arch/powerpc/kvm/book3s_hv.c

   1 /*
   2  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
   3  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
   4  *
   5  * Authors:
   6  *    Paul Mackerras <paulus@au1.ibm.com>
   7  *    Alexander Graf <agraf@suse.de>
   8  *    Kevin Wolf <mail@kevin-wolf.de>
   9  *
  10  * Description: KVM functions specific to running on Book 3S
  11  * processors in hypervisor mode (specifically POWER7 and later).
  12  *
  13  * This file is derived from arch/powerpc/kvm/book3s.c,
  14  * by Alexander Graf <agraf@suse.de>.
  15  *
  16  * This program is free software; you can redistribute it and/or modify
  17  * it under the terms of the GNU General Public License, version 2, as
  18  * published by the Free Software Foundation.
  19  */
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/err.h>
  23 #include <linux/slab.h>
  24 #include <linux/preempt.h>
  25 #include <linux/sched.h>
  26 #include <linux/delay.h>
  27 #include <linux/export.h>
  28 #include <linux/fs.h>
  29 #include <linux/anon_inodes.h>
  30 #include <linux/cpumask.h>
  31 #include <linux/spinlock.h>
  32 #include <linux/page-flags.h>
  33
  34 #include <asm/reg.h>
  35 #include <asm/cputable.h>
  36 #include <asm/cacheflush.h>
  37 #include <asm/tlbflush.h>
  38 #include <asm/uaccess.h>
  39 #include <asm/io.h>
  40 #include <asm/kvm_ppc.h>
  41 #include <asm/kvm_book3s.h>
  42 #include <asm/mmu_context.h>
  43 #include <asm/lppaca.h>
  44 #include <asm/processor.h>
  45 #include <asm/cputhreads.h>
  46 #include <asm/page.h>
  47 #include <asm/hvcall.h>
  48 #include <asm/switch_to.h>
  49 #include <linux/gfp.h>
  50 #include <linux/vmalloc.h>
  51 #include <linux/highmem.h>
  52 #include <linux/hugetlb.h>
  53
  54 /* #define EXIT_DEBUG */
  55 /* #define EXIT_DEBUG_SIMPLE */
  56 /* #define EXIT_DEBUG_INT */
  57
  58 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
  59 static int kvmppc_hv_setup_rma(struct kvm_vcpu *vcpu);
  60
  61 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
  62 {
  63         struct kvmppc_vcore *vc = vcpu->arch.vcore;
  64
  65         local_paca->kvm_hstate.kvm_vcpu = vcpu;
  66         local_paca->kvm_hstate.kvm_vcore = vc;
  67         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
  68                 vc->stolen_tb += mftb() - vc->preempt_tb;
  69 }
  70
  71 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
  72 {
  73         struct kvmppc_vcore *vc = vcpu->arch.vcore;
  74
  75         if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
  76                 vc->preempt_tb = mftb();
  77 }
  78
  79 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
  80 {
  81         vcpu->arch.shregs.msr = msr;
  82         kvmppc_end_cede(vcpu);
  83 }
  84
  85 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
  86 {
  87         vcpu->arch.pvr = pvr;
  88 }
  89
  90 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
  91 {
  92         int r;
  93
  94         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
  95         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
  96                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
  97         for (r = 0; r < 16; ++r)
  98                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
  99                        r, kvmppc_get_gpr(vcpu, r),
 100                        r+16, kvmppc_get_gpr(vcpu, r+16));
 101         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 102                vcpu->arch.ctr, vcpu->arch.lr);
 103         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 104                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 105         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 106                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 107         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 108                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 109         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 110                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 111         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 112         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 113                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 114         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 115         for (r = 0; r < vcpu->arch.slb_max; ++r)
 116                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 117                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 118         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 119                vcpu->kvm->arch.lpcr, vcpu->kvm->arch.sdr1,
 120                vcpu->arch.last_inst);
 121 }
 122
 123 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 124 {
 125         int r;
 126         struct kvm_vcpu *v, *ret = NULL;
 127
 128         mutex_lock(&kvm->lock);
 129         kvm_for_each_vcpu(r, v, kvm) {
 130                 if (v->vcpu_id == id) {
 131                         ret = v;
 132                         break;
 133                 }
 134         }
 135         mutex_unlock(&kvm->lock);
 136         return ret;
 137 }
 138
 139 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 140 {
 141         vpa->shared_proc = 1;
 142         vpa->yield_count = 1;
 143 }
 144
 145 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 146 struct reg_vpa {
 147         u32 dummy;
 148         union {
 149                 u16 hword;
 150                 u32 word;
 151         } length;
 152 };
 153
 154 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 155 {
 156         if (vpap->update_pending)
 157                 return vpap->next_gpa != 0;
 158         return vpap->pinned_addr != NULL;
 159 }
 160
