          * hold the vcore lock), so the vcpu load/put functions won't
 487          * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
 488          */
 489         if (vc->vcore_state != VCORE_INACTIVE &&
 490             vc->runner->arch.run_task != current) {
 491                 spin_lock_irq(&vc->runner->arch.tbacct_lock);
 492                 p = vc->stolen_tb;
 493                 if (vc->preempt_tb != TB_NIL)
 494                         p += now - vc->preempt_tb;
 495                 spin_unlock_irq(&vc->runner->arch.tbacct_lock);
 496         } else {
 497                 p = vc->stolen_tb;
 498         }
 499         return p;
 500 }
 501
 502 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 503                                     struct kvmppc_vcore *vc)
 504 {
 505         struct dtl_entry *dt;
 506         struct lppaca *vpa;
 507         unsigned long stolen;
 508         unsigned long core_stolen;
 509         u64 now;
 510
 511         dt = vcpu->arch.dtl_ptr;
 512         vpa = vcpu->arch.vpa.pinned_addr;
 513         now = mftb();
 514         core_stolen = vcore_stolen_time(vc, now);
 515         stolen = core_stolen - vcpu->arch.stolen_logged;
 516         vcpu->arch.stolen_logged = core_stolen;
 517         spin_lock_irq(&vcpu->arch.tbacct_lock);
 518         stolen += vcpu->arch.busy_stolen;
 519         vcpu->arch.busy_stolen = 0;
 520         spin_unlock_irq(&vcpu->arch.tbacct_lock);
 521         if (!dt || !vpa)
 522                 return;
 523         memset(dt, 0, sizeof(struct dtl_entry));
 524         dt->dispatch_reason = 7;
 525         dt->processor_id = vc->pcpu + vcpu->arch.ptid;
 526         dt->timebase = now + vc->tb_offset;
 527         dt->enqueue_to_dispatch_time = stolen;
 528         dt->srr0 = kvmppc_get_pc(vcpu);
 529         dt->srr1 = vcpu->arch.shregs.msr;
 530         ++dt;
 531         if (dt == vcpu->arch.dtl.pinned_end)
 532                 dt = vcpu->arch.dtl.pinned_addr;
 533         vcpu->arch.dtl_ptr = dt;
 534         /* order writing *dt vs. writing vpa->dtl_idx */
 535         smp_wmb();
 536         vpa->dtl_idx = ++vcpu->arch.dtl_index;
 537         vcpu->arch.dtl.dirty = true;
 538 }
 539
 540 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 541 {
 542         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 543         unsigned long target, ret = H_SUCCESS;
 544         struct kvm_vcpu *tvcpu;
 545         int idx, rc;
 546
 547         switch (req) {
 548         case H_ENTER:
 549                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 550                 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
 551                                               kvmppc_get_gpr(vcpu, 5),
 552                                               kvmppc_get_gpr(vcpu, 6),
 553                                               kvmppc_get_gpr(vcpu, 7));
 554                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 555                 break;
 556         case H_CEDE:
 557                 break;
 558         case H_PROD:
 559                 target = kvmppc_get_gpr(vcpu, 4);
 560                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 561                 if (!tvcpu) {
 562                         ret = H_PARAMETER;
 563                         break;
 564                 }
 565                 tvcpu->arch.prodded = 1;
 566                 smp_mb();
 567                 if (vcpu->arch.ceded) {
 568                         if (waitqueue_active(&vcpu->wq)) {
 569                                 wake_up_interruptible(&vcpu->wq);
 570                                 vcpu->stat.halt_wakeup++;
 571                         }
 572                 }
 573                 break;
 574         case H_CONFER:
 575                 target = kvmppc_get_gpr(vcpu, 4);
 576                 if (target == -1)
 577                         break;
 578                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 579                 if (!tvcpu) {
 580                         ret = H_PARAMETER;
 581                         break;
 582                 }
 583                 kvm_vcpu_yield_to(tvcpu);
 584                 break;
 585         case H_REGISTER_VPA:
 586                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 587                                         kvmppc_get_gpr(vcpu, 5),
 588                                         kvmppc_get_gpr(vcpu, 6));
 589                 break;
 590         case H_RTAS:
 591                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 592                         return RESUME_HOST;
 593
 594                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 595                 rc = kvmppc_rtas_hcall(vcpu);
 596                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 597
 598                 if (rc == -ENOENT)
 599                         return RESUME_HOST;
 600                 else if (rc == 0)
 601                         break;
 602
 603                 /* Send the error out to userspace via KVM_RUN */
 604                 return rc;
 605
 606         case H_XIRR:
 607         case H_CPPR:
 608         case H_EOI:
 609         case H_IPI:
 610         case H_IPOLL:
 611         case H_XIRR_X:
 612                 if (kvmppc_xics_enabled(vcpu)) {
 613                         ret = kvmppc_xics_hcall(vcpu, req);
 614                         break;
 615                 } /* fallthrough */
 616         default:
 617                 return RESUME_HOST;
 618         }
 619         kvmppc_set_gpr(vcpu, 3, ret);
 620         vcpu->arch.hcall_needed = 0;
 621         return RESUME_GUEST;
 622 }
 623
 624 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
 625                                  struct task_struct *tsk)
 626 {
 627         int r = RESUME_HOST;
 628
 629         vcpu->stat.sum_exits++;
 630
 631         run->exit_reason = KVM_EXIT_UNKNOWN;
 632         run->ready_for_interrupt_injection = 1;
 633         switch (vcpu->arch.trap) {
 634         /* We're good on these - the host merely wanted to get our attention */
 635         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 636                 vcpu->stat.dec_exits++;
 637                 r = RESUME_GUEST;
 638                 break;
 639         case BOOK3S_INTERRUPT_EXTERNAL:
 640                 vcpu->stat.ext_intr_exits++;
 641                 r = RESUME_GUEST;
 642                 break;
 643         case BOOK3S_INTERRUPT_PERFMON:
 644                 r = RESUME_GUEST;
 645                 break;
 646         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 647                 /*
 648                  * Deliver a machine check interrupt to the guest.
 649                  * We have to do this, even if the host has handled the
 650                  * machine check, because machine checks use SRR0/1 and
 651                  * the interrupt might have trashed guest state in them.
 652                  */
 653                 kvmppc_book3s_queue_irqprio(vcpu,
 654                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 655                 r = RESUME_GUEST;
 656                 break;
 657         case BOOK3S_INTERRUPT_PROGRAM:
 658         {
 659                 ulong flags;
 660                 /*
 661                  * Normally program interrupts are delivered directly
 662                  * to the guest by the hardware, but we can get here
 663                  * as a result of a hypervisor emulation interrupt
 664                  * (e40) getting turned into a 700 by BML RTAS.
 665                  */
 666                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 667                 kvmppc_core_queue_program(vcpu, flags);
 668                 r = RESUME_GUEST;
 669                 break;
 670         }
 671         case BOOK3S_INTERRUPT_SYSCALL:
 672         {
 673                 /* hcall - punt to userspace */
 674                 int i;
 675
 676                 if (vcpu->arch.shregs.msr & MSR_PR) {
 677                         /* sc 1 from userspace - reflect to guest syscall */
 678                         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_SYSCALL);
 679                         r = RESUME_GUEST;
 680                         break;
 681                 }
 682                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 683                 for (i = 0; i < 9; ++i)
 684                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 685                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 686                 vcpu->arch.hcall_needed = 1;
 687                 r = RESUME_HOST;
 688                 break;
 689         }
 690         /*
 691          * We get these next two if the guest accesses a page which it thinks
 692          * it has mapped but which is not actually present, either because
 693          * it is for an emulated I/O device or because the corresonding
 694          * host page has been paged out.  Any other HDSI/HISI interrupts
 695          * have been handled already.
