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