arch/ia64/kvm/kvm-ia64.c

   1
   2 /*
   3  * kvm_ia64.c: Basic KVM suppport On Itanium series processors
   4  *
   5  *
   6  *      Copyright (C) 2007, Intel Corporation.
   7  *      Xiantao Zhang  (xiantao.zhang@intel.com)
   8  *
   9  * This program is free software; you can redistribute it and/or modify it
  10  * under the terms and conditions of the GNU General Public License,
  11  * version 2, as published by the Free Software Foundation.
  12  *
  13  * This program is distributed in the hope it will be useful, but WITHOUT
  14  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  15  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  16  * more details.
  17  *
  18  * You should have received a copy of the GNU General Public License along with
  19  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
  20  * Place - Suite 330, Boston, MA 02111-1307 USA.
  21  *
  22  */
  23
  24 #include <linux/module.h>
  25 #include <linux/errno.h>
  26 #include <linux/percpu.h>
  27 #include <linux/gfp.h>
  28 #include <linux/fs.h>
  29 #include <linux/smp.h>
  30 #include <linux/kvm_host.h>
  31 #include <linux/kvm.h>
  32 #include <linux/bitops.h>
  33 #include <linux/hrtimer.h>
  34 #include <linux/uaccess.h>
  35
  36 #include <asm/pgtable.h>
  37 #include <asm/gcc_intrin.h>
  38 #include <asm/pal.h>
  39 #include <asm/cacheflush.h>
  40 #include <asm/div64.h>
  41 #include <asm/tlb.h>
  42
  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
  66 struct fdesc{
  67     unsigned long ip;
  68     unsigned long gp;
  69 };
  70
  71 static void kvm_flush_icache(unsigned long start, unsigned long len)
  72 {
  73         int l;
  74
  75         for (l = 0; l < (len + 32); l += 32)
  76                 ia64_fc(start + l);
  77
  78         ia64_sync_i();
  79         ia64_srlz_i();
  80 }
  81
  82 static void kvm_flush_tlb_all(void)
  83 {
  84         unsigned long i, j, count0, count1, stride0, stride1, addr;
  85         long flags;
  86
  87         addr    = local_cpu_data->ptce_base;
  88         count0  = local_cpu_data->ptce_count[0];
  89         count1  = local_cpu_data->ptce_count[1];
  90         stride0 = local_cpu_data->ptce_stride[0];
  91         stride1 = local_cpu_data->ptce_stride[1];
  92
  93         local_irq_save(flags);
  94         for (i = 0; i < count0; ++i) {
  95                 for (j = 0; j < count1; ++j) {
  96                         ia64_ptce(addr);
  97                         addr += stride1;
  98                 }
  99                 addr += stride0;
 100         }
 101         local_irq_restore(flags);
 102         ia64_srlz_i();                  /* srlz.i implies srlz.d */
 103 }
 104
 105 long ia64_pal_vp_create(u64 *vpd, u64 *host_iva, u64 *opt_handler)
 106 {
 107         struct ia64_pal_retval iprv;
 108
 109         PAL_CALL_STK(iprv, PAL_VP_CREATE, (u64)vpd, (u64)host_iva,
 110                         (u64)opt_handler);
 111
 112         return iprv.status;
 113 }
 114
 115 static  DEFINE_SPINLOCK(vp_lock);
 116
 117 void kvm_arch_hardware_enable(void *garbage)
 118 {
 119         long  status;
 120         long  tmp_base;
 121         unsigned long pte;
 122         unsigned long saved_psr;
 123         int slot;
 124
 125         pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base),
 126                                 PAGE_KERNEL));
 127         local_irq_save(saved_psr);
 128         slot = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
 129         if (slot < 0)
 130                 return;
 131         local_irq_restore(saved_psr);
 132
 133         spin_lock(&vp_lock);
 134         status = ia64_pal_vp_init_env(kvm_vsa_base ?
 135                                 VP_INIT_ENV : VP_INIT_ENV_INITALIZE,
 136                         __pa(kvm_vm_buffer), KVM_VM_BUFFER_BASE, &tmp_base);
 137         if (status != 0) {
 138                 printk(KERN_WARNING"kvm: Failed to Enable VT Support!!!!\n");
 139                 return ;
 140         }
 141
 142         if (!kvm_vsa_base) {
 143                 kvm_vsa_base = tmp_base;
 144                 printk(KERN_INFO"kvm: kvm_vsa_base:0x%lx\n", kvm_vsa_base);
 145         }
 146         spin_unlock(&vp_lock);
 147         ia64_ptr_entry(0x3, slot);
 148 }
 149
 150 void kvm_arch_hardware_disable(void *garbage)
 151 {
 152
 153         long status;
 154         int slot;
 155         unsigned long pte;
 156         unsigned long saved_psr;
 157         unsigned long host_iva = ia64_getreg(_IA64_REG_CR_IVA);
 158
 159         pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base),
 160                                 PAGE_KERNEL));
 161
 162         local_irq_save(saved_psr);
 163         slot = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
 164         if (slot < 0)
 165                 return;
 166         local_irq_restore(saved_psr);
 167
 168         status = ia64_pal_vp_exit_env(host_iva);
 169         if (status)
 170                 printk(KERN_DEBUG"kvm: Failed to disable VT support! :%ld\n",
 171                                 status);
 172         ia64_ptr_entry(0x3, slot);
 173 }
 174
 175 void kvm_arch_check_processor_compat(void *rtn)
 176 {
 177         *(int *)rtn = 0;
 178 }
 179
 180 int kvm_dev_ioctl_check_extension(long ext)
 181 {
 182
 183         int r;
 184
 185         switch (ext) {
 186         case KVM_CAP_IRQCHIP:
 187         case KVM_CAP_USER_MEMORY:
 188
 189                 r = 1;
 190                 break;
 191         default:
 192                 r = 0;
 193         }
 194         return r;
 195
 196 }
 197
 198 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
 199                                         gpa_t addr)
 200 {
 201         struct kvm_io_device *dev;
 202
 203         dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
 204
 205         return dev;
 206 }
 207
 208 static int handle_vm_error(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 209 {
 210         kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
 211         kvm_run->hw.hardware_exit_reason = 1;
 212         return 0;
 213 }
 214
 215 static int handle_mmio(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 216 {
 217         struct kvm_mmio_req *p;
 218         struct kvm_io_device *mmio_dev;
 219
 220         p = kvm_get_vcpu_ioreq(vcpu);
 221
 222         if ((p->addr & PAGE_MASK) == IOAPIC_DEFAULT_BASE_ADDRESS)
 223                 goto mmio;
 224         vcpu->mmio_needed = 1;
 225         vcpu->mmio_phys_addr = kvm_run->mmio.phys_addr = p->addr;
 226         vcpu->mmio_size = kvm_run->mmio.len = p->size;
 227         vcpu->mmio_is_write = kvm_run->mmio.is_write = !p->dir;
 228
 229         if (vcpu->mmio_is_write)
 230                 memcpy(vcpu->mmio_data, &p->data, p->size);
 231         memcpy(kvm_run->mmio.data, &p->data, p->size);
 232         kvm_run->exit_reason = KVM_EXIT_MMIO;
 233         return 0;
 234 mmio:
 235         mmio_dev = vcpu_find_mmio_dev(vcpu, p->addr);
 236         if (mmio_dev) {
 237                 if (!p->dir)
 238                         kvm_iodevice_write(mmio_dev, p->addr, p->size,
 239                                                 &p->data);
 240                 else
 241                         kvm_iodevice_read(mmio_dev, p->addr, p->size,
 242                                                 &p->data);
 243
 244         } else
 245                 printk(KERN_ERR"kvm: No iodevice found! addr:%lx\n", p->addr);
 246         p->state = STATE_IORESP_READY;
 247
 248         return 1;
 249 }
 250
 251 static int handle_pal_call(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 252 {
 253         struct exit_ctl_data *p;
 254
 255         p = kvm_get_exit_data(vcpu);
 256
 257         if (p->exit_reason == EXIT_REASON_PAL_CALL)
 258                 return kvm_pal_emul(vcpu, kvm_run);
 259         else {
 260                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
 261                 kvm_run->hw.hardware_exit_reason = 2;
 262                 return 0;
 263         }
 264 }
 265
 266 static int handle_sal_call(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 267 {
 268         struct exit_ctl_data *p;
 269
 270         p = kvm_get_exit_data(vcpu);
 271
 272         if (p->exit_reason == EXIT_REASON_SAL_CALL) {
 273                 kvm_sal_emul(vcpu);
 274                 return 1;
 275         } else {
 276                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
 277                 kvm_run->hw.hardware_exit_reason = 3;
 278                 return 0;
 279         }
 280
 281 }
 282
 283 /*
 284  *  offset: address offset to IPI space.
