SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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  *
  17  * Authors: Hollis Blanchard <hollisb@us.ibm.com>
  18  *          Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
  19  */
  20
  21 #include <linux/errno.h>
  22 #include <linux/err.h>
  23 #include <linux/kvm_host.h>
  24 #include <linux/module.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/fs.h>
  27 #include <asm/cputable.h>
  28 #include <asm/uaccess.h>
  29 #include <asm/kvm_ppc.h>
  30 #include <asm/cacheflush.h>
  31
  32 #include "booke.h"
  33 #include "44x_tlb.h"
  34
  35 unsigned long kvmppc_booke_handlers;
  36
  37 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  38 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  39
  40 struct kvm_stats_debugfs_item debugfs_entries[] = {
  41         { "exits",      VCPU_STAT(sum_exits) },
  42         { "mmio",       VCPU_STAT(mmio_exits) },
  43         { "dcr",        VCPU_STAT(dcr_exits) },
  44         { "sig",        VCPU_STAT(signal_exits) },
  45         { "light",      VCPU_STAT(light_exits) },
  46         { "itlb_r",     VCPU_STAT(itlb_real_miss_exits) },
  47         { "itlb_v",     VCPU_STAT(itlb_virt_miss_exits) },
  48         { "dtlb_r",     VCPU_STAT(dtlb_real_miss_exits) },
  49         { "dtlb_v",     VCPU_STAT(dtlb_virt_miss_exits) },
  50         { "sysc",       VCPU_STAT(syscall_exits) },
  51         { "isi",        VCPU_STAT(isi_exits) },
  52         { "dsi",        VCPU_STAT(dsi_exits) },
  53         { "inst_emu",   VCPU_STAT(emulated_inst_exits) },
  54         { "dec",        VCPU_STAT(dec_exits) },
  55         { "ext_intr",   VCPU_STAT(ext_intr_exits) },
  56         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  57         { NULL }
  58 };
  59
  60 static const u32 interrupt_msr_mask[16] = {
  61         [BOOKE_INTERRUPT_CRITICAL]      = MSR_ME,
  62         [BOOKE_INTERRUPT_MACHINE_CHECK] = 0,
  63         [BOOKE_INTERRUPT_DATA_STORAGE]  = MSR_CE|MSR_ME|MSR_DE,
  64         [BOOKE_INTERRUPT_INST_STORAGE]  = MSR_CE|MSR_ME|MSR_DE,
  65         [BOOKE_INTERRUPT_EXTERNAL]      = MSR_CE|MSR_ME|MSR_DE,
  66         [BOOKE_INTERRUPT_ALIGNMENT]     = MSR_CE|MSR_ME|MSR_DE,
  67         [BOOKE_INTERRUPT_PROGRAM]       = MSR_CE|MSR_ME|MSR_DE,
  68         [BOOKE_INTERRUPT_FP_UNAVAIL]    = MSR_CE|MSR_ME|MSR_DE,
  69         [BOOKE_INTERRUPT_SYSCALL]       = MSR_CE|MSR_ME|MSR_DE,
  70         [BOOKE_INTERRUPT_AP_UNAVAIL]    = MSR_CE|MSR_ME|MSR_DE,
  71         [BOOKE_INTERRUPT_DECREMENTER]   = MSR_CE|MSR_ME|MSR_DE,
  72         [BOOKE_INTERRUPT_FIT]           = MSR_CE|MSR_ME|MSR_DE,
  73         [BOOKE_INTERRUPT_WATCHDOG]      = MSR_ME,
  74         [BOOKE_INTERRUPT_DTLB_MISS]     = MSR_CE|MSR_ME|MSR_DE,
  75         [BOOKE_INTERRUPT_ITLB_MISS]     = MSR_CE|MSR_ME|MSR_DE,
  76         [BOOKE_INTERRUPT_DEBUG]         = MSR_ME,
  77 };
  78
  79 const unsigned char exception_priority[] = {
  80         [BOOKE_INTERRUPT_DATA_STORAGE] = 0,
  81         [BOOKE_INTERRUPT_INST_STORAGE] = 1,
  82         [BOOKE_INTERRUPT_ALIGNMENT] = 2,
  83         [BOOKE_INTERRUPT_PROGRAM] = 3,
  84         [BOOKE_INTERRUPT_FP_UNAVAIL] = 4,
  85         [BOOKE_INTERRUPT_SYSCALL] = 5,
  86         [BOOKE_INTERRUPT_AP_UNAVAIL] = 6,
  87         [BOOKE_INTERRUPT_DTLB_MISS] = 7,
  88         [BOOKE_INTERRUPT_ITLB_MISS] = 8,
