arch/ia64/kvm/kvm-ia64.c

   1 /*
   2  * kvm_ia64.c: Basic KVM suppport On Itanium series processors
   3  *
   4  *
   5  *      Copyright (C) 2007, Intel Corporation.
   6  *      Xiantao Zhang  (xiantao.zhang@intel.com)
   7  *
   8  * This program is free software; you can redistribute it and/or modify it
   9  * under the terms and conditions of the GNU General Public License,
  10  * version 2, as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope it will be useful, but WITHOUT
  13  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  14  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  15  * more details.
  16  *
  17  * You should have received a copy of the GNU General Public License along with
  18  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
  19  * Place - Suite 330, Boston, MA 02111-1307 USA.
  20  *
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/gfp.h>
  27 #include <linux/fs.h>
  28 #include <linux/smp.h>
  29 #include <linux/kvm_host.h>
  30 #include <linux/kvm.h>
  31 #include <linux/bitops.h>
  32 #include <linux/hrtimer.h>
  33 #include <linux/uaccess.h>
  34 #include <linux/intel-iommu.h>
  35
  36 #include <asm/pgtable.h>
  37 #include <asm/gcc_intrin.h>
  38 #include <asm/pal.h>
  39 #include <asm/cacheflush.h>
  40 #include <asm/div64.h>
  41 #include <asm/tlb.h>
  42 #include <asm/elf.h>
  43
  44 #include "misc.h"
  45 #include "vti.h"
  46 #include "iodev.h"
  47 #include "ioapic.h"
  48 #include "lapic.h"
  49 #include "irq.h"
  50
  51 static unsigned long kvm_vmm_base;
  52 static unsigned long kvm_vsa_base;
  53 static unsigned long kvm_vm_buffer;
  54 static unsigned long kvm_vm_buffer_size;
  55 unsigned long kvm_vmm_gp;
  56
  57 static long vp_env_info;
  58
  59 static struct kvm_vmm_info *kvm_vmm_info;
  60
  61 static DEFINE_PER_CPU(struct kvm_vcpu *, last_vcpu);
  62
  63 struct kvm_stats_debugfs_item debugfs_entries[] = {
  64         { NULL }
  65 };
  66
  67 static void kvm_flush_icache(unsigned long start, unsigned long len)
  68 {
  69         int l;
  70
  71         for (l = 0; l < (len + 32); l += 32)
  72                 ia64_fc(start + l);
  73
  74         ia64_sync_i();
  75         ia64_srlz_i();
  76 }
  77
  78 static void kvm_flush_tlb_all(void)
  79 {
  80         unsigned long i, j, count0, count1, stride0, stride1, addr;
  81         long flags;
  82
  83         addr    = local_cpu_data->ptce_base;
  84         count0  = local_cpu_data->ptce_count[0];
  85         count1  = local_cpu_data->ptce_count[1];
  86         stride0 = local_cpu_data->ptce_stride[0];
  87         stride1 = local_cpu_data->ptce_stride[1];
  88
  89         local_irq_save(flags);
  90         for (i = 0; i < count0; ++i) {
  91                 for (j = 0; j < count1; ++j) {
  92                         ia64_ptce(addr);
  93                         addr += stride1;
  94                 }
  95                 addr += stride0;
  96         }
  97         local_irq_restore(flags);
  98         ia64_srlz_i();                  /* srlz.i implies srlz.d */
  99 }
 100
 101 long ia64_pal_vp_create(u64 *vpd, u64 *host_iva, u64 *opt_handler)
 102 {
 103         struct ia64_pal_retval iprv;
 104
 105         PAL_CALL_STK(iprv, PAL_VP_CREATE, (u64)vpd, (u64)host_iva,
 106                         (u64)opt_handler);
 107
 108         return iprv.status;
 109 }
 110
 111 static  DEFINE_SPINLOCK(vp_lock);
 112
 113 void kvm_arch_hardware_enable(void *garbage)
 114 {
 115         long  status;
 116         long  tmp_base;
 117         unsigned long pte;
 118         unsigned long saved_psr;
 119         int slot;
 120
 121         pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base),
 122                                 PAGE_KERNEL));
 123         local_irq_save(saved_psr);
 124         slot = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
 125         local_irq_restore(saved_psr);
 126         if (slot < 0)
 127                 return;
 128
 129         spin_lock(&vp_lock);
 130         status = ia64_pal_vp_init_env(kvm_vsa_base ?
 131                                 VP_INIT_ENV : VP_INIT_ENV_INITALIZE,
 132                         __pa(kvm_vm_buffer), KVM_VM_BUFFER_BASE, &tmp_base);
 133         if (status != 0) {
 134                 printk(KERN_WARNING"kvm: Failed to Enable VT Support!!!!\n");
 135                 return ;
 136         }
 137
 138         if (!kvm_vsa_base) {
 139                 kvm_vsa_base = tmp_base;
 140                 printk(KERN_INFO"kvm: kvm_vsa_base:0x%lx\n", kvm_vsa_base);
 141         }
 142         spin_unlock(&vp_lock);
 143         ia64_ptr_entry(0x3, slot);
 144 }
 145
 146 void kvm_arch_hardware_disable(void *garbage)
 147 {
 148
 149         long status;
 150         int slot;
 151         unsigned long pte;
 152         unsigned long saved_psr;
 153         unsigned long host_iva = ia64_getreg(_IA64_REG_CR_IVA);
 154
 155         pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base),
 156                                 PAGE_KERNEL));
 157
 158         local_irq_save(saved_psr);
 159         slot = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
 160         local_irq_restore(saved_psr);
 161         if (slot < 0)
 162                 return;
 163
 164         status = ia64_pal_vp_exit_env(host_iva);
 165         if (status)
 166                 printk(KERN_DEBUG"kvm: Failed to disable VT support! :%ld\n",
 167                                 status);
 168         ia64_ptr_entry(0x3, slot);
 169 }
 170
 171 void kvm_arch_check_processor_compat(void *rtn)
 172 {
 173         *(int *)rtn = 0;
 174 }
 175
 176 int kvm_dev_ioctl_check_extension(long ext)
 177 {
 178
 179         int r;
 180
 181         switch (ext) {
 182         case KVM_CAP_IRQCHIP:
 183         case KVM_CAP_USER_MEMORY:
 184         case KVM_CAP_MP_STATE:
 185
 186                 r = 1;
 187                 break;
 188         case KVM_CAP_COALESCED_MMIO:
 189                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 190                 break;
 191         case KVM_CAP_IOMMU:
 192                 r = intel_iommu_found();
 193                 break;
 194         default:
 195                 r = 0;
 196         }
 197         return r;
 198
 199 }
 200
 201 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
 202                                         gpa_t addr, int len, int is_write)
 203 {
 204         struct kvm_io_device *dev;
 205
 206         dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len, is_write);
 207
 208         return dev;
 209 }
 210
 211 static int handle_vm_error(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 212 {
 213         kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
 214         kvm_run->hw.hardware_exit_reason = 1;
 215         return 0;
 216 }
 217
 218 static int handle_mmio(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 219 {
 220         struct kvm_mmio_req *p;
 221         struct kvm_io_device *mmio_dev;
 222
 223         p = kvm_get_vcpu_ioreq(vcpu);
 224
 225         if ((p->addr & PAGE_MASK) == IOAPIC_DEFAULT_BASE_ADDRESS)
 226                 goto mmio;
 227         vcpu->mmio_needed = 1;
 228         vcpu->mmio_phys_addr = kvm_run->mmio.phys_addr = p->addr;
 229         vcpu->mmio_size = kvm_run->mmio.len = p->size;
 230         vcpu->mmio_is_write = kvm_run->mmio.is_write = !p->dir;
 231
 232         if (vcpu->mmio_is_write)
 233                 memcpy(vcpu->mmio_data, &p->data, p->size);
 234         memcpy(kvm_run->mmio.data, &p->data, p->size);
 235         kvm_run->exit_reason = KVM_EXIT_MMIO;
 236         return 0;
 237 mmio:
 238         mmio_dev = vcpu_find_mmio_dev(vcpu, p->addr, p->size, !p->dir);
 239         if (mmio_dev) {
 240                 if (!p->dir)
 241                         kvm_iodevice_write(mmio_dev, p->addr, p->size,
 242                                                 &p->data);
 243                 else
 244                         kvm_iodevice_read(mmio_dev, p->addr, p->size,
 245                                                 &p->data);
 246
 247         } else
 248                 printk(KERN_ERR"kvm: No iodevice found! addr:%lx\n", p->addr);
 249         p->state = STATE_IORESP_READY;
 250
 251         return 1;
 252 }
 253
 254 static int handle_pal_call(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 255 {
 256         struct exit_ctl_data *p;
 257
 258         p = kvm_get_exit_data(vcpu);
 259
 260         if (p->exit_reason == EXIT_REASON_PAL_CALL)
 261                 return kvm_pal_emul(vcpu, kvm_run);
 262         else {
 263                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
 264                 kvm_run->hw.hardware_exit_reason = 2;
 265                 return 0;
 266         }
 267 }
 268
 269 static int handle_sal_call(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 270 {
 271         struct exit_ctl_data *p;
 272
 273         p = kvm_get_exit_data(vcpu);
 274
 275         if (p->exit_reason == EXIT_REASON_SAL_CALL) {
 276                 kvm_sal_emul(vcpu);
 277                 return 1;
 278         } else {
 279                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
 280                 kvm_run->hw.hardware_exit_reason = 3;
 281                 return 0;
 282         }
 283
 284 }
 285
 286 /*
 287  *  offset: address offset to IPI space.