 696          */
 697         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 698                 r = RESUME_PAGE_FAULT;
 699                 break;
 700         case BOOK3S_INTERRUPT_H_INST_STORAGE:
 701                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
 702                 vcpu->arch.fault_dsisr = 0;
 703                 r = RESUME_PAGE_FAULT;
 704                 break;
 705         /*
 706          * This occurs if the guest executes an illegal instruction.
 707          * We just generate a program interrupt to the guest, since
 708          * we don't emulate any guest instructions at this stage.
 709          */
 710         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 711                 kvmppc_core_queue_program(vcpu, 0x80000);
 712                 r = RESUME_GUEST;
 713                 break;
 714         default:
 715                 kvmppc_dump_regs(vcpu);
 716                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 717                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 718                         vcpu->arch.shregs.msr);
 719                 run->hw.hardware_exit_reason = vcpu->arch.trap;
 720                 r = RESUME_HOST;
 721                 break;
 722         }
 723
 724         return r;
 725 }
 726
 727 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
 728                                             struct kvm_sregs *sregs)
 729 {
 730         int i;
 731
 732         memset(sregs, 0, sizeof(struct kvm_sregs));
 733         sregs->pvr = vcpu->arch.pvr;
 734         for (i = 0; i < vcpu->arch.slb_max; i++) {
 735                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
 736                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 737         }
 738
 739         return 0;
 740 }
 741
 742 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
 743                                             struct kvm_sregs *sregs)
 744 {
 745         int i, j;
 746
 747         kvmppc_set_pvr_hv(vcpu, sregs->pvr);
 748
 749         j = 0;
 750         for (i = 0; i < vcpu->arch.slb_nr; i++) {
 751                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
 752                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
 753                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
 754                         ++j;
 755                 }
 756         }
 757         vcpu->arch.slb_max = j;
 758
 759         return 0;
 760 }
 761
 762 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr)
 763 {
 764         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 765         u64 mask;
 766
 767         spin_lock(&vc->lock);
 768         /*
 769          * Userspace can only modify DPFD (default prefetch depth),
 770          * ILE (interrupt little-endian) and TC (translation control).
 771          */
 772         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
 773         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
 774         spin_unlock(&vc->lock);
 775 }
 776
 777 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
 778                                  union kvmppc_one_reg *val)
 779 {
 780         int r = 0;
 781         long int i;
 782
 783         switch (id) {
 784         case KVM_REG_PPC_HIOR:
 785                 *val = get_reg_val(id, 0);
 786                 break;
 787         case KVM_REG_PPC_DABR:
 788                 *val = get_reg_val(id, vcpu->arch.dabr);
 789                 break;
 790         case KVM_REG_PPC_DSCR:
 791                 *val = get_reg_val(id, vcpu->arch.dscr);
 792                 break;
 793         case KVM_REG_PPC_PURR:
 794                 *val = get_reg_val(id, vcpu->arch.purr);
 795                 break;
 796         case KVM_REG_PPC_SPURR:
 797                 *val = get_reg_val(id, vcpu->arch.spurr);
 798                 break;
 799         case KVM_REG_PPC_AMR:
 800                 *val = get_reg_val(id, vcpu->arch.amr);
 801                 break;
 802         case KVM_REG_PPC_UAMOR:
 803                 *val = get_reg_val(id, vcpu->arch.uamor);
 804                 break;
 805         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
 806                 i = id - KVM_REG_PPC_MMCR0;
 807                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
 808                 break;
 809         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
 810                 i = id - KVM_REG_PPC_PMC1;
 811                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
 812                 break;
 813         case KVM_REG_PPC_SIAR:
 814                 *val = get_reg_val(id, vcpu->arch.siar);
 815                 break;
 816         case KVM_REG_PPC_SDAR:
 817                 *val = get_reg_val(id, vcpu->arch.sdar);
 818                 break;
 819 #ifdef CONFIG_VSX
 820         case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
 821                 if (cpu_has_feature(CPU_FTR_VSX)) {
 822                         /* VSX => FP reg i is stored in arch.vsr[2*i] */
 823                         long int i = id - KVM_REG_PPC_FPR0;
 824                         *val = get_reg_val(id, vcpu->arch.vsr[2 * i]);
 825                 } else {
 826                         /* let generic code handle it */
 827                         r = -EINVAL;
 828                 }
 829                 break;
 830         case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
 831                 if (cpu_has_feature(CPU_FTR_VSX)) {
 832                         long int i = id - KVM_REG_PPC_VSR0;
 833                         val->vsxval[0] = vcpu->arch.vsr[2 * i];
 834                         val->vsxval[1] = vcpu->arch.vsr[2 * i + 1];
 835                 } else {
 836                         r = -ENXIO;
 837                 }
 838                 break;
 839 #endif /* CONFIG_VSX */
 840         case KVM_REG_PPC_VPA_ADDR:
 841                 spin_lock(&vcpu->arch.vpa_update_lock);
 842                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
 843                 spin_unlock(&vcpu->arch.vpa_update_lock);
 844                 break;
 845         case KVM_REG_PPC_VPA_SLB:
 846                 spin_lock(&vcpu->arch.vpa_update_lock);
 847                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
 848                 val->vpaval.length = vcpu->arch.slb_shadow.len;
 849                 spin_unlock(&vcpu->arch.vpa_update_lock);
 850                 break;
 851         case KVM_REG_PPC_VPA_DTL:
 852                 spin_lock(&vcpu->arch.vpa_update_lock);
 853                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
 854                 val->vpaval.length = vcpu->arch.dtl.len;
 855                 spin_unlock(&vcpu->arch.vpa_update_lock);
 856                 break;
 857         case KVM_REG_PPC_TB_OFFSET:
 858                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
 859                 break;
 860         case KVM_REG_PPC_LPCR:
 861                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
 862                 break;
 863         case KVM_REG_PPC_PPR:
 864                 *val = get_reg_val(id, vcpu->arch.ppr);
 865                 break;
 866         case KVM_REG_PPC_ARCH_COMPAT:
 867                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
 868                 break;
 869         default:
 870                 r = -EINVAL;
 871                 break;
 872         }
 873
 874         return r;
 875 }
 876
 877 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
 878                                  union kvmppc_one_reg *val)
 879 {
 880         int r = 0;
 881         long int i;
 882         unsigned long addr, len;
 883
 884         switch (id) {
 885         case KVM_REG_PPC_HIOR:
 886                 /* Only allow this to be set to zero */
 887                 if (set_reg_val(id, *val))
 888                         r = -EINVAL;
 889                 break;
 890         case KVM_REG_PPC_DABR:
 891                 vcpu->arch.dabr = set_reg_val(id, *val);
 892                 break;
 893         case KVM_REG_PPC_DSCR:
 894                 vcpu->arch.dscr = set_reg_val(id, *val);
 895                 break;
 896         case KVM_REG_PPC_PURR:
 897                 vcpu->arch.purr = set_reg_val(id, *val);
 898                 break;
 899         case KVM_REG_PPC_SPURR:
 900                 vcpu->arch.spurr = set_reg_val(id, *val);
 901                 break;
 902         case KVM_REG_PPC_AMR:
 903                 vcpu->arch.amr = set_reg_val(id, *val);
 904                 break;
 905         case KVM_REG_PPC_UAMOR:
 906                 vcpu->arch.uamor = set_reg_val(id, *val);
 907                 break;
 908         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
 909                 i = id - KVM_REG_PPC_MMCR0;
 910                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
 911                 break;
 912         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
 913                 i = id - KVM_REG_PPC_PMC1;
