arch/powerpc/kvm/book3s.c

   1 /*
   2  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
   3  *
   4  * Authors:
   5  *    Alexander Graf <agraf@suse.de>
   6  *    Kevin Wolf <mail@kevin-wolf.de>
   7  *
   8  * Description:
   9  * This file is derived from arch/powerpc/kvm/44x.c,
  10  * by Hollis Blanchard <hollisb@us.ibm.com>.
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License, version 2, as
  14  * published by the Free Software Foundation.
  15  */
  16
  17 #include <linux/kvm_host.h>
  18 #include <linux/err.h>
  19 #include <linux/slab.h>
  20
  21 #include <asm/reg.h>
  22 #include <asm/cputable.h>
  23 #include <asm/cacheflush.h>
  24 #include <asm/tlbflush.h>
  25 #include <asm/uaccess.h>
  26 #include <asm/io.h>
  27 #include <asm/kvm_ppc.h>
  28 #include <asm/kvm_book3s.h>
  29 #include <asm/mmu_context.h>
  30 #include <asm/page.h>
  31 #include <linux/gfp.h>
  32 #include <linux/sched.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/highmem.h>
  35
  36 #include "trace.h"
  37
  38 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  39
  40 /* #define EXIT_DEBUG */
  41
  42 struct kvm_stats_debugfs_item debugfs_entries[] = {
  43         { "exits",       VCPU_STAT(sum_exits) },
  44         { "mmio",        VCPU_STAT(mmio_exits) },
  45         { "sig",         VCPU_STAT(signal_exits) },
  46         { "sysc",        VCPU_STAT(syscall_exits) },
  47         { "inst_emu",    VCPU_STAT(emulated_inst_exits) },
  48         { "dec",         VCPU_STAT(dec_exits) },
  49         { "ext_intr",    VCPU_STAT(ext_intr_exits) },
  50         { "queue_intr",  VCPU_STAT(queue_intr) },
  51         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  52         { "pf_storage",  VCPU_STAT(pf_storage) },
  53         { "sp_storage",  VCPU_STAT(sp_storage) },
  54         { "pf_instruc",  VCPU_STAT(pf_instruc) },
  55         { "sp_instruc",  VCPU_STAT(sp_instruc) },
  56         { "ld",          VCPU_STAT(ld) },
  57         { "ld_slow",     VCPU_STAT(ld_slow) },
  58         { "st",          VCPU_STAT(st) },
  59         { "st_slow",     VCPU_STAT(st_slow) },
  60         { NULL }
  61 };
  62
  63 void kvmppc_core_load_host_debugstate(struct kvm_vcpu *vcpu)
  64 {
  65 }
  66
  67 void kvmppc_core_load_guest_debugstate(struct kvm_vcpu *vcpu)
  68 {
  69 }
  70
  71 void kvmppc_inject_interrupt(struct kvm_vcpu *vcpu, int vec, u64 flags)
  72 {
  73         vcpu->arch.shared->srr0 = kvmppc_get_pc(vcpu);
  74         vcpu->arch.shared->srr1 = vcpu->arch.shared->msr | flags;
  75         kvmppc_set_pc(vcpu, kvmppc_interrupt_offset(vcpu) + vec);
  76         vcpu->arch.mmu.reset_msr(vcpu);
  77 }
  78
  79 static int kvmppc_book3s_vec2irqprio(unsigned int vec)
  80 {
  81         unsigned int prio;
  82
  83         switch (vec) {
  84         case 0x100: prio = BOOK3S_IRQPRIO_SYSTEM_RESET;         break;
  85         case 0x200: prio = BOOK3S_IRQPRIO_MACHINE_CHECK;        break;
  86         case 0x300: prio = BOOK3S_IRQPRIO_DATA_STORAGE;         break;
  87         case 0x380: prio = BOOK3S_IRQPRIO_DATA_SEGMENT;         break;
