arch/powerpc/kvm/booke.c

   1 /*
   2  * This program is free software; you can redistribute it and/or modify
   3  * it under the terms of the GNU General Public License, version 2, as
   4  * published by the Free Software Foundation.
   5  *
   6  * This program is distributed in the hope that it will be useful,
   7  * but WITHOUT ANY WARRANTY; without even the implied warranty of
   8  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   9  * GNU General Public License for more details.
  10  *
  11  * You should have received a copy of the GNU General Public License
  12  * along with this program; if not, write to the Free Software
  13  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  14  *
  15  * Copyright IBM Corp. 2007
  16  * Copyright 2010-2011 Freescale Semiconductor, Inc.
  17  *
  18  * Authors: Hollis Blanchard <hollisb@us.ibm.com>
  19  *          Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
  20  *          Scott Wood <scottwood@freescale.com>
  21  *          Varun Sethi <varun.sethi@freescale.com>
  22  */
  23
  24 #include <linux/errno.h>
  25 #include <linux/err.h>
  26 #include <linux/kvm_host.h>
  27 #include <linux/gfp.h>
  28 #include <linux/module.h>
  29 #include <linux/vmalloc.h>
  30 #include <linux/fs.h>
  31
  32 #include <asm/cputable.h>
  33 #include <asm/uaccess.h>
  34 #include <asm/kvm_ppc.h>
  35 #include <asm/cacheflush.h>
  36 #include <asm/dbell.h>
  37 #include <asm/hw_irq.h>
  38 #include <asm/irq.h>
  39 #include <asm/time.h>
  40
  41 #include "timing.h"
  42 #include "booke.h"
  43 #include "trace.h"
  44
  45 unsigned long kvmppc_booke_handlers;
  46
  47 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  48 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  49
  50 struct kvm_stats_debugfs_item debugfs_entries[] = {
  51         { "mmio",       VCPU_STAT(mmio_exits) },
  52         { "dcr",        VCPU_STAT(dcr_exits) },
  53         { "sig",        VCPU_STAT(signal_exits) },
  54         { "itlb_r",     VCPU_STAT(itlb_real_miss_exits) },
  55         { "itlb_v",     VCPU_STAT(itlb_virt_miss_exits) },
  56         { "dtlb_r",     VCPU_STAT(dtlb_real_miss_exits) },
  57         { "dtlb_v",     VCPU_STAT(dtlb_virt_miss_exits) },
  58         { "sysc",       VCPU_STAT(syscall_exits) },
  59         { "isi",        VCPU_STAT(isi_exits) },
  60         { "dsi",        VCPU_STAT(dsi_exits) },
  61         { "inst_emu",   VCPU_STAT(emulated_inst_exits) },
  62         { "dec",        VCPU_STAT(dec_exits) },
  63         { "ext_intr",   VCPU_STAT(ext_intr_exits) },
  64         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  65         { "doorbell", VCPU_STAT(dbell_exits) },
  66         { "guest doorbell", VCPU_STAT(gdbell_exits) },
  67         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
  68         { NULL }
  69 };
  70
  71 /* TODO: use vcpu_printf() */
  72 void kvmppc_dump_vcpu(struct kvm_vcpu *vcpu)
  73 {
  74         int i;
  75
  76         printk("pc:   %08lx msr:  %08llx\n", vcpu->arch.pc, vcpu->arch.shared->msr);
  77         printk("lr:   %08lx ctr:  %08lx\n", vcpu->arch.lr, vcpu->arch.ctr);
  78         printk("srr0: %08llx srr1: %08llx\n", vcpu->arch.shared->srr0,
  79                                             vcpu->arch.shared->srr1);
  80
  81         printk("exceptions: %08lx\n", vcpu->arch.pending_exceptions);
  82
  83         for (i = 0; i < 32; i += 4) {
  84                 printk("gpr%02d: %08lx %08lx %08lx %08lx\n", i,
  85                        kvmppc_get_gpr(vcpu, i),
  86                        kvmppc_get_gpr(vcpu, i+1),
  87                        kvmppc_get_gpr(vcpu, i+2),
  88                        kvmppc_get_gpr(vcpu, i+3));
  89         }
  90 }
  91
  92 #ifdef CONFIG_SPE
  93 void kvmppc_vcpu_disable_spe(struct kvm_vcpu *vcpu)
  94 {
  95         preempt_disable();
  96         enable_kernel_spe();
  97         kvmppc_save_guest_spe(vcpu);
  98         vcpu->arch.shadow_msr &= ~MSR_SPE;
  99         preempt_enable();
 100 }
 101
 102 static void kvmppc_vcpu_enable_spe(struct kvm_vcpu *vcpu)
 103 {
 104         preempt_disable();
 105         enable_kernel_spe();
 106         kvmppc_load_guest_spe(vcpu);
 107         vcpu->arch.shadow_msr |= MSR_SPE;
 108         preempt_enable();
 109 }
 110
 111 static void kvmppc_vcpu_sync_spe(struct kvm_vcpu *vcpu)
 112 {
 113         if (vcpu->arch.shared->msr & MSR_SPE) {
 114                 if (!(vcpu->arch.shadow_msr & MSR_SPE))
 115                         kvmppc_vcpu_enable_spe(vcpu);
 116         } else if (vcpu->arch.shadow_msr & MSR_SPE) {
 117                 kvmppc_vcpu_disable_spe(vcpu);
 118         }
 119 }
 120 #else
 121 static void kvmppc_vcpu_sync_spe(struct kvm_vcpu *vcpu)
 122 {
 123 }
 124 #endif
 125
 126 static void kvmppc_vcpu_sync_fpu(struct kvm_vcpu *vcpu)
 127 {
 128 #if defined(CONFIG_PPC_FPU) && !defined(CONFIG_KVM_BOOKE_HV)
 129         /* We always treat the FP bit as enabled from the host
 130            perspective, so only need to adjust the shadow MSR */
 131         vcpu->arch.shadow_msr &= ~MSR_FP;
 132         vcpu->arch.shadow_msr |= vcpu->arch.shared->msr & MSR_FP;
 133 #endif
 134 }
 135
 136 /*
 137  * Helper function for "full" MSR writes.  No need to call this if only
 138  * EE/CE/ME/DE/RI are changing.
 139  */
 140 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u32 new_msr)
 141 {
 142         u32 old_msr = vcpu->arch.shared->msr;
 143
 144 #ifdef CONFIG_KVM_BOOKE_HV
 145         new_msr |= MSR_GS;
 146 #endif
 147
 148         vcpu->arch.shared->msr = new_msr;
 149
 150         kvmppc_mmu_msr_notify(vcpu, old_msr);
 151         kvmppc_vcpu_sync_spe(vcpu);
 152         kvmppc_vcpu_sync_fpu(vcpu);
 153 }
 154
 155 static void kvmppc_booke_queue_irqprio(struct kvm_vcpu *vcpu,
 156                                        unsigned int priority)
 157 {
 158         trace_kvm_booke_queue_irqprio(vcpu, priority);
 159         set_bit(priority, &vcpu->arch.pending_exceptions);
 160 }
 161
 162 static void kvmppc_core_queue_dtlb_miss(struct kvm_vcpu *vcpu,
 163                                         ulong dear_flags, ulong esr_flags)
 164 {
 165         vcpu->arch.queued_dear = dear_flags;
 166         vcpu->arch.queued_esr = esr_flags;
 167         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DTLB_MISS);
 168 }
 169
 170 static void kvmppc_core_queue_data_storage(struct kvm_vcpu *vcpu,
 171                                            ulong dear_flags, ulong esr_flags)
 172 {
 173         vcpu->arch.queued_dear = dear_flags;
 174         vcpu->arch.queued_esr = esr_flags;
 175         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DATA_STORAGE);
 176 }
 177
 178 static void kvmppc_core_queue_inst_storage(struct kvm_vcpu *vcpu,
 179                                            ulong esr_flags)
 180 {
 181         vcpu->arch.queued_esr = esr_flags;
 182         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_INST_STORAGE);
 183 }
 184
 185 static void kvmppc_core_queue_alignment(struct kvm_vcpu *vcpu, ulong dear_flags,
 186                                         ulong esr_flags)
 187 {
 188         vcpu->arch.queued_dear = dear_flags;
 189         vcpu->arch.queued_esr = esr_flags;
 190         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_ALIGNMENT);
 191 }
 192
 193 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu, ulong esr_flags)
 194 {
 195         vcpu->arch.queued_esr = esr_flags;
 196         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_PROGRAM);
 197 }
 198
 199 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
 200 {
 201         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DECREMENTER);
 202 }
 203
 204 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
 205 {
 206         return test_bit(BOOKE_IRQPRIO_DECREMENTER, &vcpu->arch.pending_exceptions);
 207 }
 208
 209 void kvmppc_core_dequeue_dec(struct kvm_vcpu *vcpu)
 210 {
 211         clear_bit(BOOKE_IRQPRIO_DECREMENTER, &vcpu->arch.pending_exceptions);
 212 }
 213
 214 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
 215                                 struct kvm_interrupt *irq)
 216 {
 217         unsigned int prio = BOOKE_IRQPRIO_EXTERNAL;
 218
 219         if (irq->irq == KVM_INTERRUPT_SET_LEVEL)
 220                 prio = BOOKE_IRQPRIO_EXTERNAL_LEVEL;
 221
 222         kvmppc_booke_queue_irqprio(vcpu, prio);
 223 }
 224
 225 void kvmppc_core_dequeue_external(struct kvm_vcpu *vcpu)
 226 {
 227         clear_bit(BOOKE_IRQPRIO_EXTERNAL, &vcpu->arch.pending_exceptions);
