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
  40 #include "timing.h"
  41 #include "booke.h"
  42 #include "trace.h"
  43
  44 unsigned long kvmppc_booke_handlers;
  45
  46 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  47 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  48
  49 struct kvm_stats_debugfs_item debugfs_entries[] = {
  50         { "mmio",       VCPU_STAT(mmio_exits) },
  51         { "dcr",        VCPU_STAT(dcr_exits) },
  52         { "sig",        VCPU_STAT(signal_exits) },
  53         { "itlb_r",     VCPU_STAT(itlb_real_miss_exits) },
  54         { "itlb_v",     VCPU_STAT(itlb_virt_miss_exits) },
  55         { "dtlb_r",     VCPU_STAT(dtlb_real_miss_exits) },
  56         { "dtlb_v",     VCPU_STAT(dtlb_virt_miss_exits) },
  57         { "sysc",       VCPU_STAT(syscall_exits) },
  58         { "isi",        VCPU_STAT(isi_exits) },
  59         { "dsi",        VCPU_STAT(dsi_exits) },
  60         { "inst_emu",   VCPU_STAT(emulated_inst_exits) },
  61         { "dec",        VCPU_STAT(dec_exits) },
  62         { "ext_intr",   VCPU_STAT(ext_intr_exits) },
  63         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  64         { "doorbell", VCPU_STAT(dbell_exits) },
  65         { "guest doorbell", VCPU_STAT(gdbell_exits) },
  66         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
  67         { NULL }
  68 };
  69
  70 /* TODO: use vcpu_printf() */
  71 void kvmppc_dump_vcpu(struct kvm_vcpu *vcpu)
  72 {
  73         int i;
  74
  75         printk("pc:   %08lx msr:  %08llx\n", vcpu->arch.pc, vcpu->arch.shared->msr);
  76         printk("lr:   %08lx ctr:  %08lx\n", vcpu->arch.lr, vcpu->arch.ctr);
  77         printk("srr0: %08llx srr1: %08llx\n", vcpu->arch.shared->srr0,
  78                                             vcpu->arch.shared->srr1);
  79
  80         printk("exceptions: %08lx\n", vcpu->arch.pending_exceptions);
  81
  82         for (i = 0; i < 32; i += 4) {
  83                 printk("gpr%02d: %08lx %08lx %08lx %08lx\n", i,
  84                        kvmppc_get_gpr(vcpu, i),
  85                        kvmppc_get_gpr(vcpu, i+1),
  86                        kvmppc_get_gpr(vcpu, i+2),
  87                        kvmppc_get_gpr(vcpu, i+3));
  88         }
  89 }
  90
  91 #ifdef CONFIG_SPE
  92 void kvmppc_vcpu_disable_spe(struct kvm_vcpu *vcpu)
  93 {
  94         preempt_disable();
  95         enable_kernel_spe();
  96         kvmppc_save_guest_spe(vcpu);
  97         vcpu->arch.shadow_msr &= ~MSR_SPE;
  98         preempt_enable();
  99 }
 100
 101 static void kvmppc_vcpu_enable_spe(struct kvm_vcpu *vcpu)
 102 {
 103         preempt_disable();
 104         enable_kernel_spe();
 105         kvmppc_load_guest_spe(vcpu);
 106         vcpu->arch.shadow_msr |= MSR_SPE;
 107         preempt_enable();
 108 }
 109
 110 static void kvmppc_vcpu_sync_spe(struct kvm_vcpu *vcpu)
 111 {
 112         if (vcpu->arch.shared->msr & MSR_SPE) {
 113                 if (!(vcpu->arch.shadow_msr & MSR_SPE))
 114                         kvmppc_vcpu_enable_spe(vcpu);
 115         } else if (vcpu->arch.shadow_msr & MSR_SPE) {
 116                 kvmppc_vcpu_disable_spe(vcpu);
 117         }
 118 }
 119 #else
 120 static void kvmppc_vcpu_sync_spe(struct kvm_vcpu *vcpu)
 121 {
 122 }
 123 #endif
 124
 125 static void kvmppc_vcpu_sync_fpu(struct kvm_vcpu *vcpu)
 126 {
 127 #if defined(CONFIG_PPC_FPU) && !defined(CONFIG_KVM_BOOKE_HV)
 128         /* We always treat the FP bit as enabled from the host
 129            perspective, so only need to adjust the shadow MSR */
 130         vcpu->arch.shadow_msr &= ~MSR_FP;
 131         vcpu->arch.shadow_msr |= vcpu->arch.shared->msr & MSR_FP;
 132 #endif
 133 }
 134
 135 /*
 136  * Helper function for "full" MSR writes.  No need to call this if only
 137  * EE/CE/ME/DE/RI are changing.
 138  */
 139 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u32 new_msr)
 140 {
 141         u32 old_msr = vcpu->arch.shared->msr;
 142
 143 #ifdef CONFIG_KVM_BOOKE_HV
 144         new_msr |= MSR_GS;
 145 #endif
 146
 147         vcpu->arch.shared->msr = new_msr;
 148
 149         kvmppc_mmu_msr_notify(vcpu, old_msr);
 150         kvmppc_vcpu_sync_spe(vcpu);
 151         kvmppc_vcpu_sync_fpu(vcpu);
 152 }
 153
 154 static void kvmppc_booke_queue_irqprio(struct kvm_vcpu *vcpu,
 155                                        unsigned int priority)
 156 {
 157         trace_kvm_booke_queue_irqprio(vcpu, priority);
 158         set_bit(priority, &vcpu->arch.pending_exceptions);
 159 }
 160
 161 static void kvmppc_core_queue_dtlb_miss(struct kvm_vcpu *vcpu,
 162                                         ulong dear_flags, ulong esr_flags)
 163 {
 164         vcpu->arch.queued_dear = dear_flags;
 165         vcpu->arch.queued_esr = esr_flags;
 166         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DTLB_MISS);
 167 }
 168
 169 static void kvmppc_core_queue_data_storage(struct kvm_vcpu *vcpu,
 170                                            ulong dear_flags, ulong esr_flags)
 171 {
 172         vcpu->arch.queued_dear = dear_flags;
 173         vcpu->arch.queued_esr = esr_flags;
 174         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DATA_STORAGE);
 175 }
 176
 177 static void kvmppc_core_queue_inst_storage(struct kvm_vcpu *vcpu,
 178                                            ulong esr_flags)
 179 {
 180         vcpu->arch.queued_esr = esr_flags;
 181         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_INST_STORAGE);
 182 }
 183
 184 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu, ulong esr_flags)
 185 {
 186         vcpu->arch.queued_esr = esr_flags;
 187         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_PROGRAM);
 188 }
 189
 190 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
 191 {
 192         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DECREMENTER);
 193 }
 194
 195 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
 196 {
 197         return test_bit(BOOKE_IRQPRIO_DECREMENTER, &vcpu->arch.pending_exceptions);
 198 }
 199
 200 void kvmppc_core_dequeue_dec(struct kvm_vcpu *vcpu)
 201 {
 202         clear_bit(BOOKE_IRQPRIO_DECREMENTER, &vcpu->arch.pending_exceptions);
 203 }
 204
 205 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
 206                                 struct kvm_interrupt *irq)
 207 {
 208         unsigned int prio = BOOKE_IRQPRIO_EXTERNAL;
 209
 210         if (irq->irq == KVM_INTERRUPT_SET_LEVEL)
 211                 prio = BOOKE_IRQPRIO_EXTERNAL_LEVEL;
 212
 213         kvmppc_booke_queue_irqprio(vcpu, prio);
 214 }
 215
 216 void kvmppc_core_dequeue_external(struct kvm_vcpu *vcpu,
 217                                   struct kvm_interrupt *irq)
 218 {
 219         clear_bit(BOOKE_IRQPRIO_EXTERNAL, &vcpu->arch.pending_exceptions);
 220         clear_bit(BOOKE_IRQPRIO_EXTERNAL_LEVEL, &vcpu->arch.pending_exceptions);
 221 }
 222
 223 static void kvmppc_core_queue_watchdog(struct kvm_vcpu *vcpu)
 224 {
 225         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_WATCHDOG);
 226 }
 227
 228 static void kvmppc_core_dequeue_watchdog(struct kvm_vcpu *vcpu)
 229 {
