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
 389         call_data.ptc_g_data = p->u.ptc_g_data;
 390
 391         for (i = 0; i < KVM_MAX_VCPUS; i++) {
 392                 if (!kvm->vcpus[i] || kvm->vcpus[i]->arch.mp_state ==
 393                                                 KVM_MP_STATE_UNINITIALIZED ||
 394                                         vcpu == kvm->vcpus[i])
 395                         continue;
 396
 397                 if (waitqueue_active(&kvm->vcpus[i]->wq))
 398                         wake_up_interruptible(&kvm->vcpus[i]->wq);
 399
 400                 if (kvm->vcpus[i]->cpu != -1) {
 401                         call_data.vcpu = kvm->vcpus[i];
 402                         smp_call_function_single(kvm->vcpus[i]->cpu,
 403                                         vcpu_global_purge, &call_data, 1);
 404                 } else
 405                         printk(KERN_WARNING"kvm: Uninit vcpu received ipi!\n");
 406
 407         }
 408         return 1;
 409 }
 410
 411 static int handle_switch_rr6(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 412 {
 413         return 1;
 414 }
 415
 416 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
 417 {
 418
 419         ktime_t kt;
 420         long itc_diff;
 421         unsigned long vcpu_now_itc;
 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         if (irqchip_in_kernel(vcpu->kvm)) {
 428
 429                 vcpu_now_itc = ia64_getreg(_IA64_REG_AR_ITC) + vcpu->arch.itc_offset;
 430
 431                 if (time_after(vcpu_now_itc, vpd->itm)) {
 432                         vcpu->arch.timer_check = 1;
 433                         return 1;
 434                 }
 435                 itc_diff = vpd->itm - vcpu_now_itc;
 436                 if (itc_diff < 0)
 437                         itc_diff = -itc_diff;
 438
 439                 expires = div64_u64(itc_diff, cyc_per_usec);
 440                 kt = ktime_set(0, 1000 * expires);
 441
 442                 down_read(&vcpu->kvm->slots_lock);
 443                 vcpu->arch.ht_active = 1;
 444                 hrtimer_start(p_ht, kt, HRTIMER_MODE_ABS);
 445
 446                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
 447                 kvm_vcpu_block(vcpu);
 448                 hrtimer_cancel(p_ht);
 449                 vcpu->arch.ht_active = 0;
 450
 451                 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
 452                         if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
 453                                 vcpu->arch.mp_state =
 454                                         KVM_MP_STATE_RUNNABLE;
 455                 up_read(&vcpu->kvm->slots_lock);
 456
 457                 if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
 458                         return -EINTR;
 459                 return 1;
 460         } else {
 461                 printk(KERN_ERR"kvm: Unsupported userspace halt!");
 462                 return 0;
 463         }
 464 }
 465
 466 static int handle_vm_shutdown(struct kvm_vcpu *vcpu,
 467                 struct kvm_run *kvm_run)
 468 {
 469         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
 470         return 0;
 471 }
 472
 473 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
 474                 struct kvm_run *kvm_run)
 475 {
 476         return 1;
 477 }
 478
 479 static int (*kvm_vti_exit_handlers[])(struct kvm_vcpu *vcpu,
 480                 struct kvm_run *kvm_run) = {
 481         [EXIT_REASON_VM_PANIC]              = handle_vm_error,
 482         [EXIT_REASON_MMIO_INSTRUCTION]      = handle_mmio,
 483         [EXIT_REASON_PAL_CALL]              = handle_pal_call,
 484         [EXIT_REASON_SAL_CALL]              = handle_sal_call,
 485         [EXIT_REASON_SWITCH_RR6]            = handle_switch_rr6,
 486         [EXIT_REASON_VM_DESTROY]            = handle_vm_shutdown,
 487         [EXIT_REASON_EXTERNAL_INTERRUPT]    = handle_external_interrupt,
 488         [EXIT_REASON_IPI]                   = handle_ipi,
 489         [EXIT_REASON_PTC_G]                 = handle_global_purge,
 490
 491 };
 492
 493 static const int kvm_vti_max_exit_handlers =
 494                 sizeof(kvm_vti_exit_handlers)/sizeof(*kvm_vti_exit_handlers);
 495
 496 static uint32_t kvm_get_exit_reason(struct kvm_vcpu *vcpu)
 497 {
 498         struct exit_ctl_data *p_exit_data;
 499
 500         p_exit_data = kvm_get_exit_data(vcpu);
 501         return p_exit_data->exit_reason;
 502 }
 503
 504 /*
 505  * The guest has exited.  See if we can fix it or if we need userspace
 506  * assistance.
 507  */
 508 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
 509 {
 510         u32 exit_reason = kvm_get_exit_reason(vcpu);
 511         vcpu->arch.last_exit = exit_reason;
 512
 513         if (exit_reason < kvm_vti_max_exit_handlers
 514                         && kvm_vti_exit_handlers[exit_reason])
 515                 return kvm_vti_exit_handlers[exit_reason](vcpu, kvm_run);
 516         else {
 517                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
 518                 kvm_run->hw.hardware_exit_reason = exit_reason;
 519         }
 520         return 0;
 521 }
 522
 523 static inline void vti_set_rr6(unsigned long rr6)
 524 {
 525         ia64_set_rr(RR6, rr6);
 526         ia64_srlz_i();
 527 }
 528
 529 static int kvm_insert_vmm_mapping(struct kvm_vcpu *vcpu)
 530 {
 531         unsigned long pte;
 532         struct kvm *kvm = vcpu->kvm;
 533         int r;
 534
 535         /*Insert a pair of tr to map vmm*/
 536         pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base), PAGE_KERNEL));
 537         r = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
 538         if (r < 0)
 539                 goto out;
 540         vcpu->arch.vmm_tr_slot = r;
 541         /*Insert a pairt of tr to map data of vm*/
 542         pte = pte_val(mk_pte_phys(__pa(kvm->arch.vm_base), PAGE_KERNEL));
 543         r = ia64_itr_entry(0x3, KVM_VM_DATA_BASE,
 544                                         pte, KVM_VM_DATA_SHIFT);
 545         if (r < 0)
 546                 goto out;
 547         vcpu->arch.vm_tr_slot = r;
 548         r = 0;
 549 out:
