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  */
  21
  22 #include <linux/errno.h>
  23 #include <linux/err.h>
  24 #include <linux/kvm_host.h>
  25 #include <linux/gfp.h>
  26 #include <linux/module.h>
  27 #include <linux/vmalloc.h>
  28 #include <linux/fs.h>
  29
  30 #include <asm/cputable.h>
  31 #include <asm/uaccess.h>
  32 #include <asm/kvm_ppc.h>
  33 #include "timing.h"
  34 #include <asm/cacheflush.h>
  35
  36 #include "booke.h"
  37
  38 unsigned long kvmppc_booke_handlers;
  39
  40 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  41 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  42
  43 struct kvm_stats_debugfs_item debugfs_entries[] = {
  44         { "mmio",       VCPU_STAT(mmio_exits) },
  45         { "dcr",        VCPU_STAT(dcr_exits) },
  46         { "sig",        VCPU_STAT(signal_exits) },
  47         { "itlb_r",     VCPU_STAT(itlb_real_miss_exits) },
  48         { "itlb_v",     VCPU_STAT(itlb_virt_miss_exits) },
  49         { "dtlb_r",     VCPU_STAT(dtlb_real_miss_exits) },
  50         { "dtlb_v",     VCPU_STAT(dtlb_virt_miss_exits) },
  51         { "sysc",       VCPU_STAT(syscall_exits) },
  52         { "isi",        VCPU_STAT(isi_exits) },
  53         { "dsi",        VCPU_STAT(dsi_exits) },
  54         { "inst_emu",   VCPU_STAT(emulated_inst_exits) },
  55         { "dec",        VCPU_STAT(dec_exits) },
  56         { "ext_intr",   VCPU_STAT(ext_intr_exits) },
  57         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  58         { NULL }
  59 };
  60
  61 /* TODO: use vcpu_printf() */
  62 void kvmppc_dump_vcpu(struct kvm_vcpu *vcpu)
  63 {
  64         int i;
  65
  66         printk("pc:   %08lx msr:  %08llx\n", vcpu->arch.pc, vcpu->arch.shared->msr);
  67         printk("lr:   %08lx ctr:  %08lx\n", vcpu->arch.lr, vcpu->arch.ctr);
  68         printk("srr0: %08llx srr1: %08llx\n", vcpu->arch.shared->srr0,
  69                                             vcpu->arch.shared->srr1);
  70
  71         printk("exceptions: %08lx\n", vcpu->arch.pending_exceptions);
  72
  73         for (i = 0; i < 32; i += 4) {
  74                 printk("gpr%02d: %08lx %08lx %08lx %08lx\n", i,
  75                        kvmppc_get_gpr(vcpu, i),
  76                        kvmppc_get_gpr(vcpu, i+1),
  77                        kvmppc_get_gpr(vcpu, i+2),
  78                        kvmppc_get_gpr(vcpu, i+3));
  79         }
  80 }
  81
  82 #ifdef CONFIG_SPE
  83 void kvmppc_vcpu_disable_spe(struct kvm_vcpu *vcpu)
  84 {
  85         preempt_disable();
  86         enable_kernel_spe();
  87         kvmppc_save_guest_spe(vcpu);
  88         vcpu->arch.shadow_msr &= ~MSR_SPE;
  89         preempt_enable();
  90 }
  91
  92 static void kvmppc_vcpu_enable_spe(struct kvm_vcpu *vcpu)
  93 {
  94         preempt_disable();
  95         enable_kernel_spe();
  96         kvmppc_load_guest_spe(vcpu);
  97         vcpu->arch.shadow_msr |= MSR_SPE;
  98         preempt_enable();
  99 }
 100
 101 static void kvmppc_vcpu_sync_spe(struct kvm_vcpu *vcpu)
 102 {
 103         if (vcpu->arch.shared->msr & MSR_SPE) {
 104                 if (!(vcpu->arch.shadow_msr & MSR_SPE))
 105                         kvmppc_vcpu_enable_spe(vcpu);
 106         } else if (vcpu->arch.shadow_msr & MSR_SPE) {
 107                 kvmppc_vcpu_disable_spe(vcpu);
 108         }
 109 }
 110 #else
 111 static void kvmppc_vcpu_sync_spe(struct kvm_vcpu *vcpu)
 112 {
 113 }
 114 #endif
 115
 116 /*
 117  * Helper function for "full" MSR writes.  No need to call this if only
 118  * EE/CE/ME/DE/RI are changing.