 288  *  value:  deliver value.
 289  */
 290 static void vcpu_deliver_ipi(struct kvm_vcpu *vcpu, uint64_t dm,
 291                                 uint64_t vector)
 292 {
 293         switch (dm) {
 294         case SAPIC_FIXED:
 295                 kvm_apic_set_irq(vcpu, vector, 0);
 296                 break;
 297         case SAPIC_NMI:
 298                 kvm_apic_set_irq(vcpu, 2, 0);
 299                 break;
 300         case SAPIC_EXTINT:
 301                 kvm_apic_set_irq(vcpu, 0, 0);
 302                 break;
 303         case SAPIC_INIT:
 304         case SAPIC_PMI:
 305         default:
 306                 printk(KERN_ERR"kvm: Unimplemented Deliver reserved IPI!\n");
 307                 break;
 308         }
 309 }
 310
 311 static struct kvm_vcpu *lid_to_vcpu(struct kvm *kvm, unsigned long id,
 312                         unsigned long eid)
 313 {
 314         union ia64_lid lid;
 315         int i;
 316
 317         for (i = 0; i < KVM_MAX_VCPUS; i++) {
 318                 if (kvm->vcpus[i]) {
 319                         lid.val = VCPU_LID(kvm->vcpus[i]);
 320                         if (lid.id == id && lid.eid == eid)
 321                                 return kvm->vcpus[i];
 322                 }
 323         }
 324
 325         return NULL;
 326 }
 327
 328 static int handle_ipi(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 329 {
 330         struct exit_ctl_data *p = kvm_get_exit_data(vcpu);
 331         struct kvm_vcpu *target_vcpu;
 332         struct kvm_pt_regs *regs;
 333         union ia64_ipi_a addr = p->u.ipi_data.addr;
 334         union ia64_ipi_d data = p->u.ipi_data.data;
 335
 336         target_vcpu = lid_to_vcpu(vcpu->kvm, addr.id, addr.eid);
 337         if (!target_vcpu)
 338                 return handle_vm_error(vcpu, kvm_run);
 339
 340         if (!target_vcpu->arch.launched) {
 341                 regs = vcpu_regs(target_vcpu);
 342
 343                 regs->cr_iip = vcpu->kvm->arch.rdv_sal_data.boot_ip;
 344                 regs->r1 = vcpu->kvm->arch.rdv_sal_data.boot_gp;
 345
 346                 target_vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
 347                 if (waitqueue_active(&target_vcpu->wq))
 348                         wake_up_interruptible(&target_vcpu->wq);
 349         } else {
 350                 vcpu_deliver_ipi(target_vcpu, data.dm, data.vector);
 351                 if (target_vcpu != vcpu)
 352                         kvm_vcpu_kick(target_vcpu);
 353         }
 354
 355         return 1;
 356 }
 357
 358 struct call_data {
 359         struct kvm_ptc_g ptc_g_data;
 360         struct kvm_vcpu *vcpu;
 361 };
 362
 363 static void vcpu_global_purge(void *info)
 364 {
 365         struct call_data *p = (struct call_data *)info;
 366         struct kvm_vcpu *vcpu = p->vcpu;
 367
 368         if (test_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 369                 return;
 370
 371         set_bit(KVM_REQ_PTC_G, &vcpu->requests);
 372         if (vcpu->arch.ptc_g_count < MAX_PTC_G_NUM) {
 373                 vcpu->arch.ptc_g_data[vcpu->arch.ptc_g_count++] =
 374                                                         p->ptc_g_data;
 375         } else {
 376                 clear_bit(KVM_REQ_PTC_G, &vcpu->requests);
 377                 vcpu->arch.ptc_g_count = 0;
 378                 set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests);
 379         }
 380 }
 381
 382 static int handle_global_purge(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 383 {
 384         struct exit_ctl_data *p = kvm_get_exit_data(vcpu);
 385         struct kvm *kvm = vcpu->kvm;
 386         struct call_data call_data;
 387         int i;
 388         call_data.ptc_g_data = p->u.ptc_g_data;
 389
 390         for (i = 0; i < KVM_MAX_VCPUS; i++) {
 391                 if (!kvm->vcpus[i] || kvm->vcpus[i]->arch.mp_state ==
 392                                                 KVM_MP_STATE_UNINITIALIZED ||
 393                                         vcpu == kvm->vcpus[i])
 394                         continue;
 395
 396                 if (waitqueue_active(&kvm->vcpus[i]->wq))
 397                         wake_up_interruptible(&kvm->vcpus[i]->wq);
 398
 399                 if (kvm->vcpus[i]->cpu != -1) {
 400                         call_data.vcpu = kvm->vcpus[i];
 401                         smp_call_function_single(kvm->vcpus[i]->cpu,
 402                                         vcpu_global_purge, &call_data, 1);
 403                 } else
 404                         printk(KERN_WARNING"kvm: Uninit vcpu received ipi!\n");
 405
 406         }
 407         return 1;
 408 }
 409
 410 static int handle_switch_rr6(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 411 {
 412         return 1;
 413 }
 414
 415 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
 416 {
 417
 418         ktime_t kt;
 419         long itc_diff;
 420         unsigned long vcpu_now_itc;
 421
 422         unsigned long expires;
 423         struct hrtimer *p_ht = &vcpu->arch.hlt_timer;
 424         unsigned long cyc_per_usec = local_cpu_data->cyc_per_usec;
 425         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
 426
 427         vcpu_now_itc = ia64_getreg(_IA64_REG_AR_ITC) + vcpu->arch.itc_offset;
 428
 429         if (time_after(vcpu_now_itc, vpd->itm)) {
 430                 vcpu->arch.timer_check = 1;
 431                 return 1;
 432         }
 433         itc_diff = vpd->itm - vcpu_now_itc;
 434         if (itc_diff < 0)
 435                 itc_diff = -itc_diff;
 436
 437         expires = div64_u64(itc_diff, cyc_per_usec);
 438         kt = ktime_set(0, 1000 * expires);
 439         vcpu->arch.ht_active = 1;
 440         hrtimer_start(p_ht, kt, HRTIMER_MODE_ABS);
 441
 442         if (irqchip_in_kernel(vcpu->kvm)) {
 443                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
 444                 kvm_vcpu_block(vcpu);
 445                 hrtimer_cancel(p_ht);
 446                 vcpu->arch.ht_active = 0;
 447
 448                 if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
 449                         return -EINTR;
 450                 return 1;
 451         } else {
 452                 printk(KERN_ERR"kvm: Unsupported userspace halt!");
 453                 return 0;
 454         }
 455 }
 456
 457 static int handle_vm_shutdown(struct kvm_vcpu *vcpu,
 458                 struct kvm_run *kvm_run)
 459 {
 460         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
 461         return 0;
 462 }
 463
 464 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
 465                 struct kvm_run *kvm_run)
 466 {
 467         return 1;
 468 }
 469
 470 static int (*kvm_vti_exit_handlers[])(struct kvm_vcpu *vcpu,
 471                 struct kvm_run *kvm_run) = {
 472         [EXIT_REASON_VM_PANIC]              = handle_vm_error,
 473         [EXIT_REASON_MMIO_INSTRUCTION]      = handle_mmio,
 474         [EXIT_REASON_PAL_CALL]              = handle_pal_call,
 475         [EXIT_REASON_SAL_CALL]              = handle_sal_call,
 476         [EXIT_REASON_SWITCH_RR6]            = handle_switch_rr6,
 477         [EXIT_REASON_VM_DESTROY]            = handle_vm_shutdown,
 478         [EXIT_REASON_EXTERNAL_INTERRUPT]    = handle_external_interrupt,
 479         [EXIT_REASON_IPI]                   = handle_ipi,
 480         [EXIT_REASON_PTC_G]                 = handle_global_purge,
 481
 482 };
 483
 484 static const int kvm_vti_max_exit_handlers =
 485                 sizeof(kvm_vti_exit_handlers)/sizeof(*kvm_vti_exit_handlers);
 486
 487 static void kvm_prepare_guest_switch(struct kvm_vcpu *vcpu)
 488 {
 489 }
 490
 491 static uint32_t kvm_get_exit_reason(struct kvm_vcpu *vcpu)
 492 {
 493         struct exit_ctl_data *p_exit_data;
 494
 495         p_exit_data = kvm_get_exit_data(vcpu);
 496         return p_exit_data->exit_reason;
 497 }
 498
 499 /*
 500  * The guest has exited.  See if we can fix it or if we need userspace
 501  * assistance.