  88         case 0x400: prio = BOOK3S_IRQPRIO_INST_STORAGE;         break;
  89         case 0x480: prio = BOOK3S_IRQPRIO_INST_SEGMENT;         break;
  90         case 0x500: prio = BOOK3S_IRQPRIO_EXTERNAL;             break;
  91         case 0x501: prio = BOOK3S_IRQPRIO_EXTERNAL_LEVEL;       break;
  92         case 0x600: prio = BOOK3S_IRQPRIO_ALIGNMENT;            break;
  93         case 0x700: prio = BOOK3S_IRQPRIO_PROGRAM;              break;
  94         case 0x800: prio = BOOK3S_IRQPRIO_FP_UNAVAIL;           break;
  95         case 0x900: prio = BOOK3S_IRQPRIO_DECREMENTER;          break;
  96         case 0xc00: prio = BOOK3S_IRQPRIO_SYSCALL;              break;
  97         case 0xd00: prio = BOOK3S_IRQPRIO_DEBUG;                break;
  98         case 0xf20: prio = BOOK3S_IRQPRIO_ALTIVEC;              break;
  99         case 0xf40: prio = BOOK3S_IRQPRIO_VSX;                  break;
 100         default:    prio = BOOK3S_IRQPRIO_MAX;                  break;
 101         }
 102
 103         return prio;
 104 }
 105
 106 static void kvmppc_book3s_dequeue_irqprio(struct kvm_vcpu *vcpu,
 107                                           unsigned int vec)
 108 {
 109         unsigned long old_pending = vcpu->arch.pending_exceptions;
 110
 111         clear_bit(kvmppc_book3s_vec2irqprio(vec),
 112                   &vcpu->arch.pending_exceptions);
 113
 114         kvmppc_update_int_pending(vcpu, vcpu->arch.pending_exceptions,
 115                                   old_pending);
 116 }
 117
 118 void kvmppc_book3s_queue_irqprio(struct kvm_vcpu *vcpu, unsigned int vec)
 119 {
 120         vcpu->stat.queue_intr++;
 121
 122         set_bit(kvmppc_book3s_vec2irqprio(vec),
 123                 &vcpu->arch.pending_exceptions);
 124 #ifdef EXIT_DEBUG
 125         printk(KERN_INFO "Queueing interrupt %x\n", vec);
 126 #endif
 127 }
 128
 129
 130 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu, ulong flags)
 131 {
 132         /* might as well deliver this straight away */
 133         kvmppc_inject_interrupt(vcpu, BOOK3S_INTERRUPT_PROGRAM, flags);
 134 }
 135
 136 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
 137 {
 138         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 139 }
 140
 141 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
 142 {
 143         return test_bit(BOOK3S_IRQPRIO_DECREMENTER, &vcpu->arch.pending_exceptions);
 144 }
 145
 146 void kvmppc_core_dequeue_dec(struct kvm_vcpu *vcpu)
 147 {
 148         kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 149 }
 150
 151 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
 152                                 struct kvm_interrupt *irq)
 153 {
 154         unsigned int vec = BOOK3S_INTERRUPT_EXTERNAL;
 155
 156         if (irq->irq == KVM_INTERRUPT_SET_LEVEL)
 157                 vec = BOOK3S_INTERRUPT_EXTERNAL_LEVEL;
 158
 159         kvmppc_book3s_queue_irqprio(vcpu, vec);
 160 }
 161
 162 void kvmppc_core_dequeue_external(struct kvm_vcpu *vcpu,
 163                                   struct kvm_interrupt *irq)
 164 {
 165         kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL);