 119  */
 120 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u32 new_msr)
 121 {
 122         u32 old_msr = vcpu->arch.shared->msr;
 123
 124         vcpu->arch.shared->msr = new_msr;
 125
 126         kvmppc_mmu_msr_notify(vcpu, old_msr);
 127         kvmppc_vcpu_sync_spe(vcpu);
 128 }
 129
 130 static void kvmppc_booke_queue_irqprio(struct kvm_vcpu *vcpu,
 131                                        unsigned int priority)
 132 {
 133         set_bit(priority, &vcpu->arch.pending_exceptions);
 134 }
 135
 136 static void kvmppc_core_queue_dtlb_miss(struct kvm_vcpu *vcpu,
 137                                         ulong dear_flags, ulong esr_flags)
 138 {
 139         vcpu->arch.queued_dear = dear_flags;
 140         vcpu->arch.queued_esr = esr_flags;
 141         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DTLB_MISS);
 142 }
 143
 144 static void kvmppc_core_queue_data_storage(struct kvm_vcpu *vcpu,
 145                                            ulong dear_flags, ulong esr_flags)
 146 {
 147         vcpu->arch.queued_dear = dear_flags;
 148         vcpu->arch.queued_esr = esr_flags;
 149         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DATA_STORAGE);
 150 }
 151
 152 static void kvmppc_core_queue_inst_storage(struct kvm_vcpu *vcpu,
 153                                            ulong esr_flags)
 154 {
 155         vcpu->arch.queued_esr = esr_flags;
 156         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_INST_STORAGE);
 157 }
 158
 159 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu, ulong esr_flags)
 160 {
 161         vcpu->arch.queued_esr = esr_flags;
 162         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_PROGRAM);
 163 }
 164
 165 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
 166 {
 167         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DECREMENTER);
 168 }
 169
 170 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
 171 {
 172         return test_bit(BOOKE_IRQPRIO_DECREMENTER, &vcpu->arch.pending_exceptions);
 173 }
 174
 175 void kvmppc_core_dequeue_dec(struct kvm_vcpu *vcpu)
 176 {
 177         clear_bit(BOOKE_IRQPRIO_DECREMENTER, &vcpu->arch.pending_exceptions);
 178 }
 179
 180 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
 181                                 struct kvm_interrupt *irq)
 182 {
 183         unsigned int prio = BOOKE_IRQPRIO_EXTERNAL;
 184
 185         if (irq->irq == KVM_INTERRUPT_SET_LEVEL)
 186                 prio = BOOKE_IRQPRIO_EXTERNAL_LEVEL;
 187
 188         kvmppc_booke_queue_irqprio(vcpu, prio);
 189 }
 190
 191 void kvmppc_core_dequeue_external(struct kvm_vcpu *vcpu,
 192                                   struct kvm_interrupt *irq)
 193 {
 194         clear_bit(BOOKE_IRQPRIO_EXTERNAL, &vcpu->arch.pending_exceptions);
 195         clear_bit(BOOKE_IRQPRIO_EXTERNAL_LEVEL, &vcpu->arch.pending_exceptions);
 196 }
 197
 198 /* Deliver the interrupt of the corresponding priority, if possible. */
 199 static int kvmppc_booke_irqprio_deliver(struct kvm_vcpu *vcpu,
 200                                         unsigned int priority)
 201 {
 202         int allowed = 0;
 203         ulong uninitialized_var(msr_mask);
 204         bool update_esr = false, update_dear = false;
 205         ulong crit_raw = vcpu->arch.shared->critical;
 206         ulong crit_r1 = kvmppc_get_gpr(vcpu, 1);
 207         bool crit;
 208         bool keep_irq = false;
 209
 210         /* Truncate crit indicators in 32 bit mode */
 211         if (!(vcpu->arch.shared->msr & MSR_SF)) {
 212                 crit_raw &= 0xffffffff;
 213                 crit_r1 &= 0xffffffff;
 214         }
 215
 216         /* Critical section when crit == r1 */
 217         crit = (crit_raw == crit_r1);
 218         /* ... and we're in supervisor mode */
 219         crit = crit && !(vcpu->arch.shared->msr & MSR_PR);
 220
 221         if (priority == BOOKE_IRQPRIO_EXTERNAL_LEVEL) {
 222                 priority = BOOKE_IRQPRIO_EXTERNAL;
 223                 keep_irq = true;
 224         }
 225
 226         switch (priority) {
 227         case BOOKE_IRQPRIO_DTLB_MISS:
 228         case BOOKE_IRQPRIO_DATA_STORAGE:
 229                 update_dear = true;
 230                 /* fall through */
 231         case BOOKE_IRQPRIO_INST_STORAGE:
 232         case BOOKE_IRQPRIO_PROGRAM:
 233                 update_esr = true;
 234                 /* fall through */
 235         case BOOKE_IRQPRIO_ITLB_MISS:
 236         case BOOKE_IRQPRIO_SYSCALL:
 237         case BOOKE_IRQPRIO_FP_UNAVAIL:
 238         case BOOKE_IRQPRIO_SPE_UNAVAIL:
 239         case BOOKE_IRQPRIO_SPE_FP_DATA:
 240         case BOOKE_IRQPRIO_SPE_FP_ROUND:
 241         case BOOKE_IRQPRIO_AP_UNAVAIL:
 242         case BOOKE_IRQPRIO_ALIGNMENT:
 243                 allowed = 1;
 244                 msr_mask = MSR_CE|MSR_ME|MSR_DE;
 245                 break;
 246         case BOOKE_IRQPRIO_CRITICAL:
 247         case BOOKE_IRQPRIO_WATCHDOG:
 248                 allowed = vcpu->arch.shared->msr & MSR_CE;
 249                 msr_mask = MSR_ME;
 250                 break;
 251         case BOOKE_IRQPRIO_MACHINE_CHECK:
 252                 allowed = vcpu->arch.shared->msr & MSR_ME;
 253                 msr_mask = 0;
 254                 break;
