SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
 783         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
 784 }
 785
 786 struct  kvm *kvm_arch_create_vm(void)
 787 {
 788         struct kvm *kvm = kvm_alloc_kvm();
 789
 790         if (IS_ERR(kvm))
 791                 return ERR_PTR(-ENOMEM);
 792         kvm_init_vm(kvm);
 793
 794         return kvm;
 795
 796 }
 797
 798 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm,
 799                                         struct kvm_irqchip *chip)
 800 {
 801         int r;
 802
 803         r = 0;
 804         switch (chip->chip_id) {
 805         case KVM_IRQCHIP_IOAPIC:
 806                 memcpy(&chip->chip.ioapic, ioapic_irqchip(kvm),
 807                                 sizeof(struct kvm_ioapic_state));
 808                 break;
 809         default:
 810                 r = -EINVAL;
 811                 break;
 812         }
 813         return r;
 814 }
 815
 816 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 817 {
 818         int r;
 819
 820         r = 0;
 821         switch (chip->chip_id) {
 822         case KVM_IRQCHIP_IOAPIC:
 823                 memcpy(ioapic_irqchip(kvm),
 824                                 &chip->chip.ioapic,
 825                                 sizeof(struct kvm_ioapic_state));
 826                 break;
 827         default:
 828                 r = -EINVAL;
 829                 break;
 830         }
 831         return r;
 832 }
 833
 834 #define RESTORE_REGS(_x) vcpu->arch._x = regs->_x
 835
 836 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 837 {
 838         int i;
 839         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
 840         int r;
 841
 842         vcpu_load(vcpu);
 843
 844         for (i = 0; i < 16; i++) {
 845                 vpd->vgr[i] = regs->vpd.vgr[i];
 846                 vpd->vbgr[i] = regs->vpd.vbgr[i];
 847         }
 848         for (i = 0; i < 128; i++)
 849                 vpd->vcr[i] = regs->vpd.vcr[i];
 850         vpd->vhpi = regs->vpd.vhpi;
 851         vpd->vnat = regs->vpd.vnat;
 852         vpd->vbnat = regs->vpd.vbnat;
 853         vpd->vpsr = regs->vpd.vpsr;
 854
 855         vpd->vpr = regs->vpd.vpr;
 856
 857         r = -EFAULT;
 858         r = copy_from_user(&vcpu->arch.guest, regs->saved_guest,
 859                                                 sizeof(union context));
 860         if (r)
 861                 goto out;
 862         r = copy_from_user(vcpu + 1, regs->saved_stack +
 863                         sizeof(struct kvm_vcpu),
 864                         IA64_STK_OFFSET - sizeof(struct kvm_vcpu));
 865         if (r)
 866                 goto out;
 867         vcpu->arch.exit_data =
 868                 ((struct kvm_vcpu *)(regs->saved_stack))->arch.exit_data;
 869
 870         RESTORE_REGS(mp_state);
 871         RESTORE_REGS(vmm_rr);
 872         memcpy(vcpu->arch.itrs, regs->itrs, sizeof(struct thash_data) * NITRS);
 873         memcpy(vcpu->arch.dtrs, regs->dtrs, sizeof(struct thash_data) * NDTRS);
 874         RESTORE_REGS(itr_regions);
 875         RESTORE_REGS(dtr_regions);
 876         RESTORE_REGS(tc_regions);
 877         RESTORE_REGS(irq_check);
 878         RESTORE_REGS(itc_check);
 879         RESTORE_REGS(timer_check);
 880         RESTORE_REGS(timer_pending);
 881         RESTORE_REGS(last_itc);
 882         for (i = 0; i < 8; i++) {
 883                 vcpu->arch.vrr[i] = regs->vrr[i];
 884                 vcpu->arch.ibr[i] = regs->ibr[i];
 885                 vcpu->arch.dbr[i] = regs->dbr[i];
 886         }
 887         for (i = 0; i < 4; i++)
 888                 vcpu->arch.insvc[i] = regs->insvc[i];
 889         RESTORE_REGS(xtp);
 890         RESTORE_REGS(metaphysical_rr0);
 891         RESTORE_REGS(metaphysical_rr4);
 892         RESTORE_REGS(metaphysical_saved_rr0);
 893         RESTORE_REGS(metaphysical_saved_rr4);
 894         RESTORE_REGS(fp_psr);
 895         RESTORE_REGS(saved_gp);
 896
 897         vcpu->arch.irq_new_pending = 1;
 898         vcpu->arch.itc_offset = regs->saved_itc - ia64_getreg(_IA64_REG_AR_ITC);