 285  *  value:  deliver value.
 286  */
 287 static void vcpu_deliver_ipi(struct kvm_vcpu *vcpu, uint64_t dm,
 288                                 uint64_t vector)
 289 {
 290         switch (dm) {
 291         case SAPIC_FIXED:
 292                 kvm_apic_set_irq(vcpu, vector, 0);
 293                 break;
 294         case SAPIC_NMI:
 295                 kvm_apic_set_irq(vcpu, 2, 0);
 296                 break;
 297         case SAPIC_EXTINT:
 298                 kvm_apic_set_irq(vcpu, 0, 0);
 299                 break;
 300         case SAPIC_INIT:
 301         case SAPIC_PMI:
 302         default:
 303                 printk(KERN_ERR"kvm: Unimplemented Deliver reserved IPI!\n");
 304                 break;
 305         }
 306 }
 307
 308 static struct kvm_vcpu *lid_to_vcpu(struct kvm *kvm, unsigned long id,
 309                         unsigned long eid)
 310 {
 311         union ia64_lid lid;
 312         int i;
 313
 314         for (i = 0; i < KVM_MAX_VCPUS; i++) {
 315                 if (kvm->vcpus[i]) {
 316                         lid.val = VCPU_LID(kvm->vcpus[i]);
 317                         if (lid.id == id && lid.eid == eid)
 318                                 return kvm->vcpus[i];
 319                 }
 320         }
 321
 322         return NULL;
 323 }
 324
 325 static int handle_ipi(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 326 {
 327         struct exit_ctl_data *p = kvm_get_exit_data(vcpu);
 328         struct kvm_vcpu *target_vcpu;
 329         struct kvm_pt_regs *regs;
 330         union ia64_ipi_a addr = p->u.ipi_data.addr;
 331         union ia64_ipi_d data = p->u.ipi_data.data;
 332
 333         target_vcpu = lid_to_vcpu(vcpu->kvm, addr.id, addr.eid);
 334         if (!target_vcpu)
 335                 return handle_vm_error(vcpu, kvm_run);
 336
 337         if (!target_vcpu->arch.launched) {
 338                 regs = vcpu_regs(target_vcpu);
 339
 340                 regs->cr_iip = vcpu->kvm->arch.rdv_sal_data.boot_ip;
 341                 regs->r1 = vcpu->kvm->arch.rdv_sal_data.boot_gp;
 342
 343                 target_vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
 344                 if (waitqueue_active(&target_vcpu->wq))
 345                         wake_up_interruptible(&target_vcpu->wq);
 346         } else {
 347                 vcpu_deliver_ipi(target_vcpu, data.dm, data.vector);
 348                 if (target_vcpu != vcpu)
 349                         kvm_vcpu_kick(target_vcpu);
 350         }
 351
 352         return 1;
 353 }
 354
 355 struct call_data {
 356         struct kvm_ptc_g ptc_g_data;
 357         struct kvm_vcpu *vcpu;
 358 };
 359
 360 static void vcpu_global_purge(void *info)
 361 {
 362         struct call_data *p = (struct call_data *)info;
 363         struct kvm_vcpu *vcpu = p->vcpu;
 364
 365         if (test_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 366                 return;
 367
 368         set_bit(KVM_REQ_PTC_G, &vcpu->requests);
 369         if (vcpu->arch.ptc_g_count < MAX_PTC_G_NUM) {
 370                 vcpu->arch.ptc_g_data[vcpu->arch.ptc_g_count++] =
 371                                                         p->ptc_g_data;
 372         } else {
 373                 clear_bit(KVM_REQ_PTC_G, &vcpu->requests);
 374                 vcpu->arch.ptc_g_count = 0;
 375                 set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests);
 376         }
 377 }
 378
 379 static int handle_global_purge(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 380 {
 381         struct exit_ctl_data *p = kvm_get_exit_data(vcpu);
 382         struct kvm *kvm = vcpu->kvm;
 383         struct call_data call_data;
 384         int i;
 385         call_data.ptc_g_data = p->u.ptc_g_data;
 386
 387         for (i = 0; i < KVM_MAX_VCPUS; i++) {
 388                 if (!kvm->vcpus[i] || kvm->vcpus[i]->arch.mp_state ==
 389                                                 KVM_MP_STATE_UNINITIALIZED ||
 390                                         vcpu == kvm->vcpus[i])
 391                         continue;
 392
 393                 if (waitqueue_active(&kvm->vcpus[i]->wq))
 394                         wake_up_interruptible(&kvm->vcpus[i]->wq);
 395
 396                 if (kvm->vcpus[i]->cpu != -1) {
 397                         call_data.vcpu = kvm->vcpus[i];
 398                         smp_call_function_single(kvm->vcpus[i]->cpu,
 399                                         vcpu_global_purge, &call_data, 0, 1);
 400                 } else
 401                         printk(KERN_WARNING"kvm: Uninit vcpu received ipi!\n");
 402
 403         }
 404         return 1;
 405 }
 406
 407 static int handle_switch_rr6(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 408 {
 409         return 1;
 410 }
 411
 412 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
 413 {
 414
 415         ktime_t kt;
 416         long itc_diff;
 417         unsigned long vcpu_now_itc;
 418
 419         unsigned long expires;
 420         struct hrtimer *p_ht = &vcpu->arch.hlt_timer;
 421         unsigned long cyc_per_usec = local_cpu_data->cyc_per_usec;
 422         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
 423
 424         vcpu_now_itc = ia64_getreg(_IA64_REG_AR_ITC) + vcpu->arch.itc_offset;
 425
 426         if (time_after(vcpu_now_itc, vpd->itm)) {
 427                 vcpu->arch.timer_check = 1;
 428                 return 1;
 429         }
 430         itc_diff = vpd->itm - vcpu_now_itc;
 431         if (itc_diff < 0)
 432                 itc_diff = -itc_diff;
 433
 434         expires = div64_64(itc_diff, cyc_per_usec);
 435         kt = ktime_set(0, 1000 * expires);
 436         vcpu->arch.ht_active = 1;
 437         hrtimer_start(p_ht, kt, HRTIMER_MODE_ABS);
 438
 439         if (irqchip_in_kernel(vcpu->kvm)) {
 440                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
 441                 kvm_vcpu_block(vcpu);
 442                 hrtimer_cancel(p_ht);
 443                 vcpu->arch.ht_active = 0;
 444
 445                 if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
 446                         return -EINTR;
 447                 return 1;
 448         } else {
 449                 printk(KERN_ERR"kvm: Unsupported userspace halt!");
 450                 return 0;
 451         }
 452 }
 453
 454 static int handle_vm_shutdown(struct kvm_vcpu *vcpu,
 455                 struct kvm_run *kvm_run)
 456 {
 457         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
 458         return 0;
 459 }
 460
 461 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
 462                 struct kvm_run *kvm_run)
 463 {
 464         return 1;
 465 }
 466
 467 static int (*kvm_vti_exit_handlers[])(struct kvm_vcpu *vcpu,
 468                 struct kvm_run *kvm_run) = {
 469         [EXIT_REASON_VM_PANIC]              = handle_vm_error,
 470         [EXIT_REASON_MMIO_INSTRUCTION]      = handle_mmio,
 471         [EXIT_REASON_PAL_CALL]              = handle_pal_call,
 472         [EXIT_REASON_SAL_CALL]              = handle_sal_call,
 473         [EXIT_REASON_SWITCH_RR6]            = handle_switch_rr6,
 474         [EXIT_REASON_VM_DESTROY]            = handle_vm_shutdown,
 475         [EXIT_REASON_EXTERNAL_INTERRUPT]    = handle_external_interrupt,
 476         [EXIT_REASON_IPI]                   = handle_ipi,
 477         [EXIT_REASON_PTC_G]                 = handle_global_purge,
 478
 479 };
 480
 481 static const int kvm_vti_max_exit_handlers =
 482                 sizeof(kvm_vti_exit_handlers)/sizeof(*kvm_vti_exit_handlers);
 483
 484 static void kvm_prepare_guest_switch(struct kvm_vcpu *vcpu)
 485 {
 486 }
 487
 488 static uint32_t kvm_get_exit_reason(struct kvm_vcpu *vcpu)
 489 {
 490         struct exit_ctl_data *p_exit_data;
 491
 492         p_exit_data = kvm_get_exit_data(vcpu);
 493         return p_exit_data->exit_reason;
 494 }
 495
 496 /*
 497  * The guest has exited.  See if we can fix it or if we need userspace
 498  * assistance.