 255         case BOOKE_IRQPRIO_DECREMENTER:
 256         case BOOKE_IRQPRIO_FIT:
 257                 keep_irq = true;
 258                 /* fall through */
 259         case BOOKE_IRQPRIO_EXTERNAL:
 260                 allowed = vcpu->arch.shared->msr & MSR_EE;
 261                 allowed = allowed && !crit;
 262                 msr_mask = MSR_CE|MSR_ME|MSR_DE;
 263                 break;
 264         case BOOKE_IRQPRIO_DEBUG:
 265                 allowed = vcpu->arch.shared->msr & MSR_DE;
 266                 msr_mask = MSR_ME;
 267                 break;
 268         }
 269
 270         if (allowed) {
 271                 vcpu->arch.shared->srr0 = vcpu->arch.pc;
 272                 vcpu->arch.shared->srr1 = vcpu->arch.shared->msr;
 273                 vcpu->arch.pc = vcpu->arch.ivpr | vcpu->arch.ivor[priority];
 274                 if (update_esr == true)
 275                         vcpu->arch.shared->esr = vcpu->arch.queued_esr;
 276                 if (update_dear == true)
 277                         vcpu->arch.shared->dar = vcpu->arch.queued_dear;
 278                 kvmppc_set_msr(vcpu, vcpu->arch.shared->msr & msr_mask);
 279
 280                 if (!keep_irq)
 281                         clear_bit(priority, &vcpu->arch.pending_exceptions);
 282         }
 283
 284         return allowed;
 285 }
 286
 287 static void update_timer_ints(struct kvm_vcpu *vcpu)
 288 {
 289         if ((vcpu->arch.tcr & TCR_DIE) && (vcpu->arch.tsr & TSR_DIS))
 290                 kvmppc_core_queue_dec(vcpu);
 291         else
 292                 kvmppc_core_dequeue_dec(vcpu);
 293 }
 294
 295 static void kvmppc_core_check_exceptions(struct kvm_vcpu *vcpu)
 296 {
 297         unsigned long *pending = &vcpu->arch.pending_exceptions;
 298         unsigned int priority;
 299
 300         if (vcpu->requests) {
 301                 if (kvm_check_request(KVM_REQ_PENDING_TIMER, vcpu)) {
 302                         smp_mb();
 303                         update_timer_ints(vcpu);
 304                 }
 305         }
 306
 307         priority = __ffs(*pending);
 308         while (priority <= BOOKE_IRQPRIO_MAX) {
 309                 if (kvmppc_booke_irqprio_deliver(vcpu, priority))
 310                         break;
 311
 312                 priority = find_next_bit(pending,
 313                                          BITS_PER_BYTE * sizeof(*pending),
 314                                          priority + 1);
 315         }
 316
 317         /* Tell the guest about our interrupt status */
 318         vcpu->arch.shared->int_pending = !!*pending;
 319 }
 320
 321 /* Check pending exceptions and deliver one, if possible. */
 322 void kvmppc_core_prepare_to_enter(struct kvm_vcpu *vcpu)
 323 {
 324         WARN_ON_ONCE(!irqs_disabled());
 325
 326         kvmppc_core_check_exceptions(vcpu);
 327
 328         if (vcpu->arch.shared->msr & MSR_WE) {
 329                 local_irq_enable();
 330                 kvm_vcpu_block(vcpu);
 331                 local_irq_disable();
 332
 333                 kvmppc_set_exit_type(vcpu, EMULATED_MTMSRWE_EXITS);
 334                 kvmppc_core_check_exceptions(vcpu);
 335         };
 336 }
 337
 338 int kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
 339 {
 340         int ret;
 341
 342         if (!vcpu->arch.sane) {
 343                 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
 344                 return -EINVAL;
 345         }
 346
 347         local_irq_disable();
 348
 349         kvmppc_core_prepare_to_enter(vcpu);
 350
 351         if (signal_pending(current)) {
 352                 kvm_run->exit_reason = KVM_EXIT_INTR;
 353                 ret = -EINTR;
 354                 goto out;
 355         }
 356
 357         kvm_guest_enter();
 358         ret = __kvmppc_vcpu_run(kvm_run, vcpu);
 359         kvm_guest_exit();
 360
 361 out:
 362         local_irq_enable();
 363         return ret;
 364 }
 365
 366 /**
 367  * kvmppc_handle_exit
 368  *
 369  * Return value is in the form (errcode<<2 | RESUME_FLAG_HOST | RESUME_FLAG_NV)
 370  */
 371 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 372                        unsigned int exit_nr)
 373 {
 374         enum emulation_result er;
 375         int r = RESUME_HOST;
 376
 377         /* update before a new last_exit_type is rewritten */
 378         kvmppc_update_timing_stats(vcpu);
 379
 380         local_irq_enable();
 381
 382         run->exit_reason = KVM_EXIT_UNKNOWN;
 383         run->ready_for_interrupt_injection = 1;
 384
 385         switch (exit_nr) {
 386         case BOOKE_INTERRUPT_MACHINE_CHECK:
 387                 printk("MACHINE CHECK: %lx\n", mfspr(SPRN_MCSR));
 388                 kvmppc_dump_vcpu(vcpu);
 389                 r = RESUME_HOST;
 390                 break;
 391
 392         case BOOKE_INTERRUPT_EXTERNAL:
 393                 kvmppc_account_exit(vcpu, EXT_INTR_EXITS);
 394                 if (need_resched())
 395                         cond_resched();
 396                 r = RESUME_GUEST;
 397                 break;
 398
 399         case BOOKE_INTERRUPT_DECREMENTER:
 400                 /* Since we switched IVPR back to the host's value, the host
 401                  * handled this interrupt the moment we enabled interrupts.