 899         set_bit(KVM_REQ_RESUME, &vcpu->requests);
 900
 901         vcpu_put(vcpu);
 902         r = 0;
 903 out:
 904         return r;
 905 }
 906
 907 long kvm_arch_vm_ioctl(struct file *filp,
 908                 unsigned int ioctl, unsigned long arg)
 909 {
 910         struct kvm *kvm = filp->private_data;
 911         void __user *argp = (void __user *)arg;
 912         int r = -EINVAL;
 913
 914         switch (ioctl) {
 915         case KVM_SET_MEMORY_REGION: {
 916                 struct kvm_memory_region kvm_mem;
 917                 struct kvm_userspace_memory_region kvm_userspace_mem;
 918
 919                 r = -EFAULT;
 920                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
 921                         goto out;
 922                 kvm_userspace_mem.slot = kvm_mem.slot;
 923                 kvm_userspace_mem.flags = kvm_mem.flags;
 924                 kvm_userspace_mem.guest_phys_addr =
 925                                         kvm_mem.guest_phys_addr;
 926                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
 927                 r = kvm_vm_ioctl_set_memory_region(kvm,
 928                                         &kvm_userspace_mem, 0);
 929                 if (r)
 930                         goto out;
 931                 break;
 932                 }
 933         case KVM_CREATE_IRQCHIP:
 934                 r = -EFAULT;
 935                 r = kvm_ioapic_init(kvm);
 936                 if (r)
 937                         goto out;
 938                 break;
 939         case KVM_IRQ_LINE: {
 940                 struct kvm_irq_level irq_event;
 941
 942                 r = -EFAULT;
 943                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
 944                         goto out;
 945                 if (irqchip_in_kernel(kvm)) {
 946                         mutex_lock(&kvm->lock);
 947                         kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
 948                                     irq_event.irq, irq_event.level);
 949                         mutex_unlock(&kvm->lock);
 950                         r = 0;
 951                 }
 952                 break;
 953                 }
 954         case KVM_GET_IRQCHIP: {
 955                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
 956                 struct kvm_irqchip chip;
 957
 958                 r = -EFAULT;
 959                 if (copy_from_user(&chip, argp, sizeof chip))
 960                                 goto out;
 961                 r = -ENXIO;
 962                 if (!irqchip_in_kernel(kvm))
 963                         goto out;
 964                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
 965                 if (r)
 966                         goto out;
 967                 r = -EFAULT;
 968                 if (copy_to_user(argp, &chip, sizeof chip))
 969                                 goto out;
 970                 r = 0;
 971                 break;
 972                 }
 973         case KVM_SET_IRQCHIP: {
 974                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
 975                 struct kvm_irqchip chip;
 976
 977                 r = -EFAULT;
 978                 if (copy_from_user(&chip, argp, sizeof chip))
 979                                 goto out;
 980                 r = -ENXIO;
 981                 if (!irqchip_in_kernel(kvm))
 982                         goto out;
 983                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
 984                 if (r)
 985                         goto out;
 986                 r = 0;
 987                 break;
 988                 }
 989         default:
 990                 ;
 991         }
 992 out:
 993         return r;
 994 }
 995
 996 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 997                 struct kvm_sregs *sregs)
 998 {
 999         return -EINVAL;
1000 }
1001
1002 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1003                 struct kvm_sregs *sregs)
1004 {
1005         return -EINVAL;
1006
1007 }