 228         clear_bit(BOOKE_IRQPRIO_EXTERNAL_LEVEL, &vcpu->arch.pending_exceptions);
 229 }
 230
 231 static void kvmppc_core_queue_watchdog(struct kvm_vcpu *vcpu)
 232 {
 233         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_WATCHDOG);
 234 }
 235
 236 static void kvmppc_core_dequeue_watchdog(struct kvm_vcpu *vcpu)
 237 {
 238         clear_bit(BOOKE_IRQPRIO_WATCHDOG, &vcpu->arch.pending_exceptions);
 239 }
 240
 241 static void set_guest_srr(struct kvm_vcpu *vcpu, unsigned long srr0, u32 srr1)
 242 {
 243 #ifdef CONFIG_KVM_BOOKE_HV
 244         mtspr(SPRN_GSRR0, srr0);
 245         mtspr(SPRN_GSRR1, srr1);
 246 #else
 247         vcpu->arch.shared->srr0 = srr0;
 248         vcpu->arch.shared->srr1 = srr1;
 249 #endif
 250 }
 251
 252 static void set_guest_csrr(struct kvm_vcpu *vcpu, unsigned long srr0, u32 srr1)
 253 {
 254         vcpu->arch.csrr0 = srr0;
 255         vcpu->arch.csrr1 = srr1;
 256 }
 257
 258 static void set_guest_dsrr(struct kvm_vcpu *vcpu, unsigned long srr0, u32 srr1)
 259 {
 260         if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC)) {
 261                 vcpu->arch.dsrr0 = srr0;
 262                 vcpu->arch.dsrr1 = srr1;
 263         } else {
 264                 set_guest_csrr(vcpu, srr0, srr1);
 265         }
 266 }
 267
 268 static void set_guest_mcsrr(struct kvm_vcpu *vcpu, unsigned long srr0, u32 srr1)
 269 {
 270         vcpu->arch.mcsrr0 = srr0;
 271         vcpu->arch.mcsrr1 = srr1;
 272 }
 273
 274 static unsigned long get_guest_dear(struct kvm_vcpu *vcpu)
 275 {
 276 #ifdef CONFIG_KVM_BOOKE_HV
 277         return mfspr(SPRN_GDEAR);
 278 #else
 279         return vcpu->arch.shared->dar;
 280 #endif
 281 }
 282
 283 static void set_guest_dear(struct kvm_vcpu *vcpu, unsigned long dear)
 284 {
 285 #ifdef CONFIG_KVM_BOOKE_HV
 286         mtspr(SPRN_GDEAR, dear);
 287 #else
 288         vcpu->arch.shared->dar = dear;
 289 #endif
 290 }
 291
 292 static unsigned long get_guest_esr(struct kvm_vcpu *vcpu)
 293 {
 294 #ifdef CONFIG_KVM_BOOKE_HV
 295         return mfspr(SPRN_GESR);
 296 #else
 297         return vcpu->arch.shared->esr;
 298 #endif
 299 }
 300
 301 static void set_guest_esr(struct kvm_vcpu *vcpu, u32 esr)
 302 {
 303 #ifdef CONFIG_KVM_BOOKE_HV
 304         mtspr(SPRN_GESR, esr);
 305 #else
 306         vcpu->arch.shared->esr = esr;
 307 #endif
 308 }
 309
 310 static unsigned long get_guest_epr(struct kvm_vcpu *vcpu)
 311 {
 312 #ifdef CONFIG_KVM_BOOKE_HV
 313         return mfspr(SPRN_GEPR);
 314 #else
 315         return vcpu->arch.epr;
 316 #endif
 317 }
 318
 319 /* Deliver the interrupt of the corresponding priority, if possible. */
 320 static int kvmppc_booke_irqprio_deliver(struct kvm_vcpu *vcpu,
 321                                         unsigned int priority)
 322 {
 323         int allowed = 0;
 324         ulong msr_mask = 0;
 325         bool update_esr = false, update_dear = false, update_epr = false;
 326         ulong crit_raw = vcpu->arch.shared->critical;
 327         ulong crit_r1 = kvmppc_get_gpr(vcpu, 1);
 328         bool crit;
 329         bool keep_irq = false;
 330         enum int_class int_class;
 331         ulong new_msr = vcpu->arch.shared->msr;
 332
 333         /* Truncate crit indicators in 32 bit mode */
 334         if (!(vcpu->arch.shared->msr & MSR_SF)) {
 335                 crit_raw &= 0xffffffff;
 336                 crit_r1 &= 0xffffffff;
 337         }
 338
 339         /* Critical section when crit == r1 */
 340         crit = (crit_raw == crit_r1);
 341         /* ... and we're in supervisor mode */
 342         crit = crit && !(vcpu->arch.shared->msr & MSR_PR);
 343
 344         if (priority == BOOKE_IRQPRIO_EXTERNAL_LEVEL) {
 345                 priority = BOOKE_IRQPRIO_EXTERNAL;
 346                 keep_irq = true;
 347         }
 348
 349         if ((priority == BOOKE_IRQPRIO_EXTERNAL) && vcpu->arch.epr_flags)
 350                 update_epr = true;
 351
 352         switch (priority) {
 353         case BOOKE_IRQPRIO_DTLB_MISS:
 354         case BOOKE_IRQPRIO_DATA_STORAGE:
 355         case BOOKE_IRQPRIO_ALIGNMENT:
 356                 update_dear = true;
 357                 /* fall through */
 358         case BOOKE_IRQPRIO_INST_STORAGE:
 359         case BOOKE_IRQPRIO_PROGRAM:
 360                 update_esr = true;
 361                 /* fall through */
 362         case BOOKE_IRQPRIO_ITLB_MISS:
 363         case BOOKE_IRQPRIO_SYSCALL:
 364         case BOOKE_IRQPRIO_FP_UNAVAIL:
 365         case BOOKE_IRQPRIO_SPE_UNAVAIL:
 366         case BOOKE_IRQPRIO_SPE_FP_DATA:
 367         case BOOKE_IRQPRIO_SPE_FP_ROUND:
 368         case BOOKE_IRQPRIO_AP_UNAVAIL:
 369                 allowed = 1;
 370                 msr_mask = MSR_CE | MSR_ME | MSR_DE;
 371                 int_class = INT_CLASS_NONCRIT;
 372                 break;
 373         case BOOKE_IRQPRIO_WATCHDOG:
 374         case BOOKE_IRQPRIO_CRITICAL:
 375         case BOOKE_IRQPRIO_DBELL_CRIT:
 376                 allowed = vcpu->arch.shared->msr & MSR_CE;
 377                 allowed = allowed && !crit;
 378                 msr_mask = MSR_ME;
 379                 int_class = INT_CLASS_CRIT;
 380                 break;
 381         case BOOKE_IRQPRIO_MACHINE_CHECK:
 382                 allowed = vcpu->arch.shared->msr & MSR_ME;
 383                 allowed = allowed && !crit;
 384                 int_class = INT_CLASS_MC;
 385                 break;
 386         case BOOKE_IRQPRIO_DECREMENTER:
 387         case BOOKE_IRQPRIO_FIT:
 388                 keep_irq = true;
 389                 /* fall through */
 390         case BOOKE_IRQPRIO_EXTERNAL:
 391         case BOOKE_IRQPRIO_DBELL:
 392                 allowed = vcpu->arch.shared->msr & MSR_EE;
 393                 allowed = allowed && !crit;
 394                 msr_mask = MSR_CE | MSR_ME | MSR_DE;
 395                 int_class = INT_CLASS_NONCRIT;
 396                 break;
 397         case BOOKE_IRQPRIO_DEBUG:
 398                 allowed = vcpu->arch.shared->msr & MSR_DE;
 399                 allowed = allowed && !crit;
 400                 msr_mask = MSR_ME;
 401                 int_class = INT_CLASS_CRIT;
 402                 break;
 403         }
 404
 405         if (allowed) {
 406                 switch (int_class) {
 407                 case INT_CLASS_NONCRIT:
 408                         set_guest_srr(vcpu, vcpu->arch.pc,
 409                                       vcpu->arch.shared->msr);
 410                         break;
 411                 case INT_CLASS_CRIT:
 412                         set_guest_csrr(vcpu, vcpu->arch.pc,
 413                                        vcpu->arch.shared->msr);
 414                         break;
 415                 case INT_CLASS_DBG:
 416                         set_guest_dsrr(vcpu, vcpu->arch.pc,
 417                                        vcpu->arch.shared->msr);
 418                         break;
 419                 case INT_CLASS_MC:
 420                         set_guest_mcsrr(vcpu, vcpu->arch.pc,
 421                                         vcpu->arch.shared->msr);
 422                         break;
 423                 }
 424
 425                 vcpu->arch.pc = vcpu->arch.ivpr | vcpu->arch.ivor[priority];
 426                 if (update_esr == true)
 427                         set_guest_esr(vcpu, vcpu->arch.queued_esr);
 428                 if (update_dear == true)
 429                         set_guest_dear(vcpu, vcpu->arch.queued_dear);
 430                 if (update_epr == true) {
 431                         if (vcpu->arch.epr_flags & KVMPPC_EPR_USER)
 432                                 kvm_make_request(KVM_REQ_EPR_EXIT, vcpu);
 433                         else if (vcpu->arch.epr_flags & KVMPPC_EPR_KERNEL) {
 434                                 BUG_ON(vcpu->arch.irq_type != KVMPPC_IRQ_MPIC);
 435                                 kvmppc_mpic_set_epr(vcpu);
 436                         }
 437                 }
 438
 439                 new_msr &= msr_mask;
 440 #if defined(CONFIG_64BIT)
 441                 if (vcpu->arch.epcr & SPRN_EPCR_ICM)
 442                         new_msr |= MSR_CM;
 443 #endif
 444                 kvmppc_set_msr(vcpu, new_msr);
 445
 446                 if (!keep_irq)
 447                         clear_bit(priority, &vcpu->arch.pending_exceptions);
 448         }
 449
 450 #ifdef CONFIG_KVM_BOOKE_HV
 451         /*
 452          * If an interrupt is pending but masked, raise a guest doorbell
 453          * so that we are notified when the guest enables the relevant
 454          * MSR bit.