 502  */
 503 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
 504 {
 505         u32 exit_reason = kvm_get_exit_reason(vcpu);
 506         vcpu->arch.last_exit = exit_reason;
 507
 508         if (exit_reason < kvm_vti_max_exit_handlers
 509                         && kvm_vti_exit_handlers[exit_reason])
 510                 return kvm_vti_exit_handlers[exit_reason](vcpu, kvm_run);
 511         else {
 512                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
 513                 kvm_run->hw.hardware_exit_reason = exit_reason;
 514         }
 515         return 0;
 516 }
 517
 518 static inline void vti_set_rr6(unsigned long rr6)
 519 {
 520         ia64_set_rr(RR6, rr6);
 521         ia64_srlz_i();
 522 }
 523
 524 static int kvm_insert_vmm_mapping(struct kvm_vcpu *vcpu)
 525 {
 526         unsigned long pte;
 527         struct kvm *kvm = vcpu->kvm;
 528         int r;
 529
 530         /*Insert a pair of tr to map vmm*/
 531         pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base), PAGE_KERNEL));
 532         r = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
 533         if (r < 0)
 534                 goto out;
 535         vcpu->arch.vmm_tr_slot = r;
 536         /*Insert a pairt of tr to map data of vm*/
 537         pte = pte_val(mk_pte_phys(__pa(kvm->arch.vm_base), PAGE_KERNEL));
 538         r = ia64_itr_entry(0x3, KVM_VM_DATA_BASE,
 539                                         pte, KVM_VM_DATA_SHIFT);
 540         if (r < 0)
 541                 goto out;
 542         vcpu->arch.vm_tr_slot = r;
 543         r = 0;
 544 out:
 545         return r;
 546
 547 }
 548
 549 static void kvm_purge_vmm_mapping(struct kvm_vcpu *vcpu)
 550 {
 551
 552         ia64_ptr_entry(0x3, vcpu->arch.vmm_tr_slot);
 553         ia64_ptr_entry(0x3, vcpu->arch.vm_tr_slot);
 554
 555 }
 556
 557 static int kvm_vcpu_pre_transition(struct kvm_vcpu *vcpu)
 558 {
 559         int cpu = smp_processor_id();
 560
 561         if (vcpu->arch.last_run_cpu != cpu ||
 562                         per_cpu(last_vcpu, cpu) != vcpu) {
 563                 per_cpu(last_vcpu, cpu) = vcpu;
 564                 vcpu->arch.last_run_cpu = cpu;
 565                 kvm_flush_tlb_all();
 566         }
 567
 568         vcpu->arch.host_rr6 = ia64_get_rr(RR6);
 569         vti_set_rr6(vcpu->arch.vmm_rr);
 570         return kvm_insert_vmm_mapping(vcpu);
 571 }
 572 static void kvm_vcpu_post_transition(struct kvm_vcpu *vcpu)
 573 {
 574         kvm_purge_vmm_mapping(vcpu);
 575         vti_set_rr6(vcpu->arch.host_rr6);
 576 }
 577
 578 static int  vti_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 579 {
 580         union context *host_ctx, *guest_ctx;
 581         int r;
 582
 583         /*Get host and guest context with guest address space.*/
 584         host_ctx = kvm_get_host_context(vcpu);
 585         guest_ctx = kvm_get_guest_context(vcpu);
 586
 587         r = kvm_vcpu_pre_transition(vcpu);
 588         if (r < 0)
 589                 goto out;
 590         kvm_vmm_info->tramp_entry(host_ctx, guest_ctx);
 591         kvm_vcpu_post_transition(vcpu);
 592         r = 0;
 593 out:
 594         return r;
 595 }
 596
 597 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 598 {
 599         int r;
 600
 601 again:
 602         preempt_disable();
 603
 604         kvm_prepare_guest_switch(vcpu);
 605         local_irq_disable();
 606
 607         if (signal_pending(current)) {
 608                 local_irq_enable();
 609                 preempt_enable();
 610                 r = -EINTR;
 611                 kvm_run->exit_reason = KVM_EXIT_INTR;
 612                 goto out;
 613         }
 614
 615         vcpu->guest_mode = 1;
 616         kvm_guest_enter();
 617
 618         r = vti_vcpu_run(vcpu, kvm_run);
 619         if (r < 0) {
 620                 local_irq_enable();
 621                 preempt_enable();
 622                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
 623                 goto out;
 624         }
 625
 626         vcpu->arch.launched = 1;
 627         vcpu->guest_mode = 0;
 628         local_irq_enable();
 629
 630         /*
 631          * We must have an instruction between local_irq_enable() and
 632          * kvm_guest_exit(), so the timer interrupt isn't delayed by
 633          * the interrupt shadow.  The stat.exits increment will do nicely.
 634          * But we need to prevent reordering, hence this barrier():
 635          */
 636         barrier();
 637
 638         kvm_guest_exit();
 639
 640         preempt_enable();
 641
 642         r = kvm_handle_exit(kvm_run, vcpu);
 643
 644         if (r > 0) {
 645                 if (!need_resched())
 646                         goto again;
 647         }
 648
 649 out:
 650         if (r > 0) {
 651                 kvm_resched(vcpu);
 652                 goto again;
 653         }
 654
 655         return r;
 656 }
 657
 658 static void kvm_set_mmio_data(struct kvm_vcpu *vcpu)
 659 {
 660         struct kvm_mmio_req *p = kvm_get_vcpu_ioreq(vcpu);
 661
 662         if (!vcpu->mmio_is_write)
 663                 memcpy(&p->data, vcpu->mmio_data, 8);
 664         p->state = STATE_IORESP_READY;
 665 }
 666
 667 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 668 {
 669         int r;
 670         sigset_t sigsaved;
 671
 672         vcpu_load(vcpu);
 673
 674         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
 675                 kvm_vcpu_block(vcpu);
 676                 vcpu_put(vcpu);
 677                 return -EAGAIN;
 678         }
 679
 680         if (vcpu->sigset_active)
 681                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 682
 683         if (vcpu->mmio_needed) {
 684                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
 685                 kvm_set_mmio_data(vcpu);
 686                 vcpu->mmio_read_completed = 1;
 687                 vcpu->mmio_needed = 0;
 688         }
 689         r = __vcpu_run(vcpu, kvm_run);
 690
 691         if (vcpu->sigset_active)
 692                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 693
 694         vcpu_put(vcpu);
 695         return r;
 696 }
 697
 698 /*
 699  * Allocate 16M memory for every vm to hold its specific data.
 700  * Its memory map is defined in kvm_host.h.