 230         clear_bit(BOOKE_IRQPRIO_WATCHDOG, &vcpu->arch.pending_exceptions);
 231 }
 232
 233 static void set_guest_srr(struct kvm_vcpu *vcpu, unsigned long srr0, u32 srr1)
 234 {
 235 #ifdef CONFIG_KVM_BOOKE_HV
 236         mtspr(SPRN_GSRR0, srr0);
 237         mtspr(SPRN_GSRR1, srr1);
 238 #else
 239         vcpu->arch.shared->srr0 = srr0;
 240         vcpu->arch.shared->srr1 = srr1;
 241 #endif
 242 }
 243
 244 static void set_guest_csrr(struct kvm_vcpu *vcpu, unsigned long srr0, u32 srr1)
 245 {
 246         vcpu->arch.csrr0 = srr0;
 247         vcpu->arch.csrr1 = srr1;
 248 }
 249
 250 static void set_guest_dsrr(struct kvm_vcpu *vcpu, unsigned long srr0, u32 srr1)
 251 {
 252         if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC)) {
 253                 vcpu->arch.dsrr0 = srr0;
 254                 vcpu->arch.dsrr1 = srr1;
 255         } else {
 256                 set_guest_csrr(vcpu, srr0, srr1);
 257         }
 258 }
 259
 260 static void set_guest_mcsrr(struct kvm_vcpu *vcpu, unsigned long srr0, u32 srr1)
 261 {
 262         vcpu->arch.mcsrr0 = srr0;
 263         vcpu->arch.mcsrr1 = srr1;
 264 }
 265
 266 static unsigned long get_guest_dear(struct kvm_vcpu *vcpu)
 267 {
 268 #ifdef CONFIG_KVM_BOOKE_HV
 269         return mfspr(SPRN_GDEAR);
 270 #else
 271         return vcpu->arch.shared->dar;
 272 #endif
 273 }
 274
 275 static void set_guest_dear(struct kvm_vcpu *vcpu, unsigned long dear)
 276 {
 277 #ifdef CONFIG_KVM_BOOKE_HV
 278         mtspr(SPRN_GDEAR, dear);
 279 #else
 280         vcpu->arch.shared->dar = dear;
 281 #endif
 282 }
 283
 284 static unsigned long get_guest_esr(struct kvm_vcpu *vcpu)
 285 {
 286 #ifdef CONFIG_KVM_BOOKE_HV
 287         return mfspr(SPRN_GESR);
 288 #else
 289         return vcpu->arch.shared->esr;
 290 #endif
 291 }
 292
 293 static void set_guest_esr(struct kvm_vcpu *vcpu, u32 esr)
 294 {
 295 #ifdef CONFIG_KVM_BOOKE_HV
 296         mtspr(SPRN_GESR, esr);
 297 #else
 298         vcpu->arch.shared->esr = esr;
 299 #endif
 300 }
 301
 302 /* Deliver the interrupt of the corresponding priority, if possible. */
 303 static int kvmppc_booke_irqprio_deliver(struct kvm_vcpu *vcpu,
 304                                         unsigned int priority)
 305 {
 306         int allowed = 0;
 307         ulong msr_mask = 0;
 308         bool update_esr = false, update_dear = false;
 309         ulong crit_raw = vcpu->arch.shared->critical;
 310         ulong crit_r1 = kvmppc_get_gpr(vcpu, 1);
 311         bool crit;
 312         bool keep_irq = false;
 313         enum int_class int_class;
 314
 315         /* Truncate crit indicators in 32 bit mode */
 316         if (!(vcpu->arch.shared->msr & MSR_SF)) {
 317                 crit_raw &= 0xffffffff;
 318                 crit_r1 &= 0xffffffff;
 319         }
 320
 321         /* Critical section when crit == r1 */
 322         crit = (crit_raw == crit_r1);
 323         /* ... and we're in supervisor mode */
 324         crit = crit && !(vcpu->arch.shared->msr & MSR_PR);
 325
 326         if (priority == BOOKE_IRQPRIO_EXTERNAL_LEVEL) {
 327                 priority = BOOKE_IRQPRIO_EXTERNAL;
 328                 keep_irq = true;
 329         }
 330
 331         switch (priority) {
 332         case BOOKE_IRQPRIO_DTLB_MISS:
 333         case BOOKE_IRQPRIO_DATA_STORAGE:
 334                 update_dear = true;
 335                 /* fall through */
 336         case BOOKE_IRQPRIO_INST_STORAGE:
 337         case BOOKE_IRQPRIO_PROGRAM:
 338                 update_esr = true;
 339                 /* fall through */
 340         case BOOKE_IRQPRIO_ITLB_MISS:
 341         case BOOKE_IRQPRIO_SYSCALL:
 342         case BOOKE_IRQPRIO_FP_UNAVAIL:
 343         case BOOKE_IRQPRIO_SPE_UNAVAIL:
 344         case BOOKE_IRQPRIO_SPE_FP_DATA:
 345         case BOOKE_IRQPRIO_SPE_FP_ROUND:
 346         case BOOKE_IRQPRIO_AP_UNAVAIL:
 347         case BOOKE_IRQPRIO_ALIGNMENT:
 348                 allowed = 1;
 349                 msr_mask = MSR_CE | MSR_ME | MSR_DE;
 350                 int_class = INT_CLASS_NONCRIT;
 351                 break;
 352         case BOOKE_IRQPRIO_WATCHDOG:
 353         case BOOKE_IRQPRIO_CRITICAL:
 354         case BOOKE_IRQPRIO_DBELL_CRIT:
 355                 allowed = vcpu->arch.shared->msr & MSR_CE;
 356                 allowed = allowed && !crit;
 357                 msr_mask = MSR_ME;
 358                 int_class = INT_CLASS_CRIT;
 359                 break;
 360         case BOOKE_IRQPRIO_MACHINE_CHECK:
 361                 allowed = vcpu->arch.shared->msr & MSR_ME;
 362                 allowed = allowed && !crit;
 363                 int_class = INT_CLASS_MC;
 364                 break;
 365         case BOOKE_IRQPRIO_DECREMENTER:
 366         case BOOKE_IRQPRIO_FIT:
 367                 keep_irq = true;
 368                 /* fall through */
 369         case BOOKE_IRQPRIO_EXTERNAL:
 370         case BOOKE_IRQPRIO_DBELL:
 371                 allowed = vcpu->arch.shared->msr & MSR_EE;
 372                 allowed = allowed && !crit;
 373                 msr_mask = MSR_CE | MSR_ME | MSR_DE;
 374                 int_class = INT_CLASS_NONCRIT;
 375                 break;
 376         case BOOKE_IRQPRIO_DEBUG:
 377                 allowed = vcpu->arch.shared->msr & MSR_DE;
 378                 allowed = allowed && !crit;
 379                 msr_mask = MSR_ME;
 380                 int_class = INT_CLASS_CRIT;
 381                 break;
 382         }
 383
 384         if (allowed) {
 385                 switch (int_class) {
 386                 case INT_CLASS_NONCRIT:
 387                         set_guest_srr(vcpu, vcpu->arch.pc,
 388                                       vcpu->arch.shared->msr);
 389                         break;
 390                 case INT_CLASS_CRIT:
 391                         set_guest_csrr(vcpu, vcpu->arch.pc,
 392                                        vcpu->arch.shared->msr);
 393                         break;
 394                 case INT_CLASS_DBG:
 395                         set_guest_dsrr(vcpu, vcpu->arch.pc,
 396                                        vcpu->arch.shared->msr);
 397                         break;
 398                 case INT_CLASS_MC:
 399                         set_guest_mcsrr(vcpu, vcpu->arch.pc,
 400                                         vcpu->arch.shared->msr);
 401                         break;
 402                 }
 403
 404                 vcpu->arch.pc = vcpu->arch.ivpr | vcpu->arch.ivor[priority];
 405                 if (update_esr == true)
 406                         set_guest_esr(vcpu, vcpu->arch.queued_esr);
 407                 if (update_dear == true)
 408                         set_guest_dear(vcpu, vcpu->arch.queued_dear);
 409                 kvmppc_set_msr(vcpu, vcpu->arch.shared->msr & msr_mask);
 410
 411                 if (!keep_irq)
 412                         clear_bit(priority, &vcpu->arch.pending_exceptions);
 413         }
 414
 415 #ifdef CONFIG_KVM_BOOKE_HV
 416         /*
 417          * If an interrupt is pending but masked, raise a guest doorbell
 418          * so that we are notified when the guest enables the relevant
 419          * MSR bit.