 914                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
 915                 break;
 916         case KVM_REG_PPC_SIAR:
 917                 vcpu->arch.siar = set_reg_val(id, *val);
 918                 break;
 919         case KVM_REG_PPC_SDAR:
 920                 vcpu->arch.sdar = set_reg_val(id, *val);
 921                 break;
 922 #ifdef CONFIG_VSX
 923         case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
 924                 if (cpu_has_feature(CPU_FTR_VSX)) {
 925                         /* VSX => FP reg i is stored in arch.vsr[2*i] */
 926                         long int i = id - KVM_REG_PPC_FPR0;
 927                         vcpu->arch.vsr[2 * i] = set_reg_val(id, *val);
 928                 } else {
 929                         /* let generic code handle it */
 930                         r = -EINVAL;
 931                 }
 932                 break;
 933         case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
 934                 if (cpu_has_feature(CPU_FTR_VSX)) {
 935                         long int i = id - KVM_REG_PPC_VSR0;
 936                         vcpu->arch.vsr[2 * i] = val->vsxval[0];
 937                         vcpu->arch.vsr[2 * i + 1] = val->vsxval[1];
 938                 } else {
 939                         r = -ENXIO;
 940                 }
 941                 break;
 942 #endif /* CONFIG_VSX */
 943         case KVM_REG_PPC_VPA_ADDR:
 944                 addr = set_reg_val(id, *val);
 945                 r = -EINVAL;
 946                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
 947                               vcpu->arch.dtl.next_gpa))
 948                         break;
 949                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
 950                 break;
 951         case KVM_REG_PPC_VPA_SLB:
 952                 addr = val->vpaval.addr;
 953                 len = val->vpaval.length;
 954                 r = -EINVAL;
 955                 if (addr && !vcpu->arch.vpa.next_gpa)
 956                         break;
 957                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
 958                 break;
 959         case KVM_REG_PPC_VPA_DTL:
 960                 addr = val->vpaval.addr;
 961                 len = val->vpaval.length;
 962                 r = -EINVAL;
 963                 if (addr && (len < sizeof(struct dtl_entry) ||
 964                              !vcpu->arch.vpa.next_gpa))
 965                         break;
 966                 len -= len % sizeof(struct dtl_entry);
 967                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
 968                 break;
 969         case KVM_REG_PPC_TB_OFFSET:
 970                 /* round up to multiple of 2^24 */
 971                 vcpu->arch.vcore->tb_offset =
 972                         ALIGN(set_reg_val(id, *val), 1UL << 24);
 973                 break;
 974         case KVM_REG_PPC_LPCR:
 975                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val));
 976                 break;
 977         case KVM_REG_PPC_PPR:
 978                 vcpu->arch.ppr = set_reg_val(id, *val);
 979                 break;
 980         case KVM_REG_PPC_ARCH_COMPAT:
 981                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
 982                 break;
 983         default:
 984                 r = -EINVAL;
 985                 break;
 986         }
 987
 988         return r;
 989 }
 990
 991 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
 992                                                    unsigned int id)
 993 {
 994         struct kvm_vcpu *vcpu;
 995         int err = -EINVAL;
 996         int core;
 997         struct kvmppc_vcore *vcore;
 998
 999         core = id / threads_per_core;
1000         if (core >= KVM_MAX_VCORES)
1001                 goto out;
1002
1003         err = -ENOMEM;
1004         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1005         if (!vcpu)
1006                 goto out;
1007
1008         err = kvm_vcpu_init(vcpu, kvm, id);
1009         if (err)
1010                 goto free_vcpu;
1011
1012         vcpu->arch.shared = &vcpu->arch.shregs;
1013         vcpu->arch.mmcr[0] = MMCR0_FC;
1014         vcpu->arch.ctrl = CTRL_RUNLATCH;
1015         /* default to host PVR, since we can't spoof it */
1016         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1017         spin_lock_init(&vcpu->arch.vpa_update_lock);
1018         spin_lock_init(&vcpu->arch.tbacct_lock);
1019         vcpu->arch.busy_preempt = TB_NIL;
1020
1021         kvmppc_mmu_book3s_hv_init(vcpu);
1022
1023         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1024
1025         init_waitqueue_head(&vcpu->arch.cpu_run);
1026
1027         mutex_lock(&kvm->lock);
1028         vcore = kvm->arch.vcores[core];
1029         if (!vcore) {
1030                 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1031                 if (vcore) {
1032                         INIT_LIST_HEAD(&vcore->runnable_threads);
1033                         spin_lock_init(&vcore->lock);
1034                         init_waitqueue_head(&vcore->wq);
1035                         vcore->preempt_tb = TB_NIL;
1036                         vcore->lpcr = kvm->arch.lpcr;
1037                 }
1038                 kvm->arch.vcores[core] = vcore;
1039                 kvm->arch.online_vcores++;
1040         }
1041         mutex_unlock(&kvm->lock);
1042
1043         if (!vcore)
1044                 goto free_vcpu;
1045
1046         spin_lock(&vcore->lock);
1047         ++vcore->num_threads;
1048         spin_unlock(&vcore->lock);
1049         vcpu->arch.vcore = vcore;
1050
1051         vcpu->arch.cpu_type = KVM_CPU_3S_64;
1052         kvmppc_sanity_check(vcpu);
1053
1054         return vcpu;
1055
1056 free_vcpu:
1057         kmem_cache_free(kvm_vcpu_cache, vcpu);
1058 out:
1059         return ERR_PTR(err);
1060 }
1061
1062 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1063 {
1064         if (vpa->pinned_addr)
1065                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1066                                         vpa->dirty);
1067 }
1068
1069 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1070 {
1071         spin_lock(&vcpu->arch.vpa_update_lock);
1072         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1073         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1074         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1075         spin_unlock(&vcpu->arch.vpa_update_lock);
1076         kvm_vcpu_uninit(vcpu);
1077         kmem_cache_free(kvm_vcpu_cache, vcpu);
1078 }
1079
1080 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1081 {
1082         /* Indicate we want to get back into the guest */
1083         return 1;
1084 }
1085
1086 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1087 {
1088         unsigned long dec_nsec, now;
1089
1090         now = get_tb();
1091         if (now > vcpu->arch.dec_expires) {
1092                 /* decrementer has already gone negative */
1093                 kvmppc_core_queue_dec(vcpu);
1094                 kvmppc_core_prepare_to_enter(vcpu);
1095                 return;
1096         }
1097         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1098                    / tb_ticks_per_sec;
1099         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1100                       HRTIMER_MODE_REL);
1101         vcpu->arch.timer_running = 1;
1102 }
1103
1104 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1105 {
1106         vcpu->arch.ceded = 0;
1107         if (vcpu->arch.timer_running) {
1108                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1109                 vcpu->arch.timer_running = 0;
1110         }
1111 }
1112
1113 extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
1114
1115 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1116                                    struct kvm_vcpu *vcpu)
1117 {
1118         u64 now;
1119
1120         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1121                 return;
1122         spin_lock_irq(&vcpu->arch.tbacct_lock);
1123         now = mftb();
1124         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1125                 vcpu->arch.stolen_logged;
1126         vcpu->arch.busy_preempt = now;
1127         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1128         spin_unlock_irq(&vcpu->arch.tbacct_lock);
1129         --vc->n_runnable;
1130         list_del(&vcpu->arch.run_list);
1131 }
1132
1133 static int kvmppc_grab_hwthread(int cpu)
1134 {
1135         struct paca_struct *tpaca;
1136         long timeout = 1000;
1137
1138         tpaca = &paca[cpu];
1139
1140         /* Ensure the thread won't go into the kernel if it wakes */
1141         tpaca->kvm_hstate.hwthread_req = 1;
1142         tpaca->kvm_hstate.kvm_vcpu = NULL;
1143
1144         /*
1145          * If the thread is already executing in the kernel (e.g. handling
1146          * a stray interrupt), wait for it to get back to nap mode.