 701  */
 702 static struct kvm *kvm_alloc_kvm(void)
 703 {
 704
 705         struct kvm *kvm;
 706         uint64_t  vm_base;
 707
 708         vm_base = __get_free_pages(GFP_KERNEL, get_order(KVM_VM_DATA_SIZE));
 709
 710         if (!vm_base)
 711                 return ERR_PTR(-ENOMEM);
 712         printk(KERN_DEBUG"kvm: VM data's base Address:0x%lx\n", vm_base);
 713
 714         /* Zero all pages before use! */
 715         memset((void *)vm_base, 0, KVM_VM_DATA_SIZE);
 716
 717         kvm = (struct kvm *)(vm_base + KVM_VM_OFS);
 718         kvm->arch.vm_base = vm_base;
 719
 720         return kvm;
 721 }
 722
 723 struct kvm_io_range {
 724         unsigned long start;
 725         unsigned long size;
 726         unsigned long type;
 727 };
 728
 729 static const struct kvm_io_range io_ranges[] = {
 730         {VGA_IO_START, VGA_IO_SIZE, GPFN_FRAME_BUFFER},
 731         {MMIO_START, MMIO_SIZE, GPFN_LOW_MMIO},
 732         {LEGACY_IO_START, LEGACY_IO_SIZE, GPFN_LEGACY_IO},
 733         {IO_SAPIC_START, IO_SAPIC_SIZE, GPFN_IOSAPIC},
 734         {PIB_START, PIB_SIZE, GPFN_PIB},
 735 };
 736
 737 static void kvm_build_io_pmt(struct kvm *kvm)
 738 {
 739         unsigned long i, j;
 740
 741         /* Mark I/O ranges */
 742         for (i = 0; i < (sizeof(io_ranges) / sizeof(struct kvm_io_range));
 743                                                         i++) {
 744                 for (j = io_ranges[i].start;
 745                                 j < io_ranges[i].start + io_ranges[i].size;
 746                                 j += PAGE_SIZE)
 747                         kvm_set_pmt_entry(kvm, j >> PAGE_SHIFT,
 748                                         io_ranges[i].type, 0);
 749         }
 750
 751 }
 752
 753 /*Use unused rids to virtualize guest rid.*/
 754 #define GUEST_PHYSICAL_RR0      0x1739
 755 #define GUEST_PHYSICAL_RR4      0x2739
 756 #define VMM_INIT_RR             0x1660
 757
 758 static void kvm_init_vm(struct kvm *kvm)
 759 {
 760         long vm_base;
 761
 762         BUG_ON(!kvm);
 763
 764         kvm->arch.metaphysical_rr0 = GUEST_PHYSICAL_RR0;
 765         kvm->arch.metaphysical_rr4 = GUEST_PHYSICAL_RR4;
 766         kvm->arch.vmm_init_rr = VMM_INIT_RR;
 767
 768         vm_base = kvm->arch.vm_base;
 769         if (vm_base) {
 770                 kvm->arch.vhpt_base = vm_base + KVM_VHPT_OFS;
 771                 kvm->arch.vtlb_base = vm_base + KVM_VTLB_OFS;
 772                 kvm->arch.vpd_base  = vm_base + KVM_VPD_OFS;
 773         }
 774
 775         /*
 776          *Fill P2M entries for MMIO/IO ranges
 777          */
 778         kvm_build_io_pmt(kvm);
 779
 780         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
 781 }
 782
 783 struct  kvm *kvm_arch_create_vm(void)
 784 {
 785         struct kvm *kvm = kvm_alloc_kvm();
 786
 787         if (IS_ERR(kvm))
 788                 return ERR_PTR(-ENOMEM);
 789         kvm_init_vm(kvm);
 790
 791         return kvm;
 792
 793 }
 794
 795 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm,
 796                                         struct kvm_irqchip *chip)
 797 {
 798         int r;
 799
 800         r = 0;
 801         switch (chip->chip_id) {
 802         case KVM_IRQCHIP_IOAPIC:
 803                 memcpy(&chip->chip.ioapic, ioapic_irqchip(kvm),
 804                                 sizeof(struct kvm_ioapic_state));
 805                 break;
 806         default:
 807                 r = -EINVAL;
 808                 break;
 809         }
 810         return r;
 811 }
 812
 813 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 814 {
 815         int r;
 816
 817         r = 0;
 818         switch (chip->chip_id) {
 819         case KVM_IRQCHIP_IOAPIC:
 820                 memcpy(ioapic_irqchip(kvm),
 821                                 &chip->chip.ioapic,
 822                                 sizeof(struct kvm_ioapic_state));
 823                 break;
 824         default:
 825                 r = -EINVAL;
 826                 break;
 827         }
 828         return r;
 829 }
 830
 831 #define RESTORE_REGS(_x) vcpu->arch._x = regs->_x
 832
 833 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 834 {
 835         int i;
 836         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
 837         int r;
 838
 839         vcpu_load(vcpu);
 840
 841         for (i = 0; i < 16; i++) {
 842                 vpd->vgr[i] = regs->vpd.vgr[i];
 843                 vpd->vbgr[i] = regs->vpd.vbgr[i];
 844         }
 845         for (i = 0; i < 128; i++)
 846                 vpd->vcr[i] = regs->vpd.vcr[i];
 847         vpd->vhpi = regs->vpd.vhpi;
 848         vpd->vnat = regs->vpd.vnat;
 849         vpd->vbnat = regs->vpd.vbnat;
 850         vpd->vpsr = regs->vpd.vpsr;
 851
 852         vpd->vpr = regs->vpd.vpr;
 853
 854         r = -EFAULT;
 855         r = copy_from_user(&vcpu->arch.guest, regs->saved_guest,
 856                                                 sizeof(union context));
 857         if (r)
 858                 goto out;
 859         r = copy_from_user(vcpu + 1, regs->saved_stack +
 860                         sizeof(struct kvm_vcpu),
 861                         IA64_STK_OFFSET - sizeof(struct kvm_vcpu));
 862         if (r)
 863                 goto out;
 864         vcpu->arch.exit_data =
 865                 ((struct kvm_vcpu *)(regs->saved_stack))->arch.exit_data;
 866
 867         RESTORE_REGS(mp_state);
 868         RESTORE_REGS(vmm_rr);
 869         memcpy(vcpu->arch.itrs, regs->itrs, sizeof(struct thash_data) * NITRS);
 870         memcpy(vcpu->arch.dtrs, regs->dtrs, sizeof(struct thash_data) * NDTRS);
 871         RESTORE_REGS(itr_regions);
 872         RESTORE_REGS(dtr_regions);
 873         RESTORE_REGS(tc_regions);
 874         RESTORE_REGS(irq_check);
 875         RESTORE_REGS(itc_check);
 876         RESTORE_REGS(timer_check);
 877         RESTORE_REGS(timer_pending);
 878         RESTORE_REGS(last_itc);
 879         for (i = 0; i < 8; i++) {
 880                 vcpu->arch.vrr[i] = regs->vrr[i];
 881                 vcpu->arch.ibr[i] = regs->ibr[i];
 882                 vcpu->arch.dbr[i] = regs->dbr[i];
 883         }
 884         for (i = 0; i < 4; i++)
 885                 vcpu->arch.insvc[i] = regs->insvc[i];
 886         RESTORE_REGS(xtp);
 887         RESTORE_REGS(metaphysical_rr0);
 888         RESTORE_REGS(metaphysical_rr4);
 889         RESTORE_REGS(metaphysical_saved_rr0);
 890         RESTORE_REGS(metaphysical_saved_rr4);
 891         RESTORE_REGS(fp_psr);
 892         RESTORE_REGS(saved_gp);
 893
 894         vcpu->arch.irq_new_pending = 1;
 895         vcpu->arch.itc_offset = regs->saved_itc - ia64_getreg(_IA64_REG_AR_ITC);
 896         set_bit(KVM_REQ_RESUME, &vcpu->requests);
 897
 898         vcpu_put(vcpu);
 899         r = 0;
 900 out:
 901         return r;
 902 }
 903
 904 long kvm_arch_vm_ioctl(struct file *filp,
 905                 unsigned int ioctl, unsigned long arg)
 906 {
 907         struct kvm *kvm = filp->private_data;
 908         void __user *argp = (void __user *)arg;
 909         int r = -EINVAL;
 910
 911         switch (ioctl) {
 912         case KVM_SET_MEMORY_REGION: {
 913                 struct kvm_memory_region kvm_mem;
 914                 struct kvm_userspace_memory_region kvm_userspace_mem;
 915
 916                 r = -EFAULT;
 917                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
 918                         goto out;
 919                 kvm_userspace_mem.slot = kvm_mem.slot;
 920                 kvm_userspace_mem.flags = kvm_mem.flags;
 921                 kvm_userspace_mem.guest_phys_addr =
 922                                         kvm_mem.guest_phys_addr;