 161 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 162                                        unsigned long flags,
 163                                        unsigned long vcpuid, unsigned long vpa)
 164 {
 165         struct kvm *kvm = vcpu->kvm;
 166         unsigned long len, nb;
 167         void *va;
 168         struct kvm_vcpu *tvcpu;
 169         int err;
 170         int subfunc;
 171         struct kvmppc_vpa *vpap;
 172
 173         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 174         if (!tvcpu)
 175                 return H_PARAMETER;
 176
 177         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 178         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 179             subfunc == H_VPA_REG_SLB) {
 180                 /* Registering new area - address must be cache-line aligned */
 181                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 182                         return H_PARAMETER;
 183
 184                 /* convert logical addr to kernel addr and read length */
 185                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 186                 if (va == NULL)
 187                         return H_PARAMETER;
 188                 if (subfunc == H_VPA_REG_VPA)
 189                         len = ((struct reg_vpa *)va)->length.hword;
 190                 else
 191                         len = ((struct reg_vpa *)va)->length.word;
 192                 kvmppc_unpin_guest_page(kvm, va);
 193
 194                 /* Check length */
 195                 if (len > nb || len < sizeof(struct reg_vpa))
 196                         return H_PARAMETER;
 197         } else {
 198                 vpa = 0;
 199                 len = 0;
 200         }
 201
 202         err = H_PARAMETER;
 203         vpap = NULL;
 204         spin_lock(&tvcpu->arch.vpa_update_lock);
 205
 206         switch (subfunc) {
 207         case H_VPA_REG_VPA:             /* register VPA */
 208                 if (len < sizeof(struct lppaca))
 209                         break;
 210                 vpap = &tvcpu->arch.vpa;
 211                 err = 0;
 212                 break;
 213
 214         case H_VPA_REG_DTL:             /* register DTL */
 215                 if (len < sizeof(struct dtl_entry))
 216                         break;
 217                 len -= len % sizeof(struct dtl_entry);
 218
 219                 /* Check that they have previously registered a VPA */
 220                 err = H_RESOURCE;
 221                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 222                         break;
 223
 224                 vpap = &tvcpu->arch.dtl;
 225                 err = 0;
 226                 break;
 227
 228         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 229                 /* Check that they have previously registered a VPA */
 230                 err = H_RESOURCE;
 231                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 232                         break;
 233
 234                 vpap = &tvcpu->arch.slb_shadow;
 235                 err = 0;
 236                 break;
 237
 238         case H_VPA_DEREG_VPA:           /* deregister VPA */
 239                 /* Check they don't still have a DTL or SLB buf registered */
 240                 err = H_RESOURCE;
 241                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 242                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 243                         break;
 244
 245                 vpap = &tvcpu->arch.vpa;
 246                 err = 0;
 247                 break;
 248
 249         case H_VPA_DEREG_DTL:           /* deregister DTL */
 250                 vpap = &tvcpu->arch.dtl;
 251                 err = 0;
 252                 break;
 253
 254         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 255                 vpap = &tvcpu->arch.slb_shadow;
 256                 err = 0;
 257                 break;
 258         }
 259
 260         if (vpap) {
 261                 vpap->next_gpa = vpa;
 262                 vpap->len = len;
 263                 vpap->update_pending = 1;
 264         }
 265
 266         spin_unlock(&tvcpu->arch.vpa_update_lock);
 267
 268         return err;
 269 }
 270
 271 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 272 {
 273         struct kvm *kvm = vcpu->kvm;
 274         void *va;
 275         unsigned long nb;
 276         unsigned long gpa;
 277
 278         /*
 279          * We need to pin the page pointed to by vpap->next_gpa,
 280          * but we can't call kvmppc_pin_guest_page under the lock
 281          * as it does get_user_pages() and down_read().  So we
 282          * have to drop the lock, pin the page, then get the lock
 283          * again and check that a new area didn't get registered
 284          * in the meantime.
 285          */
 286         for (;;) {
 287                 gpa = vpap->next_gpa;
 288                 spin_unlock(&vcpu->arch.vpa_update_lock);
 289                 va = NULL;
 290                 nb = 0;
 291                 if (gpa)
 292                         va = kvmppc_pin_guest_page(kvm, vpap->next_gpa, &nb);
 293                 spin_lock(&vcpu->arch.vpa_update_lock);
 294                 if (gpa == vpap->next_gpa)
 295                         break;
 296                 /* sigh... unpin that one and try again */
 297                 if (va)
 298                         kvmppc_unpin_guest_page(kvm, va);
 299         }
 300
 301         vpap->update_pending = 0;
 302         if (va && nb < vpap->len) {
 303                 /*
 304                  * If it's now too short, it must be that userspace
 305                  * has changed the mappings underlying guest memory,
 306                  * so unregister the region.