1008 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1009                 struct kvm_translation *tr)
1010 {
1011
1012         return -EINVAL;
1013 }
1014
1015 static int kvm_alloc_vmm_area(void)
1016 {
1017         if (!kvm_vmm_base && (kvm_vm_buffer_size < KVM_VM_BUFFER_SIZE)) {
1018                 kvm_vmm_base = __get_free_pages(GFP_KERNEL,
1019                                 get_order(KVM_VMM_SIZE));
1020                 if (!kvm_vmm_base)
1021                         return -ENOMEM;
1022
1023                 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1024                 kvm_vm_buffer = kvm_vmm_base + VMM_SIZE;
1025
1026                 printk(KERN_DEBUG"kvm:VMM's Base Addr:0x%lx, vm_buffer:0x%lx\n",
1027                                 kvm_vmm_base, kvm_vm_buffer);
1028         }
1029
1030         return 0;
1031 }
1032
1033 static void kvm_free_vmm_area(void)
1034 {
1035         if (kvm_vmm_base) {
1036                 /*Zero this area before free to avoid bits leak!!*/
1037                 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1038                 free_pages(kvm_vmm_base, get_order(KVM_VMM_SIZE));
1039                 kvm_vmm_base  = 0;
1040                 kvm_vm_buffer = 0;
1041                 kvm_vsa_base = 0;
1042         }
1043 }
1044
1045 static void vti_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1046 {
1047 }
1048
1049 static int vti_init_vpd(struct kvm_vcpu *vcpu)
1050 {
1051         int i;
1052         union cpuid3_t cpuid3;
1053         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1054
1055         if (IS_ERR(vpd))
1056                 return PTR_ERR(vpd);
1057
1058         /* CPUID init */
1059         for (i = 0; i < 5; i++)
1060                 vpd->vcpuid[i] = ia64_get_cpuid(i);
1061
1062         /* Limit the CPUID number to 5 */
1063         cpuid3.value = vpd->vcpuid[3];
1064         cpuid3.number = 4;      /* 5 - 1 */
1065         vpd->vcpuid[3] = cpuid3.value;
1066
1067         /*Set vac and vdc fields*/
1068         vpd->vac.a_from_int_cr = 1;
1069         vpd->vac.a_to_int_cr = 1;
1070         vpd->vac.a_from_psr = 1;
1071         vpd->vac.a_from_cpuid = 1;
1072         vpd->vac.a_cover = 1;
1073         vpd->vac.a_bsw = 1;
1074         vpd->vac.a_int = 1;
1075         vpd->vdc.d_vmsw = 1;
1076
1077         /*Set virtual buffer*/
1078         vpd->virt_env_vaddr = KVM_VM_BUFFER_BASE;
1079
1080         return 0;
1081 }
1082
1083 static int vti_create_vp(struct kvm_vcpu *vcpu)
1084 {
1085         long ret;
1086         struct vpd *vpd = vcpu->arch.vpd;
1087         unsigned long  vmm_ivt;
1088
1089         vmm_ivt = kvm_vmm_info->vmm_ivt;
1090
1091         printk(KERN_DEBUG "kvm: vcpu:%p,ivt: 0x%lx\n", vcpu, vmm_ivt);
1092
1093         ret = ia64_pal_vp_create((u64 *)vpd, (u64 *)vmm_ivt, 0);
1094
1095         if (ret) {
1096                 printk(KERN_ERR"kvm: ia64_pal_vp_create failed!\n");
1097                 return -EINVAL;
1098         }
1099         return 0;
1100 }
1101
1102 static void init_ptce_info(struct kvm_vcpu *vcpu)
1103 {
1104         ia64_ptce_info_t ptce = {0};
1105
1106         ia64_get_ptce(&ptce);
1107         vcpu->arch.ptce_base = ptce.base;
1108         vcpu->arch.ptce_count[0] = ptce.count[0];
1109         vcpu->arch.ptce_count[1] = ptce.count[1];
1110         vcpu->arch.ptce_stride[0] = ptce.stride[0];
1111         vcpu->arch.ptce_stride[1] = ptce.stride[1];
1112 }
1113
1114 static void kvm_migrate_hlt_timer(struct kvm_vcpu *vcpu)
1115 {
1116         struct hrtimer *p_ht = &vcpu->arch.hlt_timer;
1117
1118         if (hrtimer_cancel(p_ht))
1119                 hrtimer_start_expires(p_ht, HRTIMER_MODE_ABS);
1120 }
1121
1122 static enum hrtimer_restart hlt_timer_fn(struct hrtimer *data)
1123 {
1124         struct kvm_vcpu *vcpu;
1125         wait_queue_head_t *q;
1126
1127         vcpu  = container_of(data, struct kvm_vcpu, arch.hlt_timer);
1128         if (vcpu->arch.mp_state != KVM_MP_STATE_HALTED)
1129                 goto out;
1130
1131         q = &vcpu->wq;
1132         if (waitqueue_active(q)) {
