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
  35 #include <asm/pgtable.h>
  36 #include <asm/gcc_intrin.h>
  37 #include <asm/pal.h>
  38 #include <asm/cacheflush.h>
  39 #include <asm/div64.h>
  40 #include <asm/tlb.h>
  41
  42 #include "misc.h"
  43 #include "vti.h"
  44 #include "iodev.h"
  45 #include "ioapic.h"
  46 #include "lapic.h"
  47
  48 static unsigned long kvm_vmm_base;
  49 static unsigned long kvm_vsa_base;
  50 static unsigned long kvm_vm_buffer;
  51 static unsigned long kvm_vm_buffer_size;
  52 unsigned long kvm_vmm_gp;
  53
  54 static long vp_env_info;
  55
  56 static struct kvm_vmm_info *kvm_vmm_info;
  57
  58 static DEFINE_PER_CPU(struct kvm_vcpu *, last_vcpu);
  59
  60 struct kvm_stats_debugfs_item debugfs_entries[] = {
  61         { NULL }
  62 };
  63
  64
  65 struct fdesc{
  66     unsigned long ip;
  67     unsigned long gp;
  68 };
  69
  70 static void kvm_flush_icache(unsigned long start, unsigned long len)
  71 {
  72         int l;
  73
  74         for (l = 0; l < (len + 32); l += 32)
  75                 ia64_fc(start + l);
  76
  77         ia64_sync_i();
  78         ia64_srlz_i();
  79 }
  80
  81 static void kvm_flush_tlb_all(void)
  82 {
  83         unsigned long i, j, count0, count1, stride0, stride1, addr;
  84         long flags;
  85
  86         addr    = local_cpu_data->ptce_base;
  87         count0  = local_cpu_data->ptce_count[0];
  88         count1  = local_cpu_data->ptce_count[1];
  89         stride0 = local_cpu_data->ptce_stride[0];
  90         stride1 = local_cpu_data->ptce_stride[1];
  91
  92         local_irq_save(flags);
  93         for (i = 0; i < count0; ++i) {
  94                 for (j = 0; j < count1; ++j) {
  95                         ia64_ptce(addr);
  96                         addr += stride1;
  97                 }
  98                 addr += stride0;
  99         }
 100         local_irq_restore(flags);
 101         ia64_srlz_i();                  /* srlz.i implies srlz.d */
 102 }
 103
 104 long ia64_pal_vp_create(u64 *vpd, u64 *host_iva, u64 *opt_handler)
 105 {
 106         struct ia64_pal_retval iprv;
 107
 108         PAL_CALL_STK(iprv, PAL_VP_CREATE, (u64)vpd, (u64)host_iva,
 109                         (u64)opt_handler);
 110
 111         return iprv.status;
 112 }
 113
 114 static  DEFINE_SPINLOCK(vp_lock);
 115
 116 void kvm_arch_hardware_enable(void *garbage)
 117 {
 118         long  status;
 119         long  tmp_base;
 120         unsigned long pte;
 121         unsigned long saved_psr;
 122         int slot;
 123
 124         pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base),
 125                                 PAGE_KERNEL));
 126         local_irq_save(saved_psr);
 127         slot = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
 128         if (slot < 0)
 129                 return;
 130         local_irq_restore(saved_psr);
 131
 132         spin_lock(&vp_lock);
 133         status = ia64_pal_vp_init_env(kvm_vsa_base ?
 134                                 VP_INIT_ENV : VP_INIT_ENV_INITALIZE,
 135                         __pa(kvm_vm_buffer), KVM_VM_BUFFER_BASE, &tmp_base);
 136         if (status != 0) {
 137                 printk(KERN_WARNING"kvm: Failed to Enable VT Support!!!!\n");
 138                 return ;
 139         }
 140
 141         if (!kvm_vsa_base) {
 142                 kvm_vsa_base = tmp_base;
 143                 printk(KERN_INFO"kvm: kvm_vsa_base:0x%lx\n", kvm_vsa_base);
 144         }
 145         spin_unlock(&vp_lock);
 146         ia64_ptr_entry(0x3, slot);
 147 }
 148
 149 void kvm_arch_hardware_disable(void *garbage)
 150 {
 151
 152         long status;
 153         int slot;
 154         unsigned long pte;
 155         unsigned long saved_psr;
 156         unsigned long host_iva = ia64_getreg(_IA64_REG_CR_IVA);
 157
 158         pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base),
 159                                 PAGE_KERNEL));
 160
 161         local_irq_save(saved_psr);
 162         slot = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
 163         if (slot < 0)
 164                 return;
 165         local_irq_restore(saved_psr);
 166
 167         status = ia64_pal_vp_exit_env(host_iva);
 168         if (status)
 169                 printk(KERN_DEBUG"kvm: Failed to disable VT support! :%ld\n",
 170                                 status);
 171         ia64_ptr_entry(0x3, slot);
 172 }
 173
 174 void kvm_arch_check_processor_compat(void *rtn)
 175 {
 176         *(int *)rtn = 0;
 177 }
 178
 179 int kvm_dev_ioctl_check_extension(long ext)
 180 {
 181
 182         int r;
 183
 184         switch (ext) {
 185         case KVM_CAP_IRQCHIP:
 186         case KVM_CAP_USER_MEMORY:
 187
 188                 r = 1;
 189                 break;
 190         default:
 191                 r = 0;
 192         }
 193         return r;
 194
 195 }
 196
 197 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
 198                                         gpa_t addr)
 199 {
 200         struct kvm_io_device *dev;
 201
 202         dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
 203
 204         return dev;
 205 }
 206
 207 static int handle_vm_error(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 208 {
 209         kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
 210         kvm_run->hw.hardware_exit_reason = 1;
 211         return 0;
 212 }
 213
 214 static int handle_mmio(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 215 {
 216         struct kvm_mmio_req *p;
 217         struct kvm_io_device *mmio_dev;
 218
 219         p = kvm_get_vcpu_ioreq(vcpu);
 220
 221         if ((p->addr & PAGE_MASK) == IOAPIC_DEFAULT_BASE_ADDRESS)
 222                 goto mmio;
 223         vcpu->mmio_needed = 1;
 224         vcpu->mmio_phys_addr = kvm_run->mmio.phys_addr = p->addr;
 225         vcpu->mmio_size = kvm_run->mmio.len = p->size;
 226         vcpu->mmio_is_write = kvm_run->mmio.is_write = !p->dir;
 227
 228         if (vcpu->mmio_is_write)
 229                 memcpy(vcpu->mmio_data, &p->data, p->size);
 230         memcpy(kvm_run->mmio.data, &p->data, p->size);
 231         kvm_run->exit_reason = KVM_EXIT_MMIO;
 232         return 0;
 233 mmio:
 234         mmio_dev = vcpu_find_mmio_dev(vcpu, p->addr);
 235         if (mmio_dev) {
 236                 if (!p->dir)
 237                         kvm_iodevice_write(mmio_dev, p->addr, p->size,
 238                                                 &p->data);
 239                 else
 240                         kvm_iodevice_read(mmio_dev, p->addr, p->size,
 241                                                 &p->data);
 242
 243         } else
 244                 printk(KERN_ERR"kvm: No iodevice found! addr:%lx\n", p->addr);
 245         p->state = STATE_IORESP_READY;
 246
 247         return 1;
 248 }
 249
 250 static int handle_pal_call(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 251 {
 252         struct exit_ctl_data *p;
 253
 254         p = kvm_get_exit_data(vcpu);
 255
 256         if (p->exit_reason == EXIT_REASON_PAL_CALL)
 257                 return kvm_pal_emul(vcpu, kvm_run);
 258         else {
 259                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
 260                 kvm_run->hw.hardware_exit_reason = 2;
 261                 return 0;
 262         }
 263 }
 264
 265 static int handle_sal_call(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 266 {
 267         struct exit_ctl_data *p;
 268
 269         p = kvm_get_exit_data(vcpu);
 270
 271         if (p->exit_reason == EXIT_REASON_SAL_CALL) {
 272                 kvm_sal_emul(vcpu);
 273                 return 1;
 274         } else {
 275                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
 276                 kvm_run->hw.hardware_exit_reason = 3;
 277                 return 0;
 278         }
 279
 280 }
 281
 282 /*
 283  *  offset: address offset to IPI space.