 307                  */
 308                 kvmppc_unpin_guest_page(kvm, va);
 309                 va = NULL;
 310         }
 311         if (vpap->pinned_addr)
 312                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr);
 313         vpap->pinned_addr = va;
 314         if (va)
 315                 vpap->pinned_end = va + vpap->len;
 316 }
 317
 318 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 319 {
 320         spin_lock(&vcpu->arch.vpa_update_lock);
 321         if (vcpu->arch.vpa.update_pending) {
 322                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 323                 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 324         }
 325         if (vcpu->arch.dtl.update_pending) {
 326                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 327                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 328                 vcpu->arch.dtl_index = 0;
 329         }
 330         if (vcpu->arch.slb_shadow.update_pending)
 331                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 332         spin_unlock(&vcpu->arch.vpa_update_lock);
 333 }
 334
 335 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 336                                     struct kvmppc_vcore *vc)
 337 {
 338         struct dtl_entry *dt;
 339         struct lppaca *vpa;
 340         unsigned long old_stolen;
 341
 342         dt = vcpu->arch.dtl_ptr;
 343         vpa = vcpu->arch.vpa.pinned_addr;
 344         old_stolen = vcpu->arch.stolen_logged;
 345         vcpu->arch.stolen_logged = vc->stolen_tb;
 346         if (!dt || !vpa)
 347                 return;
 348         memset(dt, 0, sizeof(struct dtl_entry));
 349         dt->dispatch_reason = 7;
 350         dt->processor_id = vc->pcpu + vcpu->arch.ptid;
 351         dt->timebase = mftb();
 352         dt->enqueue_to_dispatch_time = vc->stolen_tb - old_stolen;
 353         dt->srr0 = kvmppc_get_pc(vcpu);
 354         dt->srr1 = vcpu->arch.shregs.msr;
 355         ++dt;
 356         if (dt == vcpu->arch.dtl.pinned_end)
 357                 dt = vcpu->arch.dtl.pinned_addr;
 358         vcpu->arch.dtl_ptr = dt;
 359         /* order writing *dt vs. writing vpa->dtl_idx */
 360         smp_wmb();
 361         vpa->dtl_idx = ++vcpu->arch.dtl_index;
 362 }
 363
 364 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 365 {
 366         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 367         unsigned long target, ret = H_SUCCESS;
 368         struct kvm_vcpu *tvcpu;
 369
 370         switch (req) {
 371         case H_ENTER:
 372                 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
 373                                               kvmppc_get_gpr(vcpu, 5),
 374                                               kvmppc_get_gpr(vcpu, 6),
 375                                               kvmppc_get_gpr(vcpu, 7));
 376                 break;
 377         case H_CEDE:
 378                 break;
 379         case H_PROD:
 380                 target = kvmppc_get_gpr(vcpu, 4);
 381                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 382                 if (!tvcpu) {
 383                         ret = H_PARAMETER;
 384                         break;
 385                 }
 386                 tvcpu->arch.prodded = 1;
 387                 smp_mb();
 388                 if (vcpu->arch.ceded) {
 389                         if (waitqueue_active(&vcpu->wq)) {
 390                                 wake_up_interruptible(&vcpu->wq);
 391                                 vcpu->stat.halt_wakeup++;
 392                         }
 393                 }
 394                 break;
 395         case H_CONFER:
 396                 break;
 397         case H_REGISTER_VPA:
 398                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 399                                         kvmppc_get_gpr(vcpu, 5),
 400                                         kvmppc_get_gpr(vcpu, 6));
 401                 break;
 402         default:
 403                 return RESUME_HOST;
 404         }
 405         kvmppc_set_gpr(vcpu, 3, ret);
 406         vcpu->arch.hcall_needed = 0;
 407         return RESUME_GUEST;
 408 }
 409
 410 static int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 411                               struct task_struct *tsk)
 412 {
 413         int r = RESUME_HOST;
 414
 415         vcpu->stat.sum_exits++;
 416
 417         run->exit_reason = KVM_EXIT_UNKNOWN;
 418         run->ready_for_interrupt_injection = 1;
 419         switch (vcpu->arch.trap) {
 420         /* We're good on these - the host merely wanted to get our attention */
 421         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 422                 vcpu->stat.dec_exits++;
 423                 r = RESUME_GUEST;
 424                 break;
 425         case BOOK3S_INTERRUPT_EXTERNAL:
 426                 vcpu->stat.ext_intr_exits++;
 427                 r = RESUME_GUEST;
 428                 break;
 429         case BOOK3S_INTERRUPT_PERFMON:
 430                 r = RESUME_GUEST;
 431                 break;
 432         case BOOK3S_INTERRUPT_PROGRAM:
 433         {
 434                 ulong flags;
 435                 /*
 436                  * Normally program interrupts are delivered directly
 437                  * to the guest by the hardware, but we can get here
 438                  * as a result of a hypervisor emulation interrupt
 439                  * (e40) getting turned into a 700 by BML RTAS.