 166         kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL_LEVEL);
 167 }
 168
 169 int kvmppc_book3s_irqprio_deliver(struct kvm_vcpu *vcpu, unsigned int priority)
 170 {
 171         int deliver = 1;
 172         int vec = 0;
 173         bool crit = kvmppc_critical_section(vcpu);
 174
 175         switch (priority) {
 176         case BOOK3S_IRQPRIO_DECREMENTER:
 177                 deliver = (vcpu->arch.shared->msr & MSR_EE) && !crit;
 178                 vec = BOOK3S_INTERRUPT_DECREMENTER;
 179                 break;
 180         case BOOK3S_IRQPRIO_EXTERNAL:
 181         case BOOK3S_IRQPRIO_EXTERNAL_LEVEL:
 182                 deliver = (vcpu->arch.shared->msr & MSR_EE) && !crit;
 183                 vec = BOOK3S_INTERRUPT_EXTERNAL;
 184                 break;
 185         case BOOK3S_IRQPRIO_SYSTEM_RESET:
 186                 vec = BOOK3S_INTERRUPT_SYSTEM_RESET;
 187                 break;
 188         case BOOK3S_IRQPRIO_MACHINE_CHECK:
 189                 vec = BOOK3S_INTERRUPT_MACHINE_CHECK;
 190                 break;
 191         case BOOK3S_IRQPRIO_DATA_STORAGE:
 192                 vec = BOOK3S_INTERRUPT_DATA_STORAGE;
 193                 break;
 194         case BOOK3S_IRQPRIO_INST_STORAGE:
 195                 vec = BOOK3S_INTERRUPT_INST_STORAGE;
 196                 break;
 197         case BOOK3S_IRQPRIO_DATA_SEGMENT:
 198                 vec = BOOK3S_INTERRUPT_DATA_SEGMENT;
 199                 break;
 200         case BOOK3S_IRQPRIO_INST_SEGMENT:
 201                 vec = BOOK3S_INTERRUPT_INST_SEGMENT;
 202                 break;
 203         case BOOK3S_IRQPRIO_ALIGNMENT:
 204                 vec = BOOK3S_INTERRUPT_ALIGNMENT;
 205                 break;
 206         case BOOK3S_IRQPRIO_PROGRAM:
 207                 vec = BOOK3S_INTERRUPT_PROGRAM;
 208                 break;
 209         case BOOK3S_IRQPRIO_VSX:
 210                 vec = BOOK3S_INTERRUPT_VSX;
 211                 break;
 212         case BOOK3S_IRQPRIO_ALTIVEC:
 213                 vec = BOOK3S_INTERRUPT_ALTIVEC;
 214                 break;
 215         case BOOK3S_IRQPRIO_FP_UNAVAIL:
 216                 vec = BOOK3S_INTERRUPT_FP_UNAVAIL;
 217                 break;
 218         case BOOK3S_IRQPRIO_SYSCALL:
 219                 vec = BOOK3S_INTERRUPT_SYSCALL;
 220                 break;
 221         case BOOK3S_IRQPRIO_DEBUG:
 222                 vec = BOOK3S_INTERRUPT_TRACE;
 223                 break;
 224         case BOOK3S_IRQPRIO_PERFORMANCE_MONITOR:
 225                 vec = BOOK3S_INTERRUPT_PERFMON;
 226                 break;
 227         default:
 228                 deliver = 0;
 229                 printk(KERN_ERR "KVM: Unknown interrupt: 0x%x\n", priority);
 230                 break;
 231         }
 232
 233 #if 0
 234         printk(KERN_INFO "Deliver interrupt 0x%x? %x\n", vec, deliver);
 235 #endif
 236
 237         if (deliver)
 238                 kvmppc_inject_interrupt(vcpu, vec, 0);
 239
 240         return deliver;
 241 }
 242
 243 /*
 244  * This function determines if an irqprio should be cleared once issued.
