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/gfp.h>
  25 #include <linux/module.h>
  26 #include <linux/vmalloc.h>
  27 #include <linux/fs.h>
  28
  29 #include <asm/cputable.h>
  30 #include <asm/uaccess.h>
  31 #include <asm/kvm_ppc.h>
  32 #include "timing.h"
  33 #include <asm/cacheflush.h>
  34
  35 #include "booke.h"
  36
  37 unsigned long kvmppc_booke_handlers;
  38
  39 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  40 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  41
  42 struct kvm_stats_debugfs_item debugfs_entries[] = {
  43         { "mmio",       VCPU_STAT(mmio_exits) },
  44         { "dcr",        VCPU_STAT(dcr_exits) },
  45         { "sig",        VCPU_STAT(signal_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 /* TODO: use vcpu_printf() */
  61 void kvmppc_dump_vcpu(struct kvm_vcpu *vcpu)
  62 {
  63         int i;
  64
  65         printk("pc:   %08lx msr:  %08llx\n", vcpu->arch.pc, vcpu->arch.shared->msr);
  66         printk("lr:   %08lx ctr:  %08lx\n", vcpu->arch.lr, vcpu->arch.ctr);
  67         printk("srr0: %08llx srr1: %08llx\n", vcpu->arch.shared->srr0,
  68                                             vcpu->arch.shared->srr1);
  69
  70         printk("exceptions: %08lx\n", vcpu->arch.pending_exceptions);
  71
  72         for (i = 0; i < 32; i += 4) {
  73                 printk("gpr%02d: %08lx %08lx %08lx %08lx\n", i,
  74                        kvmppc_get_gpr(vcpu, i),
  75                        kvmppc_get_gpr(vcpu, i+1),
  76                        kvmppc_get_gpr(vcpu, i+2),
  77                        kvmppc_get_gpr(vcpu, i+3));
  78         }
  79 }
  80
  81 static void kvmppc_booke_queue_irqprio(struct kvm_vcpu *vcpu,
  82                                        unsigned int priority)
  83 {
  84         set_bit(priority, &vcpu->arch.pending_exceptions);
  85 }
  86
  87 static void kvmppc_core_queue_dtlb_miss(struct kvm_vcpu *vcpu,
  88                                         ulong dear_flags, ulong esr_flags)
  89 {
  90         vcpu->arch.queued_dear = dear_flags;
  91         vcpu->arch.queued_esr = esr_flags;
  92         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DTLB_MISS);
  93 }
  94
  95 static void kvmppc_core_queue_data_storage(struct kvm_vcpu *vcpu,
  96                                            ulong dear_flags, ulong esr_flags)
  97 {
  98         vcpu->arch.queued_dear = dear_flags;
  99         vcpu->arch.queued_esr = esr_flags;
 100         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DATA_STORAGE);
 101 }
 102
 103 static void kvmppc_core_queue_inst_storage(struct kvm_vcpu *vcpu,
 104                                            ulong esr_flags)
 105 {
 106         vcpu->arch.queued_esr = esr_flags;
 107         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_INST_STORAGE);
 108 }
 109
 110 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu, ulong esr_flags)
 111 {
 112         vcpu->arch.queued_esr = esr_flags;
 113         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_PROGRAM);
 114 }
 115
 116 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
 117 {
 118         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DECREMENTER);
 119 }
 120
 121 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
 122 {
 123         return test_bit(BOOKE_IRQPRIO_DECREMENTER, &vcpu->arch.pending_exceptions);
 124 }
 125
 126 void kvmppc_core_dequeue_dec(struct kvm_vcpu *vcpu)
 127 {
 128         clear_bit(BOOKE_IRQPRIO_DECREMENTER, &vcpu->arch.pending_exceptions);
 129 }
 130
 131 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
 132                                 struct kvm_interrupt *irq)
 133 {
 134         unsigned int prio = BOOKE_IRQPRIO_EXTERNAL;
 135
 136         if (irq->irq == KVM_INTERRUPT_SET_LEVEL)
 137                 prio = BOOKE_IRQPRIO_EXTERNAL_LEVEL;
 138
 139         kvmppc_booke_queue_irqprio(vcpu, prio);
 140 }
 141
 142 void kvmppc_core_dequeue_external(struct kvm_vcpu *vcpu,
 143                                   struct kvm_interrupt *irq)
 144 {
 145         clear_bit(BOOKE_IRQPRIO_EXTERNAL, &vcpu->arch.pending_exceptions);