 420          */
 421         if (vcpu->arch.pending_exceptions & BOOKE_IRQMASK_EE)
 422                 kvmppc_set_pending_interrupt(vcpu, INT_CLASS_NONCRIT);
 423         if (vcpu->arch.pending_exceptions & BOOKE_IRQMASK_CE)
 424                 kvmppc_set_pending_interrupt(vcpu, INT_CLASS_CRIT);
 425         if (vcpu->arch.pending_exceptions & BOOKE_IRQPRIO_MACHINE_CHECK)
 426                 kvmppc_set_pending_interrupt(vcpu, INT_CLASS_MC);
 427 #endif
 428
 429         return allowed;
 430 }
 431
 432 /*
 433  * Return the number of jiffies until the next timeout.  If the timeout is
 434  * longer than the NEXT_TIMER_MAX_DELTA, then return NEXT_TIMER_MAX_DELTA
 435  * because the larger value can break the timer APIs.
 436  */
 437 static unsigned long watchdog_next_timeout(struct kvm_vcpu *vcpu)
 438 {
 439         u64 tb, wdt_tb, wdt_ticks = 0;
 440         u64 nr_jiffies = 0;
 441         u32 period = TCR_GET_WP(vcpu->arch.tcr);
 442
 443         wdt_tb = 1ULL << (63 - period);
 444         tb = get_tb();
 445         /*
 446          * The watchdog timeout will hapeen when TB bit corresponding
 447          * to watchdog will toggle from 0 to 1.
 448          */
 449         if (tb & wdt_tb)
 450                 wdt_ticks = wdt_tb;
 451
 452         wdt_ticks += wdt_tb - (tb & (wdt_tb - 1));
 453
 454         /* Convert timebase ticks to jiffies */
 455         nr_jiffies = wdt_ticks;
 456
 457         if (do_div(nr_jiffies, tb_ticks_per_jiffy))
 458                 nr_jiffies++;
 459
 460         return min_t(unsigned long long, nr_jiffies, NEXT_TIMER_MAX_DELTA);
 461 }
 462
 463 static void arm_next_watchdog(struct kvm_vcpu *vcpu)
 464 {
 465         unsigned long nr_jiffies;
 466         unsigned long flags;
 467
 468         /*
 469          * If TSR_ENW and TSR_WIS are not set then no need to exit to
 470          * userspace, so clear the KVM_REQ_WATCHDOG request.
 471          */
 472         if ((vcpu->arch.tsr & (TSR_ENW | TSR_WIS)) != (TSR_ENW | TSR_WIS))
 473                 clear_bit(KVM_REQ_WATCHDOG, &vcpu->requests);
 474
 475         spin_lock_irqsave(&vcpu->arch.wdt_lock, flags);
 476         nr_jiffies = watchdog_next_timeout(vcpu);
 477         /*
 478          * If the number of jiffies of watchdog timer >= NEXT_TIMER_MAX_DELTA
 479          * then do not run the watchdog timer as this can break timer APIs.
 480          */
 481         if (nr_jiffies < NEXT_TIMER_MAX_DELTA)
 482                 mod_timer(&vcpu->arch.wdt_timer, jiffies + nr_jiffies);
 483         else
 484                 del_timer(&vcpu->arch.wdt_timer);
 485         spin_unlock_irqrestore(&vcpu->arch.wdt_lock, flags);
 486 }
 487
 488 void kvmppc_watchdog_func(unsigned long data)
 489 {
 490         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)data;
 491         u32 tsr, new_tsr;
 492         int final;
 493
 494         do {
 495                 new_tsr = tsr = vcpu->arch.tsr;
 496                 final = 0;
 497
 498                 /* Time out event */
 499                 if (tsr & TSR_ENW) {
 500                         if (tsr & TSR_WIS)
 501                                 final = 1;
 502                         else
 503                                 new_tsr = tsr | TSR_WIS;
 504                 } else {
 505                         new_tsr = tsr | TSR_ENW;
 506                 }
 507         } while (cmpxchg(&vcpu->arch.tsr, tsr, new_tsr) != tsr);
 508
 509         if (new_tsr & TSR_WIS) {
 510                 smp_wmb();
 511                 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
 512                 kvm_vcpu_kick(vcpu);
 513         }
 514
 515         /*
 516          * If this is final watchdog expiry and some action is required
 517          * then exit to userspace.
 518          */
 519         if (final && (vcpu->arch.tcr & TCR_WRC_MASK) &&
 520             vcpu->arch.watchdog_enabled) {
 521                 smp_wmb();
 522                 kvm_make_request(KVM_REQ_WATCHDOG, vcpu);
 523                 kvm_vcpu_kick(vcpu);
 524         }
 525
 526         /*
 527          * Stop running the watchdog timer after final expiration to
 528          * prevent the host from being flooded with timers if the
 529          * guest sets a short period.
 530          * Timers will resume when TSR/TCR is updated next time.
 531          */
 532         if (!final)
 533                 arm_next_watchdog(vcpu);
 534 }
 535
 536 static void update_timer_ints(struct kvm_vcpu *vcpu)
 537 {
 538         if ((vcpu->arch.tcr & TCR_DIE) && (vcpu->arch.tsr & TSR_DIS))
 539                 kvmppc_core_queue_dec(vcpu);
 540         else
 541                 kvmppc_core_dequeue_dec(vcpu);
 542
 543         if ((vcpu->arch.tcr & TCR_WIE) && (vcpu->arch.tsr & TSR_WIS))
 544                 kvmppc_core_queue_watchdog(vcpu);
 545         else
 546                 kvmppc_core_dequeue_watchdog(vcpu);
 547 }
 548
 549 static void kvmppc_core_check_exceptions(struct kvm_vcpu *vcpu)
 550 {
 551         unsigned long *pending = &vcpu->arch.pending_exceptions;
 552         unsigned int priority;
 553
 554         priority = __ffs(*pending);
 555         while (priority < BOOKE_IRQPRIO_MAX) {
 556                 if (kvmppc_booke_irqprio_deliver(vcpu, priority))
 557                         break;
 558
 559                 priority = find_next_bit(pending,
 560                                          BITS_PER_BYTE * sizeof(*pending),
 561                                          priority + 1);
 562         }
 563
 564         /* Tell the guest about our interrupt status */
 565         vcpu->arch.shared->int_pending = !!*pending;
 566 }
 567
 568 /* Check pending exceptions and deliver one, if possible. */
 569 int kvmppc_core_prepare_to_enter(struct kvm_vcpu *vcpu)
 570 {
 571         int r = 0;
 572         WARN_ON_ONCE(!irqs_disabled());
 573
 574         kvmppc_core_check_exceptions(vcpu);
 575
 576         if (vcpu->arch.shared->msr & MSR_WE) {
 577                 local_irq_enable();
 578                 kvm_vcpu_block(vcpu);
 579                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 580                 local_irq_disable();
 581
 582                 kvmppc_set_exit_type(vcpu, EMULATED_MTMSRWE_EXITS);
 583                 r = 1;
 584         };
 585
 586         return r;
 587 }
 588
 589 int kvmppc_core_check_requests(struct kvm_vcpu *vcpu)
 590 {
 591         int r = 1; /* Indicate we want to get back into the guest */
 592
 593         if (kvm_check_request(KVM_REQ_PENDING_TIMER, vcpu))
 594                 update_timer_ints(vcpu);
 595 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 596         if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
 597                 kvmppc_core_flush_tlb(vcpu);
 598 #endif
 599
 600         if (kvm_check_request(KVM_REQ_WATCHDOG, vcpu)) {
 601                 vcpu->run->exit_reason = KVM_EXIT_WATCHDOG;
 602                 r = 0;
 603         }
 604
 605         return r;
 606 }
 607
 608 int kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
 609 {
 610         int ret, s;
 611 #ifdef CONFIG_PPC_FPU
 612         unsigned int fpscr;
 613         int fpexc_mode;
 614         u64 fpr[32];
 615 #endif
 616
 617         if (!vcpu->arch.sane) {
 618                 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
 619                 return -EINVAL;
 620         }
 621
 622         local_irq_disable();
 623         s = kvmppc_prepare_to_enter(vcpu);
 624         if (s <= 0) {
 625                 local_irq_enable();
 626                 ret = s;
 627                 goto out;
 628         }
 629         kvmppc_lazy_ee_enable();
 630
 631         kvm_guest_enter();
 632
 633 #ifdef CONFIG_PPC_FPU
 634         /* Save userspace FPU state in stack */
 635         enable_kernel_fp();
 636         memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
 637         fpscr = current->thread.fpscr.val;
 638         fpexc_mode = current->thread.fpexc_mode;
 639
 640         /* Restore guest FPU state to thread */
 641         memcpy(current->thread.fpr, vcpu->arch.fpr, sizeof(vcpu->arch.fpr));
 642         current->thread.fpscr.val = vcpu->arch.fpscr;
 643
 644         /*
 645          * Since we can't trap on MSR_FP in GS-mode, we consider the guest
 646          * as always using the FPU.  Kernel usage of FP (via
 647          * enable_kernel_fp()) in this thread must not occur while
 648          * vcpu->fpu_active is set.