 499  */
 500 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
 501 {
 502         u32 exit_reason = kvm_get_exit_reason(vcpu);
 503         vcpu->arch.last_exit = exit_reason;
 504
 505         if (exit_reason < kvm_vti_max_exit_handlers
 506                         && kvm_vti_exit_handlers[exit_reason])
 507                 return kvm_vti_exit_handlers[exit_reason](vcpu, kvm_run);
 508         else {
 509                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
 510                 kvm_run->hw.hardware_exit_reason = exit_reason;
 511         }
 512         return 0;
 513 }
 514
 515 static inline void vti_set_rr6(unsigned long rr6)
 516 {
 517         ia64_set_rr(RR6, rr6);
 518         ia64_srlz_i();
 519 }
 520
 521 static int kvm_insert_vmm_mapping(struct kvm_vcpu *vcpu)
 522 {
 523         unsigned long pte;
 524         struct kvm *kvm = vcpu->kvm;
 525         int r;
 526
 527         /*Insert a pair of tr to map vmm*/
 528         pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base), PAGE_KERNEL));
 529         r = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
 530         if (r < 0)
 531                 goto out;
 532         vcpu->arch.vmm_tr_slot = r;
 533         /*Insert a pairt of tr to map data of vm*/
 534         pte = pte_val(mk_pte_phys(__pa(kvm->arch.vm_base), PAGE_KERNEL));
 535         r = ia64_itr_entry(0x3, KVM_VM_DATA_BASE,
 536                                         pte, KVM_VM_DATA_SHIFT);
 537         if (r < 0)
 538                 goto out;
 539         vcpu->arch.vm_tr_slot = r;
 540         r = 0;
 541 out:
 542         return r;
 543
 544 }
 545
 546 static void kvm_purge_vmm_mapping(struct kvm_vcpu *vcpu)
 547 {
 548
 549         ia64_ptr_entry(0x3, vcpu->arch.vmm_tr_slot);
 550         ia64_ptr_entry(0x3, vcpu->arch.vm_tr_slot);
 551
 552 }
 553
 554 static int kvm_vcpu_pre_transition(struct kvm_vcpu *vcpu)
 555 {
 556         int cpu = smp_processor_id();
 557
 558         if (vcpu->arch.last_run_cpu != cpu ||
 559                         per_cpu(last_vcpu, cpu) != vcpu) {
 560                 per_cpu(last_vcpu, cpu) = vcpu;
 561                 vcpu->arch.last_run_cpu = cpu;
 562                 kvm_flush_tlb_all();
 563         }
 564
 565         vcpu->arch.host_rr6 = ia64_get_rr(RR6);
 566         vti_set_rr6(vcpu->arch.vmm_rr);
 567         return kvm_insert_vmm_mapping(vcpu);
 568 }
 569 static void kvm_vcpu_post_transition(struct kvm_vcpu *vcpu)
 570 {
 571         kvm_purge_vmm_mapping(vcpu);
 572         vti_set_rr6(vcpu->arch.host_rr6);
 573 }
 574
 575 static int  vti_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 576 {
 577         union context *host_ctx, *guest_ctx;
 578         int r;
 579
 580         /*Get host and guest context with guest address space.*/
 581         host_ctx = kvm_get_host_context(vcpu);
 582         guest_ctx = kvm_get_guest_context(vcpu);
 583
 584         r = kvm_vcpu_pre_transition(vcpu);
 585         if (r < 0)
 586                 goto out;
 587         kvm_vmm_info->tramp_entry(host_ctx, guest_ctx);
 588         kvm_vcpu_post_transition(vcpu);
 589         r = 0;
 590 out:
 591         return r;
 592 }
 593
 594 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 595 {
 596         int r;
 597
 598 again:
 599         preempt_disable();
 600
 601         kvm_prepare_guest_switch(vcpu);
 602         local_irq_disable();
 603
 604         if (signal_pending(current)) {
 605                 local_irq_enable();
 606                 preempt_enable();
 607                 r = -EINTR;
 608                 kvm_run->exit_reason = KVM_EXIT_INTR;
 609                 goto out;
 610         }
 611
 612         vcpu->guest_mode = 1;
 613         kvm_guest_enter();
 614
 615         r = vti_vcpu_run(vcpu, kvm_run);
 616         if (r < 0) {
 617                 local_irq_enable();
 618                 preempt_enable();
 619                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
 620                 goto out;
 621         }
 622
 623         vcpu->arch.launched = 1;
 624         vcpu->guest_mode = 0;
 625         local_irq_enable();
 626
 627         /*
 628          * We must have an instruction between local_irq_enable() and
 629          * kvm_guest_exit(), so the timer interrupt isn't delayed by
 630          * the interrupt shadow.  The stat.exits increment will do nicely.
 631          * But we need to prevent reordering, hence this barrier():
 632          */
 633         barrier();
 634
 635         kvm_guest_exit();
 636
 637         preempt_enable();
 638
 639         r = kvm_handle_exit(kvm_run, vcpu);
 640
 641         if (r > 0) {
 642                 if (!need_resched())
 643                         goto again;
 644         }
 645
 646 out:
 647         if (r > 0) {
 648                 kvm_resched(vcpu);
 649                 goto again;
 650         }
 651
 652         return r;
 653 }
 654
 655 static void kvm_set_mmio_data(struct kvm_vcpu *vcpu)
 656 {
 657         struct kvm_mmio_req *p = kvm_get_vcpu_ioreq(vcpu);
 658
 659         if (!vcpu->mmio_is_write)
 660                 memcpy(&p->data, vcpu->mmio_data, 8);
 661         p->state = STATE_IORESP_READY;
 662 }
 663
 664 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 665 {
 666         int r;
 667         sigset_t sigsaved;
 668
 669         vcpu_load(vcpu);
 670
 671         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
 672                 kvm_vcpu_block(vcpu);
 673                 vcpu_put(vcpu);
 674                 return -EAGAIN;
 675         }
 676
 677         if (vcpu->sigset_active)
 678                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 679
 680         if (vcpu->mmio_needed) {
 681                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
 682                 kvm_set_mmio_data(vcpu);
 683                 vcpu->mmio_read_completed = 1;
 684                 vcpu->mmio_needed = 0;
 685         }
 686         r = __vcpu_run(vcpu, kvm_run);
 687
 688         if (vcpu->sigset_active)
 689                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 690
 691         vcpu_put(vcpu);
 692         return r;
 693 }
 694
 695 /*
 696  * Allocate 16M memory for every vm to hold its specific data.
 697  * Its memory map is defined in kvm_host.h.