1147          * The smp_mb() is to ensure that our setting of hwthread_req
1148          * is visible before we look at hwthread_state, so if this
1149          * races with the code at system_reset_pSeries and the thread
1150          * misses our setting of hwthread_req, we are sure to see its
1151          * setting of hwthread_state, and vice versa.
1152          */
1153         smp_mb();
1154         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1155                 if (--timeout <= 0) {
1156                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1157                         return -EBUSY;
1158                 }
1159                 udelay(1);
1160         }
1161         return 0;
1162 }
1163
1164 static void kvmppc_release_hwthread(int cpu)
1165 {
1166         struct paca_struct *tpaca;
1167
1168         tpaca = &paca[cpu];
1169         tpaca->kvm_hstate.hwthread_req = 0;
1170         tpaca->kvm_hstate.kvm_vcpu = NULL;
1171 }
1172
1173 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1174 {
1175         int cpu;
1176         struct paca_struct *tpaca;
1177         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1178
1179         if (vcpu->arch.timer_running) {
1180                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1181                 vcpu->arch.timer_running = 0;
1182         }
1183         cpu = vc->pcpu + vcpu->arch.ptid;
1184         tpaca = &paca[cpu];
1185         tpaca->kvm_hstate.kvm_vcpu = vcpu;
1186         tpaca->kvm_hstate.kvm_vcore = vc;
1187         tpaca->kvm_hstate.napping = 0;
1188         vcpu->cpu = vc->pcpu;
1189         smp_wmb();
1190 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1191         if (vcpu->arch.ptid) {
1192                 xics_wake_cpu(cpu);
1193                 ++vc->n_woken;
1194         }
1195 #endif
1196 }
1197
1198 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1199 {
1200         int i;
1201
1202         HMT_low();
1203         i = 0;
1204         while (vc->nap_count < vc->n_woken) {
1205                 if (++i >= 1000000) {
1206                         pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1207                                vc->nap_count, vc->n_woken);
1208                         break;
1209                 }
1210                 cpu_relax();
1211         }
1212         HMT_medium();
1213 }
1214
1215 /*
1216  * Check that we are on thread 0 and that any other threads in
1217  * this core are off-line.  Then grab the threads so they can't
1218  * enter the kernel.
1219  */
1220 static int on_primary_thread(void)
1221 {
1222         int cpu = smp_processor_id();
1223         int thr = cpu_thread_in_core(cpu);
1224
1225         if (thr)
1226                 return 0;
1227         while (++thr < threads_per_core)
1228                 if (cpu_online(cpu + thr))
1229                         return 0;
1230
1231         /* Grab all hw threads so they can't go into the kernel */
1232         for (thr = 1; thr < threads_per_core; ++thr) {
1233                 if (kvmppc_grab_hwthread(cpu + thr)) {
1234                         /* Couldn't grab one; let the others go */
1235                         do {
1236                                 kvmppc_release_hwthread(cpu + thr);
1237                         } while (--thr > 0);
1238                         return 0;
1239                 }
1240         }
1241         return 1;
1242 }
1243
1244 /*
1245  * Run a set of guest threads on a physical core.
1246  * Called with vc->lock held.
1247  */
1248 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1249 {
1250         struct kvm_vcpu *vcpu, *vcpu0, *vnext;
1251         long ret;
1252         u64 now;
1253         int ptid, i, need_vpa_update;
1254         int srcu_idx;
1255         struct kvm_vcpu *vcpus_to_update[threads_per_core];
1256
1257         /* don't start if any threads have a signal pending */
1258         need_vpa_update = 0;
1259         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1260                 if (signal_pending(vcpu->arch.run_task))
1261                         return;
1262                 if (vcpu->arch.vpa.update_pending ||
1263                     vcpu->arch.slb_shadow.update_pending ||
1264                     vcpu->arch.dtl.update_pending)
1265                         vcpus_to_update[need_vpa_update++] = vcpu;
1266         }
1267
1268         /*
1269          * Initialize *vc, in particular vc->vcore_state, so we can
1270          * drop the vcore lock if necessary.
1271          */
1272         vc->n_woken = 0;
1273         vc->nap_count = 0;
1274         vc->entry_exit_count = 0;
1275         vc->vcore_state = VCORE_STARTING;
1276         vc->in_guest = 0;
1277         vc->napping_threads = 0;
1278
1279         /*
1280          * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1281          * which can't be called with any spinlocks held.
1282          */
1283         if (need_vpa_update) {
1284                 spin_unlock(&vc->lock);
1285                 for (i = 0; i < need_vpa_update; ++i)
1286                         kvmppc_update_vpas(vcpus_to_update[i]);
1287                 spin_lock(&vc->lock);
1288         }
1289
1290         /*
1291          * Assign physical thread IDs, first to non-ceded vcpus
1292          * and then to ceded ones.
1293          */
1294         ptid = 0;
1295         vcpu0 = NULL;
1296         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1297                 if (!vcpu->arch.ceded) {
1298                         if (!ptid)
1299                                 vcpu0 = vcpu;
1300                         vcpu->arch.ptid = ptid++;
1301                 }
1302         }
1303         if (!vcpu0)
1304                 goto out;       /* nothing to run; should never happen */
1305         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1306                 if (vcpu->arch.ceded)
1307                         vcpu->arch.ptid = ptid++;
1308
1309         /*
1310          * Make sure we are running on thread 0, and that
1311          * secondary threads are offline.