 550         return r;
 551
 552 }
 553
 554 static void kvm_purge_vmm_mapping(struct kvm_vcpu *vcpu)
 555 {
 556
 557         ia64_ptr_entry(0x3, vcpu->arch.vmm_tr_slot);
 558         ia64_ptr_entry(0x3, vcpu->arch.vm_tr_slot);
 559
 560 }
 561
 562 static int kvm_vcpu_pre_transition(struct kvm_vcpu *vcpu)
 563 {
 564         int cpu = smp_processor_id();
 565
 566         if (vcpu->arch.last_run_cpu != cpu ||
 567                         per_cpu(last_vcpu, cpu) != vcpu) {
 568                 per_cpu(last_vcpu, cpu) = vcpu;
 569                 vcpu->arch.last_run_cpu = cpu;
 570                 kvm_flush_tlb_all();
 571         }
 572
 573         vcpu->arch.host_rr6 = ia64_get_rr(RR6);
 574         vti_set_rr6(vcpu->arch.vmm_rr);
 575         return kvm_insert_vmm_mapping(vcpu);
 576 }
 577 static void kvm_vcpu_post_transition(struct kvm_vcpu *vcpu)
 578 {
 579         kvm_purge_vmm_mapping(vcpu);
 580         vti_set_rr6(vcpu->arch.host_rr6);
 581 }
 582
 583 static int  vti_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 584 {
 585         union context *host_ctx, *guest_ctx;
 586         int r;
 587
 588         /*Get host and guest context with guest address space.*/
 589         host_ctx = kvm_get_host_context(vcpu);
 590         guest_ctx = kvm_get_guest_context(vcpu);
 591
 592         r = kvm_vcpu_pre_transition(vcpu);
 593         if (r < 0)
 594                 goto out;
 595         kvm_vmm_info->tramp_entry(host_ctx, guest_ctx);
 596         kvm_vcpu_post_transition(vcpu);
 597         r = 0;
 598 out:
 599         return r;
 600 }
 601
 602 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 603 {
 604         int r;
 605
 606 again:
 607         preempt_disable();
 608         local_irq_disable();
 609
 610         if (signal_pending(current)) {
 611                 local_irq_enable();
 612                 preempt_enable();
 613                 r = -EINTR;
 614                 kvm_run->exit_reason = KVM_EXIT_INTR;
 615                 goto out;
 616         }
 617
 618         vcpu->guest_mode = 1;
 619         kvm_guest_enter();
 620         down_read(&vcpu->kvm->slots_lock);
 621         r = vti_vcpu_run(vcpu, kvm_run);
 622         if (r < 0) {
 623                 local_irq_enable();
 624                 preempt_enable();
 625                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
 626                 goto out;
 627         }
 628
 629         vcpu->arch.launched = 1;
 630         vcpu->guest_mode = 0;
 631         local_irq_enable();
 632
 633         /*
 634          * We must have an instruction between local_irq_enable() and
 635          * kvm_guest_exit(), so the timer interrupt isn't delayed by
 636          * the interrupt shadow.  The stat.exits increment will do nicely.
 637          * But we need to prevent reordering, hence this barrier():
 638          */
 639         barrier();
 640         kvm_guest_exit();
 641         up_read(&vcpu->kvm->slots_lock);
 642         preempt_enable();
 643
 644         r = kvm_handle_exit(kvm_run, vcpu);
 645
 646         if (r > 0) {
 647                 if (!need_resched())
 648                         goto again;
 649         }
 650
 651 out:
 652         if (r > 0) {
 653                 kvm_resched(vcpu);
 654                 goto again;
 655         }
 656
 657         return r;
 658 }
 659
 660 static void kvm_set_mmio_data(struct kvm_vcpu *vcpu)
 661 {
 662         struct kvm_mmio_req *p = kvm_get_vcpu_ioreq(vcpu);
 663
 664         if (!vcpu->mmio_is_write)
 665                 memcpy(&p->data, vcpu->mmio_data, 8);
 666         p->state = STATE_IORESP_READY;
 667 }
 668
 669 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 670 {
 671         int r;
 672         sigset_t sigsaved;
 673
 674         vcpu_load(vcpu);
 675
 676         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
 677                 kvm_vcpu_block(vcpu);
 678                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 679                 vcpu_put(vcpu);
 680                 return -EAGAIN;
 681         }
 682
 683         if (vcpu->sigset_active)
 684                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 685
 686         if (vcpu->mmio_needed) {
 687                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
 688                 kvm_set_mmio_data(vcpu);
 689                 vcpu->mmio_read_completed = 1;
 690                 vcpu->mmio_needed = 0;
 691         }
 692         r = __vcpu_run(vcpu, kvm_run);
 693
 694         if (vcpu->sigset_active)
 695                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 696
 697         vcpu_put(vcpu);
 698         return r;
 699 }
 700
 701 /*
 702  * Allocate 16M memory for every vm to hold its specific data.
 703  * Its memory map is defined in kvm_host.h.
 704  */
 705 static struct kvm *kvm_alloc_kvm(void)
 706 {
 707
 708         struct kvm *kvm;
 709         uint64_t  vm_base;
 710
 711         vm_base = __get_free_pages(GFP_KERNEL, get_order(KVM_VM_DATA_SIZE));
 712
 713         if (!vm_base)
 714                 return ERR_PTR(-ENOMEM);
 715         printk(KERN_DEBUG"kvm: VM data's base Address:0x%lx\n", vm_base);
 716
 717         /* Zero all pages before use! */
 718         memset((void *)vm_base, 0, KVM_VM_DATA_SIZE);
 719
 720         kvm = (struct kvm *)(vm_base + KVM_VM_OFS);
 721         kvm->arch.vm_base = vm_base;
 722
 723         return kvm;
 724 }
 725
 726 struct kvm_io_range {
 727         unsigned long start;
 728         unsigned long size;
 729         unsigned long type;
 730 };
 731
 732 static const struct kvm_io_range io_ranges[] = {
 733         {VGA_IO_START, VGA_IO_SIZE, GPFN_FRAME_BUFFER},
 734         {MMIO_START, MMIO_SIZE, GPFN_LOW_MMIO},
 735         {LEGACY_IO_START, LEGACY_IO_SIZE, GPFN_LEGACY_IO},
 736         {IO_SAPIC_START, IO_SAPIC_SIZE, GPFN_IOSAPIC},
 737         {PIB_START, PIB_SIZE, GPFN_PIB},
 738 };
 739
 740 static void kvm_build_io_pmt(struct kvm *kvm)
 741 {
 742         unsigned long i, j;
 743
 744         /* Mark I/O ranges */
 745         for (i = 0; i < (sizeof(io_ranges) / sizeof(struct kvm_io_range));
 746                                                         i++) {