 698  */
 699 static struct kvm *kvm_alloc_kvm(void)
 700 {
 701
 702         struct kvm *kvm;
 703         uint64_t  vm_base;
 704
 705         vm_base = __get_free_pages(GFP_KERNEL, get_order(KVM_VM_DATA_SIZE));
 706
 707         if (!vm_base)
 708                 return ERR_PTR(-ENOMEM);
 709         printk(KERN_DEBUG"kvm: VM data's base Address:0x%lx\n", vm_base);
 710
 711         /* Zero all pages before use! */
 712         memset((void *)vm_base, 0, KVM_VM_DATA_SIZE);
 713
 714         kvm = (struct kvm *)(vm_base + KVM_VM_OFS);
 715         kvm->arch.vm_base = vm_base;
 716
 717         return kvm;
 718 }
 719
 720 struct kvm_io_range {
 721         unsigned long start;
 722         unsigned long size;
 723         unsigned long type;
 724 };
 725
 726 static const struct kvm_io_range io_ranges[] = {
 727         {VGA_IO_START, VGA_IO_SIZE, GPFN_FRAME_BUFFER},
 728         {MMIO_START, MMIO_SIZE, GPFN_LOW_MMIO},
 729         {LEGACY_IO_START, LEGACY_IO_SIZE, GPFN_LEGACY_IO},
 730         {IO_SAPIC_START, IO_SAPIC_SIZE, GPFN_IOSAPIC},
 731         {PIB_START, PIB_SIZE, GPFN_PIB},
 732 };
 733
 734 static void kvm_build_io_pmt(struct kvm *kvm)
 735 {
 736         unsigned long i, j;
 737
 738         /* Mark I/O ranges */
 739         for (i = 0; i < (sizeof(io_ranges) / sizeof(struct kvm_io_range));
 740                                                         i++) {
 741                 for (j = io_ranges[i].start;
 742                                 j < io_ranges[i].start + io_ranges[i].size;
 743                                 j += PAGE_SIZE)
 744                         kvm_set_pmt_entry(kvm, j >> PAGE_SHIFT,
 745                                         io_ranges[i].type, 0);
 746         }
 747
 748 }
 749
 750 /*Use unused rids to virtualize guest rid.*/
 751 #define GUEST_PHYSICAL_RR0      0x1739
 752 #define GUEST_PHYSICAL_RR4      0x2739
 753 #define VMM_INIT_RR             0x1660
 754
 755 static void kvm_init_vm(struct kvm *kvm)
 756 {
 757         long vm_base;
 758
 759         BUG_ON(!kvm);
 760
 761         kvm->arch.metaphysical_rr0 = GUEST_PHYSICAL_RR0;
 762         kvm->arch.metaphysical_rr4 = GUEST_PHYSICAL_RR4;
 763         kvm->arch.vmm_init_rr = VMM_INIT_RR;
 764
 765         vm_base = kvm->arch.vm_base;
 766         if (vm_base) {
 767                 kvm->arch.vhpt_base = vm_base + KVM_VHPT_OFS;
 768                 kvm->arch.vtlb_base = vm_base + KVM_VTLB_OFS;
 769                 kvm->arch.vpd_base  = vm_base + KVM_VPD_OFS;
 770         }
 771
 772         /*
 773          *Fill P2M entries for MMIO/IO ranges
 774          */
 775         kvm_build_io_pmt(kvm);
 776
 777 }
 778
 779 struct  kvm *kvm_arch_create_vm(void)
 780 {
 781         struct kvm *kvm = kvm_alloc_kvm();
 782
 783         if (IS_ERR(kvm))
 784                 return ERR_PTR(-ENOMEM);
 785         kvm_init_vm(kvm);
 786
 787         return kvm;
 788
 789 }
 790
 791 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm,
 792                                         struct kvm_irqchip *chip)
 793 {
 794         int r;
 795
 796         r = 0;
 797         switch (chip->chip_id) {
 798         case KVM_IRQCHIP_IOAPIC:
 799                 memcpy(&chip->chip.ioapic, ioapic_irqchip(kvm),
 800                                 sizeof(struct kvm_ioapic_state));
 801                 break;
 802         default:
 803                 r = -EINVAL;
 804                 break;
 805         }
 806         return r;
 807 }
 808
 809 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
 810 {
 811         int r;
 812
 813         r = 0;
 814         switch (chip->chip_id) {
 815         case KVM_IRQCHIP_IOAPIC:
 816                 memcpy(ioapic_irqchip(kvm),
 817                                 &chip->chip.ioapic,
 818                                 sizeof(struct kvm_ioapic_state));
 819                 break;
 820         default:
 821                 r = -EINVAL;
 822                 break;
 823         }
 824         return r;
 825 }
 826
 827 #define RESTORE_REGS(_x) vcpu->arch._x = regs->_x
 828
 829 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 830 {
 831         int i;
 832         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
 833         int r;
 834
 835         vcpu_load(vcpu);
 836
 837         for (i = 0; i < 16; i++) {
 838                 vpd->vgr[i] = regs->vpd.vgr[i];
 839                 vpd->vbgr[i] = regs->vpd.vbgr[i];
 840         }
 841         for (i = 0; i < 128; i++)
 842                 vpd->vcr[i] = regs->vpd.vcr[i];
 843         vpd->vhpi = regs->vpd.vhpi;
 844         vpd->vnat = regs->vpd.vnat;
 845         vpd->vbnat = regs->vpd.vbnat;
 846         vpd->vpsr = regs->vpd.vpsr;
 847
 848         vpd->vpr = regs->vpd.vpr;
 849
 850         r = -EFAULT;
 851         r = copy_from_user(&vcpu->arch.guest, regs->saved_guest,
 852                                                 sizeof(union context));
 853         if (r)
 854                 goto out;
 855         r = copy_from_user(vcpu + 1, regs->saved_stack +
 856                         sizeof(struct kvm_vcpu),
 857                         IA64_STK_OFFSET - sizeof(struct kvm_vcpu));
 858         if (r)
 859                 goto out;
 860         vcpu->arch.exit_data =
 861                 ((struct kvm_vcpu *)(regs->saved_stack))->arch.exit_data;
 862
 863         RESTORE_REGS(mp_state);
 864         RESTORE_REGS(vmm_rr);
 865         memcpy(vcpu->arch.itrs, regs->itrs, sizeof(struct thash_data) * NITRS);
 866         memcpy(vcpu->arch.dtrs, regs->dtrs, sizeof(struct thash_data) * NDTRS);
 867         RESTORE_REGS(itr_regions);
 868         RESTORE_REGS(dtr_regions);
 869         RESTORE_REGS(tc_regions);
 870         RESTORE_REGS(irq_check);
 871         RESTORE_REGS(itc_check);
 872         RESTORE_REGS(timer_check);
 873         RESTORE_REGS(timer_pending);
 874         RESTORE_REGS(last_itc);
 875         for (i = 0; i < 8; i++) {
 876                 vcpu->arch.vrr[i] = regs->vrr[i];
 877                 vcpu->arch.ibr[i] = regs->ibr[i];
 878                 vcpu->arch.dbr[i] = regs->dbr[i];
 879         }
 880         for (i = 0; i < 4; i++)
 881                 vcpu->arch.insvc[i] = regs->insvc[i];
 882         RESTORE_REGS(xtp);
 883         RESTORE_REGS(metaphysical_rr0);
 884         RESTORE_REGS(metaphysical_rr4);
 885         RESTORE_REGS(metaphysical_saved_rr0);
 886         RESTORE_REGS(metaphysical_saved_rr4);
 887         RESTORE_REGS(fp_psr);
 888         RESTORE_REGS(saved_gp);
 889
 890         vcpu->arch.irq_new_pending = 1;
 891         vcpu->arch.itc_offset = regs->saved_itc - ia64_getreg(_IA64_REG_AR_ITC);
 892         set_bit(KVM_REQ_RESUME, &vcpu->requests);
 893
 894         vcpu_put(vcpu);
 895         r = 0;
 896 out:
 897         return r;
 898 }
 899
 900 long kvm_arch_vm_ioctl(struct file *filp,
 901                 unsigned int ioctl, unsigned long arg)
 902 {
 903         struct kvm *kvm = filp->private_data;
 904         void __user *argp = (void __user *)arg;
 905         int r = -EINVAL;
 906
 907         switch (ioctl) {
 908         case KVM_SET_MEMORY_REGION: {
 909                 struct kvm_memory_region kvm_mem;
 910                 struct kvm_userspace_memory_region kvm_userspace_mem;
 911
 912                 r = -EFAULT;
 913                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
 914                         goto out;
 915                 kvm_userspace_mem.slot = kvm_mem.slot;