  89         [BOOKE_INTERRUPT_MACHINE_CHECK] = 9,
  90         [BOOKE_INTERRUPT_DEBUG] = 10,
  91         [BOOKE_INTERRUPT_CRITICAL] = 11,
  92         [BOOKE_INTERRUPT_WATCHDOG] = 12,
  93         [BOOKE_INTERRUPT_EXTERNAL] = 13,
  94         [BOOKE_INTERRUPT_FIT] = 14,
  95         [BOOKE_INTERRUPT_DECREMENTER] = 15,
  96 };
  97
  98 const unsigned char priority_exception[] = {
  99         BOOKE_INTERRUPT_DATA_STORAGE,
 100         BOOKE_INTERRUPT_INST_STORAGE,
 101         BOOKE_INTERRUPT_ALIGNMENT,
 102         BOOKE_INTERRUPT_PROGRAM,
 103         BOOKE_INTERRUPT_FP_UNAVAIL,
 104         BOOKE_INTERRUPT_SYSCALL,
 105         BOOKE_INTERRUPT_AP_UNAVAIL,
 106         BOOKE_INTERRUPT_DTLB_MISS,
 107         BOOKE_INTERRUPT_ITLB_MISS,
 108         BOOKE_INTERRUPT_MACHINE_CHECK,
 109         BOOKE_INTERRUPT_DEBUG,
 110         BOOKE_INTERRUPT_CRITICAL,
 111         BOOKE_INTERRUPT_WATCHDOG,
 112         BOOKE_INTERRUPT_EXTERNAL,
 113         BOOKE_INTERRUPT_FIT,
 114         BOOKE_INTERRUPT_DECREMENTER,
 115 };
 116
 117
 118 /* TODO: use vcpu_printf() */
 119 void kvmppc_dump_vcpu(struct kvm_vcpu *vcpu)
 120 {
 121         int i;
 122
 123         printk("pc:   %08x msr:  %08x\n", vcpu->arch.pc, vcpu->arch.msr);
 124         printk("lr:   %08x ctr:  %08x\n", vcpu->arch.lr, vcpu->arch.ctr);
 125         printk("srr0: %08x srr1: %08x\n", vcpu->arch.srr0, vcpu->arch.srr1);
 126
 127         printk("exceptions: %08lx\n", vcpu->arch.pending_exceptions);
 128
 129         for (i = 0; i < 32; i += 4) {
 130                 printk("gpr%02d: %08x %08x %08x %08x\n", i,
 131                        vcpu->arch.gpr[i],
 132                        vcpu->arch.gpr[i+1],
 133                        vcpu->arch.gpr[i+2],
 134                        vcpu->arch.gpr[i+3]);
 135         }
 136 }
 137
 138 static void kvmppc_booke_queue_exception(struct kvm_vcpu *vcpu, int exception)
 139 {
 140         unsigned int priority = exception_priority[exception];
 141         set_bit(priority, &vcpu->arch.pending_exceptions);
 142 }
 143
 144 static void kvmppc_booke_clear_exception(struct kvm_vcpu *vcpu, int exception)
 145 {
 146         unsigned int priority = exception_priority[exception];
 147         clear_bit(priority, &vcpu->arch.pending_exceptions);
 148 }
 149
 150 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu)
 151 {
 152         kvmppc_booke_queue_exception(vcpu, BOOKE_INTERRUPT_PROGRAM);
 153 }
 154
 155 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
 156 {
 157         kvmppc_booke_queue_exception(vcpu, BOOKE_INTERRUPT_DECREMENTER);
 158 }
 159
 160 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
 161 {
 162         unsigned int priority = exception_priority[BOOKE_INTERRUPT_DECREMENTER];
 163         return test_bit(priority, &vcpu->arch.pending_exceptions);
 164 }
 165
 166 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
 167                                 struct kvm_interrupt *irq)
 168 {
 169         kvmppc_booke_queue_exception(vcpu, BOOKE_INTERRUPT_EXTERNAL);
 170 }
 171
 172 /* Check if we are ready to deliver the interrupt */
 173 static int kvmppc_can_deliver_interrupt(struct kvm_vcpu *vcpu, int interrupt)
 174 {
 175         int r;
 176
 177         switch (interrupt) {
 178         case BOOKE_INTERRUPT_CRITICAL:
 179                 r = vcpu->arch.msr & MSR_CE;
 180                 break;