 284  *  value:  deliver value.
 285  */
 286 static void vcpu_deliver_ipi(struct kvm_vcpu *vcpu, uint64_t dm,
 287                                 uint64_t vector)
 288 {
 289         switch (dm) {
 290         case SAPIC_FIXED:
 291                 kvm_apic_set_irq(vcpu, vector, 0);
 292                 break;
 293         case SAPIC_NMI:
 294                 kvm_apic_set_irq(vcpu, 2, 0);
 295                 break;
 296         case SAPIC_EXTINT:
 297                 kvm_apic_set_irq(vcpu, 0, 0);
 298                 break;
 299         case SAPIC_INIT:
 300         case SAPIC_PMI:
 301         default:
 302                 printk(KERN_ERR"kvm: Unimplemented Deliver reserved IPI!\n");
 303                 break;
 304         }
 305 }
 306
 307 static struct kvm_vcpu *lid_to_vcpu(struct kvm *kvm, unsigned long id,
 308                         unsigned long eid)
 309 {
 310         union ia64_lid lid;
 311         int i;
 312
 313         for (i = 0; i < KVM_MAX_VCPUS; i++) {
 314                 if (kvm->vcpus[i]) {
 315                         lid.val = VCPU_LID(kvm->vcpus[i]);
 316                         if (lid.id == id && lid.eid == eid)
 317                                 return kvm->vcpus[i];
 318                 }
 319         }
 320
 321         return NULL;
 322 }
 323
 324 static int handle_ipi(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 325 {
 326         struct exit_ctl_data *p = kvm_get_exit_data(vcpu);
 327         struct kvm_vcpu *target_vcpu;
 328         struct kvm_pt_regs *regs;
 329         union ia64_ipi_a addr = p->u.ipi_data.addr;
 330         union ia64_ipi_d data = p->u.ipi_data.data;
 331
 332         target_vcpu = lid_to_vcpu(vcpu->kvm, addr.id, addr.eid);
 333         if (!target_vcpu)
 334                 return handle_vm_error(vcpu, kvm_run);
 335
 336         if (!target_vcpu->arch.launched) {
 337                 regs = vcpu_regs(target_vcpu);
 338
 339                 regs->cr_iip = vcpu->kvm->arch.rdv_sal_data.boot_ip;
 340                 regs->r1 = vcpu->kvm->arch.rdv_sal_data.boot_gp;
 341
 342                 target_vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
 343                 if (waitqueue_active(&target_vcpu->wq))
 344                         wake_up_interruptible(&target_vcpu->wq);
 345         } else {
 346                 vcpu_deliver_ipi(target_vcpu, data.dm, data.vector);
 347                 if (target_vcpu != vcpu)
 348                         kvm_vcpu_kick(target_vcpu);
 349         }
 350
 351         return 1;
 352 }
 353
 354 struct call_data {
 355         struct kvm_ptc_g ptc_g_data;
 356         struct kvm_vcpu *vcpu;
 357 };
 358
 359 static void vcpu_global_purge(void *info)
 360 {
 361         struct call_data *p = (struct call_data *)info;
 362         struct kvm_vcpu *vcpu = p->vcpu;
 363
 364         if (test_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 365                 return;
 366
 367         set_bit(KVM_REQ_PTC_G, &vcpu->requests);
 368         if (vcpu->arch.ptc_g_count < MAX_PTC_G_NUM) {
 369                 vcpu->arch.ptc_g_data[vcpu->arch.ptc_g_count++] =
 370                                                         p->ptc_g_data;
 371         } else {
 372                 clear_bit(KVM_REQ_PTC_G, &vcpu->requests);
 373                 vcpu->arch.ptc_g_count = 0;
 374                 set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests);
 375         }
 376 }
 377
 378 static int handle_global_purge(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 379 {
 380         struct exit_ctl_data *p = kvm_get_exit_data(vcpu);
 381         struct kvm *kvm = vcpu->kvm;
 382         struct call_data call_data;
 383         int i;
 384         call_data.ptc_g_data = p->u.ptc_g_data;
 385
 386         for (i = 0; i < KVM_MAX_VCPUS; i++) {
 387                 if (!kvm->vcpus[i] || kvm->vcpus[i]->arch.mp_state ==
 388                                                 KVM_MP_STATE_UNINITIALIZED ||
 389                                         vcpu == kvm->vcpus[i])
 390                         continue;
 391
 392                 if (waitqueue_active(&kvm->vcpus[i]->wq))
 393                         wake_up_interruptible(&kvm->vcpus[i]->wq);
 394
 395                 if (kvm->vcpus[i]->cpu != -1) {
 396                         call_data.vcpu = kvm->vcpus[i];
 397                         smp_call_function_single(kvm->vcpus[i]->cpu,
 398                                         vcpu_global_purge, &call_data, 1);
 399                 } else
 400                         printk(KERN_WARNING"kvm: Uninit vcpu received ipi!\n");
 401
 402         }
 403         return 1;
 404 }
 405
 406 static int handle_switch_rr6(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 407 {
 408         return 1;
 409 }
 410
 411 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
 412 {
 413
 414         ktime_t kt;
 415         long itc_diff;
 416         unsigned long vcpu_now_itc;
 417
 418         unsigned long expires;
 419         struct hrtimer *p_ht = &vcpu->arch.hlt_timer;
 420         unsigned long cyc_per_usec = local_cpu_data->cyc_per_usec;
 421         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
 422
 423         vcpu_now_itc = ia64_getreg(_IA64_REG_AR_ITC) + vcpu->arch.itc_offset;
 424
 425         if (time_after(vcpu_now_itc, vpd->itm)) {
 426                 vcpu->arch.timer_check = 1;
 427                 return 1;
 428         }
 429         itc_diff = vpd->itm - vcpu_now_itc;
 430         if (itc_diff < 0)
 431                 itc_diff = -itc_diff;
 432
 433         expires = div64_u64(itc_diff, cyc_per_usec);
 434         kt = ktime_set(0, 1000 * expires);
 435         vcpu->arch.ht_active = 1;
 436         hrtimer_start(p_ht, kt, HRTIMER_MODE_ABS);
 437
 438         if (irqchip_in_kernel(vcpu->kvm)) {
 439                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
 440                 kvm_vcpu_block(vcpu);
 441                 hrtimer_cancel(p_ht);
 442                 vcpu->arch.ht_active = 0;
 443
 444                 if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
 445                         return -EINTR;
 446                 return 1;
 447         } else {
 448                 printk(KERN_ERR"kvm: Unsupported userspace halt!");
 449                 return 0;
 450         }
 451 }
 452
 453 static int handle_vm_shutdown(struct kvm_vcpu *vcpu,
 454                 struct kvm_run *kvm_run)
 455 {
 456         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
 457         return 0;
 458 }
 459
 460 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
 461                 struct kvm_run *kvm_run)
 462 {
 463         return 1;
 464 }
 465
 466 static int (*kvm_vti_exit_handlers[])(struct kvm_vcpu *vcpu,
 467                 struct kvm_run *kvm_run) = {
 468         [EXIT_REASON_VM_PANIC]              = handle_vm_error,
 469         [EXIT_REASON_MMIO_INSTRUCTION]      = handle_mmio,
 470         [EXIT_REASON_PAL_CALL]              = handle_pal_call,
 471         [EXIT_REASON_SAL_CALL]              = handle_sal_call,
 472         [EXIT_REASON_SWITCH_RR6]            = handle_switch_rr6,
 473         [EXIT_REASON_VM_DESTROY]            = handle_vm_shutdown,
 474         [EXIT_REASON_EXTERNAL_INTERRUPT]    = handle_external_interrupt,
 475         [EXIT_REASON_IPI]                   = handle_ipi,
 476         [EXIT_REASON_PTC_G]                 = handle_global_purge,
 477
 478 };
 479
 480 static const int kvm_vti_max_exit_handlers =
 481                 sizeof(kvm_vti_exit_handlers)/sizeof(*kvm_vti_exit_handlers);
 482
 483 static void kvm_prepare_guest_switch(struct kvm_vcpu *vcpu)
 484 {
 485 }
 486
 487 static uint32_t kvm_get_exit_reason(struct kvm_vcpu *vcpu)
 488 {
 489         struct exit_ctl_data *p_exit_data;
 490
 491         p_exit_data = kvm_get_exit_data(vcpu);
 492         return p_exit_data->exit_reason;
 493 }
 494
 495 /*
 496  * The guest has exited.  See if we can fix it or if we need userspace
 497  * assistance.