 916                 kvm_userspace_mem.flags = kvm_mem.flags;
 917                 kvm_userspace_mem.guest_phys_addr =
 918                                         kvm_mem.guest_phys_addr;
 919                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
 920                 r = kvm_vm_ioctl_set_memory_region(kvm,
 921                                         &kvm_userspace_mem, 0);
 922                 if (r)
 923                         goto out;
 924                 break;
 925                 }
 926         case KVM_CREATE_IRQCHIP:
 927                 r = -EFAULT;
 928                 r = kvm_ioapic_init(kvm);
 929                 if (r)
 930                         goto out;
 931                 break;
 932         case KVM_IRQ_LINE: {
 933                 struct kvm_irq_level irq_event;
 934
 935                 r = -EFAULT;
 936                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
 937                         goto out;
 938                 if (irqchip_in_kernel(kvm)) {
 939                         mutex_lock(&kvm->lock);
 940                         kvm_ioapic_set_irq(kvm->arch.vioapic,
 941                                                 irq_event.irq,
 942                                                 irq_event.level);
 943                         mutex_unlock(&kvm->lock);
 944                         r = 0;
 945                 }
 946                 break;
 947                 }
 948         case KVM_GET_IRQCHIP: {
 949                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
 950                 struct kvm_irqchip chip;
 951
 952                 r = -EFAULT;
 953                 if (copy_from_user(&chip, argp, sizeof chip))
 954                                 goto out;
 955                 r = -ENXIO;
 956                 if (!irqchip_in_kernel(kvm))
 957                         goto out;
 958                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
 959                 if (r)
 960                         goto out;
 961                 r = -EFAULT;
 962                 if (copy_to_user(argp, &chip, sizeof chip))
 963                                 goto out;
 964                 r = 0;
 965                 break;
 966                 }
 967         case KVM_SET_IRQCHIP: {
 968                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
 969                 struct kvm_irqchip chip;
 970
 971                 r = -EFAULT;
 972                 if (copy_from_user(&chip, argp, sizeof chip))
 973                                 goto out;
 974                 r = -ENXIO;
 975                 if (!irqchip_in_kernel(kvm))
 976                         goto out;
 977                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
 978                 if (r)
 979                         goto out;
 980                 r = 0;
 981                 break;
 982                 }
 983         default:
 984                 ;
 985         }
 986 out:
 987         return r;
 988 }
 989
 990 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 991                 struct kvm_sregs *sregs)
 992 {
 993         return -EINVAL;
 994 }
 995
 996 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 997                 struct kvm_sregs *sregs)
 998 {
 999         return -EINVAL;
1000
1001 }
1002 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1003                 struct kvm_translation *tr)
1004 {
1005
1006         return -EINVAL;
1007 }
1008
1009 static int kvm_alloc_vmm_area(void)
1010 {
1011         if (!kvm_vmm_base && (kvm_vm_buffer_size < KVM_VM_BUFFER_SIZE)) {
1012                 kvm_vmm_base = __get_free_pages(GFP_KERNEL,
1013                                 get_order(KVM_VMM_SIZE));
1014                 if (!kvm_vmm_base)
1015                         return -ENOMEM;
1016
1017                 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1018                 kvm_vm_buffer = kvm_vmm_base + VMM_SIZE;
1019
1020                 printk(KERN_DEBUG"kvm:VMM's Base Addr:0x%lx, vm_buffer:0x%lx\n",
1021                                 kvm_vmm_base, kvm_vm_buffer);
1022         }
1023
1024         return 0;
1025 }
1026
1027 static void kvm_free_vmm_area(void)
1028 {
1029         if (kvm_vmm_base) {
1030                 /*Zero this area before free to avoid bits leak!!*/
1031                 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1032                 free_pages(kvm_vmm_base, get_order(KVM_VMM_SIZE));
1033                 kvm_vmm_base  = 0;
1034                 kvm_vm_buffer = 0;
1035                 kvm_vsa_base = 0;
1036         }
1037 }
1038
1039 /*
1040  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
1041  * cached on it. Leave it as blank for IA64.
1042  */
1043 void decache_vcpus_on_cpu(int cpu)
1044 {
1045 }
1046
1047 static void vti_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1048 {
1049 }
1050
1051 static int vti_init_vpd(struct kvm_vcpu *vcpu)
1052 {
1053         int i;
1054         union cpuid3_t cpuid3;
1055         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1056
1057         if (IS_ERR(vpd))
1058                 return PTR_ERR(vpd);
1059
1060         /* CPUID init */
1061         for (i = 0; i < 5; i++)
1062                 vpd->vcpuid[i] = ia64_get_cpuid(i);
1063
1064         /* Limit the CPUID number to 5 */
1065         cpuid3.value = vpd->vcpuid[3];
1066         cpuid3.number = 4;      /* 5 - 1 */
1067         vpd->vcpuid[3] = cpuid3.value;
1068
1069         /*Set vac and vdc fields*/
1070         vpd->vac.a_from_int_cr = 1;
1071         vpd->vac.a_to_int_cr = 1;
1072         vpd->vac.a_from_psr = 1;
1073         vpd->vac.a_from_cpuid = 1;
1074         vpd->vac.a_cover = 1;
1075         vpd->vac.a_bsw = 1;
1076         vpd->vac.a_int = 1;
1077         vpd->vdc.d_vmsw = 1;
1078
1079         /*Set virtual buffer*/
1080         vpd->virt_env_vaddr = KVM_VM_BUFFER_BASE;
1081
1082         return 0;
1083 }
1084
1085 static int vti_create_vp(struct kvm_vcpu *vcpu)
1086 {
1087         long ret;
1088         struct vpd *vpd = vcpu->arch.vpd;
1089         unsigned long  vmm_ivt;
1090
1091         vmm_ivt = kvm_vmm_info->vmm_ivt;
1092
1093         printk(KERN_DEBUG "kvm: vcpu:%p,ivt: 0x%lx\n", vcpu, vmm_ivt);
1094
1095         ret = ia64_pal_vp_create((u64 *)vpd, (u64 *)vmm_ivt, 0);
1096
1097         if (ret) {
1098                 printk(KERN_ERR"kvm: ia64_pal_vp_create failed!\n");
1099                 return -EINVAL;
1100         }
1101         return 0;
1102 }
1103
1104 static void init_ptce_info(struct kvm_vcpu *vcpu)
1105 {
1106         ia64_ptce_info_t ptce = {0};
1107
1108         ia64_get_ptce(&ptce);
1109         vcpu->arch.ptce_base = ptce.base;
1110         vcpu->arch.ptce_count[0] = ptce.count[0];
1111         vcpu->arch.ptce_count[1] = ptce.count[1];
1112         vcpu->arch.ptce_stride[0] = ptce.stride[0];
1113         vcpu->arch.ptce_stride[1] = ptce.stride[1];
1114 }
1115
1116 static void kvm_migrate_hlt_timer(struct kvm_vcpu *vcpu)
1117 {
1118         struct hrtimer *p_ht = &vcpu->arch.hlt_timer;
1119
1120         if (hrtimer_cancel(p_ht))
1121                 hrtimer_start(p_ht, p_ht->expires, HRTIMER_MODE_ABS);
1122 }
1123
1124 static enum hrtimer_restart hlt_timer_fn(struct hrtimer *data)
1125 {
1126         struct kvm_vcpu *vcpu;
1127         wait_queue_head_t *q;
1128
1129         vcpu  = container_of(data, struct kvm_vcpu, arch.hlt_timer);
1130         if (vcpu->arch.mp_state != KVM_MP_STATE_HALTED)
