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