1312          */
1313         if (threads_per_core > 1 && !on_primary_thread()) {
1314                 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1315                         vcpu->arch.ret = -EBUSY;
1316                 goto out;
1317         }
1318
1319         vc->pcpu = smp_processor_id();
1320         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1321                 kvmppc_start_thread(vcpu);
1322                 kvmppc_create_dtl_entry(vcpu, vc);
1323         }
1324
1325         vc->vcore_state = VCORE_RUNNING;
1326         preempt_disable();
1327         spin_unlock(&vc->lock);
1328
1329         kvm_guest_enter();
1330
1331         srcu_idx = srcu_read_lock(&vcpu0->kvm->srcu);
1332
1333         __kvmppc_vcore_entry(NULL, vcpu0);
1334
1335         spin_lock(&vc->lock);
1336         /* disable sending of IPIs on virtual external irqs */
1337         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1338                 vcpu->cpu = -1;
1339         /* wait for secondary threads to finish writing their state to memory */
1340         if (vc->nap_count < vc->n_woken)
1341                 kvmppc_wait_for_nap(vc);
1342         for (i = 0; i < threads_per_core; ++i)
1343                 kvmppc_release_hwthread(vc->pcpu + i);
1344         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
1345         vc->vcore_state = VCORE_EXITING;
1346         spin_unlock(&vc->lock);
1347
1348         srcu_read_unlock(&vcpu0->kvm->srcu, srcu_idx);
1349
1350         /* make sure updates to secondary vcpu structs are visible now */
1351         smp_mb();
1352         kvm_guest_exit();
1353
1354         preempt_enable();
1355         kvm_resched(vcpu);
1356
1357         spin_lock(&vc->lock);
1358         now = get_tb();
1359         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1360                 /* cancel pending dec exception if dec is positive */
1361                 if (now < vcpu->arch.dec_expires &&
1362                     kvmppc_core_pending_dec(vcpu))
1363                         kvmppc_core_dequeue_dec(vcpu);
1364
1365                 ret = RESUME_GUEST;
1366                 if (vcpu->arch.trap)
1367                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1368                                                     vcpu->arch.run_task);
1369
1370                 vcpu->arch.ret = ret;
1371                 vcpu->arch.trap = 0;
1372
1373                 if (vcpu->arch.ceded) {
1374                         if (ret != RESUME_GUEST)
1375                                 kvmppc_end_cede(vcpu);
1376                         else
1377                                 kvmppc_set_timer(vcpu);
1378                 }
1379         }
1380
1381  out:
1382         vc->vcore_state = VCORE_INACTIVE;
1383         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1384                                  arch.run_list) {
1385                 if (vcpu->arch.ret != RESUME_GUEST) {
1386                         kvmppc_remove_runnable(vc, vcpu);
1387                         wake_up(&vcpu->arch.cpu_run);
1388                 }
1389         }
1390 }
1391
1392 /*
1393  * Wait for some other vcpu thread to execute us, and
1394  * wake us up when we need to handle something in the host.
1395  */
1396 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1397 {
1398         DEFINE_WAIT(wait);
1399
1400         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1401         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1402                 schedule();
1403         finish_wait(&vcpu->arch.cpu_run, &wait);
1404 }
1405
1406 /*
1407  * All the vcpus in this vcore are idle, so wait for a decrementer
1408  * or external interrupt to one of the vcpus.  vc->lock is held.
1409  */
1410 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1411 {
1412         DEFINE_WAIT(wait);
1413
1414         prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1415         vc->vcore_state = VCORE_SLEEPING;
1416         spin_unlock(&vc->lock);
1417         schedule();
1418         finish_wait(&vc->wq, &wait);
1419         spin_lock(&vc->lock);
1420         vc->vcore_state = VCORE_INACTIVE;
1421 }
1422
1423 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1424 {
1425         int n_ceded;
1426         struct kvmppc_vcore *vc;
1427         struct kvm_vcpu *v, *vn;
1428
1429         kvm_run->exit_reason = 0;
1430         vcpu->arch.ret = RESUME_GUEST;
1431         vcpu->arch.trap = 0;
1432         kvmppc_update_vpas(vcpu);
1433
1434         /*
1435          * Synchronize with other threads in this virtual core
1436          */
1437         vc = vcpu->arch.vcore;
1438         spin_lock(&vc->lock);
1439         vcpu->arch.ceded = 0;
1440         vcpu->arch.run_task = current;
1441         vcpu->arch.kvm_run = kvm_run;
1442         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
1443         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1444         vcpu->arch.busy_preempt = TB_NIL;
1445         list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1446         ++vc->n_runnable;
1447
1448         /*
1449          * This happens the first time this is called for a vcpu.
1450          * If the vcore is already running, we may be able to start
1451          * this thread straight away and have it join in.
1452          */
1453         if (!signal_pending(current)) {
1454                 if (vc->vcore_state == VCORE_RUNNING &&
1455                     VCORE_EXIT_COUNT(vc) == 0) {
1456                         vcpu->arch.ptid = vc->n_runnable - 1;
1457                         kvmppc_create_dtl_entry(vcpu, vc);
1458                         kvmppc_start_thread(vcpu);
1459                 } else if (vc->vcore_state == VCORE_SLEEPING) {
1460                         wake_up(&vc->wq);
1461                 }
1462
1463         }
1464
1465         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1466                !signal_pending(current)) {
1467                 if (vc->vcore_state != VCORE_INACTIVE) {
1468                         spin_unlock(&vc->lock);
1469                         kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1470                         spin_lock(&vc->lock);
1471                         continue;
1472                 }
1473                 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1474                                          arch.run_list) {
1475                         kvmppc_core_prepare_to_enter(v);
1476                         if (signal_pending(v->arch.run_task)) {
1477                                 kvmppc_remove_runnable(vc, v);
1478                                 v->stat.signal_exits++;
1479                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1480                                 v->arch.ret = -EINTR;
1481                                 wake_up(&v->arch.cpu_run);
1482                         }
1483                 }
1484                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1485                         break;
1486                 vc->runner = vcpu;
1487                 n_ceded = 0;
1488                 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
1489                         if (!v->arch.pending_exceptions)
1490                                 n_ceded += v->arch.ceded;
1491                         else
1492                                 v->arch.ceded = 0;
1493                 }
1494                 if (n_ceded == vc->n_runnable)
1495                         kvmppc_vcore_blocked(vc);
1496                 else
1497                         kvmppc_run_core(vc);
1498                 vc->runner = NULL;
1499         }
1500
1501         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1502                (vc->vcore_state == VCORE_RUNNING ||
1503                 vc->vcore_state == VCORE_EXITING)) {
1504                 spin_unlock(&vc->lock);
1505                 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1506                 spin_lock(&vc->lock);
1507         }
1508
1509         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1510                 kvmppc_remove_runnable(vc, vcpu);
1511                 vcpu->stat.signal_exits++;
1512                 kvm_run->exit_reason = KVM_EXIT_INTR;
1513                 vcpu->arch.ret = -EINTR;
1514         }
1515
1516         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1517                 /* Wake up some vcpu to run the core */
1518                 v = list_first_entry(&vc->runnable_threads,
1519                                      struct kvm_vcpu, arch.run_list);
1520                 wake_up(&v->arch.cpu_run);
1521         }
1522
1523         spin_unlock(&vc->lock);
1524         return vcpu->arch.ret;
1525 }
1526
1527 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
1528 {
1529         int r;
1530         int srcu_idx;
1531
1532         if (!vcpu->arch.sane) {
1533                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1534                 return -EINVAL;
1535         }
1536
1537         kvmppc_core_prepare_to_enter(vcpu);
1538
1539         /* No need to go into the guest when all we'll do is come back out */
1540         if (signal_pending(current)) {
1541                 run->exit_reason = KVM_EXIT_INTR;
1542                 return -EINTR;
1543         }
1544
1545         atomic_inc(&vcpu->kvm->arch.vcpus_running);
1546         /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1547         smp_mb();
1548
1549         /* On the first time here, set up HTAB and VRMA or RMA */
1550         if (!vcpu->kvm->arch.rma_setup_done) {
1551                 r = kvmppc_hv_setup_htab_rma(vcpu);
1552                 if (r)
1553                         goto out;