 440                  */
 441                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 442                 kvmppc_core_queue_program(vcpu, flags);
 443                 r = RESUME_GUEST;
 444                 break;
 445         }
 446         case BOOK3S_INTERRUPT_SYSCALL:
 447         {
 448                 /* hcall - punt to userspace */
 449                 int i;
 450
 451                 if (vcpu->arch.shregs.msr & MSR_PR) {
 452                         /* sc 1 from userspace - reflect to guest syscall */
 453                         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_SYSCALL);
 454                         r = RESUME_GUEST;
 455                         break;
 456                 }
 457                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 458                 for (i = 0; i < 9; ++i)
 459                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 460                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 461                 vcpu->arch.hcall_needed = 1;
 462                 r = RESUME_HOST;
 463                 break;
 464         }
 465         /*
 466          * We get these next two if the guest accesses a page which it thinks
 467          * it has mapped but which is not actually present, either because
 468          * it is for an emulated I/O device or because the corresonding
 469          * host page has been paged out.  Any other HDSI/HISI interrupts
 470          * have been handled already.
 471          */
 472         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 473                 r = kvmppc_book3s_hv_page_fault(run, vcpu,
 474                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 475                 break;
 476         case BOOK3S_INTERRUPT_H_INST_STORAGE:
 477                 r = kvmppc_book3s_hv_page_fault(run, vcpu,
 478                                 kvmppc_get_pc(vcpu), 0);
 479                 break;
 480         /*
 481          * This occurs if the guest executes an illegal instruction.
 482          * We just generate a program interrupt to the guest, since
 483          * we don't emulate any guest instructions at this stage.
 484          */
 485         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 486                 kvmppc_core_queue_program(vcpu, 0x80000);
 487                 r = RESUME_GUEST;
 488                 break;
 489         default:
 490                 kvmppc_dump_regs(vcpu);
 491                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 492                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 493                         vcpu->arch.shregs.msr);
 494                 r = RESUME_HOST;
 495                 BUG();
 496                 break;
 497         }
 498
 499         return r;
 500 }
 501
 502 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 503                                   struct kvm_sregs *sregs)
 504 {
 505         int i;
 506
 507         sregs->pvr = vcpu->arch.pvr;
 508
 509         memset(sregs, 0, sizeof(struct kvm_sregs));
 510         for (i = 0; i < vcpu->arch.slb_max; i++) {
 511                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
 512                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 513         }
 514
 515         return 0;
 516 }
 517
 518 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 519                                   struct kvm_sregs *sregs)
 520 {
 521         int i, j;
 522
 523         kvmppc_set_pvr(vcpu, sregs->pvr);
 524
 525         j = 0;
 526         for (i = 0; i < vcpu->arch.slb_nr; i++) {
 527                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
 528                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
 529                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
 530                         ++j;
 531                 }
 532         }
 533         vcpu->arch.slb_max = j;
 534
 535         return 0;
 536 }
 537
 538 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
 539 {
 540         int r = -EINVAL;
 541
 542         switch (reg->id) {
 543         case KVM_REG_PPC_HIOR:
 544                 r = put_user(0, (u64 __user *)reg->addr);
 545                 break;
 546         default:
 547                 break;
 548         }
 549
 550         return r;
 551 }
 552
 553 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
 554 {
 555         int r = -EINVAL;
 556
 557         switch (reg->id) {
 558         case KVM_REG_PPC_HIOR:
 559         {
 560                 u64 hior;
 561                 /* Only allow this to be set to zero */
 562                 r = get_user(hior, (u64 __user *)reg->addr);
 563                 if (!r && (hior != 0))
 564                         r = -EINVAL;
 565                 break;
 566         }
 567         default:
 568                 break;
 569         }
 570
 571         return r;
 572 }
 573
 574 int kvmppc_core_check_processor_compat(void)
 575 {
 576         if (cpu_has_feature(CPU_FTR_HVMODE))
 577                 return 0;
 578         return -EIO;
 579 }
 580
 581 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
 582 {
 583         struct kvm_vcpu *vcpu;
 584         int err = -EINVAL;
 585         int core;
 586         struct kvmppc_vcore *vcore;
 587
 588         core = id / threads_per_core;
 589         if (core >= KVM_MAX_VCORES)
 590                 goto out;
 591
 592         err = -ENOMEM;
 593         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 594         if (!vcpu)
 595                 goto out;
 596
 597         err = kvm_vcpu_init(vcpu, kvm, id);
 598         if (err)
 599                 goto free_vcpu;
 600
 601         vcpu->arch.shared = &vcpu->arch.shregs;
 602         vcpu->arch.last_cpu = -1;
 603         vcpu->arch.mmcr[0] = MMCR0_FC;
 604         vcpu->arch.ctrl = CTRL_RUNLATCH;
 605         /* default to host PVR, since we can't spoof it */
 606         vcpu->arch.pvr = mfspr(SPRN_PVR);
 607         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
 608         spin_lock_init(&vcpu->arch.vpa_update_lock);
 609
 610         kvmppc_mmu_book3s_hv_init(vcpu);
 611
 612         /*
 613          * We consider the vcpu stopped until we see the first run ioctl for it.