 498  */
 499 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
 500 {
 501         u32 exit_reason = kvm_get_exit_reason(vcpu);
 502         vcpu->arch.last_exit = exit_reason;
 503
 504         if (exit_reason < kvm_vti_max_exit_handlers
 505                         && kvm_vti_exit_handlers[exit_reason])
 506                 return kvm_vti_exit_handlers[exit_reason](vcpu, kvm_run);
 507         else {
 508                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
 509                 kvm_run->hw.hardware_exit_reason = exit_reason;
 510         }
 511         return 0;
 512 }
 513
 514 static inline void vti_set_rr6(unsigned long rr6)
 515 {
 516         ia64_set_rr(RR6, rr6);
 517         ia64_srlz_i();
 518 }
 519
 520 static int kvm_insert_vmm_mapping(struct kvm_vcpu *vcpu)
 521 {
 522         unsigned long pte;
 523         struct kvm *kvm = vcpu->kvm;
 524         int r;
 525
 526         /*Insert a pair of tr to map vmm*/
 527         pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base), PAGE_KERNEL));
 528         r = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
 529         if (r < 0)
 530                 goto out;
 531         vcpu->arch.vmm_tr_slot = r;
 532         /*Insert a pairt of tr to map data of vm*/
 533         pte = pte_val(mk_pte_phys(__pa(kvm->arch.vm_base), PAGE_KERNEL));
 534         r = ia64_itr_entry(0x3, KVM_VM_DATA_BASE,
 535                                         pte, KVM_VM_DATA_SHIFT);
 536         if (r < 0)
 537                 goto out;
 538         vcpu->arch.vm_tr_slot = r;
 539         r = 0;
 540 out:
 541         return r;
 542
 543 }
 544
 545 static void kvm_purge_vmm_mapping(struct kvm_vcpu *vcpu)
 546 {
 547
 548         ia64_ptr_entry(0x3, vcpu->arch.vmm_tr_slot);
 549         ia64_ptr_entry(0x3, vcpu->arch.vm_tr_slot);
 550
 551 }
 552
 553 static int kvm_vcpu_pre_transition(struct kvm_vcpu *vcpu)
 554 {
 555         int cpu = smp_processor_id();
 556
 557         if (vcpu->arch.last_run_cpu != cpu ||
 558                         per_cpu(last_vcpu, cpu) != vcpu) {
 559                 per_cpu(last_vcpu, cpu) = vcpu;
 560                 vcpu->arch.last_run_cpu = cpu;
 561                 kvm_flush_tlb_all();
 562         }
 563
 564         vcpu->arch.host_rr6 = ia64_get_rr(RR6);
 565         vti_set_rr6(vcpu->arch.vmm_rr);
 566         return kvm_insert_vmm_mapping(vcpu);
 567 }
 568 static void kvm_vcpu_post_transition(struct kvm_vcpu *vcpu)
 569 {
 570         kvm_purge_vmm_mapping(vcpu);
 571         vti_set_rr6(vcpu->arch.host_rr6);
 572 }
 573
 574 static int  vti_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 575 {
 576         union context *host_ctx, *guest_ctx;
 577         int r;
 578
 579         /*Get host and guest context with guest address space.*/
 580         host_ctx = kvm_get_host_context(vcpu);
 581         guest_ctx = kvm_get_guest_context(vcpu);
 582
 583         r = kvm_vcpu_pre_transition(vcpu);
 584         if (r < 0)
 585                 goto out;
 586         kvm_vmm_info->tramp_entry(host_ctx, guest_ctx);
 587         kvm_vcpu_post_transition(vcpu);
 588         r = 0;
 589 out:
 590         return r;
 591 }
 592
 593 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 594 {
 595         int r;
 596
 597 again:
 598         preempt_disable();
 599
 600         kvm_prepare_guest_switch(vcpu);
 601         local_irq_disable();
 602
 603         if (signal_pending(current)) {
 604                 local_irq_enable();
 605                 preempt_enable();
 606                 r = -EINTR;
 607                 kvm_run->exit_reason = KVM_EXIT_INTR;
 608                 goto out;
 609         }
 610
 611         vcpu->guest_mode = 1;
 612         kvm_guest_enter();
 613
 614         r = vti_vcpu_run(vcpu, kvm_run);
 615         if (r < 0) {
 616                 local_irq_enable();
 617                 preempt_enable();
 618                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
 619                 goto out;
 620         }
 621
 622         vcpu->arch.launched = 1;
 623         vcpu->guest_mode = 0;
 624         local_irq_enable();
 625
 626         /*
 627          * We must have an instruction between local_irq_enable() and
 628          * kvm_guest_exit(), so the timer interrupt isn't delayed by
 629          * the interrupt shadow.  The stat.exits increment will do nicely.
 630          * But we need to prevent reordering, hence this barrier():
 631          */
 632         barrier();
 633
 634         kvm_guest_exit();
 635
 636         preempt_enable();
 637
 638         r = kvm_handle_exit(kvm_run, vcpu);
 639
 640         if (r > 0) {
 641                 if (!need_resched())
 642                         goto again;
 643         }
 644
 645 out:
 646         if (r > 0) {
 647                 kvm_resched(vcpu);
 648                 goto again;
 649         }
 650
 651         return r;
 652 }
 653
 654 static void kvm_set_mmio_data(struct kvm_vcpu *vcpu)
 655 {
 656         struct kvm_mmio_req *p = kvm_get_vcpu_ioreq(vcpu);
 657
 658         if (!vcpu->mmio_is_write)
 659                 memcpy(&p->data, vcpu->mmio_data, 8);
 660         p->state = STATE_IORESP_READY;
 661 }
 662
 663 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 664 {
 665         int r;
 666         sigset_t sigsaved;
 667
 668         vcpu_load(vcpu);
 669
 670         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
 671                 kvm_vcpu_block(vcpu);
 672                 vcpu_put(vcpu);
 673                 return -EAGAIN;
 674         }
 675
 676         if (vcpu->sigset_active)
 677                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 678
 679         if (vcpu->mmio_needed) {
 680                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
 681                 kvm_set_mmio_data(vcpu);
 682                 vcpu->mmio_read_completed = 1;
 683                 vcpu->mmio_needed = 0;
 684         }
 685         r = __vcpu_run(vcpu, kvm_run);
 686
 687         if (vcpu->sigset_active)
 688                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 689
 690         vcpu_put(vcpu);
 691         return r;
 692 }
 693
 694 /*
 695  * Allocate 16M memory for every vm to hold its specific data.