 245  */
 246 static bool clear_irqprio(struct kvm_vcpu *vcpu, unsigned int priority)
 247 {
 248         switch (priority) {
 249                 case BOOK3S_IRQPRIO_DECREMENTER:
 250                         /* DEC interrupts get cleared by mtdec */
 251                         return false;
 252                 case BOOK3S_IRQPRIO_EXTERNAL_LEVEL:
 253                         /* External interrupts get cleared by userspace */
 254                         return false;
 255         }
 256
 257         return true;
 258 }
 259
 260 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
 261 {
 262         unsigned long *pending = &vcpu->arch.pending_exceptions;
 263         unsigned long old_pending = vcpu->arch.pending_exceptions;
 264         unsigned int priority;
 265
 266 #ifdef EXIT_DEBUG
 267         if (vcpu->arch.pending_exceptions)
 268                 printk(KERN_EMERG "KVM: Check pending: %lx\n", vcpu->arch.pending_exceptions);
 269 #endif
 270         priority = __ffs(*pending);
 271         while (priority < BOOK3S_IRQPRIO_MAX) {
 272                 if (kvmppc_book3s_irqprio_deliver(vcpu, priority) &&
 273                     clear_irqprio(vcpu, priority)) {
 274                         clear_bit(priority, &vcpu->arch.pending_exceptions);
 275                         break;
 276                 }
 277
 278                 priority = find_next_bit(pending,
 279                                          BITS_PER_BYTE * sizeof(*pending),
 280                                          priority + 1);
 281         }
 282
 283         /* Tell the guest about our interrupt status */
 284         kvmppc_update_int_pending(vcpu, *pending, old_pending);
 285 }
 286
 287 pfn_t kvmppc_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 288 {
 289         ulong mp_pa = vcpu->arch.magic_page_pa;
 290
 291         /* Magic page override */
 292         if (unlikely(mp_pa) &&
 293             unlikely(((gfn << PAGE_SHIFT) & KVM_PAM) ==
 294                      ((mp_pa & PAGE_MASK) & KVM_PAM))) {
 295                 ulong shared_page = ((ulong)vcpu->arch.shared) & PAGE_MASK;
 296                 pfn_t pfn;
 297
 298                 pfn = (pfn_t)virt_to_phys((void*)shared_page) >> PAGE_SHIFT;
 299                 get_page(pfn_to_page(pfn));
 300                 return pfn;
 301         }
 302
 303         return gfn_to_pfn(vcpu->kvm, gfn);
 304 }
 305
 306 static int kvmppc_xlate(struct kvm_vcpu *vcpu, ulong eaddr, bool data,
 307                          struct kvmppc_pte *pte)
 308 {
 309         int relocated = (vcpu->arch.shared->msr & (data ? MSR_DR : MSR_IR));
 310         int r;
 311
 312         if (relocated) {
 313                 r = vcpu->arch.mmu.xlate(vcpu, eaddr, pte, data);
 314         } else {
 315                 pte->eaddr = eaddr;
 316                 pte->raddr = eaddr & KVM_PAM;
 317                 pte->vpage = VSID_REAL | eaddr >> 12;
 318                 pte->may_read = true;
 319                 pte->may_write = true;
 320                 pte->may_execute = true;
 321                 r = 0;
 322         }
 323
 324         return r;
 325 }
 326
 327 static hva_t kvmppc_bad_hva(void)
 328 {
 329         return PAGE_OFFSET;
 330 }
 331
 332 static hva_t kvmppc_pte_to_hva(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte,
 333                                bool read)
 334 {
 335         hva_t hpage;
 336
 337         if (read && !pte->may_read)
 338                 goto err;
 339
 340         if (!read && !pte->may_write)
 341                 goto err;
 342
 343         hpage = gfn_to_hva(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 344         if (kvm_is_error_hva(hpage))
 345                 goto err;
 346
 347         return hpage | (pte->raddr & ~PAGE_MASK);
 348 err:
 349         return kvmppc_bad_hva();
 350 }
 351
 352 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 353               bool data)
 354 {
 355         struct kvmppc_pte pte;
 356
 357         vcpu->stat.st++;
 358
 359         if (kvmppc_xlate(vcpu, *eaddr, data, &pte))
 360                 return -ENOENT;
 361
 362         *eaddr = pte.raddr;
 363
 364         if (!pte.may_write)
 365                 return -EPERM;
 366
 367         if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
 368                 return EMULATE_DO_MMIO;
 369
 370         return EMULATE_DONE;
 371 }
 372
 373 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 374                       bool data)
 375 {
 376         struct kvmppc_pte pte;