 146         clear_bit(BOOKE_IRQPRIO_EXTERNAL_LEVEL, &vcpu->arch.pending_exceptions);
 147 }
 148
 149 /* Deliver the interrupt of the corresponding priority, if possible. */
 150 static int kvmppc_booke_irqprio_deliver(struct kvm_vcpu *vcpu,
 151                                         unsigned int priority)
 152 {
 153         int allowed = 0;
 154         ulong uninitialized_var(msr_mask);
 155         bool update_esr = false, update_dear = false;
 156         ulong crit_raw = vcpu->arch.shared->critical;
 157         ulong crit_r1 = kvmppc_get_gpr(vcpu, 1);
 158         bool crit;
 159         bool keep_irq = false;
 160
 161         /* Truncate crit indicators in 32 bit mode */
 162         if (!(vcpu->arch.shared->msr & MSR_SF)) {
 163                 crit_raw &= 0xffffffff;
 164                 crit_r1 &= 0xffffffff;
 165         }
 166
 167         /* Critical section when crit == r1 */
 168         crit = (crit_raw == crit_r1);
 169         /* ... and we're in supervisor mode */
 170         crit = crit && !(vcpu->arch.shared->msr & MSR_PR);
 171
 172         if (priority == BOOKE_IRQPRIO_EXTERNAL_LEVEL) {
 173                 priority = BOOKE_IRQPRIO_EXTERNAL;
 174                 keep_irq = true;
 175         }
 176
 177         switch (priority) {
 178         case BOOKE_IRQPRIO_DTLB_MISS:
 179         case BOOKE_IRQPRIO_DATA_STORAGE:
 180                 update_dear = true;
 181                 /* fall through */
 182         case BOOKE_IRQPRIO_INST_STORAGE:
 183         case BOOKE_IRQPRIO_PROGRAM:
 184                 update_esr = true;
 185                 /* fall through */
 186         case BOOKE_IRQPRIO_ITLB_MISS:
 187         case BOOKE_IRQPRIO_SYSCALL:
 188         case BOOKE_IRQPRIO_FP_UNAVAIL:
 189         case BOOKE_IRQPRIO_SPE_UNAVAIL:
 190         case BOOKE_IRQPRIO_SPE_FP_DATA:
 191         case BOOKE_IRQPRIO_SPE_FP_ROUND:
 192         case BOOKE_IRQPRIO_AP_UNAVAIL:
 193         case BOOKE_IRQPRIO_ALIGNMENT:
 194                 allowed = 1;
 195                 msr_mask = MSR_CE|MSR_ME|MSR_DE;
 196                 break;
 197         case BOOKE_IRQPRIO_CRITICAL:
 198         case BOOKE_IRQPRIO_WATCHDOG:
 199                 allowed = vcpu->arch.shared->msr & MSR_CE;
 200                 msr_mask = MSR_ME;
 201                 break;
 202         case BOOKE_IRQPRIO_MACHINE_CHECK:
 203                 allowed = vcpu->arch.shared->msr & MSR_ME;
 204                 msr_mask = 0;
 205                 break;
 206         case BOOKE_IRQPRIO_EXTERNAL:
 207         case BOOKE_IRQPRIO_DECREMENTER:
 208         case BOOKE_IRQPRIO_FIT:
 209                 allowed = vcpu->arch.shared->msr & MSR_EE;
 210                 allowed = allowed && !crit;
 211                 msr_mask = MSR_CE|MSR_ME|MSR_DE;
 212                 break;
 213         case BOOKE_IRQPRIO_DEBUG:
 214                 allowed = vcpu->arch.shared->msr & MSR_DE;
 215                 msr_mask = MSR_ME;
 216                 break;
 217         }
 218
 219         if (allowed) {
 220                 vcpu->arch.shared->srr0 = vcpu->arch.pc;
 221                 vcpu->arch.shared->srr1 = vcpu->arch.shared->msr;
 222                 vcpu->arch.pc = vcpu->arch.ivpr | vcpu->arch.ivor[priority];
 223                 if (update_esr == true)
 224                         vcpu->arch.esr = vcpu->arch.queued_esr;
 225                 if (update_dear == true)
 226                         vcpu->arch.shared->dar = vcpu->arch.queued_dear;
 227                 kvmppc_set_msr(vcpu, vcpu->arch.shared->msr & msr_mask);
 228
 229                 if (!keep_irq)
 230                         clear_bit(priority, &vcpu->arch.pending_exceptions);
 231         }
 232
 233         return allowed;
 234 }
 235
 236 /* Check pending exceptions and deliver one, if possible. */
 237 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
 238 {
 239         unsigned long *pending = &vcpu->arch.pending_exceptions;
 240         unsigned long old_pending = vcpu->arch.pending_exceptions;
 241         unsigned int priority;
 242
 243         priority = __ffs(*pending);
 244         while (priority <= BOOKE_IRQPRIO_MAX) {