 747                 for (j = io_ranges[i].start;
 748                                 j < io_ranges[i].start + io_ranges[i].size;
 749                                 j += PAGE_SIZE)
 750                         kvm_set_pmt_entry(kvm, j >> PAGE_SHIFT,
 751                                         io_ranges[i].type, 0);
 752         }
 753
 754 }
 755
 756 /*Use unused rids to virtualize guest rid.*/
 757 #define GUEST_PHYSICAL_RR0      0x1739
 758 #define GUEST_PHYSICAL_RR4      0x2739
 759 #define VMM_INIT_RR             0x1660
 760
 761 static void kvm_init_vm(struct kvm *kvm)
 762 {
 763         long vm_base;
 764
 765         BUG_ON(!kvm);
 766
 767         kvm->arch.metaphysical_rr0 = GUEST_PHYSICAL_RR0;
 768         kvm->arch.metaphysical_rr4 = GUEST_PHYSICAL_RR4;
 769         kvm->arch.vmm_init_rr = VMM_INIT_RR;
 770
 771         vm_base = kvm->arch.vm_base;
 772         if (vm_base) {
 773                 kvm->arch.vhpt_base = vm_base + KVM_VHPT_OFS;
 774                 kvm->arch.vtlb_base = vm_base + KVM_VTLB_OFS;
 775                 kvm->arch.vpd_base  = vm_base + KVM_VPD_OFS;
 776         }
 777
 778         /*
 779          *Fill P2M entries for MMIO/IO ranges
 780          */
 781         kvm_build_io_pmt(kvm);
 782
 783         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
 784
 785         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
 786         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
 787 }
 788
 789 struct  kvm *kvm_arch_create_vm(void)
 790 {
 791         struct kvm *kvm = kvm_alloc_kvm();
 792
 793         if (IS_ERR(kvm))
 794                 return ERR_PTR(-ENOMEM);
 795         kvm_init_vm(kvm);
 796
 797         return kvm;
 798
 799 }
 800
 801 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm,
 802                                         struct kvm_irqchip *chip)
 803 {
 804         int r;
 805
 806         r = 0;
 807         switch (chip->chip_id) {
 808         case KVM_IRQCHIP_IOAPIC:
 809                 memcpy(&chip->chip.ioapic, ioapic_irqchip(kvm),
 810                                 sizeof(struct kvm_ioapic_state));
 811                 break;
 812         default:
 813                 r = -EINVAL;
 814                 break;
 815         }
 816         return r;
 817 }
 818
 819 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 820 {
 821         int r;
 822
 823         r = 0;
 824         switch (chip->chip_id) {
 825         case KVM_IRQCHIP_IOAPIC:
 826                 memcpy(ioapic_irqchip(kvm),
 827                                 &chip->chip.ioapic,
 828                                 sizeof(struct kvm_ioapic_state));
 829                 break;
 830         default:
 831                 r = -EINVAL;
 832                 break;
 833         }
 834         return r;
 835 }
 836
 837 #define RESTORE_REGS(_x) vcpu->arch._x = regs->_x
 838
 839 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 840 {
 841         int i;
 842         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
 843         int r;
 844
 845         vcpu_load(vcpu);
 846
 847         for (i = 0; i < 16; i++) {
 848                 vpd->vgr[i] = regs->vpd.vgr[i];
 849                 vpd->vbgr[i] = regs->vpd.vbgr[i];
 850         }
 851         for (i = 0; i < 128; i++)
 852                 vpd->vcr[i] = regs->vpd.vcr[i];
 853         vpd->vhpi = regs->vpd.vhpi;
 854         vpd->vnat = regs->vpd.vnat;
 855         vpd->vbnat = regs->vpd.vbnat;
 856         vpd->vpsr = regs->vpd.vpsr;
 857
 858         vpd->vpr = regs->vpd.vpr;
 859
 860         r = -EFAULT;
 861         r = copy_from_user(&vcpu->arch.guest, regs->saved_guest,
 862                                                 sizeof(union context));
 863         if (r)
 864                 goto out;
 865         r = copy_from_user(vcpu + 1, regs->saved_stack +
 866                         sizeof(struct kvm_vcpu),
 867                         IA64_STK_OFFSET - sizeof(struct kvm_vcpu));
 868         if (r)
 869                 goto out;
 870         vcpu->arch.exit_data =
 871                 ((struct kvm_vcpu *)(regs->saved_stack))->arch.exit_data;
 872
 873         RESTORE_REGS(mp_state);
 874         RESTORE_REGS(vmm_rr);
 875         memcpy(vcpu->arch.itrs, regs->itrs, sizeof(struct thash_data) * NITRS);
 876         memcpy(vcpu->arch.dtrs, regs->dtrs, sizeof(struct thash_data) * NDTRS);
 877         RESTORE_REGS(itr_regions);
 878         RESTORE_REGS(dtr_regions);
 879         RESTORE_REGS(tc_regions);
 880         RESTORE_REGS(irq_check);
 881         RESTORE_REGS(itc_check);
 882         RESTORE_REGS(timer_check);
 883         RESTORE_REGS(timer_pending);
 884         RESTORE_REGS(last_itc);
 885         for (i = 0; i < 8; i++) {
 886                 vcpu->arch.vrr[i] = regs->vrr[i];
 887                 vcpu->arch.ibr[i] = regs->ibr[i];
 888                 vcpu->arch.dbr[i] = regs->dbr[i];
 889         }
 890         for (i = 0; i < 4; i++)
 891                 vcpu->arch.insvc[i] = regs->insvc[i];
 892         RESTORE_REGS(xtp);
 893         RESTORE_REGS(metaphysical_rr0);
 894         RESTORE_REGS(metaphysical_rr4);
 895         RESTORE_REGS(metaphysical_saved_rr0);
 896         RESTORE_REGS(metaphysical_saved_rr4);
 897         RESTORE_REGS(fp_psr);
 898         RESTORE_REGS(saved_gp);
 899
 900         vcpu->arch.irq_new_pending = 1;
 901         vcpu->arch.itc_offset = regs->saved_itc - ia64_getreg(_IA64_REG_AR_ITC);
 902         set_bit(KVM_REQ_RESUME, &vcpu->requests);
 903
 904         vcpu_put(vcpu);
 905         r = 0;
 906 out:
 907         return r;
 908 }
 909
 910 long kvm_arch_vm_ioctl(struct file *filp,
 911                 unsigned int ioctl, unsigned long arg)
 912 {
 913         struct kvm *kvm = filp->private_data;
 914         void __user *argp = (void __user *)arg;
 915         int r = -EINVAL;
 916
 917         switch (ioctl) {
 918         case KVM_SET_MEMORY_REGION: {
 919                 struct kvm_memory_region kvm_mem;
 920                 struct kvm_userspace_memory_region kvm_userspace_mem;
 921
 922                 r = -EFAULT;
 923                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