 377         hva_t hva = *eaddr;
 378
 379         vcpu->stat.ld++;
 380
 381         if (kvmppc_xlate(vcpu, *eaddr, data, &pte))
 382                 goto nopte;
 383
 384         *eaddr = pte.raddr;
 385
 386         hva = kvmppc_pte_to_hva(vcpu, &pte, true);
 387         if (kvm_is_error_hva(hva))
 388                 goto mmio;
 389
 390         if (copy_from_user(ptr, (void __user *)hva, size)) {
 391                 printk(KERN_INFO "kvmppc_ld at 0x%lx failed\n", hva);
 392                 goto mmio;
 393         }
 394
 395         return EMULATE_DONE;
 396
 397 nopte:
 398         return -ENOENT;
 399 mmio:
 400         return EMULATE_DO_MMIO;
 401 }
 402
 403 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
 404 {
 405         return 0;
 406 }
 407
 408 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 409 {
 410         int i;
 411
 412         regs->pc = kvmppc_get_pc(vcpu);
 413         regs->cr = kvmppc_get_cr(vcpu);
 414         regs->ctr = kvmppc_get_ctr(vcpu);
 415         regs->lr = kvmppc_get_lr(vcpu);
 416         regs->xer = kvmppc_get_xer(vcpu);
 417         regs->msr = vcpu->arch.shared->msr;
 418         regs->srr0 = vcpu->arch.shared->srr0;
 419         regs->srr1 = vcpu->arch.shared->srr1;
 420         regs->pid = vcpu->arch.pid;
 421         regs->sprg0 = vcpu->arch.shared->sprg0;
 422         regs->sprg1 = vcpu->arch.shared->sprg1;
 423         regs->sprg2 = vcpu->arch.shared->sprg2;
 424         regs->sprg3 = vcpu->arch.shared->sprg3;
 425         regs->sprg4 = vcpu->arch.sprg4;
 426         regs->sprg5 = vcpu->arch.sprg5;
 427         regs->sprg6 = vcpu->arch.sprg6;
 428         regs->sprg7 = vcpu->arch.sprg7;
 429
 430         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
 431                 regs->gpr[i] = kvmppc_get_gpr(vcpu, i);
 432
 433         return 0;
 434 }
 435
 436 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 437 {
 438         int i;
 439
 440         kvmppc_set_pc(vcpu, regs->pc);
 441         kvmppc_set_cr(vcpu, regs->cr);
 442         kvmppc_set_ctr(vcpu, regs->ctr);
 443         kvmppc_set_lr(vcpu, regs->lr);
 444         kvmppc_set_xer(vcpu, regs->xer);
 445         kvmppc_set_msr(vcpu, regs->msr);
 446         vcpu->arch.shared->srr0 = regs->srr0;
 447         vcpu->arch.shared->srr1 = regs->srr1;
 448         vcpu->arch.shared->sprg0 = regs->sprg0;
 449         vcpu->arch.shared->sprg1 = regs->sprg1;
 450         vcpu->arch.shared->sprg2 = regs->sprg2;
 451         vcpu->arch.shared->sprg3 = regs->sprg3;
 452         vcpu->arch.sprg4 = regs->sprg4;
 453         vcpu->arch.sprg5 = regs->sprg5;
 454         vcpu->arch.sprg6 = regs->sprg6;
 455         vcpu->arch.sprg7 = regs->sprg7;
 456
 457         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
 458                 kvmppc_set_gpr(vcpu, i, regs->gpr[i]);
 459
 460         return 0;
 461 }
 462
 463 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 464 {
 465         return -ENOTSUPP;
 466 }
 467
 468 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 469 {
 470         return -ENOTSUPP;
 471 }
 472
 473 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
 474                                   struct kvm_translation *tr)
 475 {
 476         return 0;
 477 }
 478
 479 /*
 480  * Get (and clear) the dirty memory log for a memory slot.
 481  */
 482 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 483                                       struct kvm_dirty_log *log)
 484 {
 485         struct kvm_memory_slot *memslot;
 486         struct kvm_vcpu *vcpu;
 487         ulong ga, ga_end;
 488         int is_dirty = 0;
 489         int r;
 490         unsigned long n;
 491
 492         mutex_lock(&kvm->slots_lock);
 493
 494         r = kvm_get_dirty_log(kvm, log, &is_dirty);
 495         if (r)
 496                 goto out;
 497
 498         /* If nothing is dirty, don't bother messing with page tables. */
 499         if (is_dirty) {
 500                 memslot = &kvm->memslots->memslots[log->slot];
 501
 502                 ga = memslot->base_gfn << PAGE_SHIFT;
 503                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
 504
 505                 kvm_for_each_vcpu(n, vcpu, kvm)
 506                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
 507
 508                 n = kvm_dirty_bitmap_bytes(memslot);
 509                 memset(memslot->dirty_bitmap, 0, n);
 510         }
 511
 512         r = 0;
 513 out:
 514         mutex_unlock(&kvm->slots_lock);
 515         return r;
 516 }