 402                  * Now we just offer it a chance to reschedule the guest. */
 403                 kvmppc_account_exit(vcpu, DEC_EXITS);
 404                 if (need_resched())
 405                         cond_resched();
 406                 r = RESUME_GUEST;
 407                 break;
 408
 409         case BOOKE_INTERRUPT_PROGRAM:
 410                 if (vcpu->arch.shared->msr & MSR_PR) {
 411                         /* Program traps generated by user-level software must be handled
 412                          * by the guest kernel. */
 413                         kvmppc_core_queue_program(vcpu, vcpu->arch.fault_esr);
 414                         r = RESUME_GUEST;
 415                         kvmppc_account_exit(vcpu, USR_PR_INST);
 416                         break;
 417                 }
 418
 419                 er = kvmppc_emulate_instruction(run, vcpu);
 420                 switch (er) {
 421                 case EMULATE_DONE:
 422                         /* don't overwrite subtypes, just account kvm_stats */
 423                         kvmppc_account_exit_stat(vcpu, EMULATED_INST_EXITS);
 424                         /* Future optimization: only reload non-volatiles if
 425                          * they were actually modified by emulation. */
 426                         r = RESUME_GUEST_NV;
 427                         break;
 428                 case EMULATE_DO_DCR:
 429                         run->exit_reason = KVM_EXIT_DCR;
 430                         r = RESUME_HOST;
 431                         break;
 432                 case EMULATE_FAIL:
 433                         /* XXX Deliver Program interrupt to guest. */
 434                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 435                                __func__, vcpu->arch.pc, vcpu->arch.last_inst);
 436                         /* For debugging, encode the failing instruction and
 437                          * report it to userspace. */
 438                         run->hw.hardware_exit_reason = ~0ULL << 32;
 439                         run->hw.hardware_exit_reason |= vcpu->arch.last_inst;
 440                         r = RESUME_HOST;
 441                         break;
 442                 default:
 443                         BUG();
 444                 }
 445                 break;
 446
 447         case BOOKE_INTERRUPT_FP_UNAVAIL:
 448                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_FP_UNAVAIL);
 449                 kvmppc_account_exit(vcpu, FP_UNAVAIL);
 450                 r = RESUME_GUEST;
 451                 break;
 452
 453 #ifdef CONFIG_SPE
 454         case BOOKE_INTERRUPT_SPE_UNAVAIL: {
 455                 if (vcpu->arch.shared->msr & MSR_SPE)
 456                         kvmppc_vcpu_enable_spe(vcpu);
 457                 else
 458                         kvmppc_booke_queue_irqprio(vcpu,
 459                                                    BOOKE_IRQPRIO_SPE_UNAVAIL);
 460                 r = RESUME_GUEST;
 461                 break;
 462         }
 463
 464         case BOOKE_INTERRUPT_SPE_FP_DATA:
 465                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SPE_FP_DATA);
 466                 r = RESUME_GUEST;
 467                 break;
 468
 469         case BOOKE_INTERRUPT_SPE_FP_ROUND:
 470                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SPE_FP_ROUND);
 471                 r = RESUME_GUEST;
 472                 break;
 473 #else
 474         case BOOKE_INTERRUPT_SPE_UNAVAIL:
 475                 /*
 476                  * Guest wants SPE, but host kernel doesn't support it.  Send
 477                  * an "unimplemented operation" program check to the guest.
 478                  */
 479                 kvmppc_core_queue_program(vcpu, ESR_PUO | ESR_SPV);
 480                 r = RESUME_GUEST;
 481                 break;
 482
 483         /*
 484          * These really should never happen without CONFIG_SPE,
 485          * as we should never enable the real MSR[SPE] in the guest.
 486          */
 487         case BOOKE_INTERRUPT_SPE_FP_DATA:
 488         case BOOKE_INTERRUPT_SPE_FP_ROUND:
 489                 printk(KERN_CRIT "%s: unexpected SPE interrupt %u at %08lx\n",
 490                        __func__, exit_nr, vcpu->arch.pc);
 491                 run->hw.hardware_exit_reason = exit_nr;
 492                 r = RESUME_HOST;
 493                 break;
 494 #endif
 495
 496         case BOOKE_INTERRUPT_DATA_STORAGE:
 497                 kvmppc_core_queue_data_storage(vcpu, vcpu->arch.fault_dear,
 498                                                vcpu->arch.fault_esr);
 499                 kvmppc_account_exit(vcpu, DSI_EXITS);
 500                 r = RESUME_GUEST;
 501                 break;
 502
 503         case BOOKE_INTERRUPT_INST_STORAGE:
 504                 kvmppc_core_queue_inst_storage(vcpu, vcpu->arch.fault_esr);
 505                 kvmppc_account_exit(vcpu, ISI_EXITS);
 506                 r = RESUME_GUEST;
 507                 break;
 508
 509         case BOOKE_INTERRUPT_SYSCALL:
 510                 if (!(vcpu->arch.shared->msr & MSR_PR) &&