 245                 if (kvmppc_booke_irqprio_deliver(vcpu, priority))
 246                         break;
 247
 248                 priority = find_next_bit(pending,
 249                                          BITS_PER_BYTE * sizeof(*pending),
 250                                          priority + 1);
 251         }
 252
 253         /* Tell the guest about our interrupt status */
 254         if (*pending)
 255                 vcpu->arch.shared->int_pending = 1;
 256         else if (old_pending)
 257                 vcpu->arch.shared->int_pending = 0;
 258 }
 259
 260 /**
 261  * kvmppc_handle_exit
 262  *
 263  * Return value is in the form (errcode<<2 | RESUME_FLAG_HOST | RESUME_FLAG_NV)
 264  */
 265 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 266                        unsigned int exit_nr)
 267 {
 268         enum emulation_result er;
 269         int r = RESUME_HOST;
 270
 271         /* update before a new last_exit_type is rewritten */
 272         kvmppc_update_timing_stats(vcpu);
 273
 274         local_irq_enable();
 275
 276         run->exit_reason = KVM_EXIT_UNKNOWN;
 277         run->ready_for_interrupt_injection = 1;
 278
 279         switch (exit_nr) {
 280         case BOOKE_INTERRUPT_MACHINE_CHECK:
 281                 printk("MACHINE CHECK: %lx\n", mfspr(SPRN_MCSR));
 282                 kvmppc_dump_vcpu(vcpu);
 283                 r = RESUME_HOST;
 284                 break;
 285
 286         case BOOKE_INTERRUPT_EXTERNAL:
 287                 kvmppc_account_exit(vcpu, EXT_INTR_EXITS);
 288                 if (need_resched())
 289                         cond_resched();
 290                 r = RESUME_GUEST;
 291                 break;
 292
 293         case BOOKE_INTERRUPT_DECREMENTER:
 294                 /* Since we switched IVPR back to the host's value, the host
 295                  * handled this interrupt the moment we enabled interrupts.
 296                  * Now we just offer it a chance to reschedule the guest. */
 297                 kvmppc_account_exit(vcpu, DEC_EXITS);
 298                 if (need_resched())
 299                         cond_resched();
 300                 r = RESUME_GUEST;
 301                 break;
 302
 303         case BOOKE_INTERRUPT_PROGRAM:
 304                 if (vcpu->arch.shared->msr & MSR_PR) {
 305                         /* Program traps generated by user-level software must be handled
 306                          * by the guest kernel. */
 307                         kvmppc_core_queue_program(vcpu, vcpu->arch.fault_esr);
 308                         r = RESUME_GUEST;
 309                         kvmppc_account_exit(vcpu, USR_PR_INST);
 310                         break;
 311                 }
 312
 313                 er = kvmppc_emulate_instruction(run, vcpu);
 314                 switch (er) {
 315                 case EMULATE_DONE:
 316                         /* don't overwrite subtypes, just account kvm_stats */
 317                         kvmppc_account_exit_stat(vcpu, EMULATED_INST_EXITS);
 318                         /* Future optimization: only reload non-volatiles if
 319                          * they were actually modified by emulation. */
 320                         r = RESUME_GUEST_NV;
 321                         break;
 322                 case EMULATE_DO_DCR:
 323                         run->exit_reason = KVM_EXIT_DCR;
 324                         r = RESUME_HOST;
 325                         break;
 326                 case EMULATE_FAIL:
 327                         /* XXX Deliver Program interrupt to guest. */
 328                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 329                                __func__, vcpu->arch.pc, vcpu->arch.last_inst);
 330                         /* For debugging, encode the failing instruction and
 331                          * report it to userspace. */
 332                         run->hw.hardware_exit_reason = ~0ULL << 32;
 333                         run->hw.hardware_exit_reason |= vcpu->arch.last_inst;
 334                         r = RESUME_HOST;
 335                         break;
 336                 default:
 337                         BUG();