 696  * Its memory map is defined in kvm_host.h.
 697  */
 698 static struct kvm *kvm_alloc_kvm(void)
 699 {
 700
 701         struct kvm *kvm;
 702         uint64_t  vm_base;
 703
 704         vm_base = __get_free_pages(GFP_KERNEL, get_order(KVM_VM_DATA_SIZE));
 705
 706         if (!vm_base)
 707                 return ERR_PTR(-ENOMEM);
 708         printk(KERN_DEBUG"kvm: VM data's base Address:0x%lx\n", vm_base);
 709
 710         /* Zero all pages before use! */
 711         memset((void *)vm_base, 0, KVM_VM_DATA_SIZE);
 712
 713         kvm = (struct kvm *)(vm_base + KVM_VM_OFS);
 714         kvm->arch.vm_base = vm_base;
 715
 716         return kvm;
 717 }
 718
 719 struct kvm_io_range {
 720         unsigned long start;
 721         unsigned long size;
 722         unsigned long type;
 723 };
 724
 725 static const struct kvm_io_range io_ranges[] = {
 726         {VGA_IO_START, VGA_IO_SIZE, GPFN_FRAME_BUFFER},
 727         {MMIO_START, MMIO_SIZE, GPFN_LOW_MMIO},
 728         {LEGACY_IO_START, LEGACY_IO_SIZE, GPFN_LEGACY_IO},
 729         {IO_SAPIC_START, IO_SAPIC_SIZE, GPFN_IOSAPIC},
 730         {PIB_START, PIB_SIZE, GPFN_PIB},
 731 };
 732
 733 static void kvm_build_io_pmt(struct kvm *kvm)
 734 {
 735         unsigned long i, j;
 736
 737         /* Mark I/O ranges */
 738         for (i = 0; i < (sizeof(io_ranges) / sizeof(struct kvm_io_range));
 739                                                         i++) {
 740                 for (j = io_ranges[i].start;
 741                                 j < io_ranges[i].start + io_ranges[i].size;
 742                                 j += PAGE_SIZE)
 743                         kvm_set_pmt_entry(kvm, j >> PAGE_SHIFT,
 744                                         io_ranges[i].type, 0);
 745         }
 746
 747 }
 748
 749 /*Use unused rids to virtualize guest rid.*/
 750 #define GUEST_PHYSICAL_RR0      0x1739
 751 #define GUEST_PHYSICAL_RR4      0x2739
 752 #define VMM_INIT_RR             0x1660
 753
 754 static void kvm_init_vm(struct kvm *kvm)
 755 {
 756         long vm_base;
 757
 758         BUG_ON(!kvm);
 759
 760         kvm->arch.metaphysical_rr0 = GUEST_PHYSICAL_RR0;
 761         kvm->arch.metaphysical_rr4 = GUEST_PHYSICAL_RR4;
 762         kvm->arch.vmm_init_rr = VMM_INIT_RR;
 763
 764         vm_base = kvm->arch.vm_base;
 765         if (vm_base) {
 766                 kvm->arch.vhpt_base = vm_base + KVM_VHPT_OFS;
 767                 kvm->arch.vtlb_base = vm_base + KVM_VTLB_OFS;
 768                 kvm->arch.vpd_base  = vm_base + KVM_VPD_OFS;
 769         }
 770
 771         /*
 772          *Fill P2M entries for MMIO/IO ranges
 773          */
 774         kvm_build_io_pmt(kvm);
 775
 776 }
 777
 778 struct  kvm *kvm_arch_create_vm(void)
 779 {
 780         struct kvm *kvm = kvm_alloc_kvm();
 781
 782         if (IS_ERR(kvm))
 783                 return ERR_PTR(-ENOMEM);
 784         kvm_init_vm(kvm);
 785
 786         return kvm;
 787
 788 }
 789
 790 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm,
 791                                         struct kvm_irqchip *chip)
 792 {
 793         int r;
 794
 795         r = 0;
 796         switch (chip->chip_id) {
 797         case KVM_IRQCHIP_IOAPIC:
 798                 memcpy(&chip->chip.ioapic, ioapic_irqchip(kvm),
 799                                 sizeof(struct kvm_ioapic_state));
 800                 break;
 801         default:
 802                 r = -EINVAL;
 803                 break;
 804         }
 805         return r;
 806 }
 807
 808 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 809 {
 810         int r;
 811
 812         r = 0;
 813         switch (chip->chip_id) {
 814         case KVM_IRQCHIP_IOAPIC:
 815                 memcpy(ioapic_irqchip(kvm),
 816                                 &chip->chip.ioapic,
 817                                 sizeof(struct kvm_ioapic_state));
 818                 break;
 819         default:
 820                 r = -EINVAL;
 821                 break;
 822         }
 823         return r;
 824 }
 825
 826 #define RESTORE_REGS(_x) vcpu->arch._x = regs->_x
 827
 828 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 829 {
 830         int i;
 831         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
 832         int r;
 833
 834         vcpu_load(vcpu);
 835
 836         for (i = 0; i < 16; i++) {
 837                 vpd->vgr[i] = regs->vpd.vgr[i];
 838                 vpd->vbgr[i] = regs->vpd.vbgr[i];
 839         }
 840         for (i = 0; i < 128; i++)
 841                 vpd->vcr[i] = regs->vpd.vcr[i];
 842         vpd->vhpi = regs->vpd.vhpi;
 843         vpd->vnat = regs->vpd.vnat;
 844         vpd->vbnat = regs->vpd.vbnat;
 845         vpd->vpsr = regs->vpd.vpsr;
 846
 847         vpd->vpr = regs->vpd.vpr;
 848
 849         r = -EFAULT;
 850         r = copy_from_user(&vcpu->arch.guest, regs->saved_guest,
 851                                                 sizeof(union context));
 852         if (r)
 853                 goto out;
 854         r = copy_from_user(vcpu + 1, regs->saved_stack +
 855                         sizeof(struct kvm_vcpu),
 856                         IA64_STK_OFFSET - sizeof(struct kvm_vcpu));
 857         if (r)
 858                 goto out;
 859         vcpu->arch.exit_data =
 860                 ((struct kvm_vcpu *)(regs->saved_stack))->arch.exit_data;
 861
 862         RESTORE_REGS(mp_state);
 863         RESTORE_REGS(vmm_rr);
 864         memcpy(vcpu->arch.itrs, regs->itrs, sizeof(struct thash_data) * NITRS);
 865         memcpy(vcpu->arch.dtrs, regs->dtrs, sizeof(struct thash_data) * NDTRS);
 866         RESTORE_REGS(itr_regions);
 867         RESTORE_REGS(dtr_regions);
 868         RESTORE_REGS(tc_regions);
 869         RESTORE_REGS(irq_check);
 870         RESTORE_REGS(itc_check);
 871         RESTORE_REGS(timer_check);
 872         RESTORE_REGS(timer_pending);
 873         RESTORE_REGS(last_itc);
 874         for (i = 0; i < 8; i++) {
 875                 vcpu->arch.vrr[i] = regs->vrr[i];
 876                 vcpu->arch.ibr[i] = regs->ibr[i];
 877                 vcpu->arch.dbr[i] = regs->dbr[i];
 878         }
 879         for (i = 0; i < 4; i++)
 880                 vcpu->arch.insvc[i] = regs->insvc[i];
 881         RESTORE_REGS(xtp);
 882         RESTORE_REGS(metaphysical_rr0);
 883         RESTORE_REGS(metaphysical_rr4);
 884         RESTORE_REGS(metaphysical_saved_rr0);
 885         RESTORE_REGS(metaphysical_saved_rr4);
 886         RESTORE_REGS(fp_psr);
 887         RESTORE_REGS(saved_gp);
 888
 889         vcpu->arch.irq_new_pending = 1;
 890         vcpu->arch.itc_offset = regs->saved_itc - ia64_getreg(_IA64_REG_AR_ITC);
 891         set_bit(KVM_REQ_RESUME, &vcpu->requests);
 892
 893         vcpu_put(vcpu);
 894         r = 0;
 895 out:
 896         return r;
 897 }
 898
 899 long kvm_arch_vm_ioctl(struct file *filp,
 900                 unsigned int ioctl, unsigned long arg)
 901 {
 902         struct kvm *kvm = filp->private_data;