 511                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
 512                         /* KVM PV hypercalls */
 513                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
 514                         r = RESUME_GUEST;
 515                 } else {
 516                         /* Guest syscalls */
 517                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SYSCALL);
 518                 }
 519                 kvmppc_account_exit(vcpu, SYSCALL_EXITS);
 520                 r = RESUME_GUEST;
 521                 break;
 522
 523         case BOOKE_INTERRUPT_DTLB_MISS: {
 524                 unsigned long eaddr = vcpu->arch.fault_dear;
 525                 int gtlb_index;
 526                 gpa_t gpaddr;
 527                 gfn_t gfn;
 528
 529 #ifdef CONFIG_KVM_E500
 530                 if (!(vcpu->arch.shared->msr & MSR_PR) &&
 531                     (eaddr & PAGE_MASK) == vcpu->arch.magic_page_ea) {
 532                         kvmppc_map_magic(vcpu);
 533                         kvmppc_account_exit(vcpu, DTLB_VIRT_MISS_EXITS);
 534                         r = RESUME_GUEST;
 535
 536                         break;
 537                 }
 538 #endif
 539
 540                 /* Check the guest TLB. */
 541                 gtlb_index = kvmppc_mmu_dtlb_index(vcpu, eaddr);
 542                 if (gtlb_index < 0) {
 543                         /* The guest didn't have a mapping for it. */
 544                         kvmppc_core_queue_dtlb_miss(vcpu,
 545                                                     vcpu->arch.fault_dear,
 546                                                     vcpu->arch.fault_esr);
 547                         kvmppc_mmu_dtlb_miss(vcpu);
 548                         kvmppc_account_exit(vcpu, DTLB_REAL_MISS_EXITS);
 549                         r = RESUME_GUEST;
 550                         break;
 551                 }
 552
 553                 gpaddr = kvmppc_mmu_xlate(vcpu, gtlb_index, eaddr);
 554                 gfn = gpaddr >> PAGE_SHIFT;
 555
 556                 if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
 557                         /* The guest TLB had a mapping, but the shadow TLB
 558                          * didn't, and it is RAM. This could be because:
 559                          * a) the entry is mapping the host kernel, or
 560                          * b) the guest used a large mapping which we're faking
 561                          * Either way, we need to satisfy the fault without
 562                          * invoking the guest. */
 563                         kvmppc_mmu_map(vcpu, eaddr, gpaddr, gtlb_index);
 564                         kvmppc_account_exit(vcpu, DTLB_VIRT_MISS_EXITS);
 565                         r = RESUME_GUEST;
 566                 } else {
 567                         /* Guest has mapped and accessed a page which is not
 568                          * actually RAM. */
 569                         vcpu->arch.paddr_accessed = gpaddr;
 570                         r = kvmppc_emulate_mmio(run, vcpu);
 571                         kvmppc_account_exit(vcpu, MMIO_EXITS);
 572                 }
 573
 574                 break;
 575         }
 576
 577         case BOOKE_INTERRUPT_ITLB_MISS: {
 578                 unsigned long eaddr = vcpu->arch.pc;
 579                 gpa_t gpaddr;
 580                 gfn_t gfn;
 581                 int gtlb_index;
 582
 583                 r = RESUME_GUEST;
 584
 585                 /* Check the guest TLB. */
 586                 gtlb_index = kvmppc_mmu_itlb_index(vcpu, eaddr);
 587                 if (gtlb_index < 0) {
 588                         /* The guest didn't have a mapping for it. */
 589                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_ITLB_MISS);
 590                         kvmppc_mmu_itlb_miss(vcpu);
 591                         kvmppc_account_exit(vcpu, ITLB_REAL_MISS_EXITS);
 592                         break;
 593                 }
 594
 595                 kvmppc_account_exit(vcpu, ITLB_VIRT_MISS_EXITS);
 596
 597                 gpaddr = kvmppc_mmu_xlate(vcpu, gtlb_index, eaddr);
 598                 gfn = gpaddr >> PAGE_SHIFT;
 599
 600                 if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
 601                         /* The guest TLB had a mapping, but the shadow TLB
 602                          * didn't. This could be because:
 603                          * a) the entry is mapping the host kernel, or
 604                          * b) the guest used a large mapping which we're faking
 605                          * Either way, we need to satisfy the fault without
 606                          * invoking the guest. */
 607                         kvmppc_mmu_map(vcpu, eaddr, gpaddr, gtlb_index);
 608                 } else {
 609                         /* Guest mapped and leaped at non-RAM! */
 610                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_MACHINE_CHECK);
 611                 }
 612
 613                 break;
 614         }
 615
 616         case BOOKE_INTERRUPT_DEBUG: {