 338                 }
 339                 break;
 340
 341         case BOOKE_INTERRUPT_FP_UNAVAIL:
 342                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_FP_UNAVAIL);
 343                 kvmppc_account_exit(vcpu, FP_UNAVAIL);
 344                 r = RESUME_GUEST;
 345                 break;
 346
 347         case BOOKE_INTERRUPT_SPE_UNAVAIL:
 348                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SPE_UNAVAIL);
 349                 r = RESUME_GUEST;
 350                 break;
 351
 352         case BOOKE_INTERRUPT_SPE_FP_DATA:
 353                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SPE_FP_DATA);
 354                 r = RESUME_GUEST;
 355                 break;
 356
 357         case BOOKE_INTERRUPT_SPE_FP_ROUND:
 358                 kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SPE_FP_ROUND);
 359                 r = RESUME_GUEST;
 360                 break;
 361
 362         case BOOKE_INTERRUPT_DATA_STORAGE:
 363                 kvmppc_core_queue_data_storage(vcpu, vcpu->arch.fault_dear,
 364                                                vcpu->arch.fault_esr);
 365                 kvmppc_account_exit(vcpu, DSI_EXITS);
 366                 r = RESUME_GUEST;
 367                 break;
 368
 369         case BOOKE_INTERRUPT_INST_STORAGE:
 370                 kvmppc_core_queue_inst_storage(vcpu, vcpu->arch.fault_esr);
 371                 kvmppc_account_exit(vcpu, ISI_EXITS);
 372                 r = RESUME_GUEST;
 373                 break;
 374
 375         case BOOKE_INTERRUPT_SYSCALL:
 376                 if (!(vcpu->arch.shared->msr & MSR_PR) &&
 377                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
 378                         /* KVM PV hypercalls */
 379                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
 380                         r = RESUME_GUEST;
 381                 } else {
 382                         /* Guest syscalls */
 383                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SYSCALL);
 384                 }
 385                 kvmppc_account_exit(vcpu, SYSCALL_EXITS);
 386                 r = RESUME_GUEST;
 387                 break;
 388
 389         case BOOKE_INTERRUPT_DTLB_MISS: {
 390                 unsigned long eaddr = vcpu->arch.fault_dear;
 391                 int gtlb_index;
 392                 gpa_t gpaddr;
 393                 gfn_t gfn;
 394
 395                 /* Check the guest TLB. */
 396                 gtlb_index = kvmppc_mmu_dtlb_index(vcpu, eaddr);
 397                 if (gtlb_index < 0) {
 398                         /* The guest didn't have a mapping for it. */
 399                         kvmppc_core_queue_dtlb_miss(vcpu,
 400                                                     vcpu->arch.fault_dear,
 401                                                     vcpu->arch.fault_esr);
 402                         kvmppc_mmu_dtlb_miss(vcpu);
 403                         kvmppc_account_exit(vcpu, DTLB_REAL_MISS_EXITS);
 404                         r = RESUME_GUEST;
 405                         break;
 406                 }
 407
 408                 gpaddr = kvmppc_mmu_xlate(vcpu, gtlb_index, eaddr);
 409                 gfn = gpaddr >> PAGE_SHIFT;
 410
 411                 if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
 412                         /* The guest TLB had a mapping, but the shadow TLB
 413                          * didn't, and it is RAM. This could be because:
 414                          * a) the entry is mapping the host kernel, or
 415                          * b) the guest used a large mapping which we're faking
 416                          * Either way, we need to satisfy the fault without
 417                          * invoking the guest. */
 418                         kvmppc_mmu_map(vcpu, eaddr, gpaddr, gtlb_index);
 419                         kvmppc_account_exit(vcpu, DTLB_VIRT_MISS_EXITS);
 420                         r = RESUME_GUEST;
 421                 } else {
 422                         /* Guest has mapped and accessed a page which is not
 423                          * actually RAM. */