 614          */
 615         vcpu->arch.state = KVMPPC_VCPU_STOPPED;
 616
 617         init_waitqueue_head(&vcpu->arch.cpu_run);
 618
 619         mutex_lock(&kvm->lock);
 620         vcore = kvm->arch.vcores[core];
 621         if (!vcore) {
 622                 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
 623                 if (vcore) {
 624                         INIT_LIST_HEAD(&vcore->runnable_threads);
 625                         spin_lock_init(&vcore->lock);
 626                         init_waitqueue_head(&vcore->wq);
 627                         vcore->preempt_tb = mftb();
 628                 }
 629                 kvm->arch.vcores[core] = vcore;
 630         }
 631         mutex_unlock(&kvm->lock);
 632
 633         if (!vcore)
 634                 goto free_vcpu;
 635
 636         spin_lock(&vcore->lock);
 637         ++vcore->num_threads;
 638         spin_unlock(&vcore->lock);
 639         vcpu->arch.vcore = vcore;
 640         vcpu->arch.stolen_logged = vcore->stolen_tb;
 641
 642         vcpu->arch.cpu_type = KVM_CPU_3S_64;
 643         kvmppc_sanity_check(vcpu);
 644
 645         return vcpu;
 646
 647 free_vcpu:
 648         kmem_cache_free(kvm_vcpu_cache, vcpu);
 649 out:
 650         return ERR_PTR(err);
 651 }
 652
 653 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
 654 {
 655         spin_lock(&vcpu->arch.vpa_update_lock);
 656         if (vcpu->arch.dtl.pinned_addr)
 657                 kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.dtl.pinned_addr);
 658         if (vcpu->arch.slb_shadow.pinned_addr)
 659                 kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.slb_shadow.pinned_addr);
 660         if (vcpu->arch.vpa.pinned_addr)
 661                 kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.vpa.pinned_addr);
 662         spin_unlock(&vcpu->arch.vpa_update_lock);
 663         kvm_vcpu_uninit(vcpu);
 664         kmem_cache_free(kvm_vcpu_cache, vcpu);
 665 }
 666
 667 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
 668 {
 669         unsigned long dec_nsec, now;
 670
 671         now = get_tb();
 672         if (now > vcpu->arch.dec_expires) {
 673                 /* decrementer has already gone negative */
 674                 kvmppc_core_queue_dec(vcpu);
 675                 kvmppc_core_prepare_to_enter(vcpu);
 676                 return;
 677         }
 678         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
 679                    / tb_ticks_per_sec;
 680         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
 681                       HRTIMER_MODE_REL);
 682         vcpu->arch.timer_running = 1;
 683 }
 684
 685 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
 686 {
 687         vcpu->arch.ceded = 0;
 688         if (vcpu->arch.timer_running) {
 689                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
 690                 vcpu->arch.timer_running = 0;
 691         }
 692 }
 693
 694 extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
 695 extern void xics_wake_cpu(int cpu);
 696
 697 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
 698                                    struct kvm_vcpu *vcpu)
 699 {
 700         struct kvm_vcpu *v;
 701
 702         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
 703                 return;
 704         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
 705         --vc->n_runnable;
 706         ++vc->n_busy;
 707         /* decrement the physical thread id of each following vcpu */
 708         v = vcpu;
 709         list_for_each_entry_continue(v, &vc->runnable_threads, arch.run_list)
 710                 --v->arch.ptid;
 711         list_del(&vcpu->arch.run_list);
 712 }
 713
 714 static int kvmppc_grab_hwthread(int cpu)
 715 {
 716         struct paca_struct *tpaca;
 717         long timeout = 1000;
 718
 719         tpaca = &paca[cpu];
 720
 721         /* Ensure the thread won't go into the kernel if it wakes */
 722         tpaca->kvm_hstate.hwthread_req = 1;
 723
 724         /*
 725          * If the thread is already executing in the kernel (e.g. handling
 726          * a stray interrupt), wait for it to get back to nap mode.