 617                 u32 dbsr;
 618
 619                 vcpu->arch.pc = mfspr(SPRN_CSRR0);
 620
 621                 /* clear IAC events in DBSR register */
 622                 dbsr = mfspr(SPRN_DBSR);
 623                 dbsr &= DBSR_IAC1 | DBSR_IAC2 | DBSR_IAC3 | DBSR_IAC4;
 624                 mtspr(SPRN_DBSR, dbsr);
 625
 626                 run->exit_reason = KVM_EXIT_DEBUG;
 627                 kvmppc_account_exit(vcpu, DEBUG_EXITS);
 628                 r = RESUME_HOST;
 629                 break;
 630         }
 631
 632         default:
 633                 printk(KERN_EMERG "exit_nr %d\n", exit_nr);
 634                 BUG();
 635         }
 636
 637         local_irq_disable();
 638
 639         kvmppc_core_prepare_to_enter(vcpu);
 640
 641         if (!(r & RESUME_HOST)) {
 642                 /* To avoid clobbering exit_reason, only check for signals if
 643                  * we aren't already exiting to userspace for some other
 644                  * reason. */
 645                 if (signal_pending(current)) {
 646                         run->exit_reason = KVM_EXIT_INTR;
 647                         r = (-EINTR << 2) | RESUME_HOST | (r & RESUME_FLAG_NV);
 648                         kvmppc_account_exit(vcpu, SIGNAL_EXITS);
 649                 }
 650         }
 651
 652         return r;
 653 }
 654
 655 /* Initial guest state: 16MB mapping 0 -> 0, PC = 0, MSR = 0, R1 = 16MB */
 656 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
 657 {
 658         int i;
 659         int r;
 660
 661         vcpu->arch.pc = 0;
 662         vcpu->arch.shared->msr = 0;
 663         vcpu->arch.shadow_msr = MSR_USER | MSR_DE | MSR_IS | MSR_DS;
 664         vcpu->arch.shared->pir = vcpu->vcpu_id;
 665         kvmppc_set_gpr(vcpu, 1, (16<<20) - 8); /* -8 for the callee-save LR slot */
 666
 667         vcpu->arch.shadow_pid = 1;
 668
 669         /* Eye-catching numbers so we know if the guest takes an interrupt
 670          * before it's programmed its own IVPR/IVORs. */
 671         vcpu->arch.ivpr = 0x55550000;
 672         for (i = 0; i < BOOKE_IRQPRIO_MAX; i++)
 673                 vcpu->arch.ivor[i] = 0x7700 | i * 4;
 674
 675         kvmppc_init_timing_stats(vcpu);
 676
 677         r = kvmppc_core_vcpu_setup(vcpu);
 678         kvmppc_sanity_check(vcpu);
 679         return r;
 680 }
 681
 682 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 683 {
 684         int i;
 685
 686         regs->pc = vcpu->arch.pc;
 687         regs->cr = kvmppc_get_cr(vcpu);
 688         regs->ctr = vcpu->arch.ctr;
 689         regs->lr = vcpu->arch.lr;
 690         regs->xer = kvmppc_get_xer(vcpu);
 691         regs->msr = vcpu->arch.shared->msr;
 692         regs->srr0 = vcpu->arch.shared->srr0;
 693         regs->srr1 = vcpu->arch.shared->srr1;
 694         regs->pid = vcpu->arch.pid;
 695         regs->sprg0 = vcpu->arch.shared->sprg0;
 696         regs->sprg1 = vcpu->arch.shared->sprg1;
 697         regs->sprg2 = vcpu->arch.shared->sprg2;
 698         regs->sprg3 = vcpu->arch.shared->sprg3;
 699         regs->sprg4 = vcpu->arch.shared->sprg4;
 700         regs->sprg5 = vcpu->arch.shared->sprg5;
 701         regs->sprg6 = vcpu->arch.shared->sprg6;
 702         regs->sprg7 = vcpu->arch.shared->sprg7;
 703
 704         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
 705                 regs->gpr[i] = kvmppc_get_gpr(vcpu, i);
 706
 707         return 0;
 708 }
 709
 710 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 711 {
 712         int i;
 713
 714         vcpu->arch.pc = regs->pc;
 715         kvmppc_set_cr(vcpu, regs->cr);
 716         vcpu->arch.ctr = regs->ctr;
 717         vcpu->arch.lr = regs->lr;
 718         kvmppc_set_xer(vcpu, regs->xer);
 719         kvmppc_set_msr(vcpu, regs->msr);
 720         vcpu->arch.shared->srr0 = regs->srr0;
 721         vcpu->arch.shared->srr1 = regs->srr1;
 722         kvmppc_set_pid(vcpu, regs->pid);
 723         vcpu->arch.shared->sprg0 = regs->sprg0;
 724         vcpu->arch.shared->sprg1 = regs->sprg1;
 725         vcpu->arch.shared->sprg2 = regs->sprg2;
 726         vcpu->arch.shared->sprg3 = regs->sprg3;
 727         vcpu->arch.shared->sprg4 = regs->sprg4;
 728         vcpu->arch.shared->sprg5 = regs->sprg5;
 729         vcpu->arch.shared->sprg6 = regs->sprg6;
 730         vcpu->arch.shared->sprg7 = regs->sprg7;
 731
 732         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
 733                 kvmppc_set_gpr(vcpu, i, regs->gpr[i]);
 734
 735         return 0;
 736 }
 737
 738 static void get_sregs_base(struct kvm_vcpu *vcpu,
 739                            struct kvm_sregs *sregs)
 740 {
 741         u64 tb = get_tb();
 742
 743         sregs->u.e.features |= KVM_SREGS_E_BASE;
 744
 745         sregs->u.e.csrr0 = vcpu->arch.csrr0;
 746         sregs->u.e.csrr1 = vcpu->arch.csrr1;
 747         sregs->u.e.mcsr = vcpu->arch.mcsr;
 748         sregs->u.e.esr = vcpu->arch.shared->esr;
 749         sregs->u.e.dear = vcpu->arch.shared->dar;
 750         sregs->u.e.tsr = vcpu->arch.tsr;
 751         sregs->u.e.tcr = vcpu->arch.tcr;