 903         void __user *argp = (void __user *)arg;
 904         int r = -EINVAL;
 905
 906         switch (ioctl) {
 907         case KVM_SET_MEMORY_REGION: {
 908                 struct kvm_memory_region kvm_mem;
 909                 struct kvm_userspace_memory_region kvm_userspace_mem;
 910
 911                 r = -EFAULT;
 912                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
 913                         goto out;
 914                 kvm_userspace_mem.slot = kvm_mem.slot;
 915                 kvm_userspace_mem.flags = kvm_mem.flags;
 916                 kvm_userspace_mem.guest_phys_addr =
 917                                         kvm_mem.guest_phys_addr;
 918                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
 919                 r = kvm_vm_ioctl_set_memory_region(kvm,
 920                                         &kvm_userspace_mem, 0);
 921                 if (r)
 922                         goto out;
 923                 break;
 924                 }
 925         case KVM_CREATE_IRQCHIP:
 926                 r = -EFAULT;
 927                 r = kvm_ioapic_init(kvm);
 928                 if (r)
 929                         goto out;
 930                 break;
 931         case KVM_IRQ_LINE: {
 932                 struct kvm_irq_level irq_event;
 933
 934                 r = -EFAULT;
 935                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
 936                         goto out;
 937                 if (irqchip_in_kernel(kvm)) {
 938                         mutex_lock(&kvm->lock);
 939                         kvm_ioapic_set_irq(kvm->arch.vioapic,
 940                                                 irq_event.irq,
 941                                                 irq_event.level);
 942                         mutex_unlock(&kvm->lock);
 943                         r = 0;
 944                 }
 945                 break;
 946                 }
 947         case KVM_GET_IRQCHIP: {
 948                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
 949                 struct kvm_irqchip chip;
 950
 951                 r = -EFAULT;
 952                 if (copy_from_user(&chip, argp, sizeof chip))
 953                                 goto out;
 954                 r = -ENXIO;
 955                 if (!irqchip_in_kernel(kvm))
 956                         goto out;
 957                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
 958                 if (r)
 959                         goto out;
 960                 r = -EFAULT;
 961                 if (copy_to_user(argp, &chip, sizeof chip))
 962                                 goto out;
 963                 r = 0;
 964                 break;
 965                 }
 966         case KVM_SET_IRQCHIP: {
 967                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
 968                 struct kvm_irqchip chip;
 969
 970                 r = -EFAULT;
 971                 if (copy_from_user(&chip, argp, sizeof chip))
 972                                 goto out;
 973                 r = -ENXIO;
 974                 if (!irqchip_in_kernel(kvm))
 975                         goto out;
 976                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
 977                 if (r)
 978                         goto out;
 979                 r = 0;
 980                 break;
 981                 }
 982         default:
 983                 ;
 984         }
 985 out:
 986         return r;
 987 }
 988
 989 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 990                 struct kvm_sregs *sregs)
 991 {
 992         return -EINVAL;
 993 }
 994
 995 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 996                 struct kvm_sregs *sregs)
 997 {
 998         return -EINVAL;
 999
1000 }
1001 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1002                 struct kvm_translation *tr)
1003 {
1004
1005         return -EINVAL;
1006 }
1007
1008 static int kvm_alloc_vmm_area(void)
1009 {
1010         if (!kvm_vmm_base && (kvm_vm_buffer_size < KVM_VM_BUFFER_SIZE)) {
1011                 kvm_vmm_base = __get_free_pages(GFP_KERNEL,
1012                                 get_order(KVM_VMM_SIZE));
1013                 if (!kvm_vmm_base)
1014                         return -ENOMEM;
1015
1016                 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1017                 kvm_vm_buffer = kvm_vmm_base + VMM_SIZE;
1018
1019                 printk(KERN_DEBUG"kvm:VMM's Base Addr:0x%lx, vm_buffer:0x%lx\n",
1020                                 kvm_vmm_base, kvm_vm_buffer);
1021         }
1022
1023         return 0;
1024 }
1025
1026 static void kvm_free_vmm_area(void)
1027 {
1028         if (kvm_vmm_base) {
1029                 /*Zero this area before free to avoid bits leak!!*/
1030                 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1031                 free_pages(kvm_vmm_base, get_order(KVM_VMM_SIZE));
1032                 kvm_vmm_base  = 0;
1033                 kvm_vm_buffer = 0;
1034                 kvm_vsa_base = 0;
1035         }
1036 }
1037
1038 /*
1039  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
1040  * cached on it. Leave it as blank for IA64.
1041  */
1042 void decache_vcpus_on_cpu(int cpu)
1043 {
1044 }
1045
1046 static void vti_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1047 {
1048 }
1049
1050 static int vti_init_vpd(struct kvm_vcpu *vcpu)
1051 {
1052         int i;
1053         union cpuid3_t cpuid3;
1054         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1055
1056         if (IS_ERR(vpd))
1057                 return PTR_ERR(vpd);
1058
1059         /* CPUID init */
1060         for (i = 0; i < 5; i++)
1061                 vpd->vcpuid[i] = ia64_get_cpuid(i);
1062
1063         /* Limit the CPUID number to 5 */
1064         cpuid3.value = vpd->vcpuid[3];
1065         cpuid3.number = 4;      /* 5 - 1 */
1066         vpd->vcpuid[3] = cpuid3.value;
1067
1068         /*Set vac and vdc fields*/
1069         vpd->vac.a_from_int_cr = 1;
1070         vpd->vac.a_to_int_cr = 1;
1071         vpd->vac.a_from_psr = 1;
1072         vpd->vac.a_from_cpuid = 1;
1073         vpd->vac.a_cover = 1;
1074         vpd->vac.a_bsw = 1;
1075         vpd->vac.a_int = 1;
1076         vpd->vdc.d_vmsw = 1;
1077
1078         /*Set virtual buffer*/
1079         vpd->virt_env_vaddr = KVM_VM_BUFFER_BASE;
1080
1081         return 0;
1082 }
1083
1084 static int vti_create_vp(struct kvm_vcpu *vcpu)
1085 {
1086         long ret;
1087         struct vpd *vpd = vcpu->arch.vpd;
1088         unsigned long  vmm_ivt;
1089
1090         vmm_ivt = kvm_vmm_info->vmm_ivt;
1091
1092         printk(KERN_DEBUG "kvm: vcpu:%p,ivt: 0x%lx\n", vcpu, vmm_ivt);
1093
1094         ret = ia64_pal_vp_create((u64 *)vpd, (u64 *)vmm_ivt, 0);
1095
1096         if (ret) {
1097                 printk(KERN_ERR"kvm: ia64_pal_vp_create failed!\n");
1098                 return -EINVAL;
1099         }
1100         return 0;
1101 }
1102
1103 static void init_ptce_info(struct kvm_vcpu *vcpu)