 649          */
 650         vcpu->fpu_active = 1;
 651
 652         kvmppc_load_guest_fp(vcpu);
 653 #endif
 654
 655         ret = __kvmppc_vcpu_run(kvm_run, vcpu);
 656
 657         /* No need for kvm_guest_exit. It's done in handle_exit.
 658            We also get here with interrupts enabled. */
 659
 660 #ifdef CONFIG_PPC_FPU
 661         kvmppc_save_guest_fp(vcpu);
 662
 663         vcpu->fpu_active = 0;
 664
 665         /* Save guest FPU state from thread */
 666         memcpy(vcpu->arch.fpr, current->thread.fpr, sizeof(vcpu->arch.fpr));
 667         vcpu->arch.fpscr = current->thread.fpscr.val;
 668
 669         /* Restore userspace FPU state from stack */
 670         memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
 671         current->thread.fpscr.val = fpscr;
 672         current->thread.fpexc_mode = fpexc_mode;
 673 #endif
 674
 675 out:
 676         vcpu->mode = OUTSIDE_GUEST_MODE;
 677         smp_wmb();
 678         return ret;
 679 }
 680
 681 static int emulation_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
 682 {
 683         enum emulation_result er;
 684
 685         er = kvmppc_emulate_instruction(run, vcpu);
 686         switch (er) {
 687         case EMULATE_DONE:
 688                 /* don't overwrite subtypes, just account kvm_stats */
 689                 kvmppc_account_exit_stat(vcpu, EMULATED_INST_EXITS);
 690                 /* Future optimization: only reload non-volatiles if
 691                  * they were actually modified by emulation. */
 692                 return RESUME_GUEST_NV;
 693
 694         case EMULATE_DO_DCR:
 695                 run->exit_reason = KVM_EXIT_DCR;
 696                 return RESUME_HOST;
 697
 698         case EMULATE_FAIL:
 699                 printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 700                        __func__, vcpu->arch.pc, vcpu->arch.last_inst);
 701                 /* For debugging, encode the failing instruction and
 702                  * report it to userspace. */
 703                 run->hw.hardware_exit_reason = ~0ULL << 32;
 704                 run->hw.hardware_exit_reason |= vcpu->arch.last_inst;
 705                 kvmppc_core_queue_program(vcpu, ESR_PIL);
 706                 return RESUME_HOST;
 707
 708         default:
 709                 BUG();
 710         }
 711 }
 712
 713 static void kvmppc_fill_pt_regs(struct pt_regs *regs)
 714 {
 715         ulong r1, ip, msr, lr;
 716
 717         asm("mr %0, 1" : "=r"(r1));
 718         asm("mflr %0" : "=r"(lr));
 719         asm("mfmsr %0" : "=r"(msr));
 720         asm("bl 1f; 1: mflr %0" : "=r"(ip));
 721
 722         memset(regs, 0, sizeof(*regs));
 723         regs->gpr[1] = r1;
 724         regs->nip = ip;
 725         regs->msr = msr;
 726         regs->link = lr;
 727 }
 728
 729 /*
 730  * For interrupts needed to be handled by host interrupt handlers,
 731  * corresponding host handler are called from here in similar way
 732  * (but not exact) as they are called from low level handler
 733  * (such as from arch/powerpc/kernel/head_fsl_booke.S).
 734  */
 735 static void kvmppc_restart_interrupt(struct kvm_vcpu *vcpu,
 736                                      unsigned int exit_nr)
 737 {
 738         struct pt_regs regs;
 739
 740         switch (exit_nr) {
 741         case BOOKE_INTERRUPT_EXTERNAL:
 742                 kvmppc_fill_pt_regs(&regs);
 743                 do_IRQ(&regs);
 744                 break;
 745         case BOOKE_INTERRUPT_DECREMENTER:
 746                 kvmppc_fill_pt_regs(&regs);
 747                 timer_interrupt(&regs);
 748                 break;
 749 #if defined(CONFIG_PPC_FSL_BOOK3E) || defined(CONFIG_PPC_BOOK3E_64)
 750         case BOOKE_INTERRUPT_DOORBELL:
 751                 kvmppc_fill_pt_regs(&regs);
 752                 doorbell_exception(&regs);
 753                 break;
 754 #endif
 755         case BOOKE_INTERRUPT_MACHINE_CHECK:
 756                 /* FIXME */
 757                 break;
 758         case BOOKE_INTERRUPT_PERFORMANCE_MONITOR:
 759                 kvmppc_fill_pt_regs(&regs);
 760                 performance_monitor_exception(&regs);
 761                 break;
 762         case BOOKE_INTERRUPT_WATCHDOG:
 763                 kvmppc_fill_pt_regs(&regs);
 764 #ifdef CONFIG_BOOKE_WDT
 765                 WatchdogException(&regs);
 766 #else
 767                 unknown_exception(&regs);
 768 #endif
 769                 break;
 770         case BOOKE_INTERRUPT_CRITICAL:
 771                 unknown_exception(&regs);
 772                 break;
 773         }
 774 }
 775
 776 /**
 777  * kvmppc_handle_exit
 778  *
 779  * Return value is in the form (errcode<<2 | RESUME_FLAG_HOST | RESUME_FLAG_NV)
 780  */
 781 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 782                        unsigned int exit_nr)
 783 {
 784         int r = RESUME_HOST;
 785         int s;
 786
 787         /* update before a new last_exit_type is rewritten */
 788         kvmppc_update_timing_stats(vcpu);
 789
 790         /* restart interrupts if they were meant for the host */
 791         kvmppc_restart_interrupt(vcpu, exit_nr);
 792
 793         local_irq_enable();
 794
 795         trace_kvm_exit(exit_nr, vcpu);
 796         kvm_guest_exit();
 797
 798         run->exit_reason = KVM_EXIT_UNKNOWN;
 799         run->ready_for_interrupt_injection = 1;
 800
 801         switch (exit_nr) {
 802         case BOOKE_INTERRUPT_MACHINE_CHECK:
 803                 printk("MACHINE CHECK: %lx\n", mfspr(SPRN_MCSR));
 804                 kvmppc_dump_vcpu(vcpu);
 805                 /* For debugging, send invalid exit reason to user space */
 806                 run->hw.hardware_exit_reason = ~1ULL << 32;
 807                 run->hw.hardware_exit_reason |= mfspr(SPRN_MCSR);
 808                 r = RESUME_HOST;
 809                 break;
 810
 811         case BOOKE_INTERRUPT_EXTERNAL:
 812                 kvmppc_account_exit(vcpu, EXT_INTR_EXITS);
 813                 r = RESUME_GUEST;
 814                 break;
 815
 816         case BOOKE_INTERRUPT_DECREMENTER:
 817                 kvmppc_account_exit(vcpu, DEC_EXITS);
 818                 r = RESUME_GUEST;
 819                 break;
 820
 821         case BOOKE_INTERRUPT_WATCHDOG:
 822                 r = RESUME_GUEST;
 823                 break;
 824
 825         case BOOKE_INTERRUPT_DOORBELL:
 826                 kvmppc_account_exit(vcpu, DBELL_EXITS);
 827                 r = RESUME_GUEST;
 828                 break;
 829
 830         case BOOKE_INTERRUPT_GUEST_DBELL_CRIT:
 831                 kvmppc_account_exit(vcpu, GDBELL_EXITS);
 832
 833                 /*
 834                  * We are here because there is a pending guest interrupt
 835                  * which could not be delivered as MSR_CE or MSR_ME was not
 836                  * set.  Once we break from here we will retry delivery.