1133                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1134                 wake_up_interruptible(q);
1135         }
1136 out:
1137         vcpu->arch.timer_check = 1;
1138         return HRTIMER_NORESTART;
1139 }
1140
1141 #define PALE_RESET_ENTRY    0x80000000ffffffb0UL
1142
1143 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1144 {
1145         struct kvm_vcpu *v;
1146         int r;
1147         int i;
1148         long itc_offset;
1149         struct kvm *kvm = vcpu->kvm;
1150         struct kvm_pt_regs *regs = vcpu_regs(vcpu);
1151
1152         union context *p_ctx = &vcpu->arch.guest;
1153         struct kvm_vcpu *vmm_vcpu = to_guest(vcpu->kvm, vcpu);
1154
1155         /*Init vcpu context for first run.*/
1156         if (IS_ERR(vmm_vcpu))
1157                 return PTR_ERR(vmm_vcpu);
1158
1159         if (vcpu->vcpu_id == 0) {
1160                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1161
1162                 /*Set entry address for first run.*/
1163                 regs->cr_iip = PALE_RESET_ENTRY;
1164
1165                 /*Initilize itc offset for vcpus*/
1166                 itc_offset = 0UL - ia64_getreg(_IA64_REG_AR_ITC);
1167                 for (i = 0; i < MAX_VCPU_NUM; i++) {
1168                         v = (struct kvm_vcpu *)((char *)vcpu + VCPU_SIZE * i);
1169                         v->arch.itc_offset = itc_offset;
1170                         v->arch.last_itc = 0;
1171                 }
1172         } else
1173                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
1174
1175         r = -ENOMEM;
1176         vcpu->arch.apic = kzalloc(sizeof(struct kvm_lapic), GFP_KERNEL);
1177         if (!vcpu->arch.apic)
1178                 goto out;
1179         vcpu->arch.apic->vcpu = vcpu;
1180
1181         p_ctx->gr[1] = 0;
1182         p_ctx->gr[12] = (unsigned long)((char *)vmm_vcpu + IA64_STK_OFFSET);
1183         p_ctx->gr[13] = (unsigned long)vmm_vcpu;
1184         p_ctx->psr = 0x1008522000UL;
1185         p_ctx->ar[40] = FPSR_DEFAULT; /*fpsr*/
1186         p_ctx->caller_unat = 0;
1187         p_ctx->pr = 0x0;
1188         p_ctx->ar[36] = 0x0; /*unat*/
1189         p_ctx->ar[19] = 0x0; /*rnat*/
1190         p_ctx->ar[18] = (unsigned long)vmm_vcpu +
1191                                 ((sizeof(struct kvm_vcpu)+15) & ~15);
1192         p_ctx->ar[64] = 0x0; /*pfs*/
1193         p_ctx->cr[0] = 0x7e04UL;
1194         p_ctx->cr[2] = (unsigned long)kvm_vmm_info->vmm_ivt;
1195         p_ctx->cr[8] = 0x3c;
1196
1197         /*Initilize region register*/
1198         p_ctx->rr[0] = 0x30;
1199         p_ctx->rr[1] = 0x30;
1200         p_ctx->rr[2] = 0x30;
1201         p_ctx->rr[3] = 0x30;
1202         p_ctx->rr[4] = 0x30;
1203         p_ctx->rr[5] = 0x30;
1204         p_ctx->rr[7] = 0x30;
1205
1206         /*Initilize branch register 0*/
1207         p_ctx->br[0] = *(unsigned long *)kvm_vmm_info->vmm_entry;
1208
1209         vcpu->arch.vmm_rr = kvm->arch.vmm_init_rr;
1210         vcpu->arch.metaphysical_rr0 = kvm->arch.metaphysical_rr0;
1211         vcpu->arch.metaphysical_rr4 = kvm->arch.metaphysical_rr4;
1212
1213         hrtimer_init(&vcpu->arch.hlt_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1214         vcpu->arch.hlt_timer.function = hlt_timer_fn;
1215
1216         vcpu->arch.last_run_cpu = -1;
1217         vcpu->arch.vpd = (struct vpd *)VPD_ADDR(vcpu->vcpu_id);
1218         vcpu->arch.vsa_base = kvm_vsa_base;
1219         vcpu->arch.__gp = kvm_vmm_gp;
1220         vcpu->arch.dirty_log_lock_pa = __pa(&kvm->arch.dirty_log_lock);
1221         vcpu->arch.vhpt.hash = (struct thash_data *)VHPT_ADDR(vcpu->vcpu_id);
1222         vcpu->arch.vtlb.hash = (struct thash_data *)VTLB_ADDR(vcpu->vcpu_id);
1223         init_ptce_info(vcpu);
1224
1225         r = 0;
1226 out:
1227         return r;
1228 }
1229
1230 static int vti_vcpu_setup(struct kvm_vcpu *vcpu, int id)
1231 {
1232         unsigned long psr;
1233         int r;
1234
1235         local_irq_save(psr);
1236         r = kvm_insert_vmm_mapping(vcpu);
1237         if (r)
1238                 goto fail;