 923                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
 924                 r = kvm_vm_ioctl_set_memory_region(kvm,
 925                                         &kvm_userspace_mem, 0);
 926                 if (r)
 927                         goto out;
 928                 break;
 929                 }
 930         case KVM_CREATE_IRQCHIP:
 931                 r = -EFAULT;
 932                 r = kvm_ioapic_init(kvm);
 933                 if (r)
 934                         goto out;
 935                 break;
 936         case KVM_IRQ_LINE: {
 937                 struct kvm_irq_level irq_event;
 938
 939                 r = -EFAULT;
 940                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
 941                         goto out;
 942                 if (irqchip_in_kernel(kvm)) {
 943                         mutex_lock(&kvm->lock);
 944                         kvm_ioapic_set_irq(kvm->arch.vioapic,
 945                                                 irq_event.irq,
 946                                                 irq_event.level);
 947                         mutex_unlock(&kvm->lock);
 948                         r = 0;
 949                 }
 950                 break;
 951                 }
 952         case KVM_GET_IRQCHIP: {
 953                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
 954                 struct kvm_irqchip chip;
 955
 956                 r = -EFAULT;
 957                 if (copy_from_user(&chip, argp, sizeof chip))
 958                                 goto out;
 959                 r = -ENXIO;
 960                 if (!irqchip_in_kernel(kvm))
 961                         goto out;
 962                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
 963                 if (r)
 964                         goto out;
 965                 r = -EFAULT;
 966                 if (copy_to_user(argp, &chip, sizeof chip))
 967                                 goto out;
 968                 r = 0;
 969                 break;
 970                 }
 971         case KVM_SET_IRQCHIP: {
 972                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
 973                 struct kvm_irqchip chip;
 974
 975                 r = -EFAULT;
 976                 if (copy_from_user(&chip, argp, sizeof chip))
 977                                 goto out;
 978                 r = -ENXIO;
 979                 if (!irqchip_in_kernel(kvm))
 980                         goto out;
 981                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
 982                 if (r)
 983                         goto out;
 984                 r = 0;
 985                 break;
 986                 }
 987         default:
 988                 ;
 989         }
 990 out:
 991         return r;
 992 }
 993
 994 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 995                 struct kvm_sregs *sregs)
 996 {
 997         return -EINVAL;
 998 }
 999
1000 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1001                 struct kvm_sregs *sregs)
1002 {
1003         return -EINVAL;
1004
1005 }
1006 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1007                 struct kvm_translation *tr)
1008 {
1009
1010         return -EINVAL;
1011 }
1012
1013 static int kvm_alloc_vmm_area(void)
1014 {
1015         if (!kvm_vmm_base && (kvm_vm_buffer_size < KVM_VM_BUFFER_SIZE)) {
1016                 kvm_vmm_base = __get_free_pages(GFP_KERNEL,
1017                                 get_order(KVM_VMM_SIZE));
1018                 if (!kvm_vmm_base)
1019                         return -ENOMEM;
1020
1021                 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1022                 kvm_vm_buffer = kvm_vmm_base + VMM_SIZE;
1023
1024                 printk(KERN_DEBUG"kvm:VMM's Base Addr:0x%lx, vm_buffer:0x%lx\n",
1025                                 kvm_vmm_base, kvm_vm_buffer);
1026         }
1027
1028         return 0;
1029 }
1030
1031 static void kvm_free_vmm_area(void)
1032 {
1033         if (kvm_vmm_base) {
1034                 /*Zero this area before free to avoid bits leak!!*/
1035                 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1036                 free_pages(kvm_vmm_base, get_order(KVM_VMM_SIZE));
1037                 kvm_vmm_base  = 0;
1038                 kvm_vm_buffer = 0;
1039                 kvm_vsa_base = 0;
1040         }
1041 }
1042
1043 static void vti_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1044 {
1045 }
1046
1047 static int vti_init_vpd(struct kvm_vcpu *vcpu)
1048 {
1049         int i;
1050         union cpuid3_t cpuid3;
1051         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1052
1053         if (IS_ERR(vpd))
1054                 return PTR_ERR(vpd);
1055
1056         /* CPUID init */
1057         for (i = 0; i < 5; i++)
1058                 vpd->vcpuid[i] = ia64_get_cpuid(i);
1059
1060         /* Limit the CPUID number to 5 */
1061         cpuid3.value = vpd->vcpuid[3];
1062         cpuid3.number = 4;      /* 5 - 1 */
1063         vpd->vcpuid[3] = cpuid3.value;
1064
1065         /*Set vac and vdc fields*/
1066         vpd->vac.a_from_int_cr = 1;
1067         vpd->vac.a_to_int_cr = 1;
1068         vpd->vac.a_from_psr = 1;
1069         vpd->vac.a_from_cpuid = 1;
1070         vpd->vac.a_cover = 1;
1071         vpd->vac.a_bsw = 1;
1072         vpd->vac.a_int = 1;
1073         vpd->vdc.d_vmsw = 1;
1074
1075         /*Set virtual buffer*/
1076         vpd->virt_env_vaddr = KVM_VM_BUFFER_BASE;
1077
1078         return 0;
1079 }
1080
1081 static int vti_create_vp(struct kvm_vcpu *vcpu)
1082 {
1083         long ret;
1084         struct vpd *vpd = vcpu->arch.vpd;
1085         unsigned long  vmm_ivt;
1086
1087         vmm_ivt = kvm_vmm_info->vmm_ivt;
1088
1089         printk(KERN_DEBUG "kvm: vcpu:%p,ivt: 0x%lx\n", vcpu, vmm_ivt);
1090
1091         ret = ia64_pal_vp_create((u64 *)vpd, (u64 *)vmm_ivt, 0);
1092
1093         if (ret) {
1094                 printk(KERN_ERR"kvm: ia64_pal_vp_create failed!\n");
1095                 return -EINVAL;
1096         }
1097         return 0;
1098 }
1099
1100 static void init_ptce_info(struct kvm_vcpu *vcpu)
1101 {
1102         ia64_ptce_info_t ptce = {0};
1103
1104         ia64_get_ptce(&ptce);
1105         vcpu->arch.ptce_base = ptce.base;
1106         vcpu->arch.ptce_count[0] = ptce.count[0];
1107         vcpu->arch.ptce_count[1] = ptce.count[1];
1108         vcpu->arch.ptce_stride[0] = ptce.stride[0];
1109         vcpu->arch.ptce_stride[1] = ptce.stride[1];
1110 }
1111
1112 static void kvm_migrate_hlt_timer(struct kvm_vcpu *vcpu)
1113 {
1114         struct hrtimer *p_ht = &vcpu->arch.hlt_timer;
1115
1116         if (hrtimer_cancel(p_ht))
1117                 hrtimer_start_expires(p_ht, HRTIMER_MODE_ABS);
1118 }
1119
1120 static enum hrtimer_restart hlt_timer_fn(struct hrtimer *data)
1121 {
1122         struct kvm_vcpu *vcpu;
1123         wait_queue_head_t *q;
1124
1125         vcpu  = container_of(data, struct kvm_vcpu, arch.hlt_timer);
1126         if (vcpu->arch.mp_state != KVM_MP_STATE_HALTED)
1127                 goto out;
1128
1129         q = &vcpu->wq;
1130         if (waitqueue_active(q)) {
1131                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1132                 wake_up_interruptible(q);
1133         }
1134 out:
1135         vcpu->arch.timer_check = 1;
1136         return HRTIMER_NORESTART;
1137 }
1138
1139 #define PALE_RESET_ENTRY    0x80000000ffffffb0UL
1140
1141 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1142 {
1143         struct kvm_vcpu *v;
1144         int r;
1145         int i;
1146         long itc_offset;
1147         struct kvm *kvm = vcpu->kvm;
1148         struct kvm_pt_regs *regs = vcpu_regs(vcpu);
1149
1150         union context *p_ctx = &vcpu->arch.guest;