 424                         vcpu->arch.paddr_accessed = gpaddr;
 425                         r = kvmppc_emulate_mmio(run, vcpu);
 426                         kvmppc_account_exit(vcpu, MMIO_EXITS);
 427                 }
 428
 429                 break;
 430         }
 431
 432         case BOOKE_INTERRUPT_ITLB_MISS: {
 433                 unsigned long eaddr = vcpu->arch.pc;
 434                 gpa_t gpaddr;
 435                 gfn_t gfn;
 436                 int gtlb_index;
 437
 438                 r = RESUME_GUEST;
 439
 440                 /* Check the guest TLB. */
 441                 gtlb_index = kvmppc_mmu_itlb_index(vcpu, eaddr);
 442                 if (gtlb_index < 0) {
 443                         /* The guest didn't have a mapping for it. */
 444                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_ITLB_MISS);
 445                         kvmppc_mmu_itlb_miss(vcpu);
 446                         kvmppc_account_exit(vcpu, ITLB_REAL_MISS_EXITS);
 447                         break;
 448                 }
 449
 450                 kvmppc_account_exit(vcpu, ITLB_VIRT_MISS_EXITS);
 451
 452                 gpaddr = kvmppc_mmu_xlate(vcpu, gtlb_index, eaddr);
 453                 gfn = gpaddr >> PAGE_SHIFT;
 454
 455                 if (kvm_is_visible_gfn(vcpu->kvm, gfn)) {
 456                         /* The guest TLB had a mapping, but the shadow TLB
 457                          * didn't. This could be because:
 458                          * a) the entry is mapping the host kernel, or
 459                          * b) the guest used a large mapping which we're faking
 460                          * Either way, we need to satisfy the fault without
 461                          * invoking the guest. */
 462                         kvmppc_mmu_map(vcpu, eaddr, gpaddr, gtlb_index);
 463                 } else {
 464                         /* Guest mapped and leaped at non-RAM! */
 465                         kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_MACHINE_CHECK);
 466                 }
 467
 468                 break;
 469         }
 470
 471         case BOOKE_INTERRUPT_DEBUG: {
 472                 u32 dbsr;
 473
 474                 vcpu->arch.pc = mfspr(SPRN_CSRR0);
 475
 476                 /* clear IAC events in DBSR register */
 477                 dbsr = mfspr(SPRN_DBSR);
 478                 dbsr &= DBSR_IAC1 | DBSR_IAC2 | DBSR_IAC3 | DBSR_IAC4;
 479                 mtspr(SPRN_DBSR, dbsr);
 480
 481                 run->exit_reason = KVM_EXIT_DEBUG;
 482                 kvmppc_account_exit(vcpu, DEBUG_EXITS);
 483                 r = RESUME_HOST;
 484                 break;
 485         }
 486
 487         default:
 488                 printk(KERN_EMERG "exit_nr %d\n", exit_nr);
 489                 BUG();
 490         }
 491
 492         local_irq_disable();
 493
 494         kvmppc_core_deliver_interrupts(vcpu);
 495
 496         if (!(r & RESUME_HOST)) {
 497                 /* To avoid clobbering exit_reason, only check for signals if
 498                  * we aren't already exiting to userspace for some other
 499                  * reason. */
 500                 if (signal_pending(current)) {
 501                         run->exit_reason = KVM_EXIT_INTR;
 502                         r = (-EINTR << 2) | RESUME_HOST | (r & RESUME_FLAG_NV);
 503                         kvmppc_account_exit(vcpu, SIGNAL_EXITS);
 504                 }
 505         }
 506
 507         return r;
 508 }
 509
 510 /* Initial guest state: 16MB mapping 0 -> 0, PC = 0, MSR = 0, R1 = 16MB */
 511 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
 512 {
 513         int i;
 514
 515         vcpu->arch.pc = 0;
 516         vcpu->arch.shared->msr = 0;
 517         kvmppc_set_gpr(vcpu, 1, (16<<20) - 8); /* -8 for the callee-save LR slot */
 518
 519         vcpu->arch.shadow_pid = 1;
 520
 521         /* Eye-catching numbers so we know if the guest takes an interrupt
 522          * before it's programmed its own IVPR/IVORs. */
 523         vcpu->arch.ivpr = 0x55550000;
 524         for (i = 0; i < BOOKE_IRQPRIO_MAX; i++)