 727          * The smp_mb() is to ensure that our setting of hwthread_req
 728          * is visible before we look at hwthread_state, so if this
 729          * races with the code at system_reset_pSeries and the thread
 730          * misses our setting of hwthread_req, we are sure to see its
 731          * setting of hwthread_state, and vice versa.
 732          */
 733         smp_mb();
 734         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
 735                 if (--timeout <= 0) {
 736                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
 737                         return -EBUSY;
 738                 }
 739                 udelay(1);
 740         }
 741         return 0;
 742 }
 743
 744 static void kvmppc_release_hwthread(int cpu)
 745 {
 746         struct paca_struct *tpaca;
 747
 748         tpaca = &paca[cpu];
 749         tpaca->kvm_hstate.hwthread_req = 0;
 750         tpaca->kvm_hstate.kvm_vcpu = NULL;
 751 }
 752
 753 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
 754 {
 755         int cpu;
 756         struct paca_struct *tpaca;
 757         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 758
 759         if (vcpu->arch.timer_running) {
 760                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
 761                 vcpu->arch.timer_running = 0;
 762         }
 763         cpu = vc->pcpu + vcpu->arch.ptid;
 764         tpaca = &paca[cpu];
 765         tpaca->kvm_hstate.kvm_vcpu = vcpu;
 766         tpaca->kvm_hstate.kvm_vcore = vc;
 767         tpaca->kvm_hstate.napping = 0;
 768         vcpu->cpu = vc->pcpu;
 769         smp_wmb();
 770 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
 771         if (vcpu->arch.ptid) {
 772                 kvmppc_grab_hwthread(cpu);
 773                 xics_wake_cpu(cpu);
 774                 ++vc->n_woken;
 775         }
 776 #endif
 777 }
 778
 779 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
 780 {
 781         int i;
 782
 783         HMT_low();
 784         i = 0;
 785         while (vc->nap_count < vc->n_woken) {
 786                 if (++i >= 1000000) {
 787                         pr_err("kvmppc_wait_for_nap timeout %d %d\n",
 788                                vc->nap_count, vc->n_woken);
 789                         break;
 790                 }
 791                 cpu_relax();
 792         }
 793         HMT_medium();
 794 }
 795
 796 /*
 797  * Check that we are on thread 0 and that any other threads in
 798  * this core are off-line.
 799  */
 800 static int on_primary_thread(void)
 801 {
 802         int cpu = smp_processor_id();
 803         int thr = cpu_thread_in_core(cpu);
 804
 805         if (thr)
 806                 return 0;
 807         while (++thr < threads_per_core)
 808                 if (cpu_online(cpu + thr))
 809                         return 0;
 810         return 1;
 811 }
 812
 813 /*
 814  * Run a set of guest threads on a physical core.
 815  * Called with vc->lock held.
 816  */
 817 static int kvmppc_run_core(struct kvmppc_vcore *vc)
 818 {
 819         struct kvm_vcpu *vcpu, *vcpu0, *vnext;
 820         long ret;
 821         u64 now;
 822         int ptid, i, need_vpa_update;
 823
 824         /* don't start if any threads have a signal pending */
 825         need_vpa_update = 0;
 826         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
 827                 if (signal_pending(vcpu->arch.run_task))
 828                         return 0;
 829                 need_vpa_update |= vcpu->arch.vpa.update_pending |
 830                         vcpu->arch.slb_shadow.update_pending |
 831                         vcpu->arch.dtl.update_pending;
 832         }
 833
 834         /*
 835          * Initialize *vc, in particular vc->vcore_state, so we can
 836          * drop the vcore lock if necessary.
 837          */
 838         vc->n_woken = 0;
 839         vc->nap_count = 0;
 840         vc->entry_exit_count = 0;
 841         vc->vcore_state = VCORE_RUNNING;
 842         vc->in_guest = 0;
 843         vc->napping_threads = 0;
 844
 845         /*
 846          * Updating any of the vpas requires calling kvmppc_pin_guest_page,
 847          * which can't be called with any spinlocks held.
 848          */
 849         if (need_vpa_update) {
 850                 spin_unlock(&vc->lock);
 851                 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
 852                         kvmppc_update_vpas(vcpu);
 853                 spin_lock(&vc->lock);
 854         }
 855
 856         /*
 857          * Make sure we are running on thread 0, and that
 858          * secondary threads are offline.