 837                  */
 838                 r = RESUME_GUEST;
 839                 break;
 840
 841         case BOOKE_INTERRUPT_GUEST_DBELL:
 842                 kvmppc_account_exit(vcpu, GDBELL_EXITS);
 843
 844                 /*
 845                  * We are here because there is a pending guest interrupt
 846                  * which could not be delivered as MSR_EE was not set.  Once
 847                  * we break from here we will retry delivery.
 848                  */
 849                 r = RESUME_GUEST;
 850                 break;
 851
 852         case BOOKE_INTERRUPT_PERFORMANCE_MONITOR:
 853                 r = RESUME_GUEST;
 854                 break;
 855
 856         case BOOKE_INTERRUPT_HV_PRIV:
 857                 r = emulation_exit(run, vcpu);
 858                 break;
 859
 860         case BOOKE_INTERRUPT_PROGRAM:
 861                 if (vcpu->arch.shared->msr & (MSR_PR | MSR_GS)) {
 862                         /*
 863                          * Program traps generated by user-level software must
 864                          * be handled by the guest kernel.
 865                          *
 866                          * In GS mode, hypervisor privileged instructions trap
 867                          * on BOOKE_INTERRUPT_HV_PRIV, not here, so these are
 868                          * actual program interrupts, handled by the guest.
 869                          */
 870                         kvmppc_core_queue_program(vcpu, vcpu->arch.fault_esr);
 871                         r = RESUME_GUEST;
 872                         kvmppc_account_exit(vcpu, USR_PR_INST);
 873                         break;
 874                 }
 875
 876                 r = emulation_exit(run, vcpu);
 877                 break;
 878
 879         case BOOKE_INTERRUPT_FP_UNAVAIL:
 880                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_FP_UNAVAIL);
 881                 kvmppc_account_exit(vcpu, FP_UNAVAIL);
 882                 r = RESUME_GUEST;
 883                 break;
 884
 885 #ifdef CONFIG_SPE
 886         case BOOKE_INTERRUPT_SPE_UNAVAIL: {
 887                 if (vcpu->arch.shared->msr & MSR_SPE)
 888                         kvmppc_vcpu_enable_spe(vcpu);
 889                 else
 890                         kvmppc_booke_queue_irqprio(vcpu,
 891                                                    BOOKE_IRQPRIO_SPE_UNAVAIL);
 892                 r = RESUME_GUEST;
 893                 break;
 894         }
 895
 896         case BOOKE_INTERRUPT_SPE_FP_DATA:
 897                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SPE_FP_DATA);
 898                 r = RESUME_GUEST;
 899                 break;
 900
 901         case BOOKE_INTERRUPT_SPE_FP_ROUND:
 902                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SPE_FP_ROUND);
 903                 r = RESUME_GUEST;
 904                 break;
 905 #else
 906         case BOOKE_INTERRUPT_SPE_UNAVAIL:
 907                 /*
 908                  * Guest wants SPE, but host kernel doesn't support it.  Send
 909                  * an "unimplemented operation" program check to the guest.
 910                  */
 911                 kvmppc_core_queue_program(vcpu, ESR_PUO | ESR_SPV);
 912                 r = RESUME_GUEST;
 913                 break;
 914
 915         /*
 916          * These really should never happen without CONFIG_SPE,
 917          * as we should never enable the real MSR[SPE] in the guest.
 918          */
 919         case BOOKE_INTERRUPT_SPE_FP_DATA:
 920         case BOOKE_INTERRUPT_SPE_FP_ROUND:
 921                 printk(KERN_CRIT "%s: unexpected SPE interrupt %u at %08lx\n",
 922                        __func__, exit_nr, vcpu->arch.pc);
 923                 run->hw.hardware_exit_reason = exit_nr;
 924                 r = RESUME_HOST;
 925                 break;
 926 #endif
 927
 928         case BOOKE_INTERRUPT_DATA_STORAGE:
 929                 kvmppc_core_queue_data_storage(vcpu, vcpu->arch.fault_dear,
 930                                                vcpu->arch.fault_esr);
 931                 kvmppc_account_exit(vcpu, DSI_EXITS);
 932                 r = RESUME_GUEST;
 933                 break;
 934
 935         case BOOKE_INTERRUPT_INST_STORAGE:
 936                 kvmppc_core_queue_inst_storage(vcpu, vcpu->arch.fault_esr);
 937                 kvmppc_account_exit(vcpu, ISI_EXITS);
 938                 r = RESUME_GUEST;
 939                 break;
 940
 941 #ifdef CONFIG_KVM_BOOKE_HV
 942         case BOOKE_INTERRUPT_HV_SYSCALL:
 943                 if (!(vcpu->arch.shared->msr & MSR_PR)) {
 944                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
 945                 } else {
 946                         /*
 947                          * hcall from guest userspace -- send privileged
 948                          * instruction program check.
 949                          */
 950                         kvmppc_core_queue_program(vcpu, ESR_PPR);
 951                 }
 952
 953                 r = RESUME_GUEST;
 954                 break;
 955 #else
 956         case BOOKE_INTERRUPT_SYSCALL:
 957                 if (!(vcpu->arch.shared->msr & MSR_PR) &&
 958                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
 959                         /* KVM PV hypercalls */
 960                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
 961                         r = RESUME_GUEST;
 962                 } else {
 963                         /* Guest syscalls */
 964                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SYSCALL);
 965                 }
 966                 kvmppc_account_exit(vcpu, SYSCALL_EXITS);
 967                 r = RESUME_GUEST;
 968                 break;
 969 #endif
 970
 971         case BOOKE_INTERRUPT_DTLB_MISS: {
 972                 unsigned long eaddr = vcpu->arch.fault_dear;
 973                 int gtlb_index;
 974                 gpa_t gpaddr;
 975                 gfn_t gfn;
 976
 977 #ifdef CONFIG_KVM_E500V2
 978                 if (!(vcpu->arch.shared->msr & MSR_PR) &&
 979                     (eaddr & PAGE_MASK) == vcpu->arch.magic_page_ea) {
 980                         kvmppc_map_magic(vcpu);
 981                         kvmppc_account_exit(vcpu, DTLB_VIRT_MISS_EXITS);
 982                         r = RESUME_GUEST;
 983
 984                         break;
 985                 }
 986 #endif
 987
 988                 /* Check the guest TLB. */
 989                 gtlb_index = kvmppc_mmu_dtlb_index(vcpu, eaddr);
 990                 if (gtlb_index < 0) {
 991                         /* The guest didn't have a mapping for it. */
 992                         kvmppc_core_queue_dtlb_miss(vcpu,
 993                                                     vcpu->arch.fault_dear,
 994                                                     vcpu->arch.fault_esr);
 995                         kvmppc_mmu_dtlb_miss(vcpu);
 996                         kvmppc_account_exit(vcpu, DTLB_REAL_MISS_EXITS);
 997                         r = RESUME_GUEST;
 998                         break;
 999                 }
1000
1001                 gpaddr = kvmppc_mmu_xlate(vcpu, gtlb_index, eaddr);
1002                 gfn = gpaddr >> PAGE_SHIFT;
1003