 752         sregs->u.e.dec = kvmppc_get_dec(vcpu, tb);
 753         sregs->u.e.tb = tb;
 754         sregs->u.e.vrsave = vcpu->arch.vrsave;
 755 }
 756
 757 static int set_sregs_base(struct kvm_vcpu *vcpu,
 758                           struct kvm_sregs *sregs)
 759 {
 760         if (!(sregs->u.e.features & KVM_SREGS_E_BASE))
 761                 return 0;
 762
 763         vcpu->arch.csrr0 = sregs->u.e.csrr0;
 764         vcpu->arch.csrr1 = sregs->u.e.csrr1;
 765         vcpu->arch.mcsr = sregs->u.e.mcsr;
 766         vcpu->arch.shared->esr = sregs->u.e.esr;
 767         vcpu->arch.shared->dar = sregs->u.e.dear;
 768         vcpu->arch.vrsave = sregs->u.e.vrsave;
 769         kvmppc_set_tcr(vcpu, sregs->u.e.tcr);
 770
 771         if (sregs->u.e.update_special & KVM_SREGS_E_UPDATE_DEC) {
 772                 vcpu->arch.dec = sregs->u.e.dec;
 773                 kvmppc_emulate_dec(vcpu);
 774         }
 775
 776         if (sregs->u.e.update_special & KVM_SREGS_E_UPDATE_TSR) {
 777                 vcpu->arch.tsr = sregs->u.e.tsr;
 778                 update_timer_ints(vcpu);
 779         }
 780
 781         return 0;
 782 }
 783
 784 static void get_sregs_arch206(struct kvm_vcpu *vcpu,
 785                               struct kvm_sregs *sregs)
 786 {
 787         sregs->u.e.features |= KVM_SREGS_E_ARCH206;
 788
 789         sregs->u.e.pir = vcpu->vcpu_id;
 790         sregs->u.e.mcsrr0 = vcpu->arch.mcsrr0;
 791         sregs->u.e.mcsrr1 = vcpu->arch.mcsrr1;
 792         sregs->u.e.decar = vcpu->arch.decar;
 793         sregs->u.e.ivpr = vcpu->arch.ivpr;
 794 }
 795
 796 static int set_sregs_arch206(struct kvm_vcpu *vcpu,
 797                              struct kvm_sregs *sregs)
 798 {
 799         if (!(sregs->u.e.features & KVM_SREGS_E_ARCH206))
 800                 return 0;
 801
 802         if (sregs->u.e.pir != vcpu->vcpu_id)
 803                 return -EINVAL;
 804
 805         vcpu->arch.mcsrr0 = sregs->u.e.mcsrr0;
 806         vcpu->arch.mcsrr1 = sregs->u.e.mcsrr1;
 807         vcpu->arch.decar = sregs->u.e.decar;
 808         vcpu->arch.ivpr = sregs->u.e.ivpr;
 809
 810         return 0;
 811 }
 812
 813 void kvmppc_get_sregs_ivor(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
 814 {
 815         sregs->u.e.features |= KVM_SREGS_E_IVOR;
 816
 817         sregs->u.e.ivor_low[0] = vcpu->arch.ivor[BOOKE_IRQPRIO_CRITICAL];
 818         sregs->u.e.ivor_low[1] = vcpu->arch.ivor[BOOKE_IRQPRIO_MACHINE_CHECK];
 819         sregs->u.e.ivor_low[2] = vcpu->arch.ivor[BOOKE_IRQPRIO_DATA_STORAGE];
 820         sregs->u.e.ivor_low[3] = vcpu->arch.ivor[BOOKE_IRQPRIO_INST_STORAGE];
 821         sregs->u.e.ivor_low[4] = vcpu->arch.ivor[BOOKE_IRQPRIO_EXTERNAL];
 822         sregs->u.e.ivor_low[5] = vcpu->arch.ivor[BOOKE_IRQPRIO_ALIGNMENT];
 823         sregs->u.e.ivor_low[6] = vcpu->arch.ivor[BOOKE_IRQPRIO_PROGRAM];
 824         sregs->u.e.ivor_low[7] = vcpu->arch.ivor[BOOKE_IRQPRIO_FP_UNAVAIL];
 825         sregs->u.e.ivor_low[8] = vcpu->arch.ivor[BOOKE_IRQPRIO_SYSCALL];
 826         sregs->u.e.ivor_low[9] = vcpu->arch.ivor[BOOKE_IRQPRIO_AP_UNAVAIL];
 827         sregs->u.e.ivor_low[10] = vcpu->arch.ivor[BOOKE_IRQPRIO_DECREMENTER];
 828         sregs->u.e.ivor_low[11] = vcpu->arch.ivor[BOOKE_IRQPRIO_FIT];
 829         sregs->u.e.ivor_low[12] = vcpu->arch.ivor[BOOKE_IRQPRIO_WATCHDOG];
 830         sregs->u.e.ivor_low[13] = vcpu->arch.ivor[BOOKE_IRQPRIO_DTLB_MISS];
 831         sregs->u.e.ivor_low[14] = vcpu->arch.ivor[BOOKE_IRQPRIO_ITLB_MISS];
 832         sregs->u.e.ivor_low[15] = vcpu->arch.ivor[BOOKE_IRQPRIO_DEBUG];
 833 }
 834
 835 int kvmppc_set_sregs_ivor(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
 836 {
 837         if (!(sregs->u.e.features & KVM_SREGS_E_IVOR))
 838                 return 0;
 839
 840         vcpu->arch.ivor[BOOKE_IRQPRIO_CRITICAL] = sregs->u.e.ivor_low[0];
 841         vcpu->arch.ivor[BOOKE_IRQPRIO_MACHINE_CHECK] = sregs->u.e.ivor_low[1];
 842         vcpu->arch.ivor[BOOKE_IRQPRIO_DATA_STORAGE] = sregs->u.e.ivor_low[2];
 843         vcpu->arch.ivor[BOOKE_IRQPRIO_INST_STORAGE] = sregs->u.e.ivor_low[3];
 844         vcpu->arch.ivor[BOOKE_IRQPRIO_EXTERNAL] = sregs->u.e.ivor_low[4];
 845         vcpu->arch.ivor[BOOKE_IRQPRIO_ALIGNMENT] = sregs->u.e.ivor_low[5];
 846         vcpu->arch.ivor[BOOKE_IRQPRIO_PROGRAM] = sregs->u.e.ivor_low[6];
 847         vcpu->arch.ivor[BOOKE_IRQPRIO_FP_UNAVAIL] = sregs->u.e.ivor_low[7];
 848         vcpu->arch.ivor[BOOKE_IRQPRIO_SYSCALL] = sregs->u.e.ivor_low[8];
 849         vcpu->arch.ivor[BOOKE_IRQPRIO_AP_UNAVAIL] = sregs->u.e.ivor_low[9];
 850         vcpu->arch.ivor[BOOKE_IRQPRIO_DECREMENTER] = sregs->u.e.ivor_low[10];
 851         vcpu->arch.ivor[BOOKE_IRQPRIO_FIT] = sregs->u.e.ivor_low[11];
 852         vcpu->arch.ivor[BOOKE_IRQPRIO_WATCHDOG] = sregs->u.e.ivor_low[12];
 853         vcpu->arch.ivor[BOOKE_IRQPRIO_DTLB_MISS] = sregs->u.e.ivor_low[13];
 854         vcpu->arch.ivor[BOOKE_IRQPRIO_ITLB_MISS] = sregs->u.e.ivor_low[14];
 855         vcpu->arch.ivor[BOOKE_IRQPRIO_DEBUG] = sregs->u.e.ivor_low[15];
 856
 857         return 0;
 858 }
 859