1131                 goto out;
1132
1133         q = &vcpu->wq;
1134         if (waitqueue_active(q)) {
1135                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1136                 wake_up_interruptible(q);
1137         }
1138 out:
1139         vcpu->arch.timer_check = 1;
1140         return HRTIMER_NORESTART;
1141 }
1142
1143 #define PALE_RESET_ENTRY    0x80000000ffffffb0UL
1144
1145 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1146 {
1147         struct kvm_vcpu *v;
1148         int r;
1149         int i;
1150         long itc_offset;
1151         struct kvm *kvm = vcpu->kvm;
1152         struct kvm_pt_regs *regs = vcpu_regs(vcpu);
1153
1154         union context *p_ctx = &vcpu->arch.guest;
1155         struct kvm_vcpu *vmm_vcpu = to_guest(vcpu->kvm, vcpu);
1156
1157         /*Init vcpu context for first run.*/
1158         if (IS_ERR(vmm_vcpu))
1159                 return PTR_ERR(vmm_vcpu);
1160
1161         if (vcpu->vcpu_id == 0) {
1162                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1163
1164                 /*Set entry address for first run.*/
1165                 regs->cr_iip = PALE_RESET_ENTRY;
1166
1167                 /*Initilize itc offset for vcpus*/
1168                 itc_offset = 0UL - ia64_getreg(_IA64_REG_AR_ITC);
1169                 for (i = 0; i < MAX_VCPU_NUM; i++) {
1170                         v = (struct kvm_vcpu *)((char *)vcpu + VCPU_SIZE * i);
1171                         v->arch.itc_offset = itc_offset;
1172                         v->arch.last_itc = 0;
1173                 }
1174         } else
1175                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
1176
1177         r = -ENOMEM;
1178         vcpu->arch.apic = kzalloc(sizeof(struct kvm_lapic), GFP_KERNEL);
1179         if (!vcpu->arch.apic)
1180                 goto out;
1181         vcpu->arch.apic->vcpu = vcpu;
1182
1183         p_ctx->gr[1] = 0;
1184         p_ctx->gr[12] = (unsigned long)((char *)vmm_vcpu + IA64_STK_OFFSET);
1185         p_ctx->gr[13] = (unsigned long)vmm_vcpu;
1186         p_ctx->psr = 0x1008522000UL;
1187         p_ctx->ar[40] = FPSR_DEFAULT; /*fpsr*/
1188         p_ctx->caller_unat = 0;
1189         p_ctx->pr = 0x0;
1190         p_ctx->ar[36] = 0x0; /*unat*/
1191         p_ctx->ar[19] = 0x0; /*rnat*/
1192         p_ctx->ar[18] = (unsigned long)vmm_vcpu +
1193                                 ((sizeof(struct kvm_vcpu)+15) & ~15);
1194         p_ctx->ar[64] = 0x0; /*pfs*/
1195         p_ctx->cr[0] = 0x7e04UL;
1196         p_ctx->cr[2] = (unsigned long)kvm_vmm_info->vmm_ivt;
1197         p_ctx->cr[8] = 0x3c;
1198
1199         /*Initilize region register*/
1200         p_ctx->rr[0] = 0x30;
1201         p_ctx->rr[1] = 0x30;
1202         p_ctx->rr[2] = 0x30;
1203         p_ctx->rr[3] = 0x30;
1204         p_ctx->rr[4] = 0x30;
1205         p_ctx->rr[5] = 0x30;
1206         p_ctx->rr[7] = 0x30;
1207
1208         /*Initilize branch register 0*/
1209         p_ctx->br[0] = *(unsigned long *)kvm_vmm_info->vmm_entry;
1210
1211         vcpu->arch.vmm_rr = kvm->arch.vmm_init_rr;
1212         vcpu->arch.metaphysical_rr0 = kvm->arch.metaphysical_rr0;
1213         vcpu->arch.metaphysical_rr4 = kvm->arch.metaphysical_rr4;
1214
1215         hrtimer_init(&vcpu->arch.hlt_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1216         vcpu->arch.hlt_timer.function = hlt_timer_fn;
1217
1218         vcpu->arch.last_run_cpu = -1;
1219         vcpu->arch.vpd = (struct vpd *)VPD_ADDR(vcpu->vcpu_id);
1220         vcpu->arch.vsa_base = kvm_vsa_base;
1221         vcpu->arch.__gp = kvm_vmm_gp;
1222         vcpu->arch.dirty_log_lock_pa = __pa(&kvm->arch.dirty_log_lock);
1223         vcpu->arch.vhpt.hash = (struct thash_data *)VHPT_ADDR(vcpu->vcpu_id);
1224         vcpu->arch.vtlb.hash = (struct thash_data *)VTLB_ADDR(vcpu->vcpu_id);
1225         init_ptce_info(vcpu);
1226
1227         r = 0;
1228 out:
1229         return r;
1230 }
1231
1232 static int vti_vcpu_setup(struct kvm_vcpu *vcpu, int id)
1233 {
1234         unsigned long psr;
1235         int r;
1236
1237         local_irq_save(psr);
1238         r = kvm_insert_vmm_mapping(vcpu);
1239         if (r)
1240                 goto fail;
1241         r = kvm_vcpu_init(vcpu, vcpu->kvm, id);
1242         if (r)
1243                 goto fail;
1244
1245         r = vti_init_vpd(vcpu);
1246         if (r) {
1247                 printk(KERN_DEBUG"kvm: vpd init error!!\n");
1248                 goto uninit;
1249         }
1250
1251         r = vti_create_vp(vcpu);
1252         if (r)
1253                 goto uninit;
1254
1255         kvm_purge_vmm_mapping(vcpu);
1256         local_irq_restore(psr);
1257
1258         return 0;
1259 uninit:
1260         kvm_vcpu_uninit(vcpu);
1261 fail:
1262         return r;
1263 }
1264
1265 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
1266                 unsigned int id)
1267 {
1268         struct kvm_vcpu *vcpu;
1269         unsigned long vm_base = kvm->arch.vm_base;
1270         int r;
1271         int cpu;
1272
1273         r = -ENOMEM;
1274         if (!vm_base) {
1275                 printk(KERN_ERR"kvm: Create vcpu[%d] error!\n", id);
1276                 goto fail;
1277         }
1278         vcpu = (struct kvm_vcpu *)(vm_base + KVM_VCPU_OFS + VCPU_SIZE * id);
1279         vcpu->kvm = kvm;
1280
1281         cpu = get_cpu();
1282         vti_vcpu_load(vcpu, cpu);
1283         r = vti_vcpu_setup(vcpu, id);
1284         put_cpu();
1285
1286         if (r) {
1287                 printk(KERN_DEBUG"kvm: vcpu_setup error!!\n");
1288                 goto fail;
1289         }
1290
1291         return vcpu;
1292 fail:
1293         return ERR_PTR(r);
1294 }
1295
1296 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1297 {
1298         return 0;
1299 }
1300
1301 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1302 {
1303         return -EINVAL;
1304 }
1305
1306 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1307 {
1308         return -EINVAL;
1309 }
1310
1311 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
1312                 struct kvm_debug_guest *dbg)
1313 {
1314         return -EINVAL;
1315 }
1316
1317 static void free_kvm(struct kvm *kvm)
1318 {
1319         unsigned long vm_base = kvm->arch.vm_base;
1320
1321         if (vm_base) {
1322                 memset((void *)vm_base, 0, KVM_VM_DATA_SIZE);
1323                 free_pages(vm_base, get_order(KVM_VM_DATA_SIZE));
1324         }
1325
1326 }
1327
1328 static void kvm_release_vm_pages(struct kvm *kvm)
1329 {
1330         struct kvm_memory_slot *memslot;
1331         int i, j;
1332         unsigned long base_gfn;
1333
1334         for (i = 0; i < kvm->nmemslots; i++) {
1335                 memslot = &kvm->memslots[i];
1336                 base_gfn = memslot->base_gfn;
1337
1338                 for (j = 0; j < memslot->npages; j++) {
1339                         if (memslot->rmap[j])
1340                                 put_page((struct page *)memslot->rmap[j]);
1341                 }
1342         }
1343 }
1344
1345 void kvm_arch_destroy_vm(struct kvm *kvm)
1346 {
1347         kfree(kvm->arch.vioapic);
1348         kvm_release_vm_pages(kvm);
1349         kvm_free_physmem(kvm);
1350         free_kvm(kvm);
1351 }
1352
1353 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1354 {
1355 }
1356
1357 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1358 {
1359         if (cpu != vcpu->cpu) {
1360                 vcpu->cpu = cpu;