 181         case BOOKE_INTERRUPT_MACHINE_CHECK:
 182                 r = vcpu->arch.msr & MSR_ME;
 183                 break;
 184         case BOOKE_INTERRUPT_EXTERNAL:
 185                 r = vcpu->arch.msr & MSR_EE;
 186                 break;
 187         case BOOKE_INTERRUPT_DECREMENTER:
 188                 r = vcpu->arch.msr & MSR_EE;
 189                 break;
 190         case BOOKE_INTERRUPT_FIT:
 191                 r = vcpu->arch.msr & MSR_EE;
 192                 break;
 193         case BOOKE_INTERRUPT_WATCHDOG:
 194                 r = vcpu->arch.msr & MSR_CE;
 195                 break;
 196         case BOOKE_INTERRUPT_DEBUG:
 197                 r = vcpu->arch.msr & MSR_DE;
 198                 break;
 199         default:
 200                 r = 1;
 201         }
 202
 203         return r;
 204 }
 205
 206 static void kvmppc_booke_deliver_interrupt(struct kvm_vcpu *vcpu, int interrupt)
 207 {
 208         switch (interrupt) {
 209         case BOOKE_INTERRUPT_DECREMENTER:
 210                 vcpu->arch.tsr |= TSR_DIS;
 211                 break;
 212         }
 213
 214         vcpu->arch.srr0 = vcpu->arch.pc;
 215         vcpu->arch.srr1 = vcpu->arch.msr;
 216         vcpu->arch.pc = vcpu->arch.ivpr | vcpu->arch.ivor[interrupt];
 217         kvmppc_set_msr(vcpu, vcpu->arch.msr & interrupt_msr_mask[interrupt]);
 218 }
 219
 220 /* Check pending exceptions and deliver one, if possible. */
 221 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
 222 {
 223         unsigned long *pending = &vcpu->arch.pending_exceptions;
 224         unsigned int exception;
 225         unsigned int priority;
 226
 227         priority = find_first_bit(pending, BITS_PER_BYTE * sizeof(*pending));
 228         while (priority <= BOOKE_MAX_INTERRUPT) {
 229                 exception = priority_exception[priority];
 230                 if (kvmppc_can_deliver_interrupt(vcpu, exception)) {
 231                         kvmppc_booke_clear_exception(vcpu, exception);
 232                         kvmppc_booke_deliver_interrupt(vcpu, exception);
 233                         break;
 234                 }
 235
 236                 priority = find_next_bit(pending,
 237                                          BITS_PER_BYTE * sizeof(*pending),
 238                                          priority + 1);
 239         }
 240 }
 241
 242 /**
 243  * kvmppc_handle_exit
 244  *
 245  * Return value is in the form (errcode<<2 | RESUME_FLAG_HOST | RESUME_FLAG_NV)
 246  */
 247 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 248                        unsigned int exit_nr)
 249 {
 250         enum emulation_result er;
 251         int r = RESUME_HOST;
 252
 253         local_irq_enable();
 254
 255         run->exit_reason = KVM_EXIT_UNKNOWN;
 256         run->ready_for_interrupt_injection = 1;
 257
 258         switch (exit_nr) {
 259         case BOOKE_INTERRUPT_MACHINE_CHECK:
 260                 printk("MACHINE CHECK: %lx\n", mfspr(SPRN_MCSR));
 261                 kvmppc_dump_vcpu(vcpu);
 262                 r = RESUME_HOST;
 263                 break;
 264
 265         case BOOKE_INTERRUPT_EXTERNAL:
 266         case BOOKE_INTERRUPT_DECREMENTER:
 267                 /* Since we switched IVPR back to the host's value, the host
 268                  * handled this interrupt the moment we enabled interrupts.