 525                 vcpu->arch.ivor[i] = 0x7700 | i * 4;
 526
 527         kvmppc_init_timing_stats(vcpu);
 528
 529         return kvmppc_core_vcpu_setup(vcpu);
 530 }
 531
 532 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 533 {
 534         int i;
 535
 536         regs->pc = vcpu->arch.pc;
 537         regs->cr = kvmppc_get_cr(vcpu);
 538         regs->ctr = vcpu->arch.ctr;
 539         regs->lr = vcpu->arch.lr;
 540         regs->xer = kvmppc_get_xer(vcpu);
 541         regs->msr = vcpu->arch.shared->msr;
 542         regs->srr0 = vcpu->arch.shared->srr0;
 543         regs->srr1 = vcpu->arch.shared->srr1;
 544         regs->pid = vcpu->arch.pid;
 545         regs->sprg0 = vcpu->arch.shared->sprg0;
 546         regs->sprg1 = vcpu->arch.shared->sprg1;
 547         regs->sprg2 = vcpu->arch.shared->sprg2;
 548         regs->sprg3 = vcpu->arch.shared->sprg3;
 549         regs->sprg4 = vcpu->arch.sprg4;
 550         regs->sprg5 = vcpu->arch.sprg5;
 551         regs->sprg6 = vcpu->arch.sprg6;
 552         regs->sprg7 = vcpu->arch.sprg7;
 553
 554         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
 555                 regs->gpr[i] = kvmppc_get_gpr(vcpu, i);
 556
 557         return 0;
 558 }
 559
 560 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 561 {
 562         int i;
 563
 564         vcpu->arch.pc = regs->pc;
 565         kvmppc_set_cr(vcpu, regs->cr);
 566         vcpu->arch.ctr = regs->ctr;
 567         vcpu->arch.lr = regs->lr;
 568         kvmppc_set_xer(vcpu, regs->xer);
 569         kvmppc_set_msr(vcpu, regs->msr);
 570         vcpu->arch.shared->srr0 = regs->srr0;
 571         vcpu->arch.shared->srr1 = regs->srr1;
 572         kvmppc_set_pid(vcpu, regs->pid);
 573         vcpu->arch.shared->sprg0 = regs->sprg0;
 574         vcpu->arch.shared->sprg1 = regs->sprg1;
 575         vcpu->arch.shared->sprg2 = regs->sprg2;
 576         vcpu->arch.shared->sprg3 = regs->sprg3;
 577         vcpu->arch.sprg4 = regs->sprg4;
 578         vcpu->arch.sprg5 = regs->sprg5;
 579         vcpu->arch.sprg6 = regs->sprg6;
 580         vcpu->arch.sprg7 = regs->sprg7;
 581
 582         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
 583                 kvmppc_set_gpr(vcpu, i, regs->gpr[i]);
 584
 585         return 0;
 586 }
 587
 588 static void get_sregs_base(struct kvm_vcpu *vcpu,
 589                            struct kvm_sregs *sregs)
 590 {
 591         u64 tb = get_tb();
 592
 593         sregs->u.e.features |= KVM_SREGS_E_BASE;
 594
 595         sregs->u.e.csrr0 = vcpu->arch.csrr0;
 596         sregs->u.e.csrr1 = vcpu->arch.csrr1;
 597         sregs->u.e.mcsr = vcpu->arch.mcsr;
 598         sregs->u.e.esr = vcpu->arch.esr;
 599         sregs->u.e.dear = vcpu->arch.shared->dar;
 600         sregs->u.e.tsr = vcpu->arch.tsr;
 601         sregs->u.e.tcr = vcpu->arch.tcr;
 602         sregs->u.e.dec = kvmppc_get_dec(vcpu, tb);
 603         sregs->u.e.tb = tb;
 604         sregs->u.e.vrsave = vcpu->arch.vrsave;
 605 }
 606
 607 static int set_sregs_base(struct kvm_vcpu *vcpu,
 608                           struct kvm_sregs *sregs)
 609 {
 610         if (!(sregs->u.e.features & KVM_SREGS_E_BASE))
 611                 return 0;
 612
 613         vcpu->arch.csrr0 = sregs->u.e.csrr0;
 614         vcpu->arch.csrr1 = sregs->u.e.csrr1;
 615         vcpu->arch.mcsr = sregs->u.e.mcsr;
 616         vcpu->arch.esr = sregs->u.e.esr;
 617         vcpu->arch.shared->dar = sregs->u.e.dear;
 618         vcpu->arch.vrsave = sregs->u.e.vrsave;
 619         vcpu->arch.tcr = sregs->u.e.tcr;
 620
 621         if (sregs->u.e.update_special & KVM_SREGS_E_UPDATE_DEC)
 622                 vcpu->arch.dec = sregs->u.e.dec;
 623
 624         kvmppc_emulate_dec(vcpu);
 625
 626         if (sregs->u.e.update_special & KVM_SREGS_E_UPDATE_TSR) {
 627                 /*
 628                  * FIXME: existing KVM timer handling is incomplete.
 629                  * TSR cannot be read by the guest, and its value in
 630                  * vcpu->arch is always zero.  For now, just handle
 631                  * the case where the caller is trying to inject a
 632                  * decrementer interrupt.