1554         }
1555
1556         flush_fp_to_thread(current);
1557         flush_altivec_to_thread(current);
1558         flush_vsx_to_thread(current);
1559         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1560         vcpu->arch.pgdir = current->mm->pgd;
1561         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1562
1563         do {
1564                 r = kvmppc_run_vcpu(run, vcpu);
1565
1566                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1567                     !(vcpu->arch.shregs.msr & MSR_PR)) {
1568                         r = kvmppc_pseries_do_hcall(vcpu);
1569                         kvmppc_core_prepare_to_enter(vcpu);
1570                 } else if (r == RESUME_PAGE_FAULT) {
1571                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1572                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
1573                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1574                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1575                 }
1576         } while (r == RESUME_GUEST);
1577
1578  out:
1579         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1580         atomic_dec(&vcpu->kvm->arch.vcpus_running);
1581         return r;
1582 }
1583
1584
1585 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1586    Assumes POWER7 or PPC970. */
1587 static inline int lpcr_rmls(unsigned long rma_size)
1588 {
1589         switch (rma_size) {
1590         case 32ul << 20:        /* 32 MB */
1591                 if (cpu_has_feature(CPU_FTR_ARCH_206))
1592                         return 8;       /* only supported on POWER7 */
1593                 return -1;
1594         case 64ul << 20:        /* 64 MB */
1595                 return 3;
1596         case 128ul << 20:       /* 128 MB */
1597                 return 7;
1598         case 256ul << 20:       /* 256 MB */
1599                 return 4;
1600         case 1ul << 30:         /* 1 GB */
1601                 return 2;
1602         case 16ul << 30:        /* 16 GB */
1603                 return 1;
1604         case 256ul << 30:       /* 256 GB */
1605                 return 0;
1606         default:
1607                 return -1;
1608         }
1609 }
1610
1611 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1612 {
1613         struct page *page;
1614         struct kvm_rma_info *ri = vma->vm_file->private_data;
1615
1616         if (vmf->pgoff >= kvm_rma_pages)
1617                 return VM_FAULT_SIGBUS;
1618
1619         page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1620         get_page(page);
1621         vmf->page = page;
1622         return 0;
1623 }
1624
1625 static const struct vm_operations_struct kvm_rma_vm_ops = {
1626         .fault = kvm_rma_fault,
1627 };
1628
1629 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1630 {
1631         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1632         vma->vm_ops = &kvm_rma_vm_ops;
1633         return 0;
1634 }
1635
1636 static int kvm_rma_release(struct inode *inode, struct file *filp)
1637 {
1638         struct kvm_rma_info *ri = filp->private_data;
1639
1640         kvm_release_rma(ri);
1641         return 0;
1642 }
1643
1644 static const struct file_operations kvm_rma_fops = {
1645         .mmap           = kvm_rma_mmap,
1646         .release        = kvm_rma_release,
1647 };
1648
1649 static long kvm_vm_ioctl_allocate_rma(struct kvm *kvm,
1650                                       struct kvm_allocate_rma *ret)
1651 {
1652         long fd;
1653         struct kvm_rma_info *ri;
1654         /*
1655          * Only do this on PPC970 in HV mode
1656          */
1657         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
1658             !cpu_has_feature(CPU_FTR_ARCH_201))
1659                 return -EINVAL;
1660
1661         if (!kvm_rma_pages)
1662                 return -EINVAL;
1663
1664         ri = kvm_alloc_rma();
1665         if (!ri)
1666                 return -ENOMEM;
1667
1668         fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC);
1669         if (fd < 0)
1670                 kvm_release_rma(ri);
1671
1672         ret->rma_size = kvm_rma_pages << PAGE_SHIFT;
1673         return fd;
1674 }
1675
1676 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1677                                      int linux_psize)
1678 {
1679         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1680
1681         if (!def->shift)
1682                 return;
1683         (*sps)->page_shift = def->shift;
1684         (*sps)->slb_enc = def->sllp;
1685         (*sps)->enc[0].page_shift = def->shift;
1686         /*
1687          * Only return base page encoding. We don't want to return
1688          * all the supporting pte_enc, because our H_ENTER doesn't
1689          * support MPSS yet. Once they do, we can start passing all
1690          * support pte_enc here
1691          */
1692         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
1693         (*sps)++;
1694 }
1695
1696 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
1697                                          struct kvm_ppc_smmu_info *info)
1698 {
1699         struct kvm_ppc_one_seg_page_size *sps;
1700
1701         info->flags = KVM_PPC_PAGE_SIZES_REAL;
1702         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1703                 info->flags |= KVM_PPC_1T_SEGMENTS;
1704         info->slb_size = mmu_slb_size;
1705
1706         /* We only support these sizes for now, and no muti-size segments */
1707         sps = &info->sps[0];
1708         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1709         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1710         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1711
1712         return 0;
1713 }
1714
1715 /*
1716  * Get (and clear) the dirty memory log for a memory slot.
1717  */
1718 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
1719                                          struct kvm_dirty_log *log)
1720 {
1721         struct kvm_memory_slot *memslot;
1722         int r;
1723         unsigned long n;
1724
1725         mutex_lock(&kvm->slots_lock);
1726
1727         r = -EINVAL;
1728         if (log->slot >= KVM_USER_MEM_SLOTS)
1729                 goto out;
1730
1731         memslot = id_to_memslot(kvm->memslots, log->slot);
1732         r = -ENOENT;
1733         if (!memslot->dirty_bitmap)
1734                 goto out;
1735
1736         n = kvm_dirty_bitmap_bytes(memslot);
1737         memset(memslot->dirty_bitmap, 0, n);
1738
1739         r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
1740         if (r)
1741                 goto out;
1742
1743         r = -EFAULT;
1744         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1745                 goto out;
1746
1747         r = 0;
1748 out:
1749         mutex_unlock(&kvm->slots_lock);
1750         return r;
1751 }
1752
1753 static void unpin_slot(struct kvm_memory_slot *memslot)
1754 {
1755         unsigned long *physp;
1756         unsigned long j, npages, pfn;
1757         struct page *page;
1758
1759         physp = memslot->arch.slot_phys;
1760         npages = memslot->npages;
1761         if (!physp)
1762                 return;
1763         for (j = 0; j < npages; j++) {
1764                 if (!(physp[j] & KVMPPC_GOT_PAGE))
1765                         continue;
1766                 pfn = physp[j] >> PAGE_SHIFT;
1767                 page = pfn_to_page(pfn);
1768                 SetPageDirty(page);
1769                 put_page(page);
1770         }
1771 }
1772
1773 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
1774                                         struct kvm_memory_slot *dont)
1775 {
1776         if (!dont || free->arch.rmap != dont->arch.rmap) {
1777                 vfree(free->arch.rmap);
1778                 free->arch.rmap = NULL;
1779         }
1780         if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
1781                 unpin_slot(free);
1782                 vfree(free->arch.slot_phys);
1783                 free->arch.slot_phys = NULL;
1784         }
1785 }
1786
1787 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
1788                                          unsigned long npages)
1789 {
1790         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
1791         if (!slot->arch.rmap)
1792                 return -ENOMEM;
1793         slot->arch.slot_phys = NULL;
1794
1795         return 0;
1796 }
1797
1798 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
1799                                         struct kvm_memory_slot *memslot,
1800                                         struct kvm_userspace_memory_region *mem)
1801 {
1802         unsigned long *phys;
1803
1804         /* Allocate a slot_phys array if needed */
1805         phys = memslot->arch.slot_phys;
1806         if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
1807                 phys = vzalloc(memslot->npages * sizeof(unsigned long));
1808                 if (!phys)
1809                         return -ENOMEM;
1810                 memslot->arch.slot_phys = phys;
1811         }
1812
1813         return 0;
1814 }
1815
1816 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
1817                                 struct kvm_userspace_memory_region *mem,
1818                                 const struct kvm_memory_slot *old)
1819 {
1820         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
1821         struct kvm_memory_slot *memslot;
1822
1823         if (npages && old->npages) {
1824                 /*
1825                  * If modifying a memslot, reset all the rmap dirty bits.