 924                         goto out;
 925                 kvm_userspace_mem.slot = kvm_mem.slot;
 926                 kvm_userspace_mem.flags = kvm_mem.flags;
 927                 kvm_userspace_mem.guest_phys_addr =
 928                                         kvm_mem.guest_phys_addr;
 929                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
 930                 r = kvm_vm_ioctl_set_memory_region(kvm,
 931                                         &kvm_userspace_mem, 0);
 932                 if (r)
 933                         goto out;
 934                 break;
 935                 }
 936         case KVM_CREATE_IRQCHIP:
 937                 r = -EFAULT;
 938                 r = kvm_ioapic_init(kvm);
 939                 if (r)
 940                         goto out;
 941                 break;
 942         case KVM_IRQ_LINE: {
 943                 struct kvm_irq_level irq_event;
 944
 945                 r = -EFAULT;
 946                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
 947                         goto out;
 948                 if (irqchip_in_kernel(kvm)) {
 949                         mutex_lock(&kvm->lock);
 950                         kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
 951                                     irq_event.irq, irq_event.level);
 952                         mutex_unlock(&kvm->lock);
 953                         r = 0;
 954                 }
 955                 break;
 956                 }
 957         case KVM_GET_IRQCHIP: {
 958                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
 959                 struct kvm_irqchip chip;
 960
 961                 r = -EFAULT;
 962                 if (copy_from_user(&chip, argp, sizeof chip))
 963                                 goto out;
 964                 r = -ENXIO;
 965                 if (!irqchip_in_kernel(kvm))
 966                         goto out;
 967                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
 968                 if (r)
 969                         goto out;
 970                 r = -EFAULT;
 971                 if (copy_to_user(argp, &chip, sizeof chip))
 972                                 goto out;
 973                 r = 0;
 974                 break;
 975                 }
 976         case KVM_SET_IRQCHIP: {
 977                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
 978                 struct kvm_irqchip chip;
 979
 980                 r = -EFAULT;
 981                 if (copy_from_user(&chip, argp, sizeof chip))
 982                                 goto out;
 983                 r = -ENXIO;
 984                 if (!irqchip_in_kernel(kvm))
 985                         goto out;
 986                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
 987                 if (r)
 988                         goto out;
 989                 r = 0;
 990                 break;
 991                 }
 992         default:
 993                 ;
 994         }
 995 out:
 996         return r;
 997 }
 998
 999 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1000                 struct kvm_sregs *sregs)
1001 {
1002         return -EINVAL;
1003 }
1004
1005 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1006                 struct kvm_sregs *sregs)
1007 {
1008         return -EINVAL;
1009
1010 }
1011 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1012                 struct kvm_translation *tr)
1013 {
1014
1015         return -EINVAL;
1016 }
1017
1018 static int kvm_alloc_vmm_area(void)
1019 {
1020         if (!kvm_vmm_base && (kvm_vm_buffer_size < KVM_VM_BUFFER_SIZE)) {
1021                 kvm_vmm_base = __get_free_pages(GFP_KERNEL,
1022                                 get_order(KVM_VMM_SIZE));
1023                 if (!kvm_vmm_base)
1024                         return -ENOMEM;
1025
1026                 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1027                 kvm_vm_buffer = kvm_vmm_base + VMM_SIZE;
1028
1029                 printk(KERN_DEBUG"kvm:VMM's Base Addr:0x%lx, vm_buffer:0x%lx\n",
1030                                 kvm_vmm_base, kvm_vm_buffer);
1031         }
1032
1033         return 0;
1034 }
1035
1036 static void kvm_free_vmm_area(void)
1037 {
1038         if (kvm_vmm_base) {
1039                 /*Zero this area before free to avoid bits leak!!*/
1040                 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1041                 free_pages(kvm_vmm_base, get_order(KVM_VMM_SIZE));
1042                 kvm_vmm_base  = 0;
1043                 kvm_vm_buffer = 0;
1044                 kvm_vsa_base = 0;
1045         }
1046 }
1047
1048 static void vti_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1049 {
1050 }
1051
1052 static int vti_init_vpd(struct kvm_vcpu *vcpu)
1053 {
1054         int i;
1055         union cpuid3_t cpuid3;
1056         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1057
1058         if (IS_ERR(vpd))
1059                 return PTR_ERR(vpd);
1060
1061         /* CPUID init */
1062         for (i = 0; i < 5; i++)
1063                 vpd->vcpuid[i] = ia64_get_cpuid(i);
1064
1065         /* Limit the CPUID number to 5 */
1066         cpuid3.value = vpd->vcpuid[3];
1067         cpuid3.number = 4;      /* 5 - 1 */
1068         vpd->vcpuid[3] = cpuid3.value;
1069
1070         /*Set vac and vdc fields*/
1071         vpd->vac.a_from_int_cr = 1;
1072         vpd->vac.a_to_int_cr = 1;
1073         vpd->vac.a_from_psr = 1;
1074         vpd->vac.a_from_cpuid = 1;
1075         vpd->vac.a_cover = 1;
1076         vpd->vac.a_bsw = 1;
1077         vpd->vac.a_int = 1;
1078         vpd->vdc.d_vmsw = 1;
1079
1080         /*Set virtual buffer*/
1081         vpd->virt_env_vaddr = KVM_VM_BUFFER_BASE;
1082
1083         return 0;
1084 }
1085
1086 static int vti_create_vp(struct kvm_vcpu *vcpu)
1087 {
1088         long ret;
1089         struct vpd *vpd = vcpu->arch.vpd;
1090         unsigned long  vmm_ivt;
1091
1092         vmm_ivt = kvm_vmm_info->vmm_ivt;
1093
1094         printk(KERN_DEBUG "kvm: vcpu:%p,ivt: 0x%lx\n", vcpu, vmm_ivt);
1095
1096         ret = ia64_pal_vp_create((u64 *)vpd, (u64 *)vmm_ivt, 0);
1097