1239         r = kvm_vcpu_init(vcpu, vcpu->kvm, id);
1240         if (r)
1241                 goto fail;
1242
1243         r = vti_init_vpd(vcpu);
1244         if (r) {
1245                 printk(KERN_DEBUG"kvm: vpd init error!!\n");
1246                 goto uninit;
1247         }
1248
1249         r = vti_create_vp(vcpu);
1250         if (r)
1251                 goto uninit;
1252
1253         kvm_purge_vmm_mapping(vcpu);
1254         local_irq_restore(psr);
1255
1256         return 0;
1257 uninit:
1258         kvm_vcpu_uninit(vcpu);
1259 fail:
1260         local_irq_restore(psr);
1261         return r;
1262 }
1263
1264 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
1265                 unsigned int id)
1266 {
1267         struct kvm_vcpu *vcpu;
1268         unsigned long vm_base = kvm->arch.vm_base;
1269         int r;
1270         int cpu;
1271
1272         r = -ENOMEM;
1273         if (!vm_base) {
1274                 printk(KERN_ERR"kvm: Create vcpu[%d] error!\n", id);
1275                 goto fail;
1276         }
1277         vcpu = (struct kvm_vcpu *)(vm_base + KVM_VCPU_OFS + VCPU_SIZE * id);
1278         vcpu->kvm = kvm;
1279
1280         cpu = get_cpu();
1281         vti_vcpu_load(vcpu, cpu);
1282         r = vti_vcpu_setup(vcpu, id);
1283         put_cpu();
1284
1285         if (r) {
1286                 printk(KERN_DEBUG"kvm: vcpu_setup error!!\n");
1287                 goto fail;
1288         }
1289
1290         return vcpu;
1291 fail:
1292         return ERR_PTR(r);
1293 }
1294
1295 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1296 {
1297         return 0;
1298 }
1299
1300 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1301 {
1302         return -EINVAL;
1303 }
1304
1305 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1306 {
1307         return -EINVAL;
1308 }
1309
1310 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
1311                 struct kvm_debug_guest *dbg)
1312 {
1313         return -EINVAL;
1314 }
1315
1316 static void free_kvm(struct kvm *kvm)
1317 {
1318         unsigned long vm_base = kvm->arch.vm_base;
1319
1320         if (vm_base) {
1321                 memset((void *)vm_base, 0, KVM_VM_DATA_SIZE);
1322                 free_pages(vm_base, get_order(KVM_VM_DATA_SIZE));
1323         }
1324
1325 }
1326
1327 static void kvm_release_vm_pages(struct kvm *kvm)
1328 {
1329         struct kvm_memory_slot *memslot;
1330         int i, j;
1331         unsigned long base_gfn;
1332
1333         for (i = 0; i < kvm->nmemslots; i++) {
1334                 memslot = &kvm->memslots[i];
1335                 base_gfn = memslot->base_gfn;
1336
1337                 for (j = 0; j < memslot->npages; j++) {
1338                         if (memslot->rmap[j])
1339                                 put_page((struct page *)memslot->rmap[j]);
1340                 }
1341         }
1342 }
1343
1344 void kvm_arch_destroy_vm(struct kvm *kvm)
1345 {
1346         kvm_iommu_unmap_guest(kvm);
1347 #ifdef  KVM_CAP_DEVICE_ASSIGNMENT
1348         kvm_free_all_assigned_devices(kvm);
1349 #endif
1350         kfree(kvm->arch.vioapic);
1351         kvm_release_vm_pages(kvm);
1352         kvm_free_physmem(kvm);
1353         free_kvm(kvm);
1354 }
1355
1356 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1357 {
1358 }
1359
1360 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1361 {
1362         if (cpu != vcpu->cpu) {
1363                 vcpu->cpu = cpu;
1364                 if (vcpu->arch.ht_active)
1365                         kvm_migrate_hlt_timer(vcpu);
1366         }
1367 }
1368
1369 #define SAVE_REGS(_x)   regs->_x = vcpu->arch._x
1370
1371 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1372 {
1373         int i;
1374         int r;
1375         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1376         vcpu_load(vcpu);
1377
1378         for (i = 0; i < 16; i++) {
1379                 regs->vpd.vgr[i] = vpd->vgr[i];
1380                 regs->vpd.vbgr[i] = vpd->vbgr[i];
1381         }
1382         for (i = 0; i < 128; i++)
1383                 regs->vpd.vcr[i] = vpd->vcr[i];