 269                  * Now we just offer it a chance to reschedule the guest. */
 270
 271                 /* XXX At this point the TLB still holds our shadow TLB, so if
 272                  * we do reschedule the host will fault over it. Perhaps we
 273                  * should politely restore the host's entries to minimize
 274                  * misses before ceding control. */
 275                 if (need_resched())
 276                         cond_resched();
 277                 if (exit_nr == BOOKE_INTERRUPT_DECREMENTER)
 278                         vcpu->stat.dec_exits++;
 279                 else
 280                         vcpu->stat.ext_intr_exits++;
 281                 r = RESUME_GUEST;
 282                 break;
 283
 284         case BOOKE_INTERRUPT_PROGRAM:
 285                 if (vcpu->arch.msr & MSR_PR) {
 286                         /* Program traps generated by user-level software must be handled
 287                          * by the guest kernel. */
 288                         vcpu->arch.esr = vcpu->arch.fault_esr;
 289                         kvmppc_booke_queue_exception(vcpu, BOOKE_INTERRUPT_PROGRAM);
 290                         r = RESUME_GUEST;
 291                         break;
 292                 }
 293
 294                 er = kvmppc_emulate_instruction(run, vcpu);
 295                 switch (er) {
 296                 case EMULATE_DONE:
 297                         /* Future optimization: only reload non-volatiles if
 298                          * they were actually modified by emulation. */
 299                         vcpu->stat.emulated_inst_exits++;
 300                         r = RESUME_GUEST_NV;
 301                         break;
 302                 case EMULATE_DO_DCR:
 303                         run->exit_reason = KVM_EXIT_DCR;
 304                         r = RESUME_HOST;
 305                         break;
 306                 case EMULATE_FAIL:
 307                         /* XXX Deliver Program interrupt to guest. */
 308                         printk(KERN_CRIT "%s: emulation at %x failed (%08x)\n",
 309                                __func__, vcpu->arch.pc, vcpu->arch.last_inst);
 310                         /* For debugging, encode the failing instruction and
 311                          * report it to userspace. */
 312                         run->hw.hardware_exit_reason = ~0ULL << 32;
 313                         run->hw.hardware_exit_reason |= vcpu->arch.last_inst;
 314                         r = RESUME_HOST;
 315                         break;
 316                 default:
 317                         BUG();
 318                 }
 319                 break;
 320
 321         case BOOKE_INTERRUPT_FP_UNAVAIL:
 322                 kvmppc_booke_queue_exception(vcpu, exit_nr);
 323                 r = RESUME_GUEST;
 324                 break;
 325
 326         case BOOKE_INTERRUPT_DATA_STORAGE:
 327                 vcpu->arch.dear = vcpu->arch.fault_dear;
 328                 vcpu->arch.esr = vcpu->arch.fault_esr;
 329                 kvmppc_booke_queue_exception(vcpu, exit_nr);
 330                 vcpu->stat.dsi_exits++;
 331                 r = RESUME_GUEST;
 332                 break;
 333
 334         case BOOKE_INTERRUPT_INST_STORAGE:
 335                 vcpu->arch.esr = vcpu->arch.fault_esr;
 336                 kvmppc_booke_queue_exception(vcpu, exit_nr);
 337                 vcpu->stat.isi_exits++;
 338                 r = RESUME_GUEST;
 339                 break;
 340
 341         case BOOKE_INTERRUPT_SYSCALL:
 342                 kvmppc_booke_queue_exception(vcpu, exit_nr);
 343                 vcpu->stat.syscall_exits++;
 344                 r = RESUME_GUEST;
 345                 break;
 346
 347         case BOOKE_INTERRUPT_DTLB_MISS: {
 348                 struct kvmppc_44x_tlbe *gtlbe;
 349                 unsigned long eaddr = vcpu->arch.fault_dear;
 350                 gfn_t gfn;
 351
 352                 /* Check the guest TLB. */
 353                 gtlbe = kvmppc_44x_dtlb_search(vcpu, eaddr);
 354                 if (!gtlbe) {