1151         struct kvm_vcpu *vmm_vcpu = to_guest(vcpu->kvm, vcpu);
1152
1153         /*Init vcpu context for first run.*/
1154         if (IS_ERR(vmm_vcpu))
1155                 return PTR_ERR(vmm_vcpu);
1156
1157         if (vcpu->vcpu_id == 0) {
1158                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1159
1160                 /*Set entry address for first run.*/
1161                 regs->cr_iip = PALE_RESET_ENTRY;
1162
1163                 /*Initilize itc offset for vcpus*/
1164                 itc_offset = 0UL - ia64_getreg(_IA64_REG_AR_ITC);
1165                 for (i = 0; i < MAX_VCPU_NUM; i++) {
1166                         v = (struct kvm_vcpu *)((char *)vcpu + VCPU_SIZE * i);
1167                         v->arch.itc_offset = itc_offset;
1168                         v->arch.last_itc = 0;
1169                 }
1170         } else
1171                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
1172
1173         r = -ENOMEM;
1174         vcpu->arch.apic = kzalloc(sizeof(struct kvm_lapic), GFP_KERNEL);
1175         if (!vcpu->arch.apic)
1176                 goto out;
1177         vcpu->arch.apic->vcpu = vcpu;
1178
1179         p_ctx->gr[1] = 0;
1180         p_ctx->gr[12] = (unsigned long)((char *)vmm_vcpu + IA64_STK_OFFSET);
1181         p_ctx->gr[13] = (unsigned long)vmm_vcpu;
1182         p_ctx->psr = 0x1008522000UL;
1183         p_ctx->ar[40] = FPSR_DEFAULT; /*fpsr*/
1184         p_ctx->caller_unat = 0;
1185         p_ctx->pr = 0x0;
1186         p_ctx->ar[36] = 0x0; /*unat*/
1187         p_ctx->ar[19] = 0x0; /*rnat*/
1188         p_ctx->ar[18] = (unsigned long)vmm_vcpu +
1189                                 ((sizeof(struct kvm_vcpu)+15) & ~15);
1190         p_ctx->ar[64] = 0x0; /*pfs*/
1191         p_ctx->cr[0] = 0x7e04UL;
1192         p_ctx->cr[2] = (unsigned long)kvm_vmm_info->vmm_ivt;
1193         p_ctx->cr[8] = 0x3c;
1194
1195         /*Initilize region register*/
1196         p_ctx->rr[0] = 0x30;
1197         p_ctx->rr[1] = 0x30;
1198         p_ctx->rr[2] = 0x30;
1199         p_ctx->rr[3] = 0x30;
1200         p_ctx->rr[4] = 0x30;
1201         p_ctx->rr[5] = 0x30;
1202         p_ctx->rr[7] = 0x30;
1203
1204         /*Initilize branch register 0*/
1205         p_ctx->br[0] = *(unsigned long *)kvm_vmm_info->vmm_entry;
1206
1207         vcpu->arch.vmm_rr = kvm->arch.vmm_init_rr;
1208         vcpu->arch.metaphysical_rr0 = kvm->arch.metaphysical_rr0;
1209         vcpu->arch.metaphysical_rr4 = kvm->arch.metaphysical_rr4;
1210
1211         hrtimer_init(&vcpu->arch.hlt_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1212         vcpu->arch.hlt_timer.function = hlt_timer_fn;
1213
1214         vcpu->arch.last_run_cpu = -1;
1215         vcpu->arch.vpd = (struct vpd *)VPD_ADDR(vcpu->vcpu_id);
1216         vcpu->arch.vsa_base = kvm_vsa_base;
1217         vcpu->arch.__gp = kvm_vmm_gp;
1218         vcpu->arch.dirty_log_lock_pa = __pa(&kvm->arch.dirty_log_lock);
1219         vcpu->arch.vhpt.hash = (struct thash_data *)VHPT_ADDR(vcpu->vcpu_id);
1220         vcpu->arch.vtlb.hash = (struct thash_data *)VTLB_ADDR(vcpu->vcpu_id);
1221         init_ptce_info(vcpu);
1222
1223         r = 0;
1224 out:
1225         return r;
1226 }
1227
1228 static int vti_vcpu_setup(struct kvm_vcpu *vcpu, int id)
1229 {
1230         unsigned long psr;
1231         int r;
1232
1233         local_irq_save(psr);
1234         r = kvm_insert_vmm_mapping(vcpu);
1235         if (r)
1236                 goto fail;
1237         r = kvm_vcpu_init(vcpu, vcpu->kvm, id);
1238         if (r)
1239                 goto fail;
1240
1241         r = vti_init_vpd(vcpu);
1242         if (r) {
1243                 printk(KERN_DEBUG"kvm: vpd init error!!\n");
1244                 goto uninit;
1245         }
1246
1247         r = vti_create_vp(vcpu);
1248         if (r)
1249                 goto uninit;
1250
1251         kvm_purge_vmm_mapping(vcpu);
1252         local_irq_restore(psr);
1253
1254         return 0;
1255 uninit:
1256         kvm_vcpu_uninit(vcpu);
1257 fail:
1258         local_irq_restore(psr);
1259         return r;
1260 }
1261
1262 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
1263                 unsigned int id)
1264 {
1265         struct kvm_vcpu *vcpu;
1266         unsigned long vm_base = kvm->arch.vm_base;
1267         int r;
1268         int cpu;
1269
1270         r = -ENOMEM;
1271         if (!vm_base) {
1272                 printk(KERN_ERR"kvm: Create vcpu[%d] error!\n", id);
1273                 goto fail;
1274         }
1275         vcpu = (struct kvm_vcpu *)(vm_base + KVM_VCPU_OFS + VCPU_SIZE * id);
1276         vcpu->kvm = kvm;
1277
1278         cpu = get_cpu();
1279         vti_vcpu_load(vcpu, cpu);
1280         r = vti_vcpu_setup(vcpu, id);
1281         put_cpu();
1282
1283         if (r) {
1284                 printk(KERN_DEBUG"kvm: vcpu_setup error!!\n");
1285                 goto fail;
1286         }
1287
1288         return vcpu;
1289 fail:
1290         return ERR_PTR(r);
1291 }
1292
1293 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1294 {
1295         return 0;
1296 }
1297
1298 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1299 {
1300         return -EINVAL;
1301 }
1302
1303 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1304 {
1305         return -EINVAL;
1306 }
1307
1308 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
1309                 struct kvm_debug_guest *dbg)
1310 {
1311         return -EINVAL;
1312 }
1313
1314 static void free_kvm(struct kvm *kvm)
1315 {
1316         unsigned long vm_base = kvm->arch.vm_base;
1317
1318         if (vm_base) {
1319                 memset((void *)vm_base, 0, KVM_VM_DATA_SIZE);
1320                 free_pages(vm_base, get_order(KVM_VM_DATA_SIZE));
1321         }
1322
1323 }
1324
1325 static void kvm_release_vm_pages(struct kvm *kvm)
1326 {
1327         struct kvm_memory_slot *memslot;
1328         int i, j;
1329         unsigned long base_gfn;
1330
1331         for (i = 0; i < kvm->nmemslots; i++) {
1332                 memslot = &kvm->memslots[i];
1333                 base_gfn = memslot->base_gfn;
1334
1335                 for (j = 0; j < memslot->npages; j++) {
1336                         if (memslot->rmap[j])
1337                                 put_page((struct page *)memslot->rmap[j]);
1338                 }
1339         }
1340 }
1341
1342 void kvm_arch_destroy_vm(struct kvm *kvm)
1343 {
1344         kvm_iommu_unmap_guest(kvm);
1345 #ifdef  KVM_CAP_DEVICE_ASSIGNMENT
1346         kvm_free_all_assigned_devices(kvm);
1347 #endif
1348         kfree(kvm->arch.vioapic);
1349         kvm_release_vm_pages(kvm);
1350         kvm_free_physmem(kvm);
1351         free_kvm(kvm);
1352 }
1353
1354 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1355 {
1356 }
1357
1358 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1359 {
1360         if (cpu != vcpu->cpu) {
1361                 vcpu->cpu = cpu;
1362                 if (vcpu->arch.ht_active)
1363                         kvm_migrate_hlt_timer(vcpu);
1364         }
1365 }
1366
1367 #define SAVE_REGS(_x)   regs->_x = vcpu->arch._x
1368
1369 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1370 {
1371         int i;
1372         int r;
1373         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1374         vcpu_load(vcpu);
1375
1376         for (i = 0; i < 16; i++) {
1377                 regs->vpd.vgr[i] = vpd->vgr[i];
1378                 regs->vpd.vbgr[i] = vpd->vbgr[i];
1379         }
1380         for (i = 0; i < 128; i++)
1381                 regs->vpd.vcr[i] = vpd->vcr[i];
1382         regs->vpd.vhpi = vpd->vhpi;
1383         regs->vpd.vnat = vpd->vnat;
1384         regs->vpd.vbnat = vpd->vbnat;
1385         regs->vpd.vpsr = vpd->vpsr;