 455          */
 456         if (vcpu->arch.pending_exceptions & BOOKE_IRQMASK_EE)
 457                 kvmppc_set_pending_interrupt(vcpu, INT_CLASS_NONCRIT);
 458         if (vcpu->arch.pending_exceptions & BOOKE_IRQMASK_CE)
 459                 kvmppc_set_pending_interrupt(vcpu, INT_CLASS_CRIT);
 460         if (vcpu->arch.pending_exceptions & BOOKE_IRQPRIO_MACHINE_CHECK)
 461                 kvmppc_set_pending_interrupt(vcpu, INT_CLASS_MC);
 462 #endif
 463
 464         return allowed;
 465 }
 466
 467 /*
 468  * Return the number of jiffies until the next timeout.  If the timeout is
 469  * longer than the NEXT_TIMER_MAX_DELTA, then return NEXT_TIMER_MAX_DELTA
 470  * because the larger value can break the timer APIs.
 471  */
 472 static unsigned long watchdog_next_timeout(struct kvm_vcpu *vcpu)
 473 {
 474         u64 tb, wdt_tb, wdt_ticks = 0;
 475         u64 nr_jiffies = 0;
 476         u32 period = TCR_GET_WP(vcpu->arch.tcr);
 477
 478         wdt_tb = 1ULL << (63 - period);
 479         tb = get_tb();
 480         /*
 481          * The watchdog timeout will hapeen when TB bit corresponding
 482          * to watchdog will toggle from 0 to 1.
 483          */
 484         if (tb & wdt_tb)
 485                 wdt_ticks = wdt_tb;
 486
 487         wdt_ticks += wdt_tb - (tb & (wdt_tb - 1));
 488
 489         /* Convert timebase ticks to jiffies */
 490         nr_jiffies = wdt_ticks;
 491
 492         if (do_div(nr_jiffies, tb_ticks_per_jiffy))
 493                 nr_jiffies++;
 494
 495         return min_t(unsigned long long, nr_jiffies, NEXT_TIMER_MAX_DELTA);
 496 }
 497
 498 static void arm_next_watchdog(struct kvm_vcpu *vcpu)
 499 {
 500         unsigned long nr_jiffies;
 501         unsigned long flags;
 502
 503         /*
 504          * If TSR_ENW and TSR_WIS are not set then no need to exit to
 505          * userspace, so clear the KVM_REQ_WATCHDOG request.
 506          */
 507         if ((vcpu->arch.tsr & (TSR_ENW | TSR_WIS)) != (TSR_ENW | TSR_WIS))
 508                 clear_bit(KVM_REQ_WATCHDOG, &vcpu->requests);
 509
 510         spin_lock_irqsave(&vcpu->arch.wdt_lock, flags);
 511         nr_jiffies = watchdog_next_timeout(vcpu);
 512         /*
 513          * If the number of jiffies of watchdog timer >= NEXT_TIMER_MAX_DELTA
 514          * then do not run the watchdog timer as this can break timer APIs.
 515          */
 516         if (nr_jiffies < NEXT_TIMER_MAX_DELTA)
 517                 mod_timer(&vcpu->arch.wdt_timer, jiffies + nr_jiffies);
 518         else
 519                 del_timer(&vcpu->arch.wdt_timer);
 520         spin_unlock_irqrestore(&vcpu->arch.wdt_lock, flags);
 521 }
 522
 523 void kvmppc_watchdog_func(unsigned long data)
 524 {
 525         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)data;
 526         u32 tsr, new_tsr;
 527         int final;
 528
 529         do {
 530                 new_tsr = tsr = vcpu->arch.tsr;
 531                 final = 0;
 532
 533                 /* Time out event */
 534                 if (tsr & TSR_ENW) {
 535                         if (tsr & TSR_WIS)
 536                                 final = 1;
 537                         else
 538                                 new_tsr = tsr | TSR_WIS;
 539                 } else {
 540                         new_tsr = tsr | TSR_ENW;
 541                 }
 542         } while (cmpxchg(&vcpu->arch.tsr, tsr, new_tsr) != tsr);
 543
 544         if (new_tsr & TSR_WIS) {
 545                 smp_wmb();
 546                 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
 547                 kvm_vcpu_kick(vcpu);
 548         }
 549
 550         /*
 551          * If this is final watchdog expiry and some action is required
 552          * then exit to userspace.
 553          */
 554         if (final && (vcpu->arch.tcr & TCR_WRC_MASK) &&
 555             vcpu->arch.watchdog_enabled) {
 556                 smp_wmb();
 557                 kvm_make_request(KVM_REQ_WATCHDOG, vcpu);
 558                 kvm_vcpu_kick(vcpu);
 559         }
 560
 561         /*
 562          * Stop running the watchdog timer after final expiration to
 563          * prevent the host from being flooded with timers if the
 564          * guest sets a short period.
 565          * Timers will resume when TSR/TCR is updated next time.
 566          */
 567         if (!final)
 568                 arm_next_watchdog(vcpu);
 569 }
 570
 571 static void update_timer_ints(struct kvm_vcpu *vcpu)
 572 {
 573         if ((vcpu->arch.tcr & TCR_DIE) && (vcpu->arch.tsr & TSR_DIS))
 574                 kvmppc_core_queue_dec(vcpu);
 575         else
 576                 kvmppc_core_dequeue_dec(vcpu);
 577
 578         if ((vcpu->arch.tcr & TCR_WIE) && (vcpu->arch.tsr & TSR_WIS))
 579                 kvmppc_core_queue_watchdog(vcpu);
 580         else
 581                 kvmppc_core_dequeue_watchdog(vcpu);
 582 }
 583
 584 static void kvmppc_core_check_exceptions(struct kvm_vcpu *vcpu)
 585 {
 586         unsigned long *pending = &vcpu->arch.pending_exceptions;
 587         unsigned int priority;
 588
 589         priority = __ffs(*pending);
 590         while (priority < BOOKE_IRQPRIO_MAX) {
 591                 if (kvmppc_booke_irqprio_deliver(vcpu, priority))
 592                         break;
 593
 594                 priority = find_next_bit(pending,
 595                                          BITS_PER_BYTE * sizeof(*pending),
 596                                          priority + 1);
 597         }
 598
 599         /* Tell the guest about our interrupt status */
 600         vcpu->arch.shared->int_pending = !!*pending;
 601 }
 602
 603 /* Check pending exceptions and deliver one, if possible. */
 604 int kvmppc_core_prepare_to_enter(struct kvm_vcpu *vcpu)
 605 {
 606         int r = 0;
 607         WARN_ON_ONCE(!irqs_disabled());
 608
 609         kvmppc_core_check_exceptions(vcpu);
 610
 611         if (vcpu->requests) {
 612                 /* Exception delivery raised request; start over */
 613                 return 1;
 614         }
 615
 616         if (vcpu->arch.shared->msr & MSR_WE) {
 617                 local_irq_enable();
 618                 kvm_vcpu_block(vcpu);
 619                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 620                 local_irq_disable();
 621
 622                 kvmppc_set_exit_type(vcpu, EMULATED_MTMSRWE_EXITS);
 623                 r = 1;
 624         };
 625
 626         return r;
 627 }
 628
 629 int kvmppc_core_check_requests(struct kvm_vcpu *vcpu)
 630 {
 631         int r = 1; /* Indicate we want to get back into the guest */
 632
 633         if (kvm_check_request(KVM_REQ_PENDING_TIMER, vcpu))
 634                 update_timer_ints(vcpu);
 635 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 636         if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
 637                 kvmppc_core_flush_tlb(vcpu);
 638 #endif
 639
 640         if (kvm_check_request(KVM_REQ_WATCHDOG, vcpu)) {
 641                 vcpu->run->exit_reason = KVM_EXIT_WATCHDOG;
 642                 r = 0;
 643         }
 644
 645         if (kvm_check_request(KVM_REQ_EPR_EXIT, vcpu)) {
 646                 vcpu->run->epr.epr = 0;
 647                 vcpu->arch.epr_needed = true;
 648                 vcpu->run->exit_reason = KVM_EXIT_EPR;
 649                 r = 0;
 650         }
 651
 652         return r;
 653 }
 654
 655 int kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
 656 {
 657         int ret, s;
 658 #ifdef CONFIG_PPC_FPU
 659         struct thread_fp_state fp;
 660         int fpexc_mode;
 661 #endif
 662
 663         if (!vcpu->arch.sane) {
 664                 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
 665                 return -EINVAL;
 666         }
 667
 668         local_irq_disable();
 669         s = kvmppc_prepare_to_enter(vcpu);
 670         if (s <= 0) {
 671                 local_irq_enable();
 672                 ret = s;
 673                 goto out;
 674         }
 675
 676 #ifdef CONFIG_PPC_FPU
 677         /* Save userspace FPU state in stack */
 678         enable_kernel_fp();
 679         fp = current->thread.fp_state;
 680         fpexc_mode = current->thread.fpexc_mode;
 681
 682         /* Restore guest FPU state to thread */
 683         memcpy(current->thread.fp_state.fpr, vcpu->arch.fpr,
 684                sizeof(vcpu->arch.fpr));
 685         current->thread.fp_state.fpscr = vcpu->arch.fpscr;
 686
 687         /*
 688          * Since we can't trap on MSR_FP in GS-mode, we consider the guest
 689          * as always using the FPU.  Kernel usage of FP (via
 690          * enable_kernel_fp()) in this thread must not occur while
 691          * vcpu->fpu_active is set.