1361                 if (vcpu->arch.ht_active)
1362                         kvm_migrate_hlt_timer(vcpu);
1363         }
1364 }
1365
1366 #define SAVE_REGS(_x)   regs->_x = vcpu->arch._x
1367
1368 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1369 {
1370         int i;
1371         int r;
1372         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1373         vcpu_load(vcpu);
1374
1375         for (i = 0; i < 16; i++) {
1376                 regs->vpd.vgr[i] = vpd->vgr[i];
1377                 regs->vpd.vbgr[i] = vpd->vbgr[i];
1378         }
1379         for (i = 0; i < 128; i++)
1380                 regs->vpd.vcr[i] = vpd->vcr[i];
1381         regs->vpd.vhpi = vpd->vhpi;
1382         regs->vpd.vnat = vpd->vnat;
1383         regs->vpd.vbnat = vpd->vbnat;
1384         regs->vpd.vpsr = vpd->vpsr;
1385         regs->vpd.vpr = vpd->vpr;
1386
1387         r = -EFAULT;
1388         r = copy_to_user(regs->saved_guest, &vcpu->arch.guest,
1389                                         sizeof(union context));
1390         if (r)
1391                 goto out;
1392         r = copy_to_user(regs->saved_stack, (void *)vcpu, IA64_STK_OFFSET);
1393         if (r)
1394                 goto out;
1395         SAVE_REGS(mp_state);
1396         SAVE_REGS(vmm_rr);
1397         memcpy(regs->itrs, vcpu->arch.itrs, sizeof(struct thash_data) * NITRS);
1398         memcpy(regs->dtrs, vcpu->arch.dtrs, sizeof(struct thash_data) * NDTRS);
1399         SAVE_REGS(itr_regions);
1400         SAVE_REGS(dtr_regions);
1401         SAVE_REGS(tc_regions);
1402         SAVE_REGS(irq_check);
1403         SAVE_REGS(itc_check);
1404         SAVE_REGS(timer_check);
1405         SAVE_REGS(timer_pending);
1406         SAVE_REGS(last_itc);
1407         for (i = 0; i < 8; i++) {
1408                 regs->vrr[i] = vcpu->arch.vrr[i];
1409                 regs->ibr[i] = vcpu->arch.ibr[i];
1410                 regs->dbr[i] = vcpu->arch.dbr[i];
1411         }
1412         for (i = 0; i < 4; i++)
1413                 regs->insvc[i] = vcpu->arch.insvc[i];
1414         regs->saved_itc = vcpu->arch.itc_offset + ia64_getreg(_IA64_REG_AR_ITC);
1415         SAVE_REGS(xtp);
1416         SAVE_REGS(metaphysical_rr0);
1417         SAVE_REGS(metaphysical_rr4);
1418         SAVE_REGS(metaphysical_saved_rr0);
1419         SAVE_REGS(metaphysical_saved_rr4);
1420         SAVE_REGS(fp_psr);
1421         SAVE_REGS(saved_gp);
1422         vcpu_put(vcpu);
1423         r = 0;
1424 out:
1425         return r;
1426 }
1427
1428 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
1429 {
1430
1431         hrtimer_cancel(&vcpu->arch.hlt_timer);
1432         kfree(vcpu->arch.apic);
1433 }
1434
1435
1436 long kvm_arch_vcpu_ioctl(struct file *filp,
1437                 unsigned int ioctl, unsigned long arg)
1438 {
1439         return -EINVAL;
1440 }
1441
1442 int kvm_arch_set_memory_region(struct kvm *kvm,
1443                 struct kvm_userspace_memory_region *mem,
1444                 struct kvm_memory_slot old,
1445                 int user_alloc)
1446 {
1447         unsigned long i;
1448         struct page *page;
1449         int npages = mem->memory_size >> PAGE_SHIFT;
1450         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
1451         unsigned long base_gfn = memslot->base_gfn;
1452
1453         for (i = 0; i < npages; i++) {
1454                 page = gfn_to_page(kvm, base_gfn + i);
1455                 kvm_set_pmt_entry(kvm, base_gfn + i,
1456                                 page_to_pfn(page) << PAGE_SHIFT,
1457                                 _PAGE_AR_RWX|_PAGE_MA_WB);
1458                 memslot->rmap[i] = (unsigned long)page;
1459         }
1460
1461         return 0;
1462 }
1463
1464
1465 long kvm_arch_dev_ioctl(struct file *filp,
1466                 unsigned int ioctl, unsigned long arg)
1467 {
1468         return -EINVAL;
1469 }
1470
1471 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1472 {
1473         kvm_vcpu_uninit(vcpu);
1474 }
1475
1476 static int vti_cpu_has_kvm_support(void)
1477 {
1478         long  avail = 1, status = 1, control = 1;
1479         long ret;
1480
1481         ret = ia64_pal_proc_get_features(&avail, &status, &control, 0);
1482         if (ret)
1483                 goto out;
1484
1485         if (!(avail & PAL_PROC_VM_BIT))
1486                 goto out;
1487
1488         printk(KERN_DEBUG"kvm: Hardware Supports VT\n");
1489
1490         ret = ia64_pal_vp_env_info(&kvm_vm_buffer_size, &vp_env_info);
1491         if (ret)
1492                 goto out;
1493         printk(KERN_DEBUG"kvm: VM Buffer Size:0x%lx\n", kvm_vm_buffer_size);
1494
1495         if (!(vp_env_info & VP_OPCODE)) {
1496                 printk(KERN_WARNING"kvm: No opcode ability on hardware, "
1497                                 "vm_env_info:0x%lx\n", vp_env_info);
1498         }
1499
1500         return 1;
1501 out:
1502         return 0;
1503 }
1504
1505 static int kvm_relocate_vmm(struct kvm_vmm_info *vmm_info,
1506                                                 struct module *module)
1507 {
1508         unsigned long module_base;
1509         unsigned long vmm_size;
1510
1511         unsigned long vmm_offset, func_offset, fdesc_offset;
1512         struct fdesc *p_fdesc;
1513
1514         BUG_ON(!module);
1515
1516         if (!kvm_vmm_base) {
1517                 printk("kvm: kvm area hasn't been initilized yet!!\n");
1518                 return -EFAULT;
1519         }
1520
1521         /*Calculate new position of relocated vmm module.*/
1522         module_base = (unsigned long)module->module_core;
1523         vmm_size = module->core_size;
1524         if (unlikely(vmm_size > KVM_VMM_SIZE))
1525                 return -EFAULT;
1526
1527         memcpy((void *)kvm_vmm_base, (void *)module_base, vmm_size);
1528         kvm_flush_icache(kvm_vmm_base, vmm_size);
1529
1530         /*Recalculate kvm_vmm_info based on new VMM*/
1531         vmm_offset = vmm_info->vmm_ivt - module_base;
1532         kvm_vmm_info->vmm_ivt = KVM_VMM_BASE + vmm_offset;
1533         printk(KERN_DEBUG"kvm: Relocated VMM's IVT Base Addr:%lx\n",
1534                         kvm_vmm_info->vmm_ivt);
1535
1536         fdesc_offset = (unsigned long)vmm_info->vmm_entry - module_base;
1537         kvm_vmm_info->vmm_entry = (kvm_vmm_entry *)(KVM_VMM_BASE +
1538                                                         fdesc_offset);
1539         func_offset = *(unsigned long *)vmm_info->vmm_entry - module_base;
1540         p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1541         p_fdesc->ip = KVM_VMM_BASE + func_offset;
1542         p_fdesc->gp = KVM_VMM_BASE+(p_fdesc->gp - module_base);
1543
1544         printk(KERN_DEBUG"kvm: Relocated VMM's Init Entry Addr:%lx\n",
1545                         KVM_VMM_BASE+func_offset);
1546
1547         fdesc_offset = (unsigned long)vmm_info->tramp_entry - module_base;
1548         kvm_vmm_info->tramp_entry = (kvm_tramp_entry *)(KVM_VMM_BASE +
1549                         fdesc_offset);
1550         func_offset = *(unsigned long *)vmm_info->tramp_entry - module_base;
1551         p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1552         p_fdesc->ip = KVM_VMM_BASE + func_offset;
1553         p_fdesc->gp = KVM_VMM_BASE + (p_fdesc->gp - module_base);
1554
1555         kvm_vmm_gp = p_fdesc->gp;
1556
1557         printk(KERN_DEBUG"kvm: Relocated VMM's Entry IP:%p\n",
1558                                                 kvm_vmm_info->vmm_entry);