1098         if (ret) {
1099                 printk(KERN_ERR"kvm: ia64_pal_vp_create failed!\n");
1100                 return -EINVAL;
1101         }
1102         return 0;
1103 }
1104
1105 static void init_ptce_info(struct kvm_vcpu *vcpu)
1106 {
1107         ia64_ptce_info_t ptce = {0};
1108
1109         ia64_get_ptce(&ptce);
1110         vcpu->arch.ptce_base = ptce.base;
1111         vcpu->arch.ptce_count[0] = ptce.count[0];
1112         vcpu->arch.ptce_count[1] = ptce.count[1];
1113         vcpu->arch.ptce_stride[0] = ptce.stride[0];
1114         vcpu->arch.ptce_stride[1] = ptce.stride[1];
1115 }
1116
1117 static void kvm_migrate_hlt_timer(struct kvm_vcpu *vcpu)
1118 {
1119         struct hrtimer *p_ht = &vcpu->arch.hlt_timer;
1120
1121         if (hrtimer_cancel(p_ht))
1122                 hrtimer_start_expires(p_ht, HRTIMER_MODE_ABS);
1123 }
1124
1125 static enum hrtimer_restart hlt_timer_fn(struct hrtimer *data)
1126 {
1127         struct kvm_vcpu *vcpu;
1128         wait_queue_head_t *q;
1129
1130         vcpu  = container_of(data, struct kvm_vcpu, arch.hlt_timer);
1131         q = &vcpu->wq;
1132
1133         if (vcpu->arch.mp_state != KVM_MP_STATE_HALTED)
1134                 goto out;
1135
1136         if (waitqueue_active(q))
1137                 wake_up_interruptible(q);
1138
1139 out:
1140         vcpu->arch.timer_fired = 1;
1141         vcpu->arch.timer_check = 1;
1142         return HRTIMER_NORESTART;
1143 }
1144
1145 #define PALE_RESET_ENTRY    0x80000000ffffffb0UL
1146
1147 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1148 {
1149         struct kvm_vcpu *v;
1150         int r;
1151         int i;
1152         long itc_offset;
1153         struct kvm *kvm = vcpu->kvm;
1154         struct kvm_pt_regs *regs = vcpu_regs(vcpu);
1155
1156         union context *p_ctx = &vcpu->arch.guest;
1157         struct kvm_vcpu *vmm_vcpu = to_guest(vcpu->kvm, vcpu);
1158
1159         /*Init vcpu context for first run.*/
1160         if (IS_ERR(vmm_vcpu))
1161                 return PTR_ERR(vmm_vcpu);
1162
1163         if (vcpu->vcpu_id == 0) {
1164                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1165
1166                 /*Set entry address for first run.*/
1167                 regs->cr_iip = PALE_RESET_ENTRY;
1168
1169                 /*Initilize itc offset for vcpus*/
1170                 itc_offset = 0UL - ia64_getreg(_IA64_REG_AR_ITC);
1171                 for (i = 0; i < MAX_VCPU_NUM; i++) {
1172                         v = (struct kvm_vcpu *)((char *)vcpu + VCPU_SIZE * i);
1173                         v->arch.itc_offset = itc_offset;
1174                         v->arch.last_itc = 0;
1175                 }
1176         } else
1177                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
1178
1179         r = -ENOMEM;
1180         vcpu->arch.apic = kzalloc(sizeof(struct kvm_lapic), GFP_KERNEL);
1181         if (!vcpu->arch.apic)
1182                 goto out;
1183         vcpu->arch.apic->vcpu = vcpu;
1184
1185         p_ctx->gr[1] = 0;
1186         p_ctx->gr[12] = (unsigned long)((char *)vmm_vcpu + IA64_STK_OFFSET);
1187         p_ctx->gr[13] = (unsigned long)vmm_vcpu;
1188         p_ctx->psr = 0x1008522000UL;
1189         p_ctx->ar[40] = FPSR_DEFAULT; /*fpsr*/
1190         p_ctx->caller_unat = 0;
1191         p_ctx->pr = 0x0;
1192         p_ctx->ar[36] = 0x0; /*unat*/
1193         p_ctx->ar[19] = 0x0; /*rnat*/
1194         p_ctx->ar[18] = (unsigned long)vmm_vcpu +
1195                                 ((sizeof(struct kvm_vcpu)+15) & ~15);
1196         p_ctx->ar[64] = 0x0; /*pfs*/
1197         p_ctx->cr[0] = 0x7e04UL;
1198         p_ctx->cr[2] = (unsigned long)kvm_vmm_info->vmm_ivt;
1199         p_ctx->cr[8] = 0x3c;
1200
1201         /*Initilize region register*/
1202         p_ctx->rr[0] = 0x30;
1203         p_ctx->rr[1] = 0x30;
1204         p_ctx->rr[2] = 0x30;
1205         p_ctx->rr[3] = 0x30;
1206         p_ctx->rr[4] = 0x30;
1207         p_ctx->rr[5] = 0x30;
1208         p_ctx->rr[7] = 0x30;
1209
1210         /*Initilize branch register 0*/
1211         p_ctx->br[0] = *(unsigned long *)kvm_vmm_info->vmm_entry;
1212
1213         vcpu->arch.vmm_rr = kvm->arch.vmm_init_rr;
1214         vcpu->arch.metaphysical_rr0 = kvm->arch.metaphysical_rr0;
1215         vcpu->arch.metaphysical_rr4 = kvm->arch.metaphysical_rr4;
1216
1217         hrtimer_init(&vcpu->arch.hlt_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1218         vcpu->arch.hlt_timer.function = hlt_timer_fn;
1219
1220         vcpu->arch.last_run_cpu = -1;
1221         vcpu->arch.vpd = (struct vpd *)VPD_ADDR(vcpu->vcpu_id);
1222         vcpu->arch.vsa_base = kvm_vsa_base;
1223         vcpu->arch.__gp = kvm_vmm_gp;
1224         vcpu->arch.dirty_log_lock_pa = __pa(&kvm->arch.dirty_log_lock);
1225         vcpu->arch.vhpt.hash = (struct thash_data *)VHPT_ADDR(vcpu->vcpu_id);
1226         vcpu->arch.vtlb.hash = (struct thash_data *)VTLB_ADDR(vcpu->vcpu_id);
1227         init_ptce_info(vcpu);
1228
1229         r = 0;
1230 out:
1231         return r;
1232 }
1233
1234 static int vti_vcpu_setup(struct kvm_vcpu *vcpu, int id)
1235 {
1236         unsigned long psr;
1237         int r;
1238
1239         local_irq_save(psr);
1240         r = kvm_insert_vmm_mapping(vcpu);
1241         if (r)
1242                 goto fail;
1243         r = kvm_vcpu_init(vcpu, vcpu->kvm, id);
1244         if (r)
1245                 goto fail;
1246
1247         r = vti_init_vpd(vcpu);
1248         if (r) {
1249                 printk(KERN_DEBUG"kvm: vpd init error!!\n");
1250                 goto uninit;
1251         }
1252
1253         r = vti_create_vp(vcpu);
1254         if (r)
1255                 goto uninit;
1256
1257         kvm_purge_vmm_mapping(vcpu);
1258         local_irq_restore(psr);
1259
1260         return 0;
1261 uninit:
1262         kvm_vcpu_uninit(vcpu);
1263 fail:
1264         local_irq_restore(psr);
1265         return r;
1266 }
1267
1268 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
1269                 unsigned int id)
1270 {
1271         struct kvm_vcpu *vcpu;
1272         unsigned long vm_base = kvm->arch.vm_base;
1273         int r;
1274         int cpu;
1275