 859          * XXX we should also block attempts to bring any
 860          * secondary threads online.
 861          */
 862         if (threads_per_core > 1 && !on_primary_thread()) {
 863                 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
 864                         vcpu->arch.ret = -EBUSY;
 865                 goto out;
 866         }
 867
 868         /*
 869          * Assign physical thread IDs, first to non-ceded vcpus
 870          * and then to ceded ones.
 871          */
 872         ptid = 0;
 873         vcpu0 = NULL;
 874         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
 875                 if (!vcpu->arch.ceded) {
 876                         if (!ptid)
 877                                 vcpu0 = vcpu;
 878                         vcpu->arch.ptid = ptid++;
 879                 }
 880         }
 881         if (!vcpu0)
 882                 return 0;               /* nothing to run */
 883         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
 884                 if (vcpu->arch.ceded)
 885                         vcpu->arch.ptid = ptid++;
 886
 887         vc->stolen_tb += mftb() - vc->preempt_tb;
 888         vc->pcpu = smp_processor_id();
 889         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
 890                 kvmppc_start_thread(vcpu);
 891                 kvmppc_create_dtl_entry(vcpu, vc);
 892         }
 893         /* Grab any remaining hw threads so they can't go into the kernel */
 894         for (i = ptid; i < threads_per_core; ++i)
 895                 kvmppc_grab_hwthread(vc->pcpu + i);
 896
 897         preempt_disable();
 898         spin_unlock(&vc->lock);
 899
 900         kvm_guest_enter();
 901         __kvmppc_vcore_entry(NULL, vcpu0);
 902         for (i = 0; i < threads_per_core; ++i)
 903                 kvmppc_release_hwthread(vc->pcpu + i);
 904
 905         spin_lock(&vc->lock);
 906         /* disable sending of IPIs on virtual external irqs */
 907         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
 908                 vcpu->cpu = -1;
 909         /* wait for secondary threads to finish writing their state to memory */
 910         if (vc->nap_count < vc->n_woken)
 911                 kvmppc_wait_for_nap(vc);
 912         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
 913         vc->vcore_state = VCORE_EXITING;
 914         spin_unlock(&vc->lock);
 915
 916         /* make sure updates to secondary vcpu structs are visible now */
 917         smp_mb();
 918         kvm_guest_exit();
 919
 920         preempt_enable();
 921         kvm_resched(vcpu);
 922
 923         now = get_tb();
 924         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
 925                 /* cancel pending dec exception if dec is positive */
 926                 if (now < vcpu->arch.dec_expires &&
 927                     kvmppc_core_pending_dec(vcpu))
 928                         kvmppc_core_dequeue_dec(vcpu);
 929
 930                 ret = RESUME_GUEST;
 931                 if (vcpu->arch.trap)
 932                         ret = kvmppc_handle_exit(vcpu->arch.kvm_run, vcpu,
 933                                                  vcpu->arch.run_task);
 934
 935                 vcpu->arch.ret = ret;
 936                 vcpu->arch.trap = 0;
 937
 938                 if (vcpu->arch.ceded) {
 939                         if (ret != RESUME_GUEST)
 940                                 kvmppc_end_cede(vcpu);
 941                         else
 942                                 kvmppc_set_timer(vcpu);
 943                 }
 944         }
 945
 946         spin_lock(&vc->lock);
 947  out:
 948         vc->vcore_state = VCORE_INACTIVE;
 949         vc->preempt_tb = mftb();
 950         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
 951                                  arch.run_list) {
 952                 if (vcpu->arch.ret != RESUME_GUEST) {
 953                         kvmppc_remove_runnable(vc, vcpu);
 954                         wake_up(&vcpu->arch.cpu_run);
 955                 }
 956         }
 957
 958         return 1;
 959 }
 960
 961 /*
 962  * Wait for some other vcpu thread to execute us, and
 963  * wake us up when we need to handle something in the host.
 964  */
 965 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
 966 {
 967         DEFINE_WAIT(wait);
 968
 969         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
 970         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
 971                 schedule();
 972         finish_wait(&vcpu->arch.cpu_run, &wait);
 973 }
 974
 975 /*
 976  * All the vcpus in this vcore are idle, so wait for a decrementer
 977  * or external interrupt to one of the vcpus.  vc->lock is held.