 355                         /* The guest didn't have a mapping for it. */
 356                         kvmppc_booke_queue_exception(vcpu, exit_nr);
 357                         vcpu->arch.dear = vcpu->arch.fault_dear;
 358                         vcpu->arch.esr = vcpu->arch.fault_esr;
 359                         vcpu->stat.dtlb_real_miss_exits++;
 360                         r = RESUME_GUEST;
 361                         break;
 362                 }
 363
 364                 vcpu->arch.paddr_accessed = tlb_xlate(gtlbe, eaddr);
 365                 gfn = vcpu->arch.paddr_accessed >> PAGE_SHIFT;
 366
 367                 if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
 368                         /* The guest TLB had a mapping, but the shadow TLB
 369                          * didn't, and it is RAM. This could be because:
 370                          * a) the entry is mapping the host kernel, or
 371                          * b) the guest used a large mapping which we're faking
 372                          * Either way, we need to satisfy the fault without
 373                          * invoking the guest. */
 374                         kvmppc_mmu_map(vcpu, eaddr, gfn, gtlbe->tid,
 375                                        gtlbe->word2);
 376                         vcpu->stat.dtlb_virt_miss_exits++;
 377                         r = RESUME_GUEST;
 378                 } else {
 379                         /* Guest has mapped and accessed a page which is not
 380                          * actually RAM. */
 381                         r = kvmppc_emulate_mmio(run, vcpu);
 382                 }
 383
 384                 break;
 385         }
 386
 387         case BOOKE_INTERRUPT_ITLB_MISS: {
 388                 struct kvmppc_44x_tlbe *gtlbe;
 389                 unsigned long eaddr = vcpu->arch.pc;
 390                 gfn_t gfn;
 391
 392                 r = RESUME_GUEST;
 393
 394                 /* Check the guest TLB. */
 395                 gtlbe = kvmppc_44x_itlb_search(vcpu, eaddr);
 396                 if (!gtlbe) {
 397                         /* The guest didn't have a mapping for it. */
 398                         kvmppc_booke_queue_exception(vcpu, exit_nr);
 399                         vcpu->stat.itlb_real_miss_exits++;
 400                         break;
 401                 }
 402
 403                 vcpu->stat.itlb_virt_miss_exits++;
 404
 405                 gfn = tlb_xlate(gtlbe, eaddr) >> PAGE_SHIFT;
 406
 407                 if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
 408                         /* The guest TLB had a mapping, but the shadow TLB
 409                          * didn't. This could be because:
 410                          * a) the entry is mapping the host kernel, or
 411                          * b) the guest used a large mapping which we're faking
 412                          * Either way, we need to satisfy the fault without
 413                          * invoking the guest. */
 414                         kvmppc_mmu_map(vcpu, eaddr, gfn, gtlbe->tid,
 415                                        gtlbe->word2);
 416                 } else {
 417                         /* Guest mapped and leaped at non-RAM! */
 418                         kvmppc_booke_queue_exception(vcpu, BOOKE_INTERRUPT_MACHINE_CHECK);
 419                 }
 420
 421                 break;
 422         }
 423
 424         case BOOKE_INTERRUPT_DEBUG: {
 425                 u32 dbsr;
 426
 427                 vcpu->arch.pc = mfspr(SPRN_CSRR0);
 428
 429                 /* clear IAC events in DBSR register */
 430                 dbsr = mfspr(SPRN_DBSR);
 431                 dbsr &= DBSR_IAC1 | DBSR_IAC2 | DBSR_IAC3 | DBSR_IAC4;
 432                 mtspr(SPRN_DBSR, dbsr);
 433
 434                 run->exit_reason = KVM_EXIT_DEBUG;
 435                 r = RESUME_HOST;
 436                 break;
 437         }
 438
 439         default:
 440                 printk(KERN_EMERG "exit_nr %d\n", exit_nr);
 441                 BUG();
 442         }
 443
 444         local_irq_disable();
 445
 446         kvmppc_core_deliver_interrupts(vcpu);