1004                 if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
1005                         /* The guest TLB had a mapping, but the shadow TLB
1006                          * didn't, and it is RAM. This could be because:
1007                          * a) the entry is mapping the host kernel, or
1008                          * b) the guest used a large mapping which we're faking
1009                          * Either way, we need to satisfy the fault without
1010                          * invoking the guest. */
1011                         kvmppc_mmu_map(vcpu, eaddr, gpaddr, gtlb_index);
1012                         kvmppc_account_exit(vcpu, DTLB_VIRT_MISS_EXITS);
1013                         r = RESUME_GUEST;
1014                 } else {
1015                         /* Guest has mapped and accessed a page which is not
1016                          * actually RAM. */
1017                         vcpu->arch.paddr_accessed = gpaddr;
1018                         vcpu->arch.vaddr_accessed = eaddr;
1019                         r = kvmppc_emulate_mmio(run, vcpu);
1020                         kvmppc_account_exit(vcpu, MMIO_EXITS);
1021                 }
1022
1023                 break;
1024         }
1025
1026         case BOOKE_INTERRUPT_ITLB_MISS: {
1027                 unsigned long eaddr = vcpu->arch.pc;
1028                 gpa_t gpaddr;
1029                 gfn_t gfn;
1030                 int gtlb_index;
1031
1032                 r = RESUME_GUEST;
1033
1034                 /* Check the guest TLB. */
1035                 gtlb_index = kvmppc_mmu_itlb_index(vcpu, eaddr);
1036                 if (gtlb_index < 0) {
1037                         /* The guest didn't have a mapping for it. */
1038                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_ITLB_MISS);
1039                         kvmppc_mmu_itlb_miss(vcpu);
1040                         kvmppc_account_exit(vcpu, ITLB_REAL_MISS_EXITS);
1041                         break;
1042                 }
1043
1044                 kvmppc_account_exit(vcpu, ITLB_VIRT_MISS_EXITS);
1045
1046                 gpaddr = kvmppc_mmu_xlate(vcpu, gtlb_index, eaddr);
1047                 gfn = gpaddr >> PAGE_SHIFT;
1048
1049                 if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
1050                         /* The guest TLB had a mapping, but the shadow TLB
1051                          * didn't. This could be because:
1052                          * a) the entry is mapping the host kernel, or
1053                          * b) the guest used a large mapping which we're faking
1054                          * Either way, we need to satisfy the fault without
1055                          * invoking the guest. */
1056                         kvmppc_mmu_map(vcpu, eaddr, gpaddr, gtlb_index);
1057                 } else {
1058                         /* Guest mapped and leaped at non-RAM! */
1059                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_MACHINE_CHECK);
1060                 }
1061
1062                 break;
1063         }
1064
1065         case BOOKE_INTERRUPT_DEBUG: {
1066                 u32 dbsr;
1067
1068                 vcpu->arch.pc = mfspr(SPRN_CSRR0);
1069
1070                 /* clear IAC events in DBSR register */
1071                 dbsr = mfspr(SPRN_DBSR);
1072                 dbsr &= DBSR_IAC1 | DBSR_IAC2 | DBSR_IAC3 | DBSR_IAC4;
1073                 mtspr(SPRN_DBSR, dbsr);
1074
1075                 run->exit_reason = KVM_EXIT_DEBUG;
1076                 kvmppc_account_exit(vcpu, DEBUG_EXITS);
1077                 r = RESUME_HOST;
1078                 break;
1079         }
1080
1081         default:
1082                 printk(KERN_EMERG "exit_nr %d\n", exit_nr);
1083                 BUG();
1084         }
1085
1086         /*
1087          * To avoid clobbering exit_reason, only check for signals if we
1088          * aren't already exiting to userspace for some other reason.
1089          */
1090         if (!(r & RESUME_HOST)) {
1091                 local_irq_disable();
1092                 s = kvmppc_prepare_to_enter(vcpu);
1093                 if (s <= 0) {
1094                         local_irq_enable();
1095                         r = (s << 2) | RESUME_HOST | (r & RESUME_FLAG_NV);
1096                 } else {
1097                         kvmppc_lazy_ee_enable();
1098                 }
1099         }
1100
1101         return r;
1102 }
1103
1104 /* Initial guest state: 16MB mapping 0 -> 0, PC = 0, MSR = 0, R1 = 16MB */
1105 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1106 {
1107         int i;
1108         int r;
1109
1110         vcpu->arch.pc = 0;
1111         vcpu->arch.shared->pir = vcpu->vcpu_id;
1112         kvmppc_set_gpr(vcpu, 1, (16<<20) - 8); /* -8 for the callee-save LR slot */
1113         kvmppc_set_msr(vcpu, 0);
1114
1115 #ifndef CONFIG_KVM_BOOKE_HV
1116         vcpu->arch.shadow_msr = MSR_USER | MSR_DE | MSR_IS | MSR_DS;
1117         vcpu->arch.shadow_pid = 1;
1118         vcpu->arch.shared->msr = 0;
1119 #endif
1120
1121         /* Eye-catching numbers so we know if the guest takes an interrupt
1122          * before it's programmed its own IVPR/IVORs. */
1123         vcpu->arch.ivpr = 0x55550000;
1124         for (i = 0; i < BOOKE_IRQPRIO_MAX; i++)
1125                 vcpu->arch.ivor[i] = 0x7700 | i * 4;
1126
1127         kvmppc_init_timing_stats(vcpu);
1128
1129         r = kvmppc_core_vcpu_setup(vcpu);
1130         kvmppc_sanity_check(vcpu);
1131         return r;
1132 }
1133
1134 int kvmppc_subarch_vcpu_init(struct kvm_vcpu *vcpu)
1135 {
1136         /* setup watchdog timer once */
1137         spin_lock_init(&vcpu->arch.wdt_lock);
1138         setup_timer(&vcpu->arch.wdt_timer, kvmppc_watchdog_func,
1139                     (unsigned long)vcpu);
1140
1141         return 0;
1142 }
1143
1144 void kvmppc_subarch_vcpu_uninit(struct kvm_vcpu *vcpu)
1145 {
1146         del_timer_sync(&vcpu->arch.wdt_timer);
1147 }
1148
1149 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1150 {
1151         int i;
1152
1153         regs->pc = vcpu->arch.pc;
1154         regs->cr = kvmppc_get_cr(vcpu);
1155         regs->ctr = vcpu->arch.ctr;
1156         regs->lr = vcpu->arch.lr;
1157         regs->xer = kvmppc_get_xer(vcpu);
1158         regs->msr = vcpu->arch.shared->msr;
1159         regs->srr0 = vcpu->arch.shared->srr0;
1160         regs->srr1 = vcpu->arch.shared->srr1;
1161         regs->pid = vcpu->arch.pid;
1162         regs->sprg0 = vcpu->arch.shared->sprg0;
1163         regs->sprg1 = vcpu->arch.shared->sprg1;
1164         regs->sprg2 = vcpu->arch.shared->sprg2;
1165         regs->sprg3 = vcpu->arch.shared->sprg3;
1166         regs->sprg4 = vcpu->arch.shared->sprg4;
1167         regs->sprg5 = vcpu->arch.shared->sprg5;
1168         regs->sprg6 = vcpu->arch.shared->sprg6;
1169         regs->sprg7 = vcpu->arch.shared->sprg7;
1170
1171         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
1172                 regs->gpr[i] = kvmppc_get_gpr(vcpu, i);
1173
1174         return 0;
1175 }
1176
1177 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1178 {
1179         int i;
1180
1181         vcpu->arch.pc = regs->pc;
1182         kvmppc_set_cr(vcpu, regs->cr);
1183         vcpu->arch.ctr = regs->ctr;