 692          */
 693         vcpu->fpu_active = 1;
 694
 695         kvmppc_load_guest_fp(vcpu);
 696 #endif
 697
 698         kvmppc_fix_ee_before_entry();
 699
 700         ret = __kvmppc_vcpu_run(kvm_run, vcpu);
 701
 702         /* No need for kvm_guest_exit. It's done in handle_exit.
 703            We also get here with interrupts enabled. */
 704
 705 #ifdef CONFIG_PPC_FPU
 706         kvmppc_save_guest_fp(vcpu);
 707
 708         vcpu->fpu_active = 0;
 709
 710         /* Save guest FPU state from thread */
 711         memcpy(vcpu->arch.fpr, current->thread.fp_state.fpr,
 712                sizeof(vcpu->arch.fpr));
 713         vcpu->arch.fpscr = current->thread.fp_state.fpscr;
 714
 715         /* Restore userspace FPU state from stack */
 716         current->thread.fp_state = fp;
 717         current->thread.fpexc_mode = fpexc_mode;
 718 #endif
 719
 720 out:
 721         vcpu->mode = OUTSIDE_GUEST_MODE;
 722         return ret;
 723 }
 724
 725 static int emulation_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
 726 {
 727         enum emulation_result er;
 728
 729         er = kvmppc_emulate_instruction(run, vcpu);
 730         switch (er) {
 731         case EMULATE_DONE:
 732                 /* don't overwrite subtypes, just account kvm_stats */
 733                 kvmppc_account_exit_stat(vcpu, EMULATED_INST_EXITS);
 734                 /* Future optimization: only reload non-volatiles if
 735                  * they were actually modified by emulation. */
 736                 return RESUME_GUEST_NV;
 737
 738         case EMULATE_DO_DCR:
 739                 run->exit_reason = KVM_EXIT_DCR;
 740                 return RESUME_HOST;
 741
 742         case EMULATE_FAIL:
 743                 printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 744                        __func__, vcpu->arch.pc, vcpu->arch.last_inst);
 745                 /* For debugging, encode the failing instruction and
 746                  * report it to userspace. */
 747                 run->hw.hardware_exit_reason = ~0ULL << 32;
 748                 run->hw.hardware_exit_reason |= vcpu->arch.last_inst;
 749                 kvmppc_core_queue_program(vcpu, ESR_PIL);
 750                 return RESUME_HOST;
 751
 752         case EMULATE_EXIT_USER:
 753                 return RESUME_HOST;
 754
 755         default:
 756                 BUG();
 757         }
 758 }
 759
 760 static void kvmppc_fill_pt_regs(struct pt_regs *regs)
 761 {
 762         ulong r1, ip, msr, lr;
 763
 764         asm("mr %0, 1" : "=r"(r1));
 765         asm("mflr %0" : "=r"(lr));
 766         asm("mfmsr %0" : "=r"(msr));
 767         asm("bl 1f; 1: mflr %0" : "=r"(ip));
 768
 769         memset(regs, 0, sizeof(*regs));
 770         regs->gpr[1] = r1;
 771         regs->nip = ip;
 772         regs->msr = msr;
 773         regs->link = lr;
 774 }
 775
 776 /*
 777  * For interrupts needed to be handled by host interrupt handlers,
 778  * corresponding host handler are called from here in similar way
 779  * (but not exact) as they are called from low level handler
 780  * (such as from arch/powerpc/kernel/head_fsl_booke.S).
 781  */
 782 static void kvmppc_restart_interrupt(struct kvm_vcpu *vcpu,
 783                                      unsigned int exit_nr)
 784 {
 785         struct pt_regs regs;
 786
 787         switch (exit_nr) {
 788         case BOOKE_INTERRUPT_EXTERNAL:
 789                 kvmppc_fill_pt_regs(&regs);
 790                 do_IRQ(&regs);
 791                 break;
 792         case BOOKE_INTERRUPT_DECREMENTER:
 793                 kvmppc_fill_pt_regs(&regs);
 794                 timer_interrupt(&regs);
 795                 break;
 796 #if defined(CONFIG_PPC_DOORBELL)
 797         case BOOKE_INTERRUPT_DOORBELL:
 798                 kvmppc_fill_pt_regs(&regs);
 799                 doorbell_exception(&regs);
 800                 break;
 801 #endif
 802         case BOOKE_INTERRUPT_MACHINE_CHECK:
 803                 /* FIXME */
 804                 break;
 805         case BOOKE_INTERRUPT_PERFORMANCE_MONITOR:
 806                 kvmppc_fill_pt_regs(&regs);
 807                 performance_monitor_exception(&regs);
 808                 break;
 809         case BOOKE_INTERRUPT_WATCHDOG:
 810                 kvmppc_fill_pt_regs(&regs);
 811 #ifdef CONFIG_BOOKE_WDT
 812                 WatchdogException(&regs);
 813 #else
 814                 unknown_exception(&regs);
 815 #endif
 816                 break;
 817         case BOOKE_INTERRUPT_CRITICAL:
 818                 unknown_exception(&regs);
 819                 break;
 820         }
 821 }
 822
 823 /**
 824  * kvmppc_handle_exit
 825  *
 826  * Return value is in the form (errcode<<2 | RESUME_FLAG_HOST | RESUME_FLAG_NV)
 827  */
 828 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 829                        unsigned int exit_nr)
 830 {
 831         int r = RESUME_HOST;
 832         int s;
 833         int idx;
 834
 835 #ifdef CONFIG_PPC64
 836         WARN_ON(local_paca->irq_happened != 0);
 837 #endif
 838
 839         /*
 840          * We enter with interrupts disabled in hardware, but
 841          * we need to call hard_irq_disable anyway to ensure that
 842          * the software state is kept in sync.
 843          */
 844         hard_irq_disable();
 845
 846         /* update before a new last_exit_type is rewritten */
 847         kvmppc_update_timing_stats(vcpu);
 848
 849         /* restart interrupts if they were meant for the host */
 850         kvmppc_restart_interrupt(vcpu, exit_nr);
 851
 852         local_irq_enable();
 853
 854         trace_kvm_exit(exit_nr, vcpu);
 855         kvm_guest_exit();
 856
 857         run->exit_reason = KVM_EXIT_UNKNOWN;
 858         run->ready_for_interrupt_injection = 1;
 859
 860         switch (exit_nr) {
 861         case BOOKE_INTERRUPT_MACHINE_CHECK:
 862                 printk("MACHINE CHECK: %lx\n", mfspr(SPRN_MCSR));
 863                 kvmppc_dump_vcpu(vcpu);
 864                 /* For debugging, send invalid exit reason to user space */
 865                 run->hw.hardware_exit_reason = ~1ULL << 32;
 866                 run->hw.hardware_exit_reason |= mfspr(SPRN_MCSR);
 867                 r = RESUME_HOST;
 868                 break;
 869
 870         case BOOKE_INTERRUPT_EXTERNAL:
 871                 kvmppc_account_exit(vcpu, EXT_INTR_EXITS);
 872                 r = RESUME_GUEST;
 873                 break;
 874
 875         case BOOKE_INTERRUPT_DECREMENTER:
 876                 kvmppc_account_exit(vcpu, DEC_EXITS);
 877                 r = RESUME_GUEST;
 878                 break;
 879
 880         case BOOKE_INTERRUPT_WATCHDOG:
 881                 r = RESUME_GUEST;
 882                 break;
 883
 884         case BOOKE_INTERRUPT_DOORBELL:
 885                 kvmppc_account_exit(vcpu, DBELL_EXITS);
 886                 r = RESUME_GUEST;
 887                 break;
 888
 889         case BOOKE_INTERRUPT_GUEST_DBELL_CRIT:
 890                 kvmppc_account_exit(vcpu, GDBELL_EXITS);
 891
 892                 /*
 893                  * We are here because there is a pending guest interrupt
 894                  * which could not be delivered as MSR_CE or MSR_ME was not
 895                  * set.  Once we break from here we will retry delivery.