1386         regs->vpd.vpr = vpd->vpr;
1387
1388         r = -EFAULT;
1389         r = copy_to_user(regs->saved_guest, &vcpu->arch.guest,
1390                                         sizeof(union context));
1391         if (r)
1392                 goto out;
1393         r = copy_to_user(regs->saved_stack, (void *)vcpu, IA64_STK_OFFSET);
1394         if (r)
1395                 goto out;
1396         SAVE_REGS(mp_state);
1397         SAVE_REGS(vmm_rr);
1398         memcpy(regs->itrs, vcpu->arch.itrs, sizeof(struct thash_data) * NITRS);
1399         memcpy(regs->dtrs, vcpu->arch.dtrs, sizeof(struct thash_data) * NDTRS);
1400         SAVE_REGS(itr_regions);
1401         SAVE_REGS(dtr_regions);
1402         SAVE_REGS(tc_regions);
1403         SAVE_REGS(irq_check);
1404         SAVE_REGS(itc_check);
1405         SAVE_REGS(timer_check);
1406         SAVE_REGS(timer_pending);
1407         SAVE_REGS(last_itc);
1408         for (i = 0; i < 8; i++) {
1409                 regs->vrr[i] = vcpu->arch.vrr[i];
1410                 regs->ibr[i] = vcpu->arch.ibr[i];
1411                 regs->dbr[i] = vcpu->arch.dbr[i];
1412         }
1413         for (i = 0; i < 4; i++)
1414                 regs->insvc[i] = vcpu->arch.insvc[i];
1415         regs->saved_itc = vcpu->arch.itc_offset + ia64_getreg(_IA64_REG_AR_ITC);
1416         SAVE_REGS(xtp);
1417         SAVE_REGS(metaphysical_rr0);
1418         SAVE_REGS(metaphysical_rr4);
1419         SAVE_REGS(metaphysical_saved_rr0);
1420         SAVE_REGS(metaphysical_saved_rr4);
1421         SAVE_REGS(fp_psr);
1422         SAVE_REGS(saved_gp);
1423         vcpu_put(vcpu);
1424         r = 0;
1425 out:
1426         return r;
1427 }
1428
1429 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
1430 {
1431
1432         hrtimer_cancel(&vcpu->arch.hlt_timer);
1433         kfree(vcpu->arch.apic);
1434 }
1435
1436
1437 long kvm_arch_vcpu_ioctl(struct file *filp,
1438                 unsigned int ioctl, unsigned long arg)
1439 {
1440         return -EINVAL;
1441 }
1442
1443 int kvm_arch_set_memory_region(struct kvm *kvm,
1444                 struct kvm_userspace_memory_region *mem,
1445                 struct kvm_memory_slot old,
1446                 int user_alloc)
1447 {
1448         unsigned long i;
1449         unsigned long pfn;
1450         int npages = mem->memory_size >> PAGE_SHIFT;
1451         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
1452         unsigned long base_gfn = memslot->base_gfn;
1453
1454         for (i = 0; i < npages; i++) {
1455                 pfn = gfn_to_pfn(kvm, base_gfn + i);
1456                 if (!kvm_is_mmio_pfn(pfn)) {
1457                         kvm_set_pmt_entry(kvm, base_gfn + i,
1458                                         pfn << PAGE_SHIFT,
1459                                 _PAGE_AR_RWX | _PAGE_MA_WB);
1460                         memslot->rmap[i] = (unsigned long)pfn_to_page(pfn);
1461                 } else {
1462                         kvm_set_pmt_entry(kvm, base_gfn + i,
1463                                         GPFN_PHYS_MMIO | (pfn << PAGE_SHIFT),
1464                                         _PAGE_MA_UC);
1465                         memslot->rmap[i] = 0;
1466                         }
1467         }
1468
1469         return 0;
1470 }
1471
1472 void kvm_arch_flush_shadow(struct kvm *kvm)
1473 {
1474 }
1475
1476 long kvm_arch_dev_ioctl(struct file *filp,
1477                 unsigned int ioctl, unsigned long arg)
1478 {
1479         return -EINVAL;
1480 }
1481
1482 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1483 {
1484         kvm_vcpu_uninit(vcpu);
1485 }
1486
1487 static int vti_cpu_has_kvm_support(void)
1488 {
1489         long  avail = 1, status = 1, control = 1;
1490         long ret;
1491
1492         ret = ia64_pal_proc_get_features(&avail, &status, &control, 0);
1493         if (ret)
1494                 goto out;
1495
1496         if (!(avail & PAL_PROC_VM_BIT))
1497                 goto out;
1498
1499         printk(KERN_DEBUG"kvm: Hardware Supports VT\n");
1500
1501         ret = ia64_pal_vp_env_info(&kvm_vm_buffer_size, &vp_env_info);
1502         if (ret)
1503                 goto out;
1504         printk(KERN_DEBUG"kvm: VM Buffer Size:0x%lx\n", kvm_vm_buffer_size);
1505
1506         if (!(vp_env_info & VP_OPCODE)) {
1507                 printk(KERN_WARNING"kvm: No opcode ability on hardware, "
1508                                 "vm_env_info:0x%lx\n", vp_env_info);
1509         }
1510
1511         return 1;
1512 out:
1513         return 0;
1514 }
1515
1516 static int kvm_relocate_vmm(struct kvm_vmm_info *vmm_info,
1517                                                 struct module *module)
1518 {
1519         unsigned long module_base;
1520         unsigned long vmm_size;
1521
1522         unsigned long vmm_offset, func_offset, fdesc_offset;
1523         struct fdesc *p_fdesc;
1524
1525         BUG_ON(!module);
1526
1527         if (!kvm_vmm_base) {
1528                 printk("kvm: kvm area hasn't been initilized yet!!\n");
1529                 return -EFAULT;
1530         }
1531
1532         /*Calculate new position of relocated vmm module.*/
1533         module_base = (unsigned long)module->module_core;
1534         vmm_size = module->core_size;
1535         if (unlikely(vmm_size > KVM_VMM_SIZE))
1536                 return -EFAULT;
1537
1538         memcpy((void *)kvm_vmm_base, (void *)module_base, vmm_size);
1539         kvm_flush_icache(kvm_vmm_base, vmm_size);
1540
1541         /*Recalculate kvm_vmm_info based on new VMM*/
1542         vmm_offset = vmm_info->vmm_ivt - module_base;
1543         kvm_vmm_info->vmm_ivt = KVM_VMM_BASE + vmm_offset;
1544         printk(KERN_DEBUG"kvm: Relocated VMM's IVT Base Addr:%lx\n",
1545                         kvm_vmm_info->vmm_ivt);
1546
1547         fdesc_offset = (unsigned long)vmm_info->vmm_entry - module_base;
1548         kvm_vmm_info->vmm_entry = (kvm_vmm_entry *)(KVM_VMM_BASE +
1549                                                         fdesc_offset);
1550         func_offset = *(unsigned long *)vmm_info->vmm_entry - module_base;
1551         p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1552         p_fdesc->ip = KVM_VMM_BASE + func_offset;
1553         p_fdesc->gp = KVM_VMM_BASE+(p_fdesc->gp - module_base);
1554
1555         printk(KERN_DEBUG"kvm: Relocated VMM's Init Entry Addr:%lx\n",
1556                         KVM_VMM_BASE+func_offset);
1557
1558         fdesc_offset = (unsigned long)vmm_info->tramp_entry - module_base;
1559         kvm_vmm_info->tramp_entry = (kvm_tramp_entry *)(KVM_VMM_BASE +
1560                         fdesc_offset);
1561         func_offset = *(unsigned long *)vmm_info->tramp_entry - module_base;
1562         p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1563         p_fdesc->ip = KVM_VMM_BASE + func_offset;
1564         p_fdesc->gp = KVM_VMM_BASE + (p_fdesc->gp - module_base);
1565
1566         kvm_vmm_gp = p_fdesc->gp;
1567
1568         printk(KERN_DEBUG"kvm: Relocated VMM's Entry IP:%p\n",
1569                                                 kvm_vmm_info->vmm_entry);
1570         printk(KERN_DEBUG"kvm: Relocated VMM's Trampoline Entry IP:0x%lx\n",
1571                                                 KVM_VMM_BASE + func_offset);
1572
1573         return 0;
1574 }
1575
1576 int kvm_arch_init(void *opaque)
1577 {
1578         int r;
1579         struct kvm_vmm_info *vmm_info = (struct kvm_vmm_info *)opaque;
1580
1581         if (!vti_cpu_has_kvm_support()) {
1582                 printk(KERN_ERR "kvm: No Hardware Virtualization Support!\n");
1583                 r = -EOPNOTSUPP;
1584                 goto out;
1585         }
1586
1587         if (kvm_vmm_info) {