1384         regs->vpd.vhpi = vpd->vhpi;
1385         regs->vpd.vnat = vpd->vnat;
1386         regs->vpd.vbnat = vpd->vbnat;
1387         regs->vpd.vpsr = vpd->vpsr;
1388         regs->vpd.vpr = vpd->vpr;
1389
1390         r = -EFAULT;
1391         r = copy_to_user(regs->saved_guest, &vcpu->arch.guest,
1392                                         sizeof(union context));
1393         if (r)
1394                 goto out;
1395         r = copy_to_user(regs->saved_stack, (void *)vcpu, IA64_STK_OFFSET);
1396         if (r)
1397                 goto out;
1398         SAVE_REGS(mp_state);
1399         SAVE_REGS(vmm_rr);
1400         memcpy(regs->itrs, vcpu->arch.itrs, sizeof(struct thash_data) * NITRS);
1401         memcpy(regs->dtrs, vcpu->arch.dtrs, sizeof(struct thash_data) * NDTRS);
1402         SAVE_REGS(itr_regions);
1403         SAVE_REGS(dtr_regions);
1404         SAVE_REGS(tc_regions);
1405         SAVE_REGS(irq_check);
1406         SAVE_REGS(itc_check);
1407         SAVE_REGS(timer_check);
1408         SAVE_REGS(timer_pending);
1409         SAVE_REGS(last_itc);
1410         for (i = 0; i < 8; i++) {
1411                 regs->vrr[i] = vcpu->arch.vrr[i];
1412                 regs->ibr[i] = vcpu->arch.ibr[i];
1413                 regs->dbr[i] = vcpu->arch.dbr[i];
1414         }
1415         for (i = 0; i < 4; i++)
1416                 regs->insvc[i] = vcpu->arch.insvc[i];
1417         regs->saved_itc = vcpu->arch.itc_offset + ia64_getreg(_IA64_REG_AR_ITC);
1418         SAVE_REGS(xtp);
1419         SAVE_REGS(metaphysical_rr0);
1420         SAVE_REGS(metaphysical_rr4);
1421         SAVE_REGS(metaphysical_saved_rr0);
1422         SAVE_REGS(metaphysical_saved_rr4);
1423         SAVE_REGS(fp_psr);
1424         SAVE_REGS(saved_gp);
1425         vcpu_put(vcpu);
1426         r = 0;
1427 out:
1428         return r;
1429 }
1430
1431 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
1432 {
1433
1434         hrtimer_cancel(&vcpu->arch.hlt_timer);
1435         kfree(vcpu->arch.apic);
1436 }
1437
1438
1439 long kvm_arch_vcpu_ioctl(struct file *filp,
1440                 unsigned int ioctl, unsigned long arg)
1441 {
1442         return -EINVAL;
1443 }
1444
1445 int kvm_arch_set_memory_region(struct kvm *kvm,
1446                 struct kvm_userspace_memory_region *mem,
1447                 struct kvm_memory_slot old,
1448                 int user_alloc)
1449 {
1450         unsigned long i;
1451         unsigned long pfn;
1452         int npages = mem->memory_size >> PAGE_SHIFT;
1453         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
1454         unsigned long base_gfn = memslot->base_gfn;
1455
1456         for (i = 0; i < npages; i++) {
1457                 pfn = gfn_to_pfn(kvm, base_gfn + i);
1458                 if (!kvm_is_mmio_pfn(pfn)) {
1459                         kvm_set_pmt_entry(kvm, base_gfn + i,
1460                                         pfn << PAGE_SHIFT,
1461                                 _PAGE_AR_RWX | _PAGE_MA_WB);
1462                         memslot->rmap[i] = (unsigned long)pfn_to_page(pfn);
1463                 } else {
1464                         kvm_set_pmt_entry(kvm, base_gfn + i,
1465                                         GPFN_PHYS_MMIO | (pfn << PAGE_SHIFT),
1466                                         _PAGE_MA_UC);
1467                         memslot->rmap[i] = 0;
1468                         }
1469         }
1470
1471         return 0;
1472 }
1473
1474 void kvm_arch_flush_shadow(struct kvm *kvm)
1475 {
1476 }
1477
1478 long kvm_arch_dev_ioctl(struct file *filp,
1479                 unsigned int ioctl, unsigned long arg)
1480 {
1481         return -EINVAL;
1482 }
1483
1484 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1485 {
1486         kvm_vcpu_uninit(vcpu);
1487 }
1488
1489 static int vti_cpu_has_kvm_support(void)
1490 {
1491         long  avail = 1, status = 1, control = 1;
1492         long ret;
1493
1494         ret = ia64_pal_proc_get_features(&avail, &status, &control, 0);
1495         if (ret)
1496                 goto out;
1497
1498         if (!(avail & PAL_PROC_VM_BIT))