1826                  * If this is a new memslot, we don't need to do anything
1827                  * since the rmap array starts out as all zeroes,
1828                  * i.e. no pages are dirty.
1829                  */
1830                 memslot = id_to_memslot(kvm->memslots, mem->slot);
1831                 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
1832         }
1833 }
1834
1835 /*
1836  * Update LPCR values in kvm->arch and in vcores.
1837  * Caller must hold kvm->lock.
1838  */
1839 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
1840 {
1841         long int i;
1842         u32 cores_done = 0;
1843
1844         if ((kvm->arch.lpcr & mask) == lpcr)
1845                 return;
1846
1847         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
1848
1849         for (i = 0; i < KVM_MAX_VCORES; ++i) {
1850                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
1851                 if (!vc)
1852                         continue;
1853                 spin_lock(&vc->lock);
1854                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
1855                 spin_unlock(&vc->lock);
1856                 if (++cores_done >= kvm->arch.online_vcores)
1857                         break;
1858         }
1859 }
1860
1861 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
1862 {
1863         return;
1864 }
1865
1866 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
1867 {
1868         int err = 0;
1869         struct kvm *kvm = vcpu->kvm;
1870         struct kvm_rma_info *ri = NULL;
1871         unsigned long hva;
1872         struct kvm_memory_slot *memslot;
1873         struct vm_area_struct *vma;
1874         unsigned long lpcr = 0, senc;
1875         unsigned long lpcr_mask = 0;
1876         unsigned long psize, porder;
1877         unsigned long rma_size;
1878         unsigned long rmls;
1879         unsigned long *physp;
1880         unsigned long i, npages;
1881         int srcu_idx;
1882
1883         mutex_lock(&kvm->lock);
1884         if (kvm->arch.rma_setup_done)
1885                 goto out;       /* another vcpu beat us to it */
1886
1887         /* Allocate hashed page table (if not done already) and reset it */
1888         if (!kvm->arch.hpt_virt) {
1889                 err = kvmppc_alloc_hpt(kvm, NULL);
1890                 if (err) {
1891                         pr_err("KVM: Couldn't alloc HPT\n");
1892                         goto out;
1893                 }
1894         }
1895
1896         /* Look up the memslot for guest physical address 0 */
1897         srcu_idx = srcu_read_lock(&kvm->srcu);
1898         memslot = gfn_to_memslot(kvm, 0);
1899
1900         /* We must have some memory at 0 by now */
1901         err = -EINVAL;
1902         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1903                 goto out_srcu;
1904
1905         /* Look up the VMA for the start of this memory slot */
1906         hva = memslot->userspace_addr;
1907         down_read(&current->mm->mmap_sem);
1908         vma = find_vma(current->mm, hva);
1909         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
1910                 goto up_out;
1911
1912         psize = vma_kernel_pagesize(vma);
1913         porder = __ilog2(psize);
1914
1915         /* Is this one of our preallocated RMAs? */
1916         if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
1917             hva == vma->vm_start)
1918                 ri = vma->vm_file->private_data;
1919
1920         up_read(&current->mm->mmap_sem);
1921
1922         if (!ri) {
1923                 /* On POWER7, use VRMA; on PPC970, give up */
1924                 err = -EPERM;
1925                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1926                         pr_err("KVM: CPU requires an RMO\n");
1927                         goto out_srcu;
1928                 }
1929
1930                 /* We can handle 4k, 64k or 16M pages in the VRMA */
1931                 err = -EINVAL;
1932                 if (!(psize == 0x1000 || psize == 0x10000 ||
1933                       psize == 0x1000000))
1934                         goto out_srcu;
1935
1936                 /* Update VRMASD field in the LPCR */
1937                 senc = slb_pgsize_encoding(psize);
1938                 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1939                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
1940                 lpcr_mask = LPCR_VRMASD;
1941                 /* the -4 is to account for senc values starting at 0x10 */
1942                 lpcr = senc << (LPCR_VRMASD_SH - 4);
1943
1944                 /* Create HPTEs in the hash page table for the VRMA */
1945                 kvmppc_map_vrma(vcpu, memslot, porder);
1946
1947         } else {
1948                 /* Set up to use an RMO region */
1949                 rma_size = kvm_rma_pages;
1950                 if (rma_size > memslot->npages)
1951                         rma_size = memslot->npages;
1952                 rma_size <<= PAGE_SHIFT;
1953                 rmls = lpcr_rmls(rma_size);
1954                 err = -EINVAL;
1955                 if ((long)rmls < 0) {
1956                         pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
1957                         goto out_srcu;
1958                 }
1959                 atomic_inc(&ri->use_count);
1960                 kvm->arch.rma = ri;
1961
1962                 /* Update LPCR and RMOR */
1963                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1964                         /* PPC970; insert RMLS value (split field) in HID4 */
1965                         lpcr_mask = (1ul << HID4_RMLS0_SH) |
1966                                 (3ul << HID4_RMLS2_SH) | HID4_RMOR;
1967                         lpcr = ((rmls >> 2) << HID4_RMLS0_SH) |
1968                                 ((rmls & 3) << HID4_RMLS2_SH);
1969                         /* RMOR is also in HID4 */
1970                         lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
1971                                 << HID4_RMOR_SH;
1972                 } else {
1973                         /* POWER7 */
1974                         lpcr_mask = LPCR_VPM0 | LPCR_VRMA_L | LPCR_RMLS;
1975                         lpcr = rmls << LPCR_RMLS_SH;
1976                         kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT;
1977                 }
1978                 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
1979                         ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
1980
1981                 /* Initialize phys addrs of pages in RMO */
1982                 npages = kvm_rma_pages;
1983                 porder = __ilog2(npages);
1984                 physp = memslot->arch.slot_phys;
1985                 if (physp) {
1986                         if (npages > memslot->npages)
1987                                 npages = memslot->npages;
1988                         spin_lock(&kvm->arch.slot_phys_lock);
1989                         for (i = 0; i < npages; ++i)
1990                                 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
1991                                         porder;
1992                         spin_unlock(&kvm->arch.slot_phys_lock);
1993                 }
1994         }
1995
1996         kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
1997
1998         /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
1999         smp_wmb();
2000         kvm->arch.rma_setup_done = 1;
2001         err = 0;
2002  out_srcu:
2003         srcu_read_unlock(&kvm->srcu, srcu_idx);
2004  out:
2005         mutex_unlock(&kvm->lock);
2006         return err;
2007
2008  up_out:
2009         up_read(&current->mm->mmap_sem);
2010         goto out_srcu;
2011 }
2012
2013 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2014 {
2015         unsigned long lpcr, lpid;
2016
2017         /* Allocate the guest's logical partition ID */
2018
2019         lpid = kvmppc_alloc_lpid();
2020         if ((long)lpid < 0)
2021                 return -ENOMEM;
2022         kvm->arch.lpid = lpid;
2023
2024         /*
2025          * Since we don't flush the TLB when tearing down a VM,
2026          * and this lpid might have previously been used,
2027          * make sure we flush on each core before running the new VM.