1276         r = -ENOMEM;
1277         if (!vm_base) {
1278                 printk(KERN_ERR"kvm: Create vcpu[%d] error!\n", id);
1279                 goto fail;
1280         }
1281         vcpu = (struct kvm_vcpu *)(vm_base + KVM_VCPU_OFS + VCPU_SIZE * id);
1282         vcpu->kvm = kvm;
1283
1284         cpu = get_cpu();
1285         vti_vcpu_load(vcpu, cpu);
1286         r = vti_vcpu_setup(vcpu, id);
1287         put_cpu();
1288
1289         if (r) {
1290                 printk(KERN_DEBUG"kvm: vcpu_setup error!!\n");
1291                 goto fail;
1292         }
1293
1294         return vcpu;
1295 fail:
1296         return ERR_PTR(r);
1297 }
1298
1299 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1300 {
1301         return 0;
1302 }
1303
1304 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1305 {
1306         return -EINVAL;
1307 }
1308
1309 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1310 {
1311         return -EINVAL;
1312 }
1313
1314 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
1315                 struct kvm_debug_guest *dbg)
1316 {
1317         return -EINVAL;
1318 }
1319
1320 static void free_kvm(struct kvm *kvm)
1321 {
1322         unsigned long vm_base = kvm->arch.vm_base;
1323
1324         if (vm_base) {
1325                 memset((void *)vm_base, 0, KVM_VM_DATA_SIZE);
1326                 free_pages(vm_base, get_order(KVM_VM_DATA_SIZE));
1327         }
1328
1329 }
1330
1331 static void kvm_release_vm_pages(struct kvm *kvm)
1332 {
1333         struct kvm_memory_slot *memslot;
1334         int i, j;
1335         unsigned long base_gfn;
1336
1337         for (i = 0; i < kvm->nmemslots; i++) {
1338                 memslot = &kvm->memslots[i];
1339                 base_gfn = memslot->base_gfn;
1340
1341                 for (j = 0; j < memslot->npages; j++) {
1342                         if (memslot->rmap[j])
1343                                 put_page((struct page *)memslot->rmap[j]);
1344                 }
1345         }
1346 }
1347
1348 void kvm_arch_destroy_vm(struct kvm *kvm)
1349 {
1350         kvm_iommu_unmap_guest(kvm);
1351 #ifdef  KVM_CAP_DEVICE_ASSIGNMENT
1352         kvm_free_all_assigned_devices(kvm);
1353 #endif
1354         kfree(kvm->arch.vioapic);
1355         kvm_release_vm_pages(kvm);
1356         kvm_free_physmem(kvm);
1357         free_kvm(kvm);
1358 }
1359
1360 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1361 {
1362 }
1363
1364 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1365 {
1366         if (cpu != vcpu->cpu) {
1367                 vcpu->cpu = cpu;
1368                 if (vcpu->arch.ht_active)
1369                         kvm_migrate_hlt_timer(vcpu);
1370         }
1371 }
1372
1373 #define SAVE_REGS(_x)   regs->_x = vcpu->arch._x
1374
1375 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1376 {
1377         int i;
1378         int r;
1379         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1380         vcpu_load(vcpu);
1381
1382         for (i = 0; i < 16; i++) {
1383                 regs->vpd.vgr[i] = vpd->vgr[i];
1384                 regs->vpd.vbgr[i] = vpd->vbgr[i];
1385         }
1386         for (i = 0; i < 128; i++)
1387                 regs->vpd.vcr[i] = vpd->vcr[i];
1388         regs->vpd.vhpi = vpd->vhpi;
1389         regs->vpd.vnat = vpd->vnat;
1390         regs->vpd.vbnat = vpd->vbnat;
1391         regs->vpd.vpsr = vpd->vpsr;
1392         regs->vpd.vpr = vpd->vpr;
1393
1394         r = -EFAULT;
1395         r = copy_to_user(regs->saved_guest, &vcpu->arch.guest,
1396                                         sizeof(union context));
1397         if (r)
1398                 goto out;
1399         r = copy_to_user(regs->saved_stack, (void *)vcpu, IA64_STK_OFFSET);
1400         if (r)
1401                 goto out;
1402         SAVE_REGS(mp_state);
1403         SAVE_REGS(vmm_rr);
1404         memcpy(regs->itrs, vcpu->arch.itrs, sizeof(struct thash_data) * NITRS);
1405         memcpy(regs->dtrs, vcpu->arch.dtrs, sizeof(struct thash_data) * NDTRS);
1406         SAVE_REGS(itr_regions);
1407         SAVE_REGS(dtr_regions);
1408         SAVE_REGS(tc_regions);
1409         SAVE_REGS(irq_check);
1410         SAVE_REGS(itc_check);
1411         SAVE_REGS(timer_check);
1412         SAVE_REGS(timer_pending);
1413         SAVE_REGS(last_itc);
1414         for (i = 0; i < 8; i++) {
1415                 regs->vrr[i] = vcpu->arch.vrr[i];
1416                 regs->ibr[i] = vcpu->arch.ibr[i];
1417                 regs->dbr[i] = vcpu->arch.dbr[i];
1418         }
1419         for (i = 0; i < 4; i++)
1420                 regs->insvc[i] = vcpu->arch.insvc[i];
1421         regs->saved_itc = vcpu->arch.itc_offset + ia64_getreg(_IA64_REG_AR_ITC);
1422         SAVE_REGS(xtp);
1423         SAVE_REGS(metaphysical_rr0);
1424         SAVE_REGS(metaphysical_rr4);
1425         SAVE_REGS(metaphysical_saved_rr0);
1426         SAVE_REGS(metaphysical_saved_rr4);
1427         SAVE_REGS(fp_psr);
1428         SAVE_REGS(saved_gp);
1429         vcpu_put(vcpu);
1430         r = 0;
1431 out:
1432         return r;
1433 }
1434
1435 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
1436 {
1437
1438         hrtimer_cancel(&vcpu->arch.hlt_timer);
1439         kfree(vcpu->arch.apic);
1440 }
1441
1442
1443 long kvm_arch_vcpu_ioctl(struct file *filp,
1444                 unsigned int ioctl, unsigned long arg)
1445 {
1446         return -EINVAL;
1447 }
1448
1449 int kvm_arch_set_memory_region(struct kvm *kvm,
1450                 struct kvm_userspace_memory_region *mem,
1451                 struct kvm_memory_slot old,
1452                 int user_alloc)
1453 {
1454         unsigned long i;
1455         unsigned long pfn;
1456         int npages = mem->memory_size >> PAGE_SHIFT;
1457         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
1458         unsigned long base_gfn = memslot->base_gfn;
1459
1460         for (i = 0; i < npages; i++) {
1461                 pfn = gfn_to_pfn(kvm, base_gfn + i);
1462                 if (!kvm_is_mmio_pfn(pfn)) {
1463                         kvm_set_pmt_entry(kvm, base_gfn + i,
1464                                         pfn << PAGE_SHIFT,
1465                                 _PAGE_AR_RWX | _PAGE_MA_WB);