1499                 goto out;
1500
1501         printk(KERN_DEBUG"kvm: Hardware Supports VT\n");
1502
1503         ret = ia64_pal_vp_env_info(&kvm_vm_buffer_size, &vp_env_info);
1504         if (ret)
1505                 goto out;
1506         printk(KERN_DEBUG"kvm: VM Buffer Size:0x%lx\n", kvm_vm_buffer_size);
1507
1508         if (!(vp_env_info & VP_OPCODE)) {
1509                 printk(KERN_WARNING"kvm: No opcode ability on hardware, "
1510                                 "vm_env_info:0x%lx\n", vp_env_info);
1511         }
1512
1513         return 1;
1514 out:
1515         return 0;
1516 }
1517
1518 static int kvm_relocate_vmm(struct kvm_vmm_info *vmm_info,
1519                                                 struct module *module)
1520 {
1521         unsigned long module_base;
1522         unsigned long vmm_size;
1523
1524         unsigned long vmm_offset, func_offset, fdesc_offset;
1525         struct fdesc *p_fdesc;
1526
1527         BUG_ON(!module);
1528
1529         if (!kvm_vmm_base) {
1530                 printk("kvm: kvm area hasn't been initilized yet!!\n");
1531                 return -EFAULT;
1532         }
1533
1534         /*Calculate new position of relocated vmm module.*/
1535         module_base = (unsigned long)module->module_core;
1536         vmm_size = module->core_size;
1537         if (unlikely(vmm_size > KVM_VMM_SIZE))
1538                 return -EFAULT;
1539
1540         memcpy((void *)kvm_vmm_base, (void *)module_base, vmm_size);
1541         kvm_flush_icache(kvm_vmm_base, vmm_size);
1542
1543         /*Recalculate kvm_vmm_info based on new VMM*/
1544         vmm_offset = vmm_info->vmm_ivt - module_base;
1545         kvm_vmm_info->vmm_ivt = KVM_VMM_BASE + vmm_offset;
1546         printk(KERN_DEBUG"kvm: Relocated VMM's IVT Base Addr:%lx\n",
1547                         kvm_vmm_info->vmm_ivt);
1548
1549         fdesc_offset = (unsigned long)vmm_info->vmm_entry - module_base;
1550         kvm_vmm_info->vmm_entry = (kvm_vmm_entry *)(KVM_VMM_BASE +
1551                                                         fdesc_offset);
1552         func_offset = *(unsigned long *)vmm_info->vmm_entry - module_base;
1553         p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1554         p_fdesc->ip = KVM_VMM_BASE + func_offset;
1555         p_fdesc->gp = KVM_VMM_BASE+(p_fdesc->gp - module_base);
1556
1557         printk(KERN_DEBUG"kvm: Relocated VMM's Init Entry Addr:%lx\n",
1558                         KVM_VMM_BASE+func_offset);
1559
1560         fdesc_offset = (unsigned long)vmm_info->tramp_entry - module_base;
1561         kvm_vmm_info->tramp_entry = (kvm_tramp_entry *)(KVM_VMM_BASE +
1562                         fdesc_offset);
1563         func_offset = *(unsigned long *)vmm_info->tramp_entry - module_base;
1564         p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1565         p_fdesc->ip = KVM_VMM_BASE + func_offset;
1566         p_fdesc->gp = KVM_VMM_BASE + (p_fdesc->gp - module_base);
1567
1568         kvm_vmm_gp = p_fdesc->gp;
1569
1570         printk(KERN_DEBUG"kvm: Relocated VMM's Entry IP:%p\n",
1571                                                 kvm_vmm_info->vmm_entry);
1572         printk(KERN_DEBUG"kvm: Relocated VMM's Trampoline Entry IP:0x%lx\n",
1573                                                 KVM_VMM_BASE + func_offset);
1574
1575         return 0;
1576 }
1577
1578 int kvm_arch_init(void *opaque)
1579 {
1580         int r;
1581         struct kvm_vmm_info *vmm_info = (struct kvm_vmm_info *)opaque;
1582
1583         if (!vti_cpu_has_kvm_support()) {
1584                 printk(KERN_ERR "kvm: No Hardware Virtualization Support!\n");
1585                 r = -EOPNOTSUPP;
1586                 goto out;
1587         }
1588
1589         if (kvm_vmm_info) {
1590                 printk(KERN_ERR "kvm: Already loaded VMM module!\n");
1591                 r = -EEXIST;
1592                 goto out;
1593         }
1594
1595         r = -ENOMEM;
1596         kvm_vmm_info = kzalloc(sizeof(struct kvm_vmm_info), GFP_KERNEL);
1597         if (!kvm_vmm_info)
1598                 goto out;