 633                  */
 634
 635                 if ((sregs->u.e.tsr & TSR_DIS) &&
 636                     (vcpu->arch.tcr & TCR_DIE))
 637                         kvmppc_core_queue_dec(vcpu);
 638         }
 639
 640         return 0;
 641 }
 642
 643 static void get_sregs_arch206(struct kvm_vcpu *vcpu,
 644                               struct kvm_sregs *sregs)
 645 {
 646         sregs->u.e.features |= KVM_SREGS_E_ARCH206;
 647
 648         sregs->u.e.pir = 0;
 649         sregs->u.e.mcsrr0 = vcpu->arch.mcsrr0;
 650         sregs->u.e.mcsrr1 = vcpu->arch.mcsrr1;
 651         sregs->u.e.decar = vcpu->arch.decar;
 652         sregs->u.e.ivpr = vcpu->arch.ivpr;
 653 }
 654
 655 static int set_sregs_arch206(struct kvm_vcpu *vcpu,
 656                              struct kvm_sregs *sregs)
 657 {
 658         if (!(sregs->u.e.features & KVM_SREGS_E_ARCH206))
 659                 return 0;
 660
 661         if (sregs->u.e.pir != 0)
 662                 return -EINVAL;
 663
 664         vcpu->arch.mcsrr0 = sregs->u.e.mcsrr0;
 665         vcpu->arch.mcsrr1 = sregs->u.e.mcsrr1;
 666         vcpu->arch.decar = sregs->u.e.decar;
 667         vcpu->arch.ivpr = sregs->u.e.ivpr;
 668
 669         return 0;
 670 }
 671
 672 void kvmppc_get_sregs_ivor(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
 673 {
 674         sregs->u.e.features |= KVM_SREGS_E_IVOR;
 675
 676         sregs->u.e.ivor_low[0] = vcpu->arch.ivor[BOOKE_IRQPRIO_CRITICAL];
 677         sregs->u.e.ivor_low[1] = vcpu->arch.ivor[BOOKE_IRQPRIO_MACHINE_CHECK];
 678         sregs->u.e.ivor_low[2] = vcpu->arch.ivor[BOOKE_IRQPRIO_DATA_STORAGE];
 679         sregs->u.e.ivor_low[3] = vcpu->arch.ivor[BOOKE_IRQPRIO_INST_STORAGE];
 680         sregs->u.e.ivor_low[4] = vcpu->arch.ivor[BOOKE_IRQPRIO_EXTERNAL];
 681         sregs->u.e.ivor_low[5] = vcpu->arch.ivor[BOOKE_IRQPRIO_ALIGNMENT];
 682         sregs->u.e.ivor_low[6] = vcpu->arch.ivor[BOOKE_IRQPRIO_PROGRAM];
 683         sregs->u.e.ivor_low[7] = vcpu->arch.ivor[BOOKE_IRQPRIO_FP_UNAVAIL];
 684         sregs->u.e.ivor_low[8] = vcpu->arch.ivor[BOOKE_IRQPRIO_SYSCALL];
 685         sregs->u.e.ivor_low[9] = vcpu->arch.ivor[BOOKE_IRQPRIO_AP_UNAVAIL];
 686         sregs->u.e.ivor_low[10] = vcpu->arch.ivor[BOOKE_IRQPRIO_DECREMENTER];
 687         sregs->u.e.ivor_low[11] = vcpu->arch.ivor[BOOKE_IRQPRIO_FIT];
 688         sregs->u.e.ivor_low[12] = vcpu->arch.ivor[BOOKE_IRQPRIO_WATCHDOG];
 689         sregs->u.e.ivor_low[13] = vcpu->arch.ivor[BOOKE_IRQPRIO_DTLB_MISS];
 690         sregs->u.e.ivor_low[14] = vcpu->arch.ivor[BOOKE_IRQPRIO_ITLB_MISS];
 691         sregs->u.e.ivor_low[15] = vcpu->arch.ivor[BOOKE_IRQPRIO_DEBUG];
 692 }
 693
 694 int kvmppc_set_sregs_ivor(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
 695 {
 696         if (!(sregs->u.e.features & KVM_SREGS_E_IVOR))
 697                 return 0;
 698
 699         vcpu->arch.ivor[BOOKE_IRQPRIO_CRITICAL] = sregs->u.e.ivor_low[0];
 700         vcpu->arch.ivor[BOOKE_IRQPRIO_MACHINE_CHECK] = sregs->u.e.ivor_low[1];
 701         vcpu->arch.ivor[BOOKE_IRQPRIO_DATA_STORAGE] = sregs->u.e.ivor_low[2];
 702         vcpu->arch.ivor[BOOKE_IRQPRIO_INST_STORAGE] = sregs->u.e.ivor_low[3];
 703         vcpu->arch.ivor[BOOKE_IRQPRIO_EXTERNAL] = sregs->u.e.ivor_low[4];
 704         vcpu->arch.ivor[BOOKE_IRQPRIO_ALIGNMENT] = sregs->u.e.ivor_low[5];
 705         vcpu->arch.ivor[BOOKE_IRQPRIO_PROGRAM] = sregs->u.e.ivor_low[6];
 706         vcpu->arch.ivor[BOOKE_IRQPRIO_FP_UNAVAIL] = sregs->u.e.ivor_low[7];
 707         vcpu->arch.ivor[BOOKE_IRQPRIO_SYSCALL] = sregs->u.e.ivor_low[8];
 708         vcpu->arch.ivor[BOOKE_IRQPRIO_AP_UNAVAIL] = sregs->u.e.ivor_low[9];
 709         vcpu->arch.ivor[BOOKE_IRQPRIO_DECREMENTER] = sregs->u.e.ivor_low[10];
 710         vcpu->arch.ivor[BOOKE_IRQPRIO_FIT] = sregs->u.e.ivor_low[11];
 711         vcpu->arch.ivor[BOOKE_IRQPRIO_WATCHDOG] = sregs->u.e.ivor_low[12];