1588                 printk(KERN_ERR "kvm: Already loaded VMM module!\n");
1589                 r = -EEXIST;
1590                 goto out;
1591         }
1592
1593         r = -ENOMEM;
1594         kvm_vmm_info = kzalloc(sizeof(struct kvm_vmm_info), GFP_KERNEL);
1595         if (!kvm_vmm_info)
1596                 goto out;
1597
1598         if (kvm_alloc_vmm_area())
1599                 goto out_free0;
1600
1601         r = kvm_relocate_vmm(vmm_info, vmm_info->module);
1602         if (r)
1603                 goto out_free1;
1604
1605         return 0;
1606
1607 out_free1:
1608         kvm_free_vmm_area();
1609 out_free0:
1610         kfree(kvm_vmm_info);
1611 out:
1612         return r;
1613 }
1614
1615 void kvm_arch_exit(void)
1616 {
1617         kvm_free_vmm_area();
1618         kfree(kvm_vmm_info);
1619         kvm_vmm_info = NULL;
1620 }
1621
1622 static int kvm_ia64_sync_dirty_log(struct kvm *kvm,
1623                 struct kvm_dirty_log *log)
1624 {
1625         struct kvm_memory_slot *memslot;
1626         int r, i;
1627         long n, base;
1628         unsigned long *dirty_bitmap = (unsigned long *)((void *)kvm - KVM_VM_OFS
1629                                         + KVM_MEM_DIRTY_LOG_OFS);
1630
1631         r = -EINVAL;
1632         if (log->slot >= KVM_MEMORY_SLOTS)
1633                 goto out;
1634
1635         memslot = &kvm->memslots[log->slot];
1636         r = -ENOENT;
1637         if (!memslot->dirty_bitmap)
1638                 goto out;
1639
1640         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1641         base = memslot->base_gfn / BITS_PER_LONG;
1642
1643         for (i = 0; i < n/sizeof(long); ++i) {
1644                 memslot->dirty_bitmap[i] = dirty_bitmap[base + i];
1645                 dirty_bitmap[base + i] = 0;
1646         }
1647         r = 0;
1648 out:
1649         return r;
1650 }
1651
1652 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1653                 struct kvm_dirty_log *log)
1654 {
1655         int r;
1656         int n;
1657         struct kvm_memory_slot *memslot;
1658         int is_dirty = 0;
1659
1660         spin_lock(&kvm->arch.dirty_log_lock);
1661
1662         r = kvm_ia64_sync_dirty_log(kvm, log);
1663         if (r)
1664                 goto out;
1665
1666         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1667         if (r)
1668                 goto out;
1669
1670         /* If nothing is dirty, don't bother messing with page tables. */
1671         if (is_dirty) {
1672                 kvm_flush_remote_tlbs(kvm);
1673                 memslot = &kvm->memslots[log->slot];
1674                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1675                 memset(memslot->dirty_bitmap, 0, n);
1676         }
1677         r = 0;
1678 out:
1679         spin_unlock(&kvm->arch.dirty_log_lock);
1680         return r;
1681 }
1682
1683 int kvm_arch_hardware_setup(void)
1684 {
1685         return 0;
1686 }
1687
1688 void kvm_arch_hardware_unsetup(void)
1689 {
1690 }
1691
1692 static void vcpu_kick_intr(void *info)
1693 {
1694 #ifdef DEBUG
1695         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
1696         printk(KERN_DEBUG"vcpu_kick_intr %p \n", vcpu);
1697 #endif
1698 }
1699
1700 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1701 {
1702         int ipi_pcpu = vcpu->cpu;
1703
1704         if (waitqueue_active(&vcpu->wq))
1705                 wake_up_interruptible(&vcpu->wq);
1706
1707         if (vcpu->guest_mode)
1708                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
1709 }
1710
1711 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, u8 vec, u8 trig)
1712 {
1713
1714         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1715
1716         if (!test_and_set_bit(vec, &vpd->irr[0])) {
1717                 vcpu->arch.irq_new_pending = 1;
1718                  if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
1719                         kvm_vcpu_kick(vcpu);
1720                 else if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED) {
1721                         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1722                         if (waitqueue_active(&vcpu->wq))
1723                                 wake_up_interruptible(&vcpu->wq);
1724                 }
1725                 return 1;
1726         }
1727         return 0;
1728 }
1729
1730 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
1731 {
1732         return apic->vcpu->vcpu_id == dest;
1733 }
1734
1735 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
1736 {
1737         return 0;
1738 }
1739
1740 struct kvm_vcpu *kvm_get_lowest_prio_vcpu(struct kvm *kvm, u8 vector,
1741                                        unsigned long bitmap)
1742 {
1743         struct kvm_vcpu *lvcpu = kvm->vcpus[0];
1744         int i;
1745
1746         for (i = 1; i < KVM_MAX_VCPUS; i++) {
1747                 if (!kvm->vcpus[i])
1748                         continue;
1749                 if (lvcpu->arch.xtp > kvm->vcpus[i]->arch.xtp)
1750                         lvcpu = kvm->vcpus[i];
1751         }
1752
1753         return lvcpu;
1754 }
1755
1756 static int find_highest_bits(int *dat)
1757 {
1758         u32  bits, bitnum;
1759         int i;
1760
1761         /* loop for all 256 bits */
1762         for (i = 7; i >= 0 ; i--) {
1763                 bits = dat[i];
1764                 if (bits) {
1765                         bitnum = fls(bits);
1766                         return i * 32 + bitnum - 1;
1767                 }
1768         }
1769
1770         return -1;
1771 }
1772
1773 int kvm_highest_pending_irq(struct kvm_vcpu *vcpu)
1774 {
1775     struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1776
1777     if (vpd->irr[0] & (1UL << NMI_VECTOR))
1778                 return NMI_VECTOR;
1779     if (vpd->irr[0] & (1UL << ExtINT_VECTOR))
1780                 return ExtINT_VECTOR;
1781
1782     return find_highest_bits((int *)&vpd->irr[0]);
1783 }
1784
1785 int kvm_cpu_has_interrupt(struct kvm_vcpu *vcpu)
1786 {
1787         if (kvm_highest_pending_irq(vcpu) != -1)
1788                 return 1;
1789         return 0;
1790 }
1791
1792 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1793 {
1794         return 0;
1795 }
1796
1797 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1798 {
1799         return gfn;
1800 }
1801
1802 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
1803 {
1804         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE;
1805 }
1806
1807 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1808                                     struct kvm_mp_state *mp_state)
1809 {
1810         vcpu_load(vcpu);
1811         mp_state->mp_state = vcpu->arch.mp_state;
1812         vcpu_put(vcpu);
1813         return 0;
1814 }
1815
1816 static int vcpu_reset(struct kvm_vcpu *vcpu)
1817 {
1818         int r;
1819         long psr;
1820         local_irq_save(psr);
1821         r = kvm_insert_vmm_mapping(vcpu);
1822         if (r)
1823                 goto fail;
1824
1825         vcpu->arch.launched = 0;
1826         kvm_arch_vcpu_uninit(vcpu);
1827         r = kvm_arch_vcpu_init(vcpu);
1828         if (r)
1829                 goto fail;
1830
1831         kvm_purge_vmm_mapping(vcpu);
1832         r = 0;
1833 fail:
1834         local_irq_restore(psr);
1835         return r;
1836 }
1837
1838 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1839                                     struct kvm_mp_state *mp_state)
1840 {
1841         int r = 0;
1842
1843         vcpu_load(vcpu);
1844         vcpu->arch.mp_state = mp_state->mp_state;
1845         if (vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)
1846                 r = vcpu_reset(vcpu);
1847         vcpu_put(vcpu);
1848         return r;
1849 }