1599
1600         if (kvm_alloc_vmm_area())
1601                 goto out_free0;
1602
1603         r = kvm_relocate_vmm(vmm_info, vmm_info->module);
1604         if (r)
1605                 goto out_free1;
1606
1607         return 0;
1608
1609 out_free1:
1610         kvm_free_vmm_area();
1611 out_free0:
1612         kfree(kvm_vmm_info);
1613 out:
1614         return r;
1615 }
1616
1617 void kvm_arch_exit(void)
1618 {
1619         kvm_free_vmm_area();
1620         kfree(kvm_vmm_info);
1621         kvm_vmm_info = NULL;
1622 }
1623
1624 static int kvm_ia64_sync_dirty_log(struct kvm *kvm,
1625                 struct kvm_dirty_log *log)
1626 {
1627         struct kvm_memory_slot *memslot;
1628         int r, i;
1629         long n, base;
1630         unsigned long *dirty_bitmap = (unsigned long *)((void *)kvm - KVM_VM_OFS
1631                                         + KVM_MEM_DIRTY_LOG_OFS);
1632
1633         r = -EINVAL;
1634         if (log->slot >= KVM_MEMORY_SLOTS)
1635                 goto out;
1636
1637         memslot = &kvm->memslots[log->slot];
1638         r = -ENOENT;
1639         if (!memslot->dirty_bitmap)
1640                 goto out;
1641
1642         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1643         base = memslot->base_gfn / BITS_PER_LONG;
1644
1645         for (i = 0; i < n/sizeof(long); ++i) {
1646                 memslot->dirty_bitmap[i] = dirty_bitmap[base + i];
1647                 dirty_bitmap[base + i] = 0;
1648         }
1649         r = 0;
1650 out:
1651         return r;
1652 }
1653
1654 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1655                 struct kvm_dirty_log *log)
1656 {
1657         int r;
1658         int n;
1659         struct kvm_memory_slot *memslot;
1660         int is_dirty = 0;
1661
1662         spin_lock(&kvm->arch.dirty_log_lock);
1663
1664         r = kvm_ia64_sync_dirty_log(kvm, log);
1665         if (r)
1666                 goto out;
1667
1668         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1669         if (r)
1670                 goto out;
1671
1672         /* If nothing is dirty, don't bother messing with page tables. */
1673         if (is_dirty) {
1674                 kvm_flush_remote_tlbs(kvm);
1675                 memslot = &kvm->memslots[log->slot];
1676                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1677                 memset(memslot->dirty_bitmap, 0, n);
1678         }
1679         r = 0;
1680 out:
1681         spin_unlock(&kvm->arch.dirty_log_lock);
1682         return r;
1683 }
1684
1685 int kvm_arch_hardware_setup(void)
1686 {
1687         return 0;
1688 }
1689
1690 void kvm_arch_hardware_unsetup(void)
1691 {
1692 }
1693
1694 static void vcpu_kick_intr(void *info)
1695 {
1696 #ifdef DEBUG
1697         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
1698         printk(KERN_DEBUG"vcpu_kick_intr %p \n", vcpu);
1699 #endif
1700 }
1701
1702 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1703 {
1704         int ipi_pcpu = vcpu->cpu;
1705
1706         if (waitqueue_active(&vcpu->wq))
1707                 wake_up_interruptible(&vcpu->wq);
1708
1709         if (vcpu->guest_mode)
1710                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
1711 }
1712
1713 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, u8 vec, u8 trig)
1714 {
1715
1716         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1717
1718         if (!test_and_set_bit(vec, &vpd->irr[0])) {
1719                 vcpu->arch.irq_new_pending = 1;
1720                  if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
1721                         kvm_vcpu_kick(vcpu);
1722                 else if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED) {
1723                         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1724                         if (waitqueue_active(&vcpu->wq))
1725                                 wake_up_interruptible(&vcpu->wq);
1726                 }
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 0;
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 }