 447
 448         /* Do some exit accounting. */
 449         vcpu->stat.sum_exits++;
 450         if (!(r & RESUME_HOST)) {
 451                 /* To avoid clobbering exit_reason, only check for signals if
 452                  * we aren't already exiting to userspace for some other
 453                  * reason. */
 454                 if (signal_pending(current)) {
 455                         run->exit_reason = KVM_EXIT_INTR;
 456                         r = (-EINTR << 2) | RESUME_HOST | (r & RESUME_FLAG_NV);
 457
 458                         vcpu->stat.signal_exits++;
 459                 } else {
 460                         vcpu->stat.light_exits++;
 461                 }
 462         } else {
 463                 switch (run->exit_reason) {
 464                 case KVM_EXIT_MMIO:
 465                         vcpu->stat.mmio_exits++;
 466                         break;
 467                 case KVM_EXIT_DCR:
 468                         vcpu->stat.dcr_exits++;
 469                         break;
 470                 case KVM_EXIT_INTR:
 471                         vcpu->stat.signal_exits++;
 472                         break;
 473                 }
 474         }
 475
 476         return r;
 477 }
 478
 479 /* Initial guest state: 16MB mapping 0 -> 0, PC = 0, MSR = 0, R1 = 16MB */
 480 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
 481 {
 482         struct kvmppc_44x_tlbe *tlbe = &vcpu->arch.guest_tlb[0];
 483
 484         tlbe->tid = 0;
 485         tlbe->word0 = PPC44x_TLB_16M | PPC44x_TLB_VALID;
 486         tlbe->word1 = 0;
 487         tlbe->word2 = PPC44x_TLB_SX | PPC44x_TLB_SW | PPC44x_TLB_SR;
 488
 489         tlbe++;
 490         tlbe->tid = 0;
 491         tlbe->word0 = 0xef600000 | PPC44x_TLB_4K | PPC44x_TLB_VALID;
 492         tlbe->word1 = 0xef600000;
 493         tlbe->word2 = PPC44x_TLB_SX | PPC44x_TLB_SW | PPC44x_TLB_SR
 494                       | PPC44x_TLB_I | PPC44x_TLB_G;
 495
 496         vcpu->arch.pc = 0;
 497         vcpu->arch.msr = 0;
 498         vcpu->arch.gpr[1] = (16<<20) - 8; /* -8 for the callee-save LR slot */
 499
 500         vcpu->arch.shadow_pid = 1;
 501
 502         /* Eye-catching number so we know if the guest takes an interrupt
 503          * before it's programmed its own IVPR. */
 504         vcpu->arch.ivpr = 0x55550000;
 505
 506         /* Since the guest can directly access the timebase, it must know the
 507          * real timebase frequency. Accordingly, it must see the state of
 508          * CCR1[TCS]. */
 509         vcpu->arch.ccr1 = mfspr(SPRN_CCR1);
 510
 511         return 0;
 512 }
 513
 514 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 515 {
 516         int i;
 517
 518         regs->pc = vcpu->arch.pc;
 519         regs->cr = vcpu->arch.cr;
 520         regs->ctr = vcpu->arch.ctr;
 521         regs->lr = vcpu->arch.lr;
 522         regs->xer = vcpu->arch.xer;
 523         regs->msr = vcpu->arch.msr;
 524         regs->srr0 = vcpu->arch.srr0;
 525         regs->srr1 = vcpu->arch.srr1;
 526         regs->pid = vcpu->arch.pid;
 527         regs->sprg0 = vcpu->arch.sprg0;
 528         regs->sprg1 = vcpu->arch.sprg1;
 529         regs->sprg2 = vcpu->arch.sprg2;
 530         regs->sprg3 = vcpu->arch.sprg3;
 531         regs->sprg5 = vcpu->arch.sprg4;
 532         regs->sprg6 = vcpu->arch.sprg5;
 533         regs->sprg7 = vcpu->arch.sprg6;
 534
 535         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
 536                 regs->gpr[i] = vcpu->arch.gpr[i];
 537
 538         return 0;
 539 }
 540
 541 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 542 {
 543         int i;
 544
 545         vcpu->arch.pc = regs->pc;
 546         vcpu->arch.cr = regs->cr;
 547         vcpu->arch.ctr = regs->ctr;
 548         vcpu->arch.lr = regs->lr;
 549         vcpu->arch.xer = regs->xer;
 550         vcpu->arch.msr = regs->msr;
 551         vcpu->arch.srr0 = regs->srr0;
 552         vcpu->arch.srr1 = regs->srr1;
 553         vcpu->arch.sprg0 = regs->sprg0;
 554         vcpu->arch.sprg1 = regs->sprg1;
 555         vcpu->arch.sprg2 = regs->sprg2;