1184         vcpu->arch.lr = regs->lr;
1185         kvmppc_set_xer(vcpu, regs->xer);
1186         kvmppc_set_msr(vcpu, regs->msr);
1187         vcpu->arch.shared->srr0 = regs->srr0;
1188         vcpu->arch.shared->srr1 = regs->srr1;
1189         kvmppc_set_pid(vcpu, regs->pid);
1190         vcpu->arch.shared->sprg0 = regs->sprg0;
1191         vcpu->arch.shared->sprg1 = regs->sprg1;
1192         vcpu->arch.shared->sprg2 = regs->sprg2;
1193         vcpu->arch.shared->sprg3 = regs->sprg3;
1194         vcpu->arch.shared->sprg4 = regs->sprg4;
1195         vcpu->arch.shared->sprg5 = regs->sprg5;
1196         vcpu->arch.shared->sprg6 = regs->sprg6;
1197         vcpu->arch.shared->sprg7 = regs->sprg7;
1198
1199         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
1200                 kvmppc_set_gpr(vcpu, i, regs->gpr[i]);
1201
1202         return 0;
1203 }
1204
1205 static void get_sregs_base(struct kvm_vcpu *vcpu,
1206                            struct kvm_sregs *sregs)
1207 {
1208         u64 tb = get_tb();
1209
1210         sregs->u.e.features |= KVM_SREGS_E_BASE;
1211
1212         sregs->u.e.csrr0 = vcpu->arch.csrr0;
1213         sregs->u.e.csrr1 = vcpu->arch.csrr1;
1214         sregs->u.e.mcsr = vcpu->arch.mcsr;
1215         sregs->u.e.esr = get_guest_esr(vcpu);
1216         sregs->u.e.dear = get_guest_dear(vcpu);
1217         sregs->u.e.tsr = vcpu->arch.tsr;
1218         sregs->u.e.tcr = vcpu->arch.tcr;
1219         sregs->u.e.dec = kvmppc_get_dec(vcpu, tb);
1220         sregs->u.e.tb = tb;
1221         sregs->u.e.vrsave = vcpu->arch.vrsave;
1222 }
1223
1224 static int set_sregs_base(struct kvm_vcpu *vcpu,
1225                           struct kvm_sregs *sregs)
1226 {
1227         if (!(sregs->u.e.features & KVM_SREGS_E_BASE))
1228                 return 0;
1229
1230         vcpu->arch.csrr0 = sregs->u.e.csrr0;
1231         vcpu->arch.csrr1 = sregs->u.e.csrr1;
1232         vcpu->arch.mcsr = sregs->u.e.mcsr;
1233         set_guest_esr(vcpu, sregs->u.e.esr);
1234         set_guest_dear(vcpu, sregs->u.e.dear);
1235         vcpu->arch.vrsave = sregs->u.e.vrsave;
1236         kvmppc_set_tcr(vcpu, sregs->u.e.tcr);
1237
1238         if (sregs->u.e.update_special & KVM_SREGS_E_UPDATE_DEC) {
1239                 vcpu->arch.dec = sregs->u.e.dec;
1240                 kvmppc_emulate_dec(vcpu);
1241         }
1242
1243         if (sregs->u.e.update_special & KVM_SREGS_E_UPDATE_TSR) {
1244                 u32 old_tsr = vcpu->arch.tsr;
1245
1246                 vcpu->arch.tsr = sregs->u.e.tsr;
1247
1248                 if ((old_tsr ^ vcpu->arch.tsr) & (TSR_ENW | TSR_WIS))
1249                         arm_next_watchdog(vcpu);
1250
1251                 update_timer_ints(vcpu);
1252         }
1253
1254         return 0;
1255 }
1256
1257 static void get_sregs_arch206(struct kvm_vcpu *vcpu,
1258                               struct kvm_sregs *sregs)
1259 {
1260         sregs->u.e.features |= KVM_SREGS_E_ARCH206;
1261
1262         sregs->u.e.pir = vcpu->vcpu_id;
1263         sregs->u.e.mcsrr0 = vcpu->arch.mcsrr0;
1264         sregs->u.e.mcsrr1 = vcpu->arch.mcsrr1;
1265         sregs->u.e.decar = vcpu->arch.decar;
1266         sregs->u.e.ivpr = vcpu->arch.ivpr;
1267 }
1268
1269 static int set_sregs_arch206(struct kvm_vcpu *vcpu,
1270                              struct kvm_sregs *sregs)
1271 {
1272         if (!(sregs->u.e.features & KVM_SREGS_E_ARCH206))
1273                 return 0;
1274
1275         if (sregs->u.e.pir != vcpu->vcpu_id)
1276                 return -EINVAL;
1277
1278         vcpu->arch.mcsrr0 = sregs->u.e.mcsrr0;
1279         vcpu->arch.mcsrr1 = sregs->u.e.mcsrr1;
1280         vcpu->arch.decar = sregs->u.e.decar;
1281         vcpu->arch.ivpr = sregs->u.e.ivpr;
1282
1283         return 0;
1284 }
1285
1286 void kvmppc_get_sregs_ivor(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
1287 {
1288         sregs->u.e.features |= KVM_SREGS_E_IVOR;
1289
1290         sregs->u.e.ivor_low[0] = vcpu->arch.ivor[BOOKE_IRQPRIO_CRITICAL];
1291         sregs->u.e.ivor_low[1] = vcpu->arch.ivor[BOOKE_IRQPRIO_MACHINE_CHECK];
1292         sregs->u.e.ivor_low[2] = vcpu->arch.ivor[BOOKE_IRQPRIO_DATA_STORAGE];
1293         sregs->u.e.ivor_low[3] = vcpu->arch.ivor[BOOKE_IRQPRIO_INST_STORAGE];
1294         sregs->u.e.ivor_low[4] = vcpu->arch.ivor[BOOKE_IRQPRIO_EXTERNAL];
1295         sregs->u.e.ivor_low[5] = vcpu->arch.ivor[BOOKE_IRQPRIO_ALIGNMENT];
1296         sregs->u.e.ivor_low[6] = vcpu->arch.ivor[BOOKE_IRQPRIO_PROGRAM];
1297         sregs->u.e.ivor_low[7] = vcpu->arch.ivor[BOOKE_IRQPRIO_FP_UNAVAIL];
1298         sregs->u.e.ivor_low[8] = vcpu->arch.ivor[BOOKE_IRQPRIO_SYSCALL];
1299         sregs->u.e.ivor_low[9] = vcpu->arch.ivor[BOOKE_IRQPRIO_AP_UNAVAIL];
1300         sregs->u.e.ivor_low[10] = vcpu->arch.ivor[BOOKE_IRQPRIO_DECREMENTER];
1301         sregs->u.e.ivor_low[11] = vcpu->arch.ivor[BOOKE_IRQPRIO_FIT];
1302         sregs->u.e.ivor_low[12] = vcpu->arch.ivor[BOOKE_IRQPRIO_WATCHDOG];
1303         sregs->u.e.ivor_low[13] = vcpu->arch.ivor[BOOKE_IRQPRIO_DTLB_MISS];
1304         sregs->u.e.ivor_low[14] = vcpu->arch.ivor[BOOKE_IRQPRIO_ITLB_MISS];
1305         sregs->u.e.ivor_low[15] = vcpu->arch.ivor[BOOKE_IRQPRIO_DEBUG];
1306 }
1307
1308 int kvmppc_set_sregs_ivor(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
1309 {
1310         if (!(sregs->u.e.features & KVM_SREGS_E_IVOR))
1311                 return 0;
1312
1313         vcpu->arch.ivor[BOOKE_IRQPRIO_CRITICAL] = sregs->u.e.ivor_low[0];
1314         vcpu->arch.ivor[BOOKE_IRQPRIO_MACHINE_CHECK] = sregs->u.e.ivor_low[1];
1315         vcpu->arch.ivor[BOOKE_IRQPRIO_DATA_STORAGE] = sregs->u.e.ivor_low[2];
1316         vcpu->arch.ivor[BOOKE_IRQPRIO_INST_STORAGE] = sregs->u.e.ivor_low[3];
1317         vcpu->arch.ivor[BOOKE_IRQPRIO_EXTERNAL] = sregs->u.e.ivor_low[4];
1318         vcpu->arch.ivor[BOOKE_IRQPRIO_ALIGNMENT] = sregs->u.e.ivor_low[5];
1319         vcpu->arch.ivor[BOOKE_IRQPRIO_PROGRAM] = sregs->u.e.ivor_low[6];
1320         vcpu->arch.ivor[BOOKE_IRQPRIO_FP_UNAVAIL] = sregs->u.e.ivor_low[7];
1321         vcpu->arch.ivor[BOOKE_IRQPRIO_SYSCALL] = sregs->u.e.ivor_low[8];
1322         vcpu->arch.ivor[BOOKE_IRQPRIO_AP_UNAVAIL] = sregs->u.e.ivor_low[9];
1323         vcpu->arch.ivor[BOOKE_IRQPRIO_DECREMENTER] = sregs->u.e.ivor_low[10];
1324         vcpu->arch.ivor[BOOKE_IRQPRIO_FIT] = sregs->u.e.ivor_low[11];
1325         vcpu->arch.ivor[BOOKE_IRQPRIO_WATCHDOG] = sregs->u.e.ivor_low[12];
1326         vcpu->arch.ivor[BOOKE_IRQPRIO_DTLB_MISS] = sregs->u.e.ivor_low[13];
1327         vcpu->arch.ivor[BOOKE_IRQPRIO_ITLB_MISS] = sregs->u.e.ivor_low[14];
1328         vcpu->arch.ivor[BOOKE_IRQPRIO_DEBUG] = sregs->u.e.ivor_low[15];
1329
1330         return 0;
1331 }
1332
1333 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1334                                   struct kvm_sregs *sregs)
1335 {
1336         sregs->pvr = vcpu->arch.pvr;
1337
1338         get_sregs_base(vcpu, sregs);
1339         get_sregs_arch206(vcpu, sregs);
1340         kvmppc_core_get_sregs(vcpu, sregs);
1341         return 0;
1342 }
1343
1344 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1345                                   struct kvm_sregs *sregs)
1346 {
1347         int ret;
1348
1349         if (vcpu->arch.pvr != sregs->pvr)