 896                  */
 897                 r = RESUME_GUEST;
 898                 break;
 899
 900         case BOOKE_INTERRUPT_GUEST_DBELL:
 901                 kvmppc_account_exit(vcpu, GDBELL_EXITS);
 902
 903                 /*
 904                  * We are here because there is a pending guest interrupt
 905                  * which could not be delivered as MSR_EE was not set.  Once
 906                  * we break from here we will retry delivery.
 907                  */
 908                 r = RESUME_GUEST;
 909                 break;
 910
 911         case BOOKE_INTERRUPT_PERFORMANCE_MONITOR:
 912                 r = RESUME_GUEST;
 913                 break;
 914
 915         case BOOKE_INTERRUPT_HV_PRIV:
 916                 r = emulation_exit(run, vcpu);
 917                 break;
 918
 919         case BOOKE_INTERRUPT_PROGRAM:
 920                 if (vcpu->arch.shared->msr & (MSR_PR | MSR_GS)) {
 921                         /*
 922                          * Program traps generated by user-level software must
 923                          * be handled by the guest kernel.
 924                          *
 925                          * In GS mode, hypervisor privileged instructions trap
 926                          * on BOOKE_INTERRUPT_HV_PRIV, not here, so these are
 927                          * actual program interrupts, handled by the guest.
 928                          */
 929                         kvmppc_core_queue_program(vcpu, vcpu->arch.fault_esr);
 930                         r = RESUME_GUEST;
 931                         kvmppc_account_exit(vcpu, USR_PR_INST);
 932                         break;
 933                 }
 934
 935                 r = emulation_exit(run, vcpu);
 936                 break;
 937
 938         case BOOKE_INTERRUPT_FP_UNAVAIL:
 939                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_FP_UNAVAIL);
 940                 kvmppc_account_exit(vcpu, FP_UNAVAIL);
 941                 r = RESUME_GUEST;
 942                 break;
 943
 944 #ifdef CONFIG_SPE
 945         case BOOKE_INTERRUPT_SPE_UNAVAIL: {
 946                 if (vcpu->arch.shared->msr & MSR_SPE)
 947                         kvmppc_vcpu_enable_spe(vcpu);
 948                 else
 949                         kvmppc_booke_queue_irqprio(vcpu,
 950                                                    BOOKE_IRQPRIO_SPE_UNAVAIL);
 951                 r = RESUME_GUEST;
 952                 break;
 953         }
 954
 955         case BOOKE_INTERRUPT_SPE_FP_DATA:
 956                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SPE_FP_DATA);
 957                 r = RESUME_GUEST;
 958                 break;
 959
 960         case BOOKE_INTERRUPT_SPE_FP_ROUND:
 961                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SPE_FP_ROUND);
 962                 r = RESUME_GUEST;
 963                 break;
 964 #else
 965         case BOOKE_INTERRUPT_SPE_UNAVAIL:
 966                 /*
 967                  * Guest wants SPE, but host kernel doesn't support it.  Send
 968                  * an "unimplemented operation" program check to the guest.
 969                  */
 970                 kvmppc_core_queue_program(vcpu, ESR_PUO | ESR_SPV);
 971                 r = RESUME_GUEST;
 972                 break;
 973
 974         /*
 975          * These really should never happen without CONFIG_SPE,
 976          * as we should never enable the real MSR[SPE] in the guest.
 977          */
 978         case BOOKE_INTERRUPT_SPE_FP_DATA:
 979         case BOOKE_INTERRUPT_SPE_FP_ROUND:
 980                 printk(KERN_CRIT "%s: unexpected SPE interrupt %u at %08lx\n",
 981                        __func__, exit_nr, vcpu->arch.pc);
 982                 run->hw.hardware_exit_reason = exit_nr;
 983                 r = RESUME_HOST;
 984                 break;
 985 #endif
 986
 987         case BOOKE_INTERRUPT_DATA_STORAGE:
 988                 kvmppc_core_queue_data_storage(vcpu, vcpu->arch.fault_dear,
 989                                                vcpu->arch.fault_esr);
 990                 kvmppc_account_exit(vcpu, DSI_EXITS);
 991                 r = RESUME_GUEST;
 992                 break;
 993
 994         case BOOKE_INTERRUPT_INST_STORAGE:
 995                 kvmppc_core_queue_inst_storage(vcpu, vcpu->arch.fault_esr);
 996                 kvmppc_account_exit(vcpu, ISI_EXITS);
 997                 r = RESUME_GUEST;
 998                 break;
 999
1000         case BOOKE_INTERRUPT_ALIGNMENT:
1001                 kvmppc_core_queue_alignment(vcpu, vcpu->arch.fault_dear,
1002                                             vcpu->arch.fault_esr);
1003                 r = RESUME_GUEST;
1004                 break;
1005
1006 #ifdef CONFIG_KVM_BOOKE_HV
1007         case BOOKE_INTERRUPT_HV_SYSCALL:
1008                 if (!(vcpu->arch.shared->msr & MSR_PR)) {
1009                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
1010                 } else {
1011                         /*
1012                          * hcall from guest userspace -- send privileged
1013                          * instruction program check.
1014                          */
1015                         kvmppc_core_queue_program(vcpu, ESR_PPR);
1016                 }
1017
1018                 r = RESUME_GUEST;
1019                 break;
1020 #else
1021         case BOOKE_INTERRUPT_SYSCALL:
1022                 if (!(vcpu->arch.shared->msr & MSR_PR) &&
1023                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
1024                         /* KVM PV hypercalls */
1025                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
1026                         r = RESUME_GUEST;
1027                 } else {
1028                         /* Guest syscalls */
1029                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SYSCALL);
1030                 }
1031                 kvmppc_account_exit(vcpu, SYSCALL_EXITS);
1032                 r = RESUME_GUEST;
1033                 break;
1034 #endif
1035
1036         case BOOKE_INTERRUPT_DTLB_MISS: {
1037                 unsigned long eaddr = vcpu->arch.fault_dear;
1038                 int gtlb_index;
1039                 gpa_t gpaddr;
1040                 gfn_t gfn;
1041
1042 #ifdef CONFIG_KVM_E500V2
1043                 if (!(vcpu->arch.shared->msr & MSR_PR) &&
1044                     (eaddr & PAGE_MASK) == vcpu->arch.magic_page_ea) {
1045                         kvmppc_map_magic(vcpu);
1046                         kvmppc_account_exit(vcpu, DTLB_VIRT_MISS_EXITS);
1047                         r = RESUME_GUEST;
1048
1049                         break;
1050                 }
1051 #endif
1052
1053                 /* Check the guest TLB. */
1054                 gtlb_index = kvmppc_mmu_dtlb_index(vcpu, eaddr);
1055                 if (gtlb_index < 0) {
1056                         /* The guest didn't have a mapping for it. */
1057                         kvmppc_core_queue_dtlb_miss(vcpu,
1058                                                     vcpu->arch.fault_dear,
1059                                                     vcpu->arch.fault_esr);
1060                         kvmppc_mmu_dtlb_miss(vcpu);
1061                         kvmppc_account_exit(vcpu, DTLB_REAL_MISS_EXITS);
1062                         r = RESUME_GUEST;
1063                         break;
1064                 }
1065
1066                 idx = srcu_read_lock(&vcpu->kvm->srcu);
1067
1068                 gpaddr = kvmppc_mmu_xlate(vcpu, gtlb_index, eaddr);
1069                 gfn = gpaddr >> PAGE_SHIFT;
1070
1071                 if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
1072                         /* The guest TLB had a mapping, but the shadow TLB
1073                          * didn't, and it is RAM. This could be because:
1074                          * a) the entry is mapping the host kernel, or
1075                          * b) the guest used a large mapping which we're faking
1076                          * Either way, we need to satisfy the fault without
1077                          * invoking the guest. */
1078                         kvmppc_mmu_map(vcpu, eaddr, gpaddr, gtlb_index);
1079                         kvmppc_account_exit(vcpu, DTLB_VIRT_MISS_EXITS);
1080                         r = RESUME_GUEST;
1081                 } else {
1082                         /* Guest has mapped and accessed a page which is not
1083                          * actually RAM. */
1084                         vcpu->arch.paddr_accessed = gpaddr;
1085                         vcpu->arch.vaddr_accessed = eaddr;
1086                         r = kvmppc_emulate_mmio(run, vcpu);
1087                         kvmppc_account_exit(vcpu, MMIO_EXITS);
1088                 }
1089
1090                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1091                 break;
1092         }
1093
1094         case BOOKE_INTERRUPT_ITLB_MISS: {
1095                 unsigned long eaddr = vcpu->arch.pc;
1096                 gpa_t gpaddr;
1097                 gfn_t gfn;
1098                 int gtlb_index;
1099
1100                 r = RESUME_GUEST;
1101
1102                 /* Check the guest TLB. */
1103                 gtlb_index = kvmppc_mmu_itlb_index(vcpu, eaddr);
1104                 if (gtlb_index < 0) {
1105                         /* The guest didn't have a mapping for it. */
1106                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_ITLB_MISS);
1107                         kvmppc_mmu_itlb_miss(vcpu);
1108                         kvmppc_account_exit(vcpu, ITLB_REAL_MISS_EXITS);
1109                         break;
1110                 }
1111
1112                 kvmppc_account_exit(vcpu, ITLB_VIRT_MISS_EXITS);
1113
1114                 idx = srcu_read_lock(&vcpu->kvm->srcu);
1115
1116                 gpaddr = kvmppc_mmu_xlate(vcpu, gtlb_index, eaddr);
1117                 gfn = gpaddr >> PAGE_SHIFT;
1118
1119                 if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
1120                         /* The guest TLB had a mapping, but the shadow TLB
1121                          * didn't. This could be because:
1122                          * a) the entry is mapping the host kernel, or
1123                          * b) the guest used a large mapping which we're faking
1124                          * Either way, we need to satisfy the fault without
1125                          * invoking the guest. */
1126                         kvmppc_mmu_map(vcpu, eaddr, gpaddr, gtlb_index);
1127                 } else {
1128                         /* Guest mapped and leaped at non-RAM! */
1129                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_MACHINE_CHECK);
1130                 }
1131
1132                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1133                 break;
1134         }
1135
1136         case BOOKE_INTERRUPT_DEBUG: {
1137                 u32 dbsr;
1138
1139                 vcpu->arch.pc = mfspr(SPRN_CSRR0);
1140
1141                 /* clear IAC events in DBSR register */
1142                 dbsr = mfspr(SPRN_DBSR);
1143                 dbsr &= DBSR_IAC1 | DBSR_IAC2 | DBSR_IAC3 | DBSR_IAC4;
1144                 mtspr(SPRN_DBSR, dbsr);
1145
1146                 run->exit_reason = KVM_EXIT_DEBUG;
1147                 kvmppc_account_exit(vcpu, DEBUG_EXITS);
1148                 r = RESUME_HOST;
1149                 break;
1150         }
1151
1152         default:
1153                 printk(KERN_EMERG "exit_nr %d\n", exit_nr);
1154                 BUG();
1155         }
1156
1157         /*
1158          * To avoid clobbering exit_reason, only check for signals if we
1159          * aren't already exiting to userspace for some other reason.