1104 {
1105         ia64_ptce_info_t ptce = {0};
1106
1107         ia64_get_ptce(&ptce);
1108         vcpu->arch.ptce_base = ptce.base;
1109         vcpu->arch.ptce_count[0] = ptce.count[0];
1110         vcpu->arch.ptce_count[1] = ptce.count[1];
1111         vcpu->arch.ptce_stride[0] = ptce.stride[0];
1112         vcpu->arch.ptce_stride[1] = ptce.stride[1];
1113 }
1114
1115 static void kvm_migrate_hlt_timer(struct kvm_vcpu *vcpu)
1116 {
1117         struct hrtimer *p_ht = &vcpu->arch.hlt_timer;
1118
1119         if (hrtimer_cancel(p_ht))
1120                 hrtimer_start(p_ht, p_ht->expires, HRTIMER_MODE_ABS);
1121 }
1122
1123 static enum hrtimer_restart hlt_timer_fn(struct hrtimer *data)
1124 {
1125         struct kvm_vcpu *vcpu;
1126         wait_queue_head_t *q;
1127
1128         vcpu  = container_of(data, struct kvm_vcpu, arch.hlt_timer);
1129         if (vcpu->arch.mp_state != KVM_MP_STATE_HALTED)
1130                 goto out;
1131
1132         q = &vcpu->wq;
1133         if (waitqueue_active(q)) {
1134                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1135                 wake_up_interruptible(q);
1136         }
1137 out:
1138         vcpu->arch.timer_check = 1;
1139         return HRTIMER_NORESTART;
1140 }
1141
1142 #define PALE_RESET_ENTRY    0x80000000ffffffb0UL
1143
1144 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1145 {
1146         struct kvm_vcpu *v;
1147         int r;
1148         int i;
1149         long itc_offset;
1150         struct kvm *kvm = vcpu->kvm;
1151         struct kvm_pt_regs *regs = vcpu_regs(vcpu);
1152
1153         union context *p_ctx = &vcpu->arch.guest;
1154         struct kvm_vcpu *vmm_vcpu = to_guest(vcpu->kvm, vcpu);
1155
1156         /*Init vcpu context for first run.*/
1157         if (IS_ERR(vmm_vcpu))
1158                 return PTR_ERR(vmm_vcpu);
1159
1160         if (vcpu->vcpu_id == 0) {
1161                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1162
1163                 /*Set entry address for first run.*/
1164                 regs->cr_iip = PALE_RESET_ENTRY;
1165
1166                 /*Initilize itc offset for vcpus*/
1167                 itc_offset = 0UL - ia64_getreg(_IA64_REG_AR_ITC);
1168                 for (i = 0; i < MAX_VCPU_NUM; i++) {
1169                         v = (struct kvm_vcpu *)((char *)vcpu + VCPU_SIZE * i);
1170                         v->arch.itc_offset = itc_offset;
1171                         v->arch.last_itc = 0;
1172                 }
1173         } else
1174                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
1175
1176         r = -ENOMEM;
1177         vcpu->arch.apic = kzalloc(sizeof(struct kvm_lapic), GFP_KERNEL);
1178         if (!vcpu->arch.apic)
1179                 goto out;
1180         vcpu->arch.apic->vcpu = vcpu;
1181
1182         p_ctx->gr[1] = 0;
1183         p_ctx->gr[12] = (unsigned long)((char *)vmm_vcpu + IA64_STK_OFFSET);
1184         p_ctx->gr[13] = (unsigned long)vmm_vcpu;
1185         p_ctx->psr = 0x1008522000UL;
1186         p_ctx->ar[40] = FPSR_DEFAULT; /*fpsr*/
1187         p_ctx->caller_unat = 0;
1188         p_ctx->pr = 0x0;
1189         p_ctx->ar[36] = 0x0; /*unat*/
1190         p_ctx->ar[19] = 0x0; /*rnat*/
1191         p_ctx->ar[18] = (unsigned long)vmm_vcpu +
1192                                 ((sizeof(struct kvm_vcpu)+15) & ~15);
1193         p_ctx->ar[64] = 0x0; /*pfs*/
1194         p_ctx->cr[0] = 0x7e04UL;
1195         p_ctx->cr[2] = (unsigned long)kvm_vmm_info->vmm_ivt;
1196         p_ctx->cr[8] = 0x3c;
1197
1198         /*Initilize region register*/
1199         p_ctx->rr[0] = 0x30;
1200         p_ctx->rr[1] = 0x30;
1201         p_ctx->rr[2] = 0x30;
1202         p_ctx->rr[3] = 0x30;
1203         p_ctx->rr[4] = 0x30;
1204         p_ctx->rr[5] = 0x30;
1205         p_ctx->rr[7] = 0x30;
1206
1207         /*Initilize branch register 0*/
1208         p_ctx->br[0] = *(unsigned long *)kvm_vmm_info->vmm_entry;
1209
1210         vcpu->arch.vmm_rr = kvm->arch.vmm_init_rr;
1211         vcpu->arch.metaphysical_rr0 = kvm->arch.metaphysical_rr0;
1212         vcpu->arch.metaphysical_rr4 = kvm->arch.metaphysical_rr4;
1213
1214         hrtimer_init(&vcpu->arch.hlt_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1215         vcpu->arch.hlt_timer.function = hlt_timer_fn;
1216
1217         vcpu->arch.last_run_cpu = -1;
1218         vcpu->arch.vpd = (struct vpd *)VPD_ADDR(vcpu->vcpu_id);
1219         vcpu->arch.vsa_base = kvm_vsa_base;
1220         vcpu->arch.__gp = kvm_vmm_gp;
1221         vcpu->arch.dirty_log_lock_pa = __pa(&kvm->arch.dirty_log_lock);
1222         vcpu->arch.vhpt.hash = (struct thash_data *)VHPT_ADDR(vcpu->vcpu_id);
1223         vcpu->arch.vtlb.hash = (struct thash_data *)VTLB_ADDR(vcpu->vcpu_id);
1224         init_ptce_info(vcpu);
1225
1226         r = 0;
1227 out:
1228         return r;
1229 }
1230
1231 static int vti_vcpu_setup(struct kvm_vcpu *vcpu, int id)
1232 {
1233         unsigned long psr;
1234         int r;
1235
1236         local_irq_save(psr);
1237         r = kvm_insert_vmm_mapping(vcpu);
1238         if (r)
1239                 goto fail;
1240         r = kvm_vcpu_init(vcpu, vcpu->kvm, id);
1241         if (r)
1242                 goto fail;
1243
1244         r = vti_init_vpd(vcpu);
1245         if (r) {
1246                 printk(KERN_DEBUG"kvm: vpd init error!!\n");
1247                 goto uninit;
1248         }
1249
1250         r = vti_create_vp(vcpu);
1251         if (r)
1252                 goto uninit;
1253
1254         kvm_purge_vmm_mapping(vcpu);
1255         local_irq_restore(psr);
1256
1257         return 0;
1258 uninit:
1259         kvm_vcpu_uninit(vcpu);
1260 fail:
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         kfree(kvm->arch.vioapic);
1347         kvm_release_vm_pages(kvm);
1348         kvm_free_physmem(kvm);
1349         free_kvm(kvm);
1350 }
1351
1352 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1353 {
1354 }
1355
1356 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1357 {
1358         if (cpu != vcpu->cpu) {
1359                 vcpu->cpu = cpu;
1360                 if (vcpu->arch.ht_active)
1361                         kvm_migrate_hlt_timer(vcpu);
1362         }
1363 }
1364
1365 #define SAVE_REGS(_x)   regs->_x = vcpu->arch._x
1366
1367 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1368 {
1369         int i;
1370         int r;
1371         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1372         vcpu_load(vcpu);
1373
1374         for (i = 0; i < 16; i++) {
1375                 regs->vpd.vgr[i] = vpd->vgr[i];
1376                 regs->vpd.vbgr[i] = vpd->vbgr[i];
1377         }
1378         for (i = 0; i < 128; i++)
1379                 regs->vpd.vcr[i] = vpd->vcr[i];
1380         regs->vpd.vhpi = vpd->vhpi;
1381         regs->vpd.vnat = vpd->vnat;
1382         regs->vpd.vbnat = vpd->vbnat;
1383         regs->vpd.vpsr = vpd->vpsr;
1384         regs->vpd.vpr = vpd->vpr;
1385
1386         r = -EFAULT;