 556         vcpu->arch.sprg3 = regs->sprg3;
 557         vcpu->arch.sprg5 = regs->sprg4;
 558         vcpu->arch.sprg6 = regs->sprg5;
 559         vcpu->arch.sprg7 = regs->sprg6;
 560
 561         for (i = 0; i < ARRAY_SIZE(vcpu->arch.gpr); i++)
 562                 vcpu->arch.gpr[i] = regs->gpr[i];
 563
 564         return 0;
 565 }
 566
 567 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 568                                   struct kvm_sregs *sregs)
 569 {
 570         return -ENOTSUPP;
 571 }
 572
 573 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 574                                   struct kvm_sregs *sregs)
 575 {
 576         return -ENOTSUPP;
 577 }
 578
 579 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 580 {
 581         return -ENOTSUPP;
 582 }
 583
 584 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 585 {
 586         return -ENOTSUPP;
 587 }
 588
 589 /* 'linear_address' is actually an encoding of AS|PID|EADDR . */
 590 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
 591                                   struct kvm_translation *tr)
 592 {
 593         struct kvmppc_44x_tlbe *gtlbe;
 594         int index;
 595         gva_t eaddr;
 596         u8 pid;
 597         u8 as;
 598
 599         eaddr = tr->linear_address;
 600         pid = (tr->linear_address >> 32) & 0xff;
 601         as = (tr->linear_address >> 40) & 0x1;
 602
 603         index = kvmppc_44x_tlb_index(vcpu, eaddr, pid, as);
 604         if (index == -1) {
 605                 tr->valid = 0;
 606                 return 0;
 607         }
 608
 609         gtlbe = &vcpu->arch.guest_tlb[index];
 610
 611         tr->physical_address = tlb_xlate(gtlbe, eaddr);
 612         /* XXX what does "writeable" and "usermode" even mean? */
 613         tr->valid = 1;
 614
 615         return 0;
 616 }
 617
 618 static int kvmppc_booke_init(void)
 619 {
 620         unsigned long ivor[16];
 621         unsigned long max_ivor = 0;
 622         int i;
 623
 624         /* We install our own exception handlers by hijacking IVPR. IVPR must
 625          * be 16-bit aligned, so we need a 64KB allocation. */
 626         kvmppc_booke_handlers = __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 627                                                  VCPU_SIZE_ORDER);
 628         if (!kvmppc_booke_handlers)
 629                 return -ENOMEM;
 630
 631         /* XXX make sure our handlers are smaller than Linux's */
 632
 633         /* Copy our interrupt handlers to match host IVORs. That way we don't
 634          * have to swap the IVORs on every guest/host transition. */
 635         ivor[0] = mfspr(SPRN_IVOR0);
 636         ivor[1] = mfspr(SPRN_IVOR1);
 637         ivor[2] = mfspr(SPRN_IVOR2);
 638         ivor[3] = mfspr(SPRN_IVOR3);
 639         ivor[4] = mfspr(SPRN_IVOR4);
 640         ivor[5] = mfspr(SPRN_IVOR5);
 641         ivor[6] = mfspr(SPRN_IVOR6);
 642         ivor[7] = mfspr(SPRN_IVOR7);
 643         ivor[8] = mfspr(SPRN_IVOR8);
 644         ivor[9] = mfspr(SPRN_IVOR9);
 645         ivor[10] = mfspr(SPRN_IVOR10);
 646         ivor[11] = mfspr(SPRN_IVOR11);
 647         ivor[12] = mfspr(SPRN_IVOR12);
 648         ivor[13] = mfspr(SPRN_IVOR13);
 649         ivor[14] = mfspr(SPRN_IVOR14);
 650         ivor[15] = mfspr(SPRN_IVOR15);
 651
 652         for (i = 0; i < 16; i++) {
 653                 if (ivor[i] > max_ivor)
 654                         max_ivor = ivor[i];
 655
 656                 memcpy((void *)kvmppc_booke_handlers + ivor[i],
 657                        kvmppc_handlers_start + i * kvmppc_handler_len,
 658                        kvmppc_handler_len);
 659         }
 660         flush_icache_range(kvmppc_booke_handlers,
 661                            kvmppc_booke_handlers + max_ivor + kvmppc_handler_len);
 662
 663         return kvm_init(NULL, sizeof(struct kvm_vcpu), THIS_MODULE);
 664 }
 665
 666 static void __exit kvmppc_booke_exit(void)
 667 {
 668         free_pages(kvmppc_booke_handlers, VCPU_SIZE_ORDER);
 669         kvm_exit();
 670 }
 671
 672 module_init(kvmppc_booke_init)
 673 module_exit(kvmppc_booke_exit)