1160          */
1161         if (!(r & RESUME_HOST)) {
1162                 local_irq_disable();
1163                 s = kvmppc_prepare_to_enter(vcpu);
1164                 if (s <= 0) {
1165                         local_irq_enable();
1166                         r = (s << 2) | RESUME_HOST | (r & RESUME_FLAG_NV);
1167                 } else {
1168                         kvmppc_fix_ee_before_entry();
1169                 }
1170         }
1171
1172         return r;
1173 }
1174
1175 static void kvmppc_set_tsr(struct kvm_vcpu *vcpu, u32 new_tsr)
1176 {
1177         u32 old_tsr = vcpu->arch.tsr;
1178
1179         vcpu->arch.tsr = new_tsr;
1180
1181         if ((old_tsr ^ vcpu->arch.tsr) & (TSR_ENW | TSR_WIS))
1182                 arm_next_watchdog(vcpu);
1183
1184         update_timer_ints(vcpu);
1185 }
1186
1187 /* Initial guest state: 16MB mapping 0 -> 0, PC = 0, MSR = 0, R1 = 16MB */
1188 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1189 {
1190         int i;
1191         int r;
1192
1193         vcpu->arch.pc = 0;
1194         vcpu->arch.shared->pir = vcpu->vcpu_id;
1195         kvmppc_set_gpr(vcpu, 1, (16<<20) - 8); /* -8 for the callee-save LR slot */
1196         kvmppc_set_msr(vcpu, 0);
1197
1198 #ifndef CONFIG_KVM_BOOKE_HV
1199         vcpu->arch.shadow_msr = MSR_USER | MSR_DE | MSR_IS | MSR_DS;
1200         vcpu->arch.shadow_pid = 1;
1201         vcpu->arch.shared->msr = 0;
1202 #endif
1203
1204         /* Eye-catching numbers so we know if the guest takes an interrupt
1205          * before it's programmed its own IVPR/IVORs. */
1206         vcpu->arch.ivpr = 0x55550000;
1207         for (i = 0; i < BOOKE_IRQPRIO_MAX; i++)
1208                 vcpu->arch.ivor[i] = 0x7700 | i * 4;
1209
1210         kvmppc_init_timing_stats(vcpu);
1211
1212         r = kvmppc_core_vcpu_setup(vcpu);
1213         kvmppc_sanity_check(vcpu);
1214         return r;
1215 }
1216
1217 int kvmppc_subarch_vcpu_init(struct kvm_vcpu *vcpu)
1218 {
1219         /* setup watchdog timer once */
1220         spin_lock_init(&vcpu->arch.wdt_lock);
1221         setup_timer(&vcpu->arch.wdt_timer, kvmppc_watchdog_func,
1222                     (unsigned long)vcpu);
1223
1224         return 0;
1225 }
1226
1227 void kvmppc_subarch_vcpu_uninit(struct kvm_vcpu *vcpu)
1228 {
1229         del_timer_sync(&vcpu->arch.wdt_timer);
1230 }
1231
1232 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1233 {
1234         int i;
1235
1236         regs->pc = vcpu->arch.pc;
1237         regs->cr = kvmppc_get_cr(vcpu);
1238         regs->ctr = vcpu->arch.ctr;
1239         regs->lr = vcpu->arch.lr;
1240         regs->xer = kvmppc_get_xer(vcpu);
1241         regs->msr = vcpu->arch.shared->msr;
1242         regs->srr0 = vcpu->arch.shared->srr0;
1243         regs->srr1 = vcpu->arch.shared->srr1;
1244         regs->pid = vcpu->arch.pid;
1245         regs->sprg0 = vcpu->arch.shared->sprg0;
1246         regs->sprg1 = vcpu->arch.shared->sprg1;
1247         regs->sprg2 = vcpu->arch.shared->sprg2;
1248         regs->sprg3 = vcpu->arch.shared->sprg3;
1249         regs->sprg4 = vcpu->arch.shared->sprg4;
1250         regs->sprg5 = vcpu->arch.shared->sprg5;
1251         regs->sprg6 = vcpu->arch.shared->sprg6;
1252         regs->sprg7 = vcpu->arch.shared->sprg7;
1253
1254         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
1255                 regs->gpr[i] = kvmppc_get_gpr(vcpu, i);
1256
1257         return 0;
1258 }
1259
1260 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1261 {
1262         int i;
1263
1264         vcpu->arch.pc = regs->pc;
1265         kvmppc_set_cr(vcpu, regs->cr);
1266         vcpu->arch.ctr = regs->ctr;
1267         vcpu->arch.lr = regs->lr;
1268         kvmppc_set_xer(vcpu, regs->xer);
1269         kvmppc_set_msr(vcpu, regs->msr);
1270         vcpu->arch.shared->srr0 = regs->srr0;
1271         vcpu->arch.shared->srr1 = regs->srr1;
1272         kvmppc_set_pid(vcpu, regs->pid);
1273         vcpu->arch.shared->sprg0 = regs->sprg0;
1274         vcpu->arch.shared->sprg1 = regs->sprg1;
1275         vcpu->arch.shared->sprg2 = regs->sprg2;
1276         vcpu->arch.shared->sprg3 = regs->sprg3;
1277         vcpu->arch.shared->sprg4 = regs->sprg4;
1278         vcpu->arch.shared->sprg5 = regs->sprg5;
1279         vcpu->arch.shared->sprg6 = regs->sprg6;
1280         vcpu->arch.shared->sprg7 = regs->sprg7;
1281
1282         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
1283                 kvmppc_set_gpr(vcpu, i, regs->gpr[i]);
1284
1285         return 0;
1286 }
1287
1288 static void get_sregs_base(struct kvm_vcpu *vcpu,
1289                            struct kvm_sregs *sregs)
1290 {
1291         u64 tb = get_tb();
1292
1293         sregs->u.e.features |= KVM_SREGS_E_BASE;
1294
1295         sregs->u.e.csrr0 = vcpu->arch.csrr0;
1296         sregs->u.e.csrr1 = vcpu->arch.csrr1;
1297         sregs->u.e.mcsr = vcpu->arch.mcsr;
1298         sregs->u.e.esr = get_guest_esr(vcpu);
1299         sregs->u.e.dear = get_guest_dear(vcpu);
1300         sregs->u.e.tsr = vcpu->arch.tsr;
1301         sregs->u.e.tcr = vcpu->arch.tcr;
1302         sregs->u.e.dec = kvmppc_get_dec(vcpu, tb);
1303         sregs->u.e.tb = tb;
1304         sregs->u.e.vrsave = vcpu->arch.vrsave;
1305 }
1306
1307 static int set_sregs_base(struct kvm_vcpu *vcpu,
1308                           struct kvm_sregs *sregs)
1309 {
1310         if (!(sregs->u.e.features & KVM_SREGS_E_BASE))
1311                 return 0;
1312
1313         vcpu->arch.csrr0 = sregs->u.e.csrr0;
1314         vcpu->arch.csrr1 = sregs->u.e.csrr1;
1315         vcpu->arch.mcsr = sregs->u.e.mcsr;
1316         set_guest_esr(vcpu, sregs->u.e.esr);
1317         set_guest_dear(vcpu, sregs->u.e.dear);
1318         vcpu->arch.vrsave = sregs->u.e.vrsave;
1319         kvmppc_set_tcr(vcpu, sregs->u.e.tcr);
1320
1321         if (sregs->u.e.update_special & KVM_SREGS_E_UPDATE_DEC) {
1322                 vcpu->arch.dec = sregs->u.e.dec;
1323                 kvmppc_emulate_dec(vcpu);
1324         }
1325
1326         if (sregs->u.e.update_special & KVM_SREGS_E_UPDATE_TSR)
1327                 kvmppc_set_tsr(vcpu, sregs->u.e.tsr);
1328
1329         return 0;
1330 }
1331
1332 static void get_sregs_arch206(struct kvm_vcpu *vcpu,
1333                               struct kvm_sregs *sregs)
1334 {
1335         sregs->u.e.features |= KVM_SREGS_E_ARCH206;
1336
1337         sregs->u.e.pir = vcpu->vcpu_id;
1338         sregs->u.e.mcsrr0 = vcpu->arch.mcsrr0;
1339         sregs->u.e.mcsrr1 = vcpu->arch.mcsrr1;
1340         sregs->u.e.decar = vcpu->arch.decar;
1341         sregs->u.e.ivpr = vcpu->arch.ivpr;
1342 }
1343
1344 static int set_sregs_arch206(struct kvm_vcpu *vcpu,
1345                              struct kvm_sregs *sregs)
1346 {
1347         if (!(sregs->u.e.features & KVM_SREGS_E_ARCH206))
1348                 return 0;
1349
1350         if (sregs->u.e.pir != vcpu->vcpu_id)
1351                 return -EINVAL;
1352
1353         vcpu->arch.mcsrr0 = sregs->u.e.mcsrr0;
1354         vcpu->arch.mcsrr1 = sregs->u.e.mcsrr1;
1355         vcpu->arch.decar = sregs->u.e.decar;