1387         r = copy_to_user(regs->saved_guest, &vcpu->arch.guest,
1388                                         sizeof(union context));
1389         if (r)
1390                 goto out;
1391         r = copy_to_user(regs->saved_stack, (void *)vcpu, IA64_STK_OFFSET);
1392         if (r)
1393                 goto out;
1394         SAVE_REGS(mp_state);
1395         SAVE_REGS(vmm_rr);
1396         memcpy(regs->itrs, vcpu->arch.itrs, sizeof(struct thash_data) * NITRS);
1397         memcpy(regs->dtrs, vcpu->arch.dtrs, sizeof(struct thash_data) * NDTRS);
1398         SAVE_REGS(itr_regions);
1399         SAVE_REGS(dtr_regions);
1400         SAVE_REGS(tc_regions);
1401         SAVE_REGS(irq_check);
1402         SAVE_REGS(itc_check);
1403         SAVE_REGS(timer_check);
1404         SAVE_REGS(timer_pending);
1405         SAVE_REGS(last_itc);
1406         for (i = 0; i < 8; i++) {
1407                 regs->vrr[i] = vcpu->arch.vrr[i];
1408                 regs->ibr[i] = vcpu->arch.ibr[i];
1409                 regs->dbr[i] = vcpu->arch.dbr[i];
1410         }
1411         for (i = 0; i < 4; i++)
1412                 regs->insvc[i] = vcpu->arch.insvc[i];
1413         regs->saved_itc = vcpu->arch.itc_offset + ia64_getreg(_IA64_REG_AR_ITC);
1414         SAVE_REGS(xtp);
1415         SAVE_REGS(metaphysical_rr0);
1416         SAVE_REGS(metaphysical_rr4);
1417         SAVE_REGS(metaphysical_saved_rr0);
1418         SAVE_REGS(metaphysical_saved_rr4);
1419         SAVE_REGS(fp_psr);
1420         SAVE_REGS(saved_gp);
1421         vcpu_put(vcpu);
1422         r = 0;
1423 out:
1424         return r;
1425 }
1426
1427 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
1428 {
1429
1430         hrtimer_cancel(&vcpu->arch.hlt_timer);
1431         kfree(vcpu->arch.apic);
1432 }
1433
1434
1435 long kvm_arch_vcpu_ioctl(struct file *filp,
1436                 unsigned int ioctl, unsigned long arg)
1437 {
1438         return -EINVAL;
1439 }
1440
1441 int kvm_arch_set_memory_region(struct kvm *kvm,
1442                 struct kvm_userspace_memory_region *mem,
1443                 struct kvm_memory_slot old,
1444                 int user_alloc)
1445 {
1446         unsigned long i;
1447         struct page *page;
1448         int npages = mem->memory_size >> PAGE_SHIFT;
1449         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
1450         unsigned long base_gfn = memslot->base_gfn;
1451
1452         for (i = 0; i < npages; i++) {
1453                 page = gfn_to_page(kvm, base_gfn + i);
1454                 kvm_set_pmt_entry(kvm, base_gfn + i,
1455                                 page_to_pfn(page) << PAGE_SHIFT,
1456                                 _PAGE_AR_RWX|_PAGE_MA_WB);
1457                 memslot->rmap[i] = (unsigned long)page;
1458         }
1459
1460         return 0;
1461 }
1462
1463
1464 long kvm_arch_dev_ioctl(struct file *filp,
1465                 unsigned int ioctl, unsigned long arg)
1466 {
1467         return -EINVAL;
1468 }
1469
1470 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1471 {
1472         kvm_vcpu_uninit(vcpu);
1473 }
1474
1475 static int vti_cpu_has_kvm_support(void)
1476 {
1477         long  avail = 1, status = 1, control = 1;
1478         long ret;
1479
1480         ret = ia64_pal_proc_get_features(&avail, &status, &control, 0);
1481         if (ret)
1482                 goto out;
1483
1484         if (!(avail & PAL_PROC_VM_BIT))
1485                 goto out;
1486
1487         printk(KERN_DEBUG"kvm: Hardware Supports VT\n");
1488
1489         ret = ia64_pal_vp_env_info(&kvm_vm_buffer_size, &vp_env_info);
1490         if (ret)
1491                 goto out;
1492         printk(KERN_DEBUG"kvm: VM Buffer Size:0x%lx\n", kvm_vm_buffer_size);
1493
1494         if (!(vp_env_info & VP_OPCODE)) {
1495                 printk(KERN_WARNING"kvm: No opcode ability on hardware, "
1496                                 "vm_env_info:0x%lx\n", vp_env_info);
1497         }
1498
1499         return 1;
1500 out:
1501         return 0;
1502 }
1503
1504 static int kvm_relocate_vmm(struct kvm_vmm_info *vmm_info,
1505                                                 struct module *module)
1506 {
1507         unsigned long module_base;
1508         unsigned long vmm_size;
1509
1510         unsigned long vmm_offset, func_offset, fdesc_offset;
1511         struct fdesc *p_fdesc;
1512
1513         BUG_ON(!module);
1514
1515         if (!kvm_vmm_base) {
1516                 printk("kvm: kvm area hasn't been initilized yet!!\n");
1517                 return -EFAULT;
1518         }
1519
1520         /*Calculate new position of relocated vmm module.*/
1521         module_base = (unsigned long)module->module_core;
1522         vmm_size = module->core_size;
1523         if (unlikely(vmm_size > KVM_VMM_SIZE))
1524                 return -EFAULT;
1525
1526         memcpy((void *)kvm_vmm_base, (void *)module_base, vmm_size);
1527         kvm_flush_icache(kvm_vmm_base, vmm_size);
1528
1529         /*Recalculate kvm_vmm_info based on new VMM*/
1530         vmm_offset = vmm_info->vmm_ivt - module_base;
1531         kvm_vmm_info->vmm_ivt = KVM_VMM_BASE + vmm_offset;
1532         printk(KERN_DEBUG"kvm: Relocated VMM's IVT Base Addr:%lx\n",
1533                         kvm_vmm_info->vmm_ivt);
1534
1535         fdesc_offset = (unsigned long)vmm_info->vmm_entry - module_base;
1536         kvm_vmm_info->vmm_entry = (kvm_vmm_entry *)(KVM_VMM_BASE +
1537                                                         fdesc_offset);
1538         func_offset = *(unsigned long *)vmm_info->vmm_entry - module_base;
1539         p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1540         p_fdesc->ip = KVM_VMM_BASE + func_offset;
1541         p_fdesc->gp = KVM_VMM_BASE+(p_fdesc->gp - module_base);
1542
1543         printk(KERN_DEBUG"kvm: Relocated VMM's Init Entry Addr:%lx\n",
1544                         KVM_VMM_BASE+func_offset);
1545
1546         fdesc_offset = (unsigned long)vmm_info->tramp_entry - module_base;
1547         kvm_vmm_info->tramp_entry = (kvm_tramp_entry *)(KVM_VMM_BASE +
1548                         fdesc_offset);
1549         func_offset = *(unsigned long *)vmm_info->tramp_entry - module_base;
1550         p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1551         p_fdesc->ip = KVM_VMM_BASE + func_offset;
1552         p_fdesc->gp = KVM_VMM_BASE + (p_fdesc->gp - module_base);
1553
1554         kvm_vmm_gp = p_fdesc->gp;
1555
1556         printk(KERN_DEBUG"kvm: Relocated VMM's Entry IP:%p\n",
1557                                                 kvm_vmm_info->vmm_entry);
1558         printk(KERN_DEBUG"kvm: Relocated VMM's Trampoline Entry IP:0x%lx\n",
1559                                                 KVM_VMM_BASE + func_offset);
1560
1561         return 0;
1562 }
1563
1564 int kvm_arch_init(void *opaque)
1565 {
1566         int r;
1567         struct kvm_vmm_info *vmm_info = (struct kvm_vmm_info *)opaque;
1568
1569         if (!vti_cpu_has_kvm_support()) {
1570                 printk(KERN_ERR "kvm: No Hardware Virtualization Support!\n");