 860 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 861                                   struct kvm_sregs *sregs)
 862 {
 863         sregs->pvr = vcpu->arch.pvr;
 864
 865         get_sregs_base(vcpu, sregs);
 866         get_sregs_arch206(vcpu, sregs);
 867         kvmppc_core_get_sregs(vcpu, sregs);
 868         return 0;
 869 }
 870
 871 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 872                                   struct kvm_sregs *sregs)
 873 {
 874         int ret;
 875
 876         if (vcpu->arch.pvr != sregs->pvr)
 877                 return -EINVAL;
 878
 879         ret = set_sregs_base(vcpu, sregs);
 880         if (ret < 0)
 881                 return ret;
 882
 883         ret = set_sregs_arch206(vcpu, sregs);
 884         if (ret < 0)
 885                 return ret;
 886
 887         return kvmppc_core_set_sregs(vcpu, sregs);
 888 }
 889
 890 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
 891 {
 892         return -EINVAL;
 893 }
 894
 895 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
 896 {
 897         return -EINVAL;
 898 }
 899
 900 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 901 {
 902         return -ENOTSUPP;
 903 }
 904
 905 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 906 {
 907         return -ENOTSUPP;
 908 }
 909
 910 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
 911                                   struct kvm_translation *tr)
 912 {
 913         int r;
 914
 915         r = kvmppc_core_vcpu_translate(vcpu, tr);
 916         return r;
 917 }
 918
 919 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 920 {
 921         return -ENOTSUPP;
 922 }
 923
 924 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
 925                                       struct kvm_userspace_memory_region *mem)
 926 {
 927         return 0;
 928 }
 929
 930 void kvmppc_core_commit_memory_region(struct kvm *kvm,
 931                                 struct kvm_userspace_memory_region *mem)
 932 {
 933 }
 934
 935 int kvmppc_core_init_vm(struct kvm *kvm)
 936 {
 937         return 0;
 938 }
 939
 940 void kvmppc_core_destroy_vm(struct kvm *kvm)
 941 {
 942 }
 943
 944 void kvmppc_set_tcr(struct kvm_vcpu *vcpu, u32 new_tcr)
 945 {
 946         vcpu->arch.tcr = new_tcr;
 947         update_timer_ints(vcpu);
 948 }
 949
 950 void kvmppc_set_tsr_bits(struct kvm_vcpu *vcpu, u32 tsr_bits)
 951 {
 952         set_bits(tsr_bits, &vcpu->arch.tsr);
 953         smp_wmb();
 954         kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
 955         kvm_vcpu_kick(vcpu);
 956 }
 957
 958 void kvmppc_clr_tsr_bits(struct kvm_vcpu *vcpu, u32 tsr_bits)
 959 {
 960         clear_bits(tsr_bits, &vcpu->arch.tsr);
 961         update_timer_ints(vcpu);
 962 }
 963
 964 void kvmppc_decrementer_func(unsigned long data)
 965 {
 966         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)data;
 967
 968         kvmppc_set_tsr_bits(vcpu, TSR_DIS);
 969 }
 970
 971 int __init kvmppc_booke_init(void)
 972 {
 973         unsigned long ivor[16];
 974         unsigned long max_ivor = 0;
 975         int i;
 976
 977         /* We install our own exception handlers by hijacking IVPR. IVPR must
 978          * be 16-bit aligned, so we need a 64KB allocation. */
 979         kvmppc_booke_handlers = __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 980                                                  VCPU_SIZE_ORDER);
 981         if (!kvmppc_booke_handlers)
 982                 return -ENOMEM;
 983
 984         /* XXX make sure our handlers are smaller than Linux's */
 985
 986         /* Copy our interrupt handlers to match host IVORs. That way we don't
 987          * have to swap the IVORs on every guest/host transition. */
 988         ivor[0] = mfspr(SPRN_IVOR0);
 989         ivor[1] = mfspr(SPRN_IVOR1);
 990         ivor[2] = mfspr(SPRN_IVOR2);
 991         ivor[3] = mfspr(SPRN_IVOR3);
 992         ivor[4] = mfspr(SPRN_IVOR4);
 993         ivor[5] = mfspr(SPRN_IVOR5);
 994         ivor[6] = mfspr(SPRN_IVOR6);
 995         ivor[7] = mfspr(SPRN_IVOR7);
 996         ivor[8] = mfspr(SPRN_IVOR8);
 997         ivor[9] = mfspr(SPRN_IVOR9);
 998         ivor[10] = mfspr(SPRN_IVOR10);
 999         ivor[11] = mfspr(SPRN_IVOR11);
1000         ivor[12] = mfspr(SPRN_IVOR12);
1001         ivor[13] = mfspr(SPRN_IVOR13);
1002         ivor[14] = mfspr(SPRN_IVOR14);
1003         ivor[15] = mfspr(SPRN_IVOR15);
1004
1005         for (i = 0; i < 16; i++) {
1006                 if (ivor[i] > max_ivor)
1007                         max_ivor = ivor[i];
1008
1009                 memcpy((void *)kvmppc_booke_handlers + ivor[i],
1010                        kvmppc_handlers_start + i * kvmppc_handler_len,
1011                        kvmppc_handler_len);
1012         }
1013         flush_icache_range(kvmppc_booke_handlers,
1014                            kvmppc_booke_handlers + max_ivor + kvmppc_handler_len);
1015
1016         return 0;
1017 }
1018
1019 void __exit kvmppc_booke_exit(void)
1020 {
1021         free_pages(kvmppc_booke_handlers, VCPU_SIZE_ORDER);
1022         kvm_exit();
1023 }