1356         vcpu->arch.ivpr = sregs->u.e.ivpr;
1357
1358         return 0;
1359 }
1360
1361 void kvmppc_get_sregs_ivor(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
1362 {
1363         sregs->u.e.features |= KVM_SREGS_E_IVOR;
1364
1365         sregs->u.e.ivor_low[0] = vcpu->arch.ivor[BOOKE_IRQPRIO_CRITICAL];
1366         sregs->u.e.ivor_low[1] = vcpu->arch.ivor[BOOKE_IRQPRIO_MACHINE_CHECK];
1367         sregs->u.e.ivor_low[2] = vcpu->arch.ivor[BOOKE_IRQPRIO_DATA_STORAGE];
1368         sregs->u.e.ivor_low[3] = vcpu->arch.ivor[BOOKE_IRQPRIO_INST_STORAGE];
1369         sregs->u.e.ivor_low[4] = vcpu->arch.ivor[BOOKE_IRQPRIO_EXTERNAL];
1370         sregs->u.e.ivor_low[5] = vcpu->arch.ivor[BOOKE_IRQPRIO_ALIGNMENT];
1371         sregs->u.e.ivor_low[6] = vcpu->arch.ivor[BOOKE_IRQPRIO_PROGRAM];
1372         sregs->u.e.ivor_low[7] = vcpu->arch.ivor[BOOKE_IRQPRIO_FP_UNAVAIL];
1373         sregs->u.e.ivor_low[8] = vcpu->arch.ivor[BOOKE_IRQPRIO_SYSCALL];
1374         sregs->u.e.ivor_low[9] = vcpu->arch.ivor[BOOKE_IRQPRIO_AP_UNAVAIL];
1375         sregs->u.e.ivor_low[10] = vcpu->arch.ivor[BOOKE_IRQPRIO_DECREMENTER];
1376         sregs->u.e.ivor_low[11] = vcpu->arch.ivor[BOOKE_IRQPRIO_FIT];
1377         sregs->u.e.ivor_low[12] = vcpu->arch.ivor[BOOKE_IRQPRIO_WATCHDOG];
1378         sregs->u.e.ivor_low[13] = vcpu->arch.ivor[BOOKE_IRQPRIO_DTLB_MISS];
1379         sregs->u.e.ivor_low[14] = vcpu->arch.ivor[BOOKE_IRQPRIO_ITLB_MISS];
1380         sregs->u.e.ivor_low[15] = vcpu->arch.ivor[BOOKE_IRQPRIO_DEBUG];
1381 }
1382
1383 int kvmppc_set_sregs_ivor(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
1384 {
1385         if (!(sregs->u.e.features & KVM_SREGS_E_IVOR))
1386                 return 0;
1387
1388         vcpu->arch.ivor[BOOKE_IRQPRIO_CRITICAL] = sregs->u.e.ivor_low[0];
1389         vcpu->arch.ivor[BOOKE_IRQPRIO_MACHINE_CHECK] = sregs->u.e.ivor_low[1];
1390         vcpu->arch.ivor[BOOKE_IRQPRIO_DATA_STORAGE] = sregs->u.e.ivor_low[2];
1391         vcpu->arch.ivor[BOOKE_IRQPRIO_INST_STORAGE] = sregs->u.e.ivor_low[3];
1392         vcpu->arch.ivor[BOOKE_IRQPRIO_EXTERNAL] = sregs->u.e.ivor_low[4];
1393         vcpu->arch.ivor[BOOKE_IRQPRIO_ALIGNMENT] = sregs->u.e.ivor_low[5];
1394         vcpu->arch.ivor[BOOKE_IRQPRIO_PROGRAM] = sregs->u.e.ivor_low[6];
1395         vcpu->arch.ivor[BOOKE_IRQPRIO_FP_UNAVAIL] = sregs->u.e.ivor_low[7];
1396         vcpu->arch.ivor[BOOKE_IRQPRIO_SYSCALL] = sregs->u.e.ivor_low[8];
1397         vcpu->arch.ivor[BOOKE_IRQPRIO_AP_UNAVAIL] = sregs->u.e.ivor_low[9];
1398         vcpu->arch.ivor[BOOKE_IRQPRIO_DECREMENTER] = sregs->u.e.ivor_low[10];
1399         vcpu->arch.ivor[BOOKE_IRQPRIO_FIT] = sregs->u.e.ivor_low[11];
1400         vcpu->arch.ivor[BOOKE_IRQPRIO_WATCHDOG] = sregs->u.e.ivor_low[12];
1401         vcpu->arch.ivor[BOOKE_IRQPRIO_DTLB_MISS] = sregs->u.e.ivor_low[13];
1402         vcpu->arch.ivor[BOOKE_IRQPRIO_ITLB_MISS] = sregs->u.e.ivor_low[14];
1403         vcpu->arch.ivor[BOOKE_IRQPRIO_DEBUG] = sregs->u.e.ivor_low[15];
1404
1405         return 0;
1406 }
1407
1408 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1409                                   struct kvm_sregs *sregs)
1410 {
1411         sregs->pvr = vcpu->arch.pvr;
1412
1413         get_sregs_base(vcpu, sregs);
1414         get_sregs_arch206(vcpu, sregs);
1415         kvmppc_core_get_sregs(vcpu, sregs);
1416         return 0;
1417 }
1418
1419 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1420                                   struct kvm_sregs *sregs)
1421 {
1422         int ret;
1423
1424         if (vcpu->arch.pvr != sregs->pvr)
1425                 return -EINVAL;
1426
1427         ret = set_sregs_base(vcpu, sregs);
1428         if (ret < 0)
1429                 return ret;
1430
1431         ret = set_sregs_arch206(vcpu, sregs);
1432         if (ret < 0)
1433                 return ret;
1434
1435         return kvmppc_core_set_sregs(vcpu, sregs);
1436 }
1437
1438 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1439 {
1440         int r = 0;
1441         union kvmppc_one_reg val;
1442         int size;
1443         long int i;
1444
1445         size = one_reg_size(reg->id);
1446         if (size > sizeof(val))
1447                 return -EINVAL;
1448
1449         switch (reg->id) {
1450         case KVM_REG_PPC_IAC1:
1451         case KVM_REG_PPC_IAC2:
1452         case KVM_REG_PPC_IAC3:
1453         case KVM_REG_PPC_IAC4:
1454                 i = reg->id - KVM_REG_PPC_IAC1;
1455                 val = get_reg_val(reg->id, vcpu->arch.dbg_reg.iac[i]);
1456                 break;
1457         case KVM_REG_PPC_DAC1:
1458         case KVM_REG_PPC_DAC2:
1459                 i = reg->id - KVM_REG_PPC_DAC1;
1460                 val = get_reg_val(reg->id, vcpu->arch.dbg_reg.dac[i]);
1461                 break;
1462         case KVM_REG_PPC_EPR: {
1463                 u32 epr = get_guest_epr(vcpu);
1464                 val = get_reg_val(reg->id, epr);
1465                 break;
1466         }
1467 #if defined(CONFIG_64BIT)
1468         case KVM_REG_PPC_EPCR:
1469                 val = get_reg_val(reg->id, vcpu->arch.epcr);
1470                 break;
1471 #endif
1472         case KVM_REG_PPC_TCR:
1473                 val = get_reg_val(reg->id, vcpu->arch.tcr);
1474                 break;
1475         case KVM_REG_PPC_TSR:
1476                 val = get_reg_val(reg->id, vcpu->arch.tsr);
1477                 break;
1478         case KVM_REG_PPC_DEBUG_INST:
1479                 val = get_reg_val(reg->id, KVMPPC_INST_EHPRIV);
1480                 break;
1481         default:
1482                 r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1483                 break;
1484         }
1485
1486         if (r)
1487                 return r;
1488
1489         if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1490                 r = -EFAULT;
1491
1492         return r;
1493 }
1494
1495 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1496 {
1497         int r = 0;
1498         union kvmppc_one_reg val;
1499         int size;
1500         long int i;
1501
1502         size = one_reg_size(reg->id);
1503         if (size > sizeof(val))
1504                 return -EINVAL;
1505
1506         if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1507                 return -EFAULT;
1508
1509         switch (reg->id) {
1510         case KVM_REG_PPC_IAC1:
1511         case KVM_REG_PPC_IAC2:
1512         case KVM_REG_PPC_IAC3:
1513         case KVM_REG_PPC_IAC4:
1514                 i = reg->id - KVM_REG_PPC_IAC1;
1515                 vcpu->arch.dbg_reg.iac[i] = set_reg_val(reg->id, val);
1516                 break;
1517         case KVM_REG_PPC_DAC1:
1518         case KVM_REG_PPC_DAC2:
1519                 i = reg->id - KVM_REG_PPC_DAC1;
1520                 vcpu->arch.dbg_reg.dac[i] = set_reg_val(reg->id, val);
1521                 break;
1522         case KVM_REG_PPC_EPR: {
1523                 u32 new_epr = set_reg_val(reg->id, val);
1524                 kvmppc_set_epr(vcpu, new_epr);
1525                 break;
1526         }
1527 #if defined(CONFIG_64BIT)
1528         case KVM_REG_PPC_EPCR: {