2028          */
2029         cpumask_setall(&kvm->arch.need_tlb_flush);
2030
2031         kvm->arch.rma = NULL;
2032
2033         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
2034
2035         if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2036                 /* PPC970; HID4 is effectively the LPCR */
2037                 kvm->arch.host_lpid = 0;
2038                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
2039                 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
2040                 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
2041                         ((lpid & 0xf) << HID4_LPID5_SH);
2042         } else {
2043                 /* POWER7; init LPCR for virtual RMA mode */
2044                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
2045                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2046                 lpcr &= LPCR_PECE | LPCR_LPES;
2047                 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
2048                         LPCR_VPM0 | LPCR_VPM1;
2049                 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2050                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
2051         }
2052         kvm->arch.lpcr = lpcr;
2053
2054         kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
2055         spin_lock_init(&kvm->arch.slot_phys_lock);
2056
2057         /*
2058          * Don't allow secondary CPU threads to come online
2059          * while any KVM VMs exist.
2060          */
2061         inhibit_secondary_onlining();
2062
2063         return 0;
2064 }
2065
2066 static void kvmppc_free_vcores(struct kvm *kvm)
2067 {
2068         long int i;
2069
2070         for (i = 0; i < KVM_MAX_VCORES; ++i)
2071                 kfree(kvm->arch.vcores[i]);
2072         kvm->arch.online_vcores = 0;
2073 }
2074
2075 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
2076 {
2077         uninhibit_secondary_onlining();
2078
2079         kvmppc_free_vcores(kvm);
2080         if (kvm->arch.rma) {
2081                 kvm_release_rma(kvm->arch.rma);
2082                 kvm->arch.rma = NULL;
2083         }
2084
2085         kvmppc_free_hpt(kvm);
2086 }
2087
2088 /* We don't need to emulate any privileged instructions or dcbz */
2089 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2090                                      unsigned int inst, int *advance)
2091 {
2092         return EMULATE_FAIL;
2093 }
2094
2095 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2096                                         ulong spr_val)
2097 {
2098         return EMULATE_FAIL;
2099 }
2100
2101 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2102                                         ulong *spr_val)
2103 {
2104         return EMULATE_FAIL;
2105 }
2106
2107 static int kvmppc_core_check_processor_compat_hv(void)
2108 {
2109         if (!cpu_has_feature(CPU_FTR_HVMODE))
2110                 return -EIO;
2111         return 0;
2112 }
2113
2114 static long kvm_arch_vm_ioctl_hv(struct file *filp,
2115                                  unsigned int ioctl, unsigned long arg)
2116 {
2117         struct kvm *kvm __maybe_unused = filp->private_data;
2118         void __user *argp = (void __user *)arg;
2119         long r;
2120
2121         switch (ioctl) {
2122
2123         case KVM_ALLOCATE_RMA: {
2124                 struct kvm_allocate_rma rma;
2125                 struct kvm *kvm = filp->private_data;
2126
2127                 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
2128                 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
2129                         r = -EFAULT;
2130                 break;
2131         }
2132
2133         case KVM_PPC_ALLOCATE_HTAB: {
2134                 u32 htab_order;
2135
2136                 r = -EFAULT;
2137                 if (get_user(htab_order, (u32 __user *)argp))
2138                         break;
2139                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2140                 if (r)
2141                         break;
2142                 r = -EFAULT;
2143                 if (put_user(htab_order, (u32 __user *)argp))
2144                         break;
2145                 r = 0;
2146                 break;
2147         }
2148
2149         case KVM_PPC_GET_HTAB_FD: {
2150                 struct kvm_get_htab_fd ghf;
2151
2152                 r = -EFAULT;
2153                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
2154                         break;
2155                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2156                 break;
2157         }
2158
2159         default:
2160                 r = -ENOTTY;
2161         }
2162
2163         return r;
2164 }
2165
2166 static struct kvmppc_ops kvm_ops_hv = {
2167         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2168         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2169         .get_one_reg = kvmppc_get_one_reg_hv,
2170         .set_one_reg = kvmppc_set_one_reg_hv,
2171         .vcpu_load   = kvmppc_core_vcpu_load_hv,
2172         .vcpu_put    = kvmppc_core_vcpu_put_hv,
2173         .set_msr     = kvmppc_set_msr_hv,
2174         .vcpu_run    = kvmppc_vcpu_run_hv,
2175         .vcpu_create = kvmppc_core_vcpu_create_hv,
2176         .vcpu_free   = kvmppc_core_vcpu_free_hv,
2177         .check_requests = kvmppc_core_check_requests_hv,
2178         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
2179         .flush_memslot  = kvmppc_core_flush_memslot_hv,
2180         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2181         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
2182         .unmap_hva = kvm_unmap_hva_hv,
2183         .unmap_hva_range = kvm_unmap_hva_range_hv,
2184         .age_hva  = kvm_age_hva_hv,
2185         .test_age_hva = kvm_test_age_hva_hv,
2186         .set_spte_hva = kvm_set_spte_hva_hv,
2187         .mmu_destroy  = kvmppc_mmu_destroy_hv,
2188         .free_memslot = kvmppc_core_free_memslot_hv,
2189         .create_memslot = kvmppc_core_create_memslot_hv,
2190         .init_vm =  kvmppc_core_init_vm_hv,
2191         .destroy_vm = kvmppc_core_destroy_vm_hv,
2192         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2193         .emulate_op = kvmppc_core_emulate_op_hv,
2194         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2195         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2196         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2197         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
2198 };
2199
2200 static int kvmppc_book3s_init_hv(void)
2201 {
2202         int r;
2203         /*
2204          * FIXME!! Do we need to check on all cpus ?
2205          */
2206         r = kvmppc_core_check_processor_compat_hv();
2207         if (r < 0)
2208                 return r;
2209
2210         kvm_ops_hv.owner = THIS_MODULE;
2211         kvmppc_hv_ops = &kvm_ops_hv;
2212
2213         r = kvmppc_mmu_hv_init();
2214         return r;
2215 }
2216
2217 static void kvmppc_book3s_exit_hv(void)
2218 {
2219         kvmppc_hv_ops = NULL;
2220 }
2221
2222 module_init(kvmppc_book3s_init_hv);
2223 module_exit(kvmppc_book3s_exit_hv);
2224 MODULE_LICENSE("GPL");