1571                 r = -EOPNOTSUPP;
1572                 goto out;
1573         }
1574
1575         if (kvm_vmm_info) {
1576                 printk(KERN_ERR "kvm: Already loaded VMM module!\n");
1577                 r = -EEXIST;
1578                 goto out;
1579         }
1580
1581         r = -ENOMEM;
1582         kvm_vmm_info = kzalloc(sizeof(struct kvm_vmm_info), GFP_KERNEL);
1583         if (!kvm_vmm_info)
1584                 goto out;
1585
1586         if (kvm_alloc_vmm_area())
1587                 goto out_free0;
1588
1589         r = kvm_relocate_vmm(vmm_info, vmm_info->module);
1590         if (r)
1591                 goto out_free1;
1592
1593         return 0;
1594
1595 out_free1:
1596         kvm_free_vmm_area();
1597 out_free0:
1598         kfree(kvm_vmm_info);
1599 out:
1600         return r;
1601 }
1602
1603 void kvm_arch_exit(void)
1604 {
1605         kvm_free_vmm_area();
1606         kfree(kvm_vmm_info);
1607         kvm_vmm_info = NULL;
1608 }
1609
1610 static int kvm_ia64_sync_dirty_log(struct kvm *kvm,
1611                 struct kvm_dirty_log *log)
1612 {
1613         struct kvm_memory_slot *memslot;
1614         int r, i;
1615         long n, base;
1616         unsigned long *dirty_bitmap = (unsigned long *)((void *)kvm - KVM_VM_OFS
1617                                         + KVM_MEM_DIRTY_LOG_OFS);
1618
1619         r = -EINVAL;
1620         if (log->slot >= KVM_MEMORY_SLOTS)
1621                 goto out;
1622
1623         memslot = &kvm->memslots[log->slot];
1624         r = -ENOENT;
1625         if (!memslot->dirty_bitmap)
1626                 goto out;
1627
1628         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1629         base = memslot->base_gfn / BITS_PER_LONG;
1630
1631         for (i = 0; i < n/sizeof(long); ++i) {
1632                 memslot->dirty_bitmap[i] = dirty_bitmap[base + i];
1633                 dirty_bitmap[base + i] = 0;
1634         }
1635         r = 0;
1636 out:
1637         return r;
1638 }
1639
1640 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1641                 struct kvm_dirty_log *log)
1642 {
1643         int r;
1644         int n;
1645         struct kvm_memory_slot *memslot;
1646         int is_dirty = 0;
1647
1648         spin_lock(&kvm->arch.dirty_log_lock);
1649
1650         r = kvm_ia64_sync_dirty_log(kvm, log);
1651         if (r)
1652                 goto out;
1653
1654         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1655         if (r)
1656                 goto out;
1657
1658         /* If nothing is dirty, don't bother messing with page tables. */
1659         if (is_dirty) {
1660                 kvm_flush_remote_tlbs(kvm);
1661                 memslot = &kvm->memslots[log->slot];
1662                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1663                 memset(memslot->dirty_bitmap, 0, n);
1664         }
1665         r = 0;
1666 out:
1667         spin_unlock(&kvm->arch.dirty_log_lock);
1668         return r;
1669 }
1670
1671 int kvm_arch_hardware_setup(void)
1672 {
1673         return 0;
1674 }
1675
1676 void kvm_arch_hardware_unsetup(void)
1677 {
1678 }
1679
1680 static void vcpu_kick_intr(void *info)
1681 {
1682 #ifdef DEBUG
1683         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
1684         printk(KERN_DEBUG"vcpu_kick_intr %p \n", vcpu);
1685 #endif
1686 }
1687
1688 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1689 {
1690         int ipi_pcpu = vcpu->cpu;
1691
1692         if (waitqueue_active(&vcpu->wq))
1693                 wake_up_interruptible(&vcpu->wq);
1694
1695         if (vcpu->guest_mode)
1696                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
1697 }
1698
1699 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, u8 vec, u8 trig)
1700 {
1701
1702         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1703
1704         if (!test_and_set_bit(vec, &vpd->irr[0])) {
1705                 vcpu->arch.irq_new_pending = 1;
1706                  if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
1707                         kvm_vcpu_kick(vcpu);
1708                 else if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED) {
1709                         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1710                         if (waitqueue_active(&vcpu->wq))
1711                                 wake_up_interruptible(&vcpu->wq);
1712                 }
1713                 return 1;
1714         }
1715         return 0;
1716 }
1717
1718 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
1719 {
1720         return apic->vcpu->vcpu_id == dest;
1721 }
1722
1723 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
1724 {
1725         return 0;
1726 }
1727
1728 struct kvm_vcpu *kvm_get_lowest_prio_vcpu(struct kvm *kvm, u8 vector,
1729                                        unsigned long bitmap)
1730 {
1731         struct kvm_vcpu *lvcpu = kvm->vcpus[0];
1732         int i;
1733
1734         for (i = 1; i < KVM_MAX_VCPUS; i++) {
1735                 if (!kvm->vcpus[i])
1736                         continue;
1737                 if (lvcpu->arch.xtp > kvm->vcpus[i]->arch.xtp)
1738                         lvcpu = kvm->vcpus[i];
1739         }
1740
1741         return lvcpu;
1742 }
1743
1744 static int find_highest_bits(int *dat)
1745 {
1746         u32  bits, bitnum;
1747         int i;
1748
1749         /* loop for all 256 bits */
1750         for (i = 7; i >= 0 ; i--) {
1751                 bits = dat[i];
1752                 if (bits) {
1753                         bitnum = fls(bits);
1754                         return i * 32 + bitnum - 1;
1755                 }
1756         }
1757
1758         return -1;
1759 }
1760
1761 int kvm_highest_pending_irq(struct kvm_vcpu *vcpu)
1762 {
1763     struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1764
1765     if (vpd->irr[0] & (1UL << NMI_VECTOR))
1766                 return NMI_VECTOR;
1767     if (vpd->irr[0] & (1UL << ExtINT_VECTOR))
1768                 return ExtINT_VECTOR;
1769
1770     return find_highest_bits((int *)&vpd->irr[0]);
1771 }
1772
1773 int kvm_cpu_has_interrupt(struct kvm_vcpu *vcpu)
1774 {
1775         if (kvm_highest_pending_irq(vcpu) != -1)
1776                 return 1;
1777         return 0;
1778 }
1779
1780 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1781 {
1782         return 0;
1783 }
1784
1785 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1786 {
1787         return gfn;
1788 }
1789
1790 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
1791 {
1792         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE;
1793 }
1794
1795 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1796                                     struct kvm_mp_state *mp_state)
1797 {
1798         return -EINVAL;
1799 }
1800
1801 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1802                                     struct kvm_mp_state *mp_state)
1803 {
1804         return -EINVAL;
1805 }