1466                         memslot->rmap[i] = (unsigned long)pfn_to_page(pfn);
1467                 } else {
1468                         kvm_set_pmt_entry(kvm, base_gfn + i,
1469                                         GPFN_PHYS_MMIO | (pfn << PAGE_SHIFT),
1470                                         _PAGE_MA_UC);
1471                         memslot->rmap[i] = 0;
1472                         }
1473         }
1474
1475         return 0;
1476 }
1477
1478 void kvm_arch_flush_shadow(struct kvm *kvm)
1479 {
1480 }
1481
1482 long kvm_arch_dev_ioctl(struct file *filp,
1483                 unsigned int ioctl, unsigned long arg)
1484 {
1485         return -EINVAL;
1486 }
1487
1488 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1489 {
1490         kvm_vcpu_uninit(vcpu);
1491 }
1492
1493 static int vti_cpu_has_kvm_support(void)
1494 {
1495         long  avail = 1, status = 1, control = 1;
1496         long ret;
1497
1498         ret = ia64_pal_proc_get_features(&avail, &status, &control, 0);
1499         if (ret)
1500                 goto out;
1501
1502         if (!(avail & PAL_PROC_VM_BIT))
1503                 goto out;
1504
1505         printk(KERN_DEBUG"kvm: Hardware Supports VT\n");
1506
1507         ret = ia64_pal_vp_env_info(&kvm_vm_buffer_size, &vp_env_info);
1508         if (ret)
1509                 goto out;
1510         printk(KERN_DEBUG"kvm: VM Buffer Size:0x%lx\n", kvm_vm_buffer_size);
1511
1512         if (!(vp_env_info & VP_OPCODE)) {
1513                 printk(KERN_WARNING"kvm: No opcode ability on hardware, "
1514                                 "vm_env_info:0x%lx\n", vp_env_info);
1515         }
1516
1517         return 1;
1518 out:
1519         return 0;
1520 }
1521
1522 static int kvm_relocate_vmm(struct kvm_vmm_info *vmm_info,
1523                                                 struct module *module)
1524 {
1525         unsigned long module_base;
1526         unsigned long vmm_size;
1527
1528         unsigned long vmm_offset, func_offset, fdesc_offset;
1529         struct fdesc *p_fdesc;
1530
1531         BUG_ON(!module);
1532
1533         if (!kvm_vmm_base) {
1534                 printk("kvm: kvm area hasn't been initilized yet!!\n");
1535                 return -EFAULT;
1536         }
1537
1538         /*Calculate new position of relocated vmm module.*/
1539         module_base = (unsigned long)module->module_core;
1540         vmm_size = module->core_size;
1541         if (unlikely(vmm_size > KVM_VMM_SIZE))
1542                 return -EFAULT;
1543
1544         memcpy((void *)kvm_vmm_base, (void *)module_base, vmm_size);
1545         kvm_flush_icache(kvm_vmm_base, vmm_size);
1546
1547         /*Recalculate kvm_vmm_info based on new VMM*/
1548         vmm_offset = vmm_info->vmm_ivt - module_base;
1549         kvm_vmm_info->vmm_ivt = KVM_VMM_BASE + vmm_offset;
1550         printk(KERN_DEBUG"kvm: Relocated VMM's IVT Base Addr:%lx\n",
1551                         kvm_vmm_info->vmm_ivt);
1552
1553         fdesc_offset = (unsigned long)vmm_info->vmm_entry - module_base;
1554         kvm_vmm_info->vmm_entry = (kvm_vmm_entry *)(KVM_VMM_BASE +
1555                                                         fdesc_offset);
1556         func_offset = *(unsigned long *)vmm_info->vmm_entry - module_base;
1557         p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1558         p_fdesc->ip = KVM_VMM_BASE + func_offset;
1559         p_fdesc->gp = KVM_VMM_BASE+(p_fdesc->gp - module_base);
1560
1561         printk(KERN_DEBUG"kvm: Relocated VMM's Init Entry Addr:%lx\n",
1562                         KVM_VMM_BASE+func_offset);
1563
1564         fdesc_offset = (unsigned long)vmm_info->tramp_entry - module_base;
1565         kvm_vmm_info->tramp_entry = (kvm_tramp_entry *)(KVM_VMM_BASE +
1566                         fdesc_offset);
1567         func_offset = *(unsigned long *)vmm_info->tramp_entry - module_base;
1568         p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1569         p_fdesc->ip = KVM_VMM_BASE + func_offset;
1570         p_fdesc->gp = KVM_VMM_BASE + (p_fdesc->gp - module_base);
1571
1572         kvm_vmm_gp = p_fdesc->gp;
1573
1574         printk(KERN_DEBUG"kvm: Relocated VMM's Entry IP:%p\n",
1575                                                 kvm_vmm_info->vmm_entry);
1576         printk(KERN_DEBUG"kvm: Relocated VMM's Trampoline Entry IP:0x%lx\n",
1577                                                 KVM_VMM_BASE + func_offset);
1578
1579         return 0;
1580 }
1581
1582 int kvm_arch_init(void *opaque)
1583 {
1584         int r;
1585         struct kvm_vmm_info *vmm_info = (struct kvm_vmm_info *)opaque;
1586
1587         if (!vti_cpu_has_kvm_support()) {
1588                 printk(KERN_ERR "kvm: No Hardware Virtualization Support!\n");
1589                 r = -EOPNOTSUPP;
1590                 goto out;
1591         }
1592
1593         if (kvm_vmm_info) {
1594                 printk(KERN_ERR "kvm: Already loaded VMM module!\n");
1595                 r = -EEXIST;
1596                 goto out;
1597         }
1598
1599         r = -ENOMEM;
1600         kvm_vmm_info = kzalloc(sizeof(struct kvm_vmm_info), GFP_KERNEL);
1601         if (!kvm_vmm_info)
1602                 goto out;
1603
1604         if (kvm_alloc_vmm_area())
1605                 goto out_free0;
1606
1607         r = kvm_relocate_vmm(vmm_info, vmm_info->module);
1608         if (r)
1609                 goto out_free1;
1610
1611         return 0;
1612
1613 out_free1:
1614         kvm_free_vmm_area();
1615 out_free0:
1616         kfree(kvm_vmm_info);
1617 out:
1618         return r;
1619 }
1620
1621 void kvm_arch_exit(void)
1622 {
1623         kvm_free_vmm_area();
1624         kfree(kvm_vmm_info);
1625         kvm_vmm_info = NULL;
1626 }
1627
1628 static int kvm_ia64_sync_dirty_log(struct kvm *kvm,
1629                 struct kvm_dirty_log *log)
1630 {
1631         struct kvm_memory_slot *memslot;
1632         int r, i;
1633         long n, base;
1634         unsigned long *dirty_bitmap = (unsigned long *)((void *)kvm - KVM_VM_OFS
1635                                         + KVM_MEM_DIRTY_LOG_OFS);
1636
1637         r = -EINVAL;