1529                 u32 new_epcr = set_reg_val(reg->id, val);
1530                 kvmppc_set_epcr(vcpu, new_epcr);
1531                 break;
1532         }
1533 #endif
1534         case KVM_REG_PPC_OR_TSR: {
1535                 u32 tsr_bits = set_reg_val(reg->id, val);
1536                 kvmppc_set_tsr_bits(vcpu, tsr_bits);
1537                 break;
1538         }
1539         case KVM_REG_PPC_CLEAR_TSR: {
1540                 u32 tsr_bits = set_reg_val(reg->id, val);
1541                 kvmppc_clr_tsr_bits(vcpu, tsr_bits);
1542                 break;
1543         }
1544         case KVM_REG_PPC_TSR: {
1545                 u32 tsr = set_reg_val(reg->id, val);
1546                 kvmppc_set_tsr(vcpu, tsr);
1547                 break;
1548         }
1549         case KVM_REG_PPC_TCR: {
1550                 u32 tcr = set_reg_val(reg->id, val);
1551                 kvmppc_set_tcr(vcpu, tcr);
1552                 break;
1553         }
1554         default:
1555                 r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1556                 break;
1557         }
1558
1559         return r;
1560 }
1561
1562 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
1563                                          struct kvm_guest_debug *dbg)
1564 {
1565         return -EINVAL;
1566 }
1567
1568 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1569 {
1570         return -ENOTSUPP;
1571 }
1572
1573 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1574 {
1575         return -ENOTSUPP;
1576 }
1577
1578 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1579                                   struct kvm_translation *tr)
1580 {
1581         int r;
1582
1583         r = kvmppc_core_vcpu_translate(vcpu, tr);
1584         return r;
1585 }
1586
1587 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1588 {
1589         return -ENOTSUPP;
1590 }
1591
1592 void kvmppc_core_free_memslot(struct kvm_memory_slot *free,
1593                               struct kvm_memory_slot *dont)
1594 {
1595 }
1596
1597 int kvmppc_core_create_memslot(struct kvm_memory_slot *slot,
1598                                unsigned long npages)
1599 {
1600         return 0;
1601 }
1602
1603 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1604                                       struct kvm_memory_slot *memslot,
1605                                       struct kvm_userspace_memory_region *mem)
1606 {
1607         return 0;
1608 }
1609
1610 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1611                                 struct kvm_userspace_memory_region *mem,
1612                                 const struct kvm_memory_slot *old)
1613 {
1614 }
1615
1616 void kvmppc_core_flush_memslot(struct kvm *kvm, struct kvm_memory_slot *memslot)
1617 {
1618 }
1619
1620 void kvmppc_set_epcr(struct kvm_vcpu *vcpu, u32 new_epcr)
1621 {
1622 #if defined(CONFIG_64BIT)
1623         vcpu->arch.epcr = new_epcr;
1624 #ifdef CONFIG_KVM_BOOKE_HV
1625         vcpu->arch.shadow_epcr &= ~SPRN_EPCR_GICM;
1626         if (vcpu->arch.epcr  & SPRN_EPCR_ICM)
1627                 vcpu->arch.shadow_epcr |= SPRN_EPCR_GICM;
1628 #endif
1629 #endif
1630 }
1631
1632 void kvmppc_set_tcr(struct kvm_vcpu *vcpu, u32 new_tcr)
1633 {
1634         vcpu->arch.tcr = new_tcr;
1635         arm_next_watchdog(vcpu);
1636         update_timer_ints(vcpu);
1637 }
1638
1639 void kvmppc_set_tsr_bits(struct kvm_vcpu *vcpu, u32 tsr_bits)
1640 {
1641         set_bits(tsr_bits, &vcpu->arch.tsr);
1642         smp_wmb();
1643         kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1644         kvm_vcpu_kick(vcpu);
1645 }
1646
1647 void kvmppc_clr_tsr_bits(struct kvm_vcpu *vcpu, u32 tsr_bits)
1648 {
1649         clear_bits(tsr_bits, &vcpu->arch.tsr);
1650
1651         /*
1652          * We may have stopped the watchdog due to
1653          * being stuck on final expiration.
1654          */
1655         if (tsr_bits & (TSR_ENW | TSR_WIS))
1656                 arm_next_watchdog(vcpu);
1657
1658         update_timer_ints(vcpu);
1659 }
1660
1661 void kvmppc_decrementer_func(unsigned long data)
1662 {
1663         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)data;
1664
1665         if (vcpu->arch.tcr & TCR_ARE) {
1666                 vcpu->arch.dec = vcpu->arch.decar;
1667                 kvmppc_emulate_dec(vcpu);
1668         }
1669
1670         kvmppc_set_tsr_bits(vcpu, TSR_DIS);
1671 }
1672
1673 void kvmppc_booke_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1674 {
1675         vcpu->cpu = smp_processor_id();
1676         current->thread.kvm_vcpu = vcpu;
1677 }
1678
1679 void kvmppc_booke_vcpu_put(struct kvm_vcpu *vcpu)
1680 {
1681         current->thread.kvm_vcpu = NULL;
1682         vcpu->cpu = -1;
1683 }
1684
1685 int __init kvmppc_booke_init(void)
1686 {
1687 #ifndef CONFIG_KVM_BOOKE_HV
1688         unsigned long ivor[16];
1689         unsigned long *handler = kvmppc_booke_handler_addr;
1690         unsigned long max_ivor = 0;
1691         unsigned long handler_len;
1692         int i;
1693
1694         /* We install our own exception handlers by hijacking IVPR. IVPR must
1695          * be 16-bit aligned, so we need a 64KB allocation. */
1696         kvmppc_booke_handlers = __get_free_pages(GFP_KERNEL | __GFP_ZERO,
1697                                                  VCPU_SIZE_ORDER);
1698         if (!kvmppc_booke_handlers)
1699                 return -ENOMEM;
1700
1701         /* XXX make sure our handlers are smaller than Linux's */
1702
1703         /* Copy our interrupt handlers to match host IVORs. That way we don't
1704          * have to swap the IVORs on every guest/host transition. */
1705         ivor[0] = mfspr(SPRN_IVOR0);
1706         ivor[1] = mfspr(SPRN_IVOR1);
1707         ivor[2] = mfspr(SPRN_IVOR2);
1708         ivor[3] = mfspr(SPRN_IVOR3);
1709         ivor[4] = mfspr(SPRN_IVOR4);
1710         ivor[5] = mfspr(SPRN_IVOR5);
1711         ivor[6] = mfspr(SPRN_IVOR6);
1712         ivor[7] = mfspr(SPRN_IVOR7);
1713         ivor[8] = mfspr(SPRN_IVOR8);
1714         ivor[9] = mfspr(SPRN_IVOR9);
1715         ivor[10] = mfspr(SPRN_IVOR10);
1716         ivor[11] = mfspr(SPRN_IVOR11);
1717         ivor[12] = mfspr(SPRN_IVOR12);
1718         ivor[13] = mfspr(SPRN_IVOR13);
1719         ivor[14] = mfspr(SPRN_IVOR14);
1720         ivor[15] = mfspr(SPRN_IVOR15);
1721
1722         for (i = 0; i < 16; i++) {
1723                 if (ivor[i] > max_ivor)
1724                         max_ivor = i;
1725
1726                 handler_len = handler[i + 1] - handler[i];
1727                 memcpy((void *)kvmppc_booke_handlers + ivor[i],
1728                        (void *)handler[i], handler_len);
1729         }
1730
1731         handler_len = handler[max_ivor + 1] - handler[max_ivor];
1732         flush_icache_range(kvmppc_booke_handlers, kvmppc_booke_handlers +
1733                            ivor[max_ivor] + handler_len);
1734 #endif /* !BOOKE_HV */
1735         return 0;
1736 }
1737
1738 void __exit kvmppc_booke_exit(void)
1739 {
1740         free_pages(kvmppc_booke_handlers, VCPU_SIZE_ORDER);
1741         kvm_exit();
1742 }