1559         printk(KERN_DEBUG"kvm: Relocated VMM's Trampoline Entry IP:0x%lx\n",
1560                                                 KVM_VMM_BASE + func_offset);
1561
1562         return 0;
1563 }
1564
1565 int kvm_arch_init(void *opaque)
1566 {
1567         int r;
1568         struct kvm_vmm_info *vmm_info = (struct kvm_vmm_info *)opaque;
1569
1570         if (!vti_cpu_has_kvm_support()) {
1571                 printk(KERN_ERR "kvm: No Hardware Virtualization Support!\n");
1572                 r = -EOPNOTSUPP;
1573                 goto out;
1574         }
1575
1576         if (kvm_vmm_info) {
1577                 printk(KERN_ERR "kvm: Already loaded VMM module!\n");
1578                 r = -EEXIST;
1579                 goto out;
1580         }
1581
1582         r = -ENOMEM;
1583         kvm_vmm_info = kzalloc(sizeof(struct kvm_vmm_info), GFP_KERNEL);
1584         if (!kvm_vmm_info)
1585                 goto out;
1586
1587         if (kvm_alloc_vmm_area())
1588                 goto out_free0;
1589
1590         r = kvm_relocate_vmm(vmm_info, vmm_info->module);
1591         if (r)
1592                 goto out_free1;
1593
1594         return 0;
1595
1596 out_free1:
1597         kvm_free_vmm_area();
1598 out_free0:
1599         kfree(kvm_vmm_info);
1600 out:
1601         return r;
1602 }
1603
1604 void kvm_arch_exit(void)
1605 {
1606         kvm_free_vmm_area();
1607         kfree(kvm_vmm_info);
1608         kvm_vmm_info = NULL;
1609 }
1610
1611 static int kvm_ia64_sync_dirty_log(struct kvm *kvm,
1612                 struct kvm_dirty_log *log)
1613 {
1614         struct kvm_memory_slot *memslot;
1615         int r, i;
1616         long n, base;
1617         unsigned long *dirty_bitmap = (unsigned long *)((void *)kvm - KVM_VM_OFS
1618                                         + KVM_MEM_DIRTY_LOG_OFS);
1619
1620         r = -EINVAL;
1621         if (log->slot >= KVM_MEMORY_SLOTS)
1622                 goto out;
1623
1624         memslot = &kvm->memslots[log->slot];
1625         r = -ENOENT;
1626         if (!memslot->dirty_bitmap)
1627                 goto out;
1628
1629         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1630         base = memslot->base_gfn / BITS_PER_LONG;
1631
1632         for (i = 0; i < n/sizeof(long); ++i) {
1633                 memslot->dirty_bitmap[i] = dirty_bitmap[base + i];
1634                 dirty_bitmap[base + i] = 0;
1635         }
1636         r = 0;
1637 out:
1638         return r;
1639 }
1640
1641 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1642                 struct kvm_dirty_log *log)
1643 {
1644         int r;
1645         int n;
1646         struct kvm_memory_slot *memslot;
1647         int is_dirty = 0;
1648
1649         spin_lock(&kvm->arch.dirty_log_lock);
1650
1651         r = kvm_ia64_sync_dirty_log(kvm, log);
1652         if (r)
1653                 goto out;
1654
1655         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1656         if (r)
1657                 goto out;
1658
1659         /* If nothing is dirty, don't bother messing with page tables. */
1660         if (is_dirty) {
1661                 kvm_flush_remote_tlbs(kvm);
1662                 memslot = &kvm->memslots[log->slot];
1663                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1664                 memset(memslot->dirty_bitmap, 0, n);
1665         }
1666         r = 0;
1667 out:
1668         spin_unlock(&kvm->arch.dirty_log_lock);
1669         return r;
1670 }
1671
1672 int kvm_arch_hardware_setup(void)
1673 {
1674         return 0;
1675 }
1676
1677 void kvm_arch_hardware_unsetup(void)
1678 {
1679 }
1680
1681 static void vcpu_kick_intr(void *info)
1682 {
1683 #ifdef DEBUG
1684         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
1685         printk(KERN_DEBUG"vcpu_kick_intr %p \n", vcpu);
1686 #endif
1687 }
1688
1689 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1690 {
1691         int ipi_pcpu = vcpu->cpu;
1692
1693         if (waitqueue_active(&vcpu->wq))
1694                 wake_up_interruptible(&vcpu->wq);
1695
1696         if (vcpu->guest_mode)
1697                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0, 0);
1698 }
1699
1700 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, u8 vec, u8 trig)
1701 {
1702
1703         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1704
1705         if (!test_and_set_bit(vec, &vpd->irr[0])) {
1706                 vcpu->arch.irq_new_pending = 1;
1707                  if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
1708                         kvm_vcpu_kick(vcpu);
1709                 else if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED) {
1710                         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1711                         if (waitqueue_active(&vcpu->wq))
1712                                 wake_up_interruptible(&vcpu->wq);
1713                 }
1714                 return 1;
1715         }
1716         return 0;
1717 }
1718
1719 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
1720 {
1721         return apic->vcpu->vcpu_id == dest;
1722 }
1723
1724 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
1725 {
1726         return 0;
1727 }
1728
1729 struct kvm_vcpu *kvm_get_lowest_prio_vcpu(struct kvm *kvm, u8 vector,
1730                                        unsigned long bitmap)
1731 {
1732         struct kvm_vcpu *lvcpu = kvm->vcpus[0];
1733         int i;
1734
1735         for (i = 1; i < KVM_MAX_VCPUS; i++) {
1736                 if (!kvm->vcpus[i])
1737                         continue;
1738                 if (lvcpu->arch.xtp > kvm->vcpus[i]->arch.xtp)
1739                         lvcpu = kvm->vcpus[i];
1740         }
1741
1742         return lvcpu;
1743 }
1744
1745 static int find_highest_bits(int *dat)
1746 {
1747         u32  bits, bitnum;
1748         int i;
1749
1750         /* loop for all 256 bits */
1751         for (i = 7; i >= 0 ; i--) {
1752                 bits = dat[i];
1753                 if (bits) {
1754                         bitnum = fls(bits);
1755                         return i * 32 + bitnum - 1;
1756                 }
1757         }
1758
1759         return -1;
1760 }
1761
1762 int kvm_highest_pending_irq(struct kvm_vcpu *vcpu)
1763 {
1764     struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1765
1766     if (vpd->irr[0] & (1UL << NMI_VECTOR))
1767                 return NMI_VECTOR;
1768     if (vpd->irr[0] & (1UL << ExtINT_VECTOR))
1769                 return ExtINT_VECTOR;
1770
1771     return find_highest_bits((int *)&vpd->irr[0]);
1772 }
1773
1774 int kvm_cpu_has_interrupt(struct kvm_vcpu *vcpu)
1775 {
1776         if (kvm_highest_pending_irq(vcpu) != -1)
1777                 return 1;
1778         return 0;
1779 }
1780
1781 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1782 {
1783         return 0;
1784 }
1785
1786 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1787 {
1788         return gfn;
1789 }
1790
1791 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
1792 {
1793         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE;
1794 }
1795
1796 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1797                                     struct kvm_mp_state *mp_state)
1798 {
1799         return -EINVAL;
1800 }
1801
1802 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1803                                     struct kvm_mp_state *mp_state)
1804 {
1805         return -EINVAL;
1806 }