1638         if (log->slot >= KVM_MEMORY_SLOTS)
1639                 goto out;
1640
1641         memslot = &kvm->memslots[log->slot];
1642         r = -ENOENT;
1643         if (!memslot->dirty_bitmap)
1644                 goto out;
1645
1646         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1647         base = memslot->base_gfn / BITS_PER_LONG;
1648
1649         for (i = 0; i < n/sizeof(long); ++i) {
1650                 memslot->dirty_bitmap[i] = dirty_bitmap[base + i];
1651                 dirty_bitmap[base + i] = 0;
1652         }
1653         r = 0;
1654 out:
1655         return r;
1656 }
1657
1658 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1659                 struct kvm_dirty_log *log)
1660 {
1661         int r;
1662         int n;
1663         struct kvm_memory_slot *memslot;
1664         int is_dirty = 0;
1665
1666         spin_lock(&kvm->arch.dirty_log_lock);
1667
1668         r = kvm_ia64_sync_dirty_log(kvm, log);
1669         if (r)
1670                 goto out;
1671
1672         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1673         if (r)
1674                 goto out;
1675
1676         /* If nothing is dirty, don't bother messing with page tables. */
1677         if (is_dirty) {
1678                 kvm_flush_remote_tlbs(kvm);
1679                 memslot = &kvm->memslots[log->slot];
1680                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1681                 memset(memslot->dirty_bitmap, 0, n);
1682         }
1683         r = 0;
1684 out:
1685         spin_unlock(&kvm->arch.dirty_log_lock);
1686         return r;
1687 }
1688
1689 int kvm_arch_hardware_setup(void)
1690 {
1691         return 0;
1692 }
1693
1694 void kvm_arch_hardware_unsetup(void)
1695 {
1696 }
1697
1698 static void vcpu_kick_intr(void *info)
1699 {
1700 #ifdef DEBUG
1701         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
1702         printk(KERN_DEBUG"vcpu_kick_intr %p \n", vcpu);
1703 #endif
1704 }
1705
1706 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1707 {
1708         int ipi_pcpu = vcpu->cpu;
1709         int cpu = get_cpu();
1710
1711         if (waitqueue_active(&vcpu->wq))
1712                 wake_up_interruptible(&vcpu->wq);
1713
1714         if (vcpu->guest_mode && cpu != ipi_pcpu)
1715                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
1716         put_cpu();
1717 }
1718
1719 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, u8 vec, u8 trig)
1720 {
1721
1722         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1723
1724         if (!test_and_set_bit(vec, &vpd->irr[0])) {
1725                 vcpu->arch.irq_new_pending = 1;
1726                 kvm_vcpu_kick(vcpu);
1727                 return 1;
1728         }
1729         return 0;
1730 }
1731
1732 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
1733 {
1734         return apic->vcpu->vcpu_id == dest;
1735 }
1736
1737 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
1738 {
1739         return 0;
1740 }
1741
1742 struct kvm_vcpu *kvm_get_lowest_prio_vcpu(struct kvm *kvm, u8 vector,
1743                                        unsigned long bitmap)
1744 {
1745         struct kvm_vcpu *lvcpu = kvm->vcpus[0];
1746         int i;
1747
1748         for (i = 1; i < KVM_MAX_VCPUS; i++) {
1749                 if (!kvm->vcpus[i])
1750                         continue;
1751                 if (lvcpu->arch.xtp > kvm->vcpus[i]->arch.xtp)
1752                         lvcpu = kvm->vcpus[i];
1753         }
1754
1755         return lvcpu;
1756 }
1757
1758 static int find_highest_bits(int *dat)
1759 {
1760         u32  bits, bitnum;
1761         int i;
1762
1763         /* loop for all 256 bits */
1764         for (i = 7; i >= 0 ; i--) {
1765                 bits = dat[i];
1766                 if (bits) {
1767                         bitnum = fls(bits);
1768                         return i * 32 + bitnum - 1;
1769                 }
1770         }
1771
1772         return -1;
1773 }
1774
1775 int kvm_highest_pending_irq(struct kvm_vcpu *vcpu)
1776 {
1777     struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1778
1779     if (vpd->irr[0] & (1UL << NMI_VECTOR))
1780                 return NMI_VECTOR;
1781     if (vpd->irr[0] & (1UL << ExtINT_VECTOR))
1782                 return ExtINT_VECTOR;
1783
1784     return find_highest_bits((int *)&vpd->irr[0]);
1785 }
1786
1787 int kvm_cpu_has_interrupt(struct kvm_vcpu *vcpu)
1788 {
1789         if (kvm_highest_pending_irq(vcpu) != -1)
1790                 return 1;
1791         return 0;
1792 }
1793
1794 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1795 {
1796         return vcpu->arch.timer_fired;
1797 }
1798
1799 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1800 {
1801         return gfn;
1802 }
1803
1804 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
1805 {
1806         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE;
1807 }
1808
1809 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1810                                     struct kvm_mp_state *mp_state)
1811 {
1812         vcpu_load(vcpu);
1813         mp_state->mp_state = vcpu->arch.mp_state;
1814         vcpu_put(vcpu);
1815         return 0;
1816 }
1817
1818 static int vcpu_reset(struct kvm_vcpu *vcpu)
1819 {
1820         int r;
1821         long psr;
1822         local_irq_save(psr);
1823         r = kvm_insert_vmm_mapping(vcpu);
1824         if (r)
1825                 goto fail;
1826
1827         vcpu->arch.launched = 0;
1828         kvm_arch_vcpu_uninit(vcpu);
1829         r = kvm_arch_vcpu_init(vcpu);
1830         if (r)
1831                 goto fail;
1832
1833         kvm_purge_vmm_mapping(vcpu);
1834         r = 0;
1835 fail:
1836         local_irq_restore(psr);
1837         return r;
1838 }
1839
1840 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1841                                     struct kvm_mp_state *mp_state)
1842 {
1843         int r = 0;
1844
1845         vcpu_load(vcpu);
1846         vcpu->arch.mp_state = mp_state->mp_state;
1847         if (vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)
1848                 r = vcpu_reset(vcpu);
1849         vcpu_put(vcpu);
1850         return r;
1851 }