 712         vcpu->arch.ivor[BOOKE_IRQPRIO_DTLB_MISS] = sregs->u.e.ivor_low[13];
 713         vcpu->arch.ivor[BOOKE_IRQPRIO_ITLB_MISS] = sregs->u.e.ivor_low[14];
 714         vcpu->arch.ivor[BOOKE_IRQPRIO_DEBUG] = sregs->u.e.ivor_low[15];
 715
 716         return 0;
 717 }
 718
 719 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 720                                   struct kvm_sregs *sregs)
 721 {
 722         sregs->pvr = vcpu->arch.pvr;
 723
 724         get_sregs_base(vcpu, sregs);
 725         get_sregs_arch206(vcpu, sregs);
 726         kvmppc_core_get_sregs(vcpu, sregs);
 727         return 0;
 728 }
 729
 730 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 731                                   struct kvm_sregs *sregs)
 732 {
 733         int ret;
 734
 735         if (vcpu->arch.pvr != sregs->pvr)
 736                 return -EINVAL;
 737
 738         ret = set_sregs_base(vcpu, sregs);
 739         if (ret < 0)
 740                 return ret;
 741
 742         ret = set_sregs_arch206(vcpu, sregs);
 743         if (ret < 0)
 744                 return ret;
 745
 746         return kvmppc_core_set_sregs(vcpu, sregs);
 747 }
 748
 749 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 750 {
 751         return -ENOTSUPP;
 752 }
 753
 754 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 755 {
 756         return -ENOTSUPP;
 757 }
 758
 759 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
 760                                   struct kvm_translation *tr)
 761 {
 762         int r;
 763
 764         r = kvmppc_core_vcpu_translate(vcpu, tr);
 765         return r;
 766 }
 767
 768 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 769 {
 770         return -ENOTSUPP;
 771 }
 772
 773 int __init kvmppc_booke_init(void)
 774 {
 775         unsigned long ivor[16];
 776         unsigned long max_ivor = 0;
 777         int i;
 778
 779         /* We install our own exception handlers by hijacking IVPR. IVPR must
 780          * be 16-bit aligned, so we need a 64KB allocation. */
 781         kvmppc_booke_handlers = __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 782                                                  VCPU_SIZE_ORDER);
 783         if (!kvmppc_booke_handlers)
 784                 return -ENOMEM;
 785
 786         /* XXX make sure our handlers are smaller than Linux's */
 787
 788         /* Copy our interrupt handlers to match host IVORs. That way we don't
 789          * have to swap the IVORs on every guest/host transition. */
 790         ivor[0] = mfspr(SPRN_IVOR0);
 791         ivor[1] = mfspr(SPRN_IVOR1);
 792         ivor[2] = mfspr(SPRN_IVOR2);
 793         ivor[3] = mfspr(SPRN_IVOR3);
 794         ivor[4] = mfspr(SPRN_IVOR4);
 795         ivor[5] = mfspr(SPRN_IVOR5);
 796         ivor[6] = mfspr(SPRN_IVOR6);
 797         ivor[7] = mfspr(SPRN_IVOR7);
 798         ivor[8] = mfspr(SPRN_IVOR8);
 799         ivor[9] = mfspr(SPRN_IVOR9);
 800         ivor[10] = mfspr(SPRN_IVOR10);
 801         ivor[11] = mfspr(SPRN_IVOR11);
 802         ivor[12] = mfspr(SPRN_IVOR12);
 803         ivor[13] = mfspr(SPRN_IVOR13);
 804         ivor[14] = mfspr(SPRN_IVOR14);
 805         ivor[15] = mfspr(SPRN_IVOR15);
 806
 807         for (i = 0; i < 16; i++) {
 808                 if (ivor[i] > max_ivor)
 809                         max_ivor = ivor[i];
 810
 811                 memcpy((void *)kvmppc_booke_handlers + ivor[i],
 812                        kvmppc_handlers_start + i * kvmppc_handler_len,
 813                        kvmppc_handler_len);
 814         }
 815         flush_icache_range(kvmppc_booke_handlers,
 816                            kvmppc_booke_handlers + max_ivor + kvmppc_handler_len);
 817
 818         return 0;
 819 }
 820
 821 void __exit kvmppc_booke_exit(void)
 822 {
 823         free_pages(kvmppc_booke_handlers, VCPU_SIZE_ORDER);
 824         kvm_exit();
 825 }