1350                 return -EINVAL;
1351
1352         ret = set_sregs_base(vcpu, sregs);
1353         if (ret < 0)
1354                 return ret;
1355
1356         ret = set_sregs_arch206(vcpu, sregs);
1357         if (ret < 0)
1358                 return ret;
1359
1360         return kvmppc_core_set_sregs(vcpu, sregs);
1361 }
1362
1363 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1364 {
1365         int r = -EINVAL;
1366
1367         switch (reg->id) {
1368         case KVM_REG_PPC_IAC1:
1369         case KVM_REG_PPC_IAC2:
1370         case KVM_REG_PPC_IAC3:
1371         case KVM_REG_PPC_IAC4: {
1372                 int iac = reg->id - KVM_REG_PPC_IAC1;
1373                 r = copy_to_user((u64 __user *)(long)reg->addr,
1374                                  &vcpu->arch.dbg_reg.iac[iac], sizeof(u64));
1375                 break;
1376         }
1377         case KVM_REG_PPC_DAC1:
1378         case KVM_REG_PPC_DAC2: {
1379                 int dac = reg->id - KVM_REG_PPC_DAC1;
1380                 r = copy_to_user((u64 __user *)(long)reg->addr,
1381                                  &vcpu->arch.dbg_reg.dac[dac], sizeof(u64));
1382                 break;
1383         }
1384         default:
1385                 break;
1386         }
1387         return r;
1388 }
1389
1390 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1391 {
1392         int r = -EINVAL;
1393
1394         switch (reg->id) {
1395         case KVM_REG_PPC_IAC1:
1396         case KVM_REG_PPC_IAC2:
1397         case KVM_REG_PPC_IAC3:
1398         case KVM_REG_PPC_IAC4: {
1399                 int iac = reg->id - KVM_REG_PPC_IAC1;
1400                 r = copy_from_user(&vcpu->arch.dbg_reg.iac[iac],
1401                              (u64 __user *)(long)reg->addr, sizeof(u64));
1402                 break;
1403         }
1404         case KVM_REG_PPC_DAC1:
1405         case KVM_REG_PPC_DAC2: {
1406                 int dac = reg->id - KVM_REG_PPC_DAC1;
1407                 r = copy_from_user(&vcpu->arch.dbg_reg.dac[dac],
1408                              (u64 __user *)(long)reg->addr, sizeof(u64));
1409                 break;
1410         }
1411         default:
1412                 break;
1413         }
1414         return r;
1415 }
1416
1417 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1418 {
1419         return -ENOTSUPP;
1420 }
1421
1422 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1423 {
1424         return -ENOTSUPP;
1425 }
1426
1427 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1428                                   struct kvm_translation *tr)
1429 {
1430         int r;
1431
1432         r = kvmppc_core_vcpu_translate(vcpu, tr);
1433         return r;
1434 }
1435
1436 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1437 {
1438         return -ENOTSUPP;
1439 }
1440
1441 void kvmppc_core_free_memslot(struct kvm_memory_slot *free,
1442                               struct kvm_memory_slot *dont)
1443 {
1444 }
1445
1446 int kvmppc_core_create_memslot(struct kvm_memory_slot *slot,
1447                                unsigned long npages)
1448 {
1449         return 0;
1450 }
1451
1452 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1453                                       struct kvm_memory_slot *memslot,
1454                                       struct kvm_userspace_memory_region *mem)
1455 {
1456         return 0;
1457 }
1458
1459 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1460                                 struct kvm_userspace_memory_region *mem)
1461 {
1462 }
1463
1464 void kvmppc_set_tcr(struct kvm_vcpu *vcpu, u32 new_tcr)
1465 {
1466         vcpu->arch.tcr = new_tcr;
1467         arm_next_watchdog(vcpu);
1468         update_timer_ints(vcpu);
1469 }
1470
1471 void kvmppc_set_tsr_bits(struct kvm_vcpu *vcpu, u32 tsr_bits)
1472 {
1473         set_bits(tsr_bits, &vcpu->arch.tsr);
1474         smp_wmb();
1475         kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1476         kvm_vcpu_kick(vcpu);
1477 }
1478
1479 void kvmppc_clr_tsr_bits(struct kvm_vcpu *vcpu, u32 tsr_bits)
1480 {
1481         clear_bits(tsr_bits, &vcpu->arch.tsr);
1482
1483         /*
1484          * We may have stopped the watchdog due to
1485          * being stuck on final expiration.
1486          */
1487         if (tsr_bits & (TSR_ENW | TSR_WIS))
1488                 arm_next_watchdog(vcpu);
1489
1490         update_timer_ints(vcpu);
1491 }
1492
1493 void kvmppc_decrementer_func(unsigned long data)
1494 {
1495         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)data;
1496
1497         if (vcpu->arch.tcr & TCR_ARE) {
1498                 vcpu->arch.dec = vcpu->arch.decar;
1499                 kvmppc_emulate_dec(vcpu);
1500         }
1501
1502         kvmppc_set_tsr_bits(vcpu, TSR_DIS);
1503 }
1504
1505 void kvmppc_booke_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1506 {
1507         current->thread.kvm_vcpu = vcpu;
1508 }
1509
1510 void kvmppc_booke_vcpu_put(struct kvm_vcpu *vcpu)
1511 {
1512         current->thread.kvm_vcpu = NULL;
1513 }
1514
1515 int __init kvmppc_booke_init(void)
1516 {
1517 #ifndef CONFIG_KVM_BOOKE_HV
1518         unsigned long ivor[16];
1519         unsigned long max_ivor = 0;
1520         int i;
1521
1522         /* We install our own exception handlers by hijacking IVPR. IVPR must
1523          * be 16-bit aligned, so we need a 64KB allocation. */
1524         kvmppc_booke_handlers = __get_free_pages(GFP_KERNEL | __GFP_ZERO,
1525                                                  VCPU_SIZE_ORDER);
1526         if (!kvmppc_booke_handlers)
1527                 return -ENOMEM;
1528
1529         /* XXX make sure our handlers are smaller than Linux's */
1530
1531         /* Copy our interrupt handlers to match host IVORs. That way we don't
1532          * have to swap the IVORs on every guest/host transition. */
1533         ivor[0] = mfspr(SPRN_IVOR0);
1534         ivor[1] = mfspr(SPRN_IVOR1);
1535         ivor[2] = mfspr(SPRN_IVOR2);
1536         ivor[3] = mfspr(SPRN_IVOR3);
1537         ivor[4] = mfspr(SPRN_IVOR4);
1538         ivor[5] = mfspr(SPRN_IVOR5);
1539         ivor[6] = mfspr(SPRN_IVOR6);
1540         ivor[7] = mfspr(SPRN_IVOR7);
1541         ivor[8] = mfspr(SPRN_IVOR8);
1542         ivor[9] = mfspr(SPRN_IVOR9);
1543         ivor[10] = mfspr(SPRN_IVOR10);
1544         ivor[11] = mfspr(SPRN_IVOR11);
1545         ivor[12] = mfspr(SPRN_IVOR12);
1546         ivor[13] = mfspr(SPRN_IVOR13);
1547         ivor[14] = mfspr(SPRN_IVOR14);
1548         ivor[15] = mfspr(SPRN_IVOR15);
1549
1550         for (i = 0; i < 16; i++) {
1551                 if (ivor[i] > max_ivor)
1552                         max_ivor = ivor[i];
1553
1554                 memcpy((void *)kvmppc_booke_handlers + ivor[i],
1555                        kvmppc_handlers_start + i * kvmppc_handler_len,
1556                        kvmppc_handler_len);
1557         }
1558         flush_icache_range(kvmppc_booke_handlers,
1559                            kvmppc_booke_handlers + max_ivor + kvmppc_handler_len);
1560 #endif /* !BOOKE_HV */
1561         return 0;
1562 }
1563
1564 void __exit kvmppc_booke_exit(void)
1565 {
1566         free_pages(kvmppc_booke_handlers, VCPU_SIZE_ORDER);
1567         kvm_exit();
1568 }