 978  */
 979 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
 980 {
 981         DEFINE_WAIT(wait);
 982         struct kvm_vcpu *v;
 983         int all_idle = 1;
 984
 985         prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
 986         vc->vcore_state = VCORE_SLEEPING;
 987         spin_unlock(&vc->lock);
 988         list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
 989                 if (!v->arch.ceded || v->arch.pending_exceptions) {
 990                         all_idle = 0;
 991                         break;
 992                 }
 993         }
 994         if (all_idle)
 995                 schedule();
 996         finish_wait(&vc->wq, &wait);
 997         spin_lock(&vc->lock);
 998         vc->vcore_state = VCORE_INACTIVE;
 999 }
1000
1001 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1002 {
1003         int n_ceded;
1004         int prev_state;
1005         struct kvmppc_vcore *vc;
1006         struct kvm_vcpu *v, *vn;
1007
1008         kvm_run->exit_reason = 0;
1009         vcpu->arch.ret = RESUME_GUEST;
1010         vcpu->arch.trap = 0;
1011
1012         /*
1013          * Synchronize with other threads in this virtual core
1014          */
1015         vc = vcpu->arch.vcore;
1016         spin_lock(&vc->lock);
1017         vcpu->arch.ceded = 0;
1018         vcpu->arch.run_task = current;
1019         vcpu->arch.kvm_run = kvm_run;
1020         prev_state = vcpu->arch.state;
1021         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1022         list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1023         ++vc->n_runnable;
1024
1025         /*
1026          * This happens the first time this is called for a vcpu.
1027          * If the vcore is already running, we may be able to start
1028          * this thread straight away and have it join in.
1029          */
1030         if (prev_state == KVMPPC_VCPU_STOPPED) {
1031                 if (vc->vcore_state == VCORE_RUNNING &&
1032                     VCORE_EXIT_COUNT(vc) == 0) {
1033                         vcpu->arch.ptid = vc->n_runnable - 1;
1034                         kvmppc_start_thread(vcpu);
1035                 }
1036
1037         } else if (prev_state == KVMPPC_VCPU_BUSY_IN_HOST)
1038                 --vc->n_busy;
1039
1040         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1041                !signal_pending(current)) {
1042                 if (vc->n_busy || vc->vcore_state != VCORE_INACTIVE) {
1043                         spin_unlock(&vc->lock);
1044                         kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1045                         spin_lock(&vc->lock);
1046                         continue;
1047                 }
1048                 vc->runner = vcpu;
1049                 n_ceded = 0;
1050                 list_for_each_entry(v, &vc->runnable_threads, arch.run_list)
1051                         n_ceded += v->arch.ceded;
1052                 if (n_ceded == vc->n_runnable)
1053                         kvmppc_vcore_blocked(vc);
1054                 else
1055                         kvmppc_run_core(vc);
1056
1057                 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1058                                          arch.run_list) {
1059                         kvmppc_core_prepare_to_enter(v);
1060                         if (signal_pending(v->arch.run_task)) {
1061                                 kvmppc_remove_runnable(vc, v);
1062                                 v->stat.signal_exits++;
1063                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1064                                 v->arch.ret = -EINTR;
1065                                 wake_up(&v->arch.cpu_run);
1066                         }
1067                 }
1068                 vc->runner = NULL;
1069         }
1070
1071         if (signal_pending(current)) {
1072                 if (vc->vcore_state == VCORE_RUNNING ||
1073                     vc->vcore_state == VCORE_EXITING) {
1074                         spin_unlock(&vc->lock);
1075                         kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1076                         spin_lock(&vc->lock);
1077                 }
1078                 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1079                         kvmppc_remove_runnable(vc, vcpu);
1080                         vcpu->stat.signal_exits++;
1081                         kvm_run->exit_reason = KVM_EXIT_INTR;
1082                         vcpu->arch.ret = -EINTR;
1083                 }
1084         }
1085
1086         spin_unlock(&vc->lock);
1087         return vcpu->arch.ret;
1088 }
1089
1090 int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
1091 {
1092         int r;
1093
1094         if (!vcpu->arch.sane) {
1095                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1096                 return -EINVAL;
1097         }
1098
1099         kvmppc_core_prepare_to_enter(vcpu);
1100
1101         /* No need to go into the guest when all we'll do is come back out */
1102         if (signal_pending(current)) {
1103                 run->exit_reason = KVM_EXIT_INTR;
1104                 return -EINTR;
1105         }
1106
1107         /* On the first time here, set up VRMA or RMA */
1108         if (!vcpu->kvm->arch.rma_setup_done) {
1109                 r = kvmppc_hv_setup_rma(vcpu);
1110                 if (r)
1111                         return r;
1112         }
1113
1114         flush_fp_to_thread(current);
1115         flush_altivec_to_thread(current);
1116         flush_vsx_to_thread(current);
1117         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1118         vcpu->arch.pgdir = current->mm->pgd;
1119
1120         do {
1121                 r = kvmppc_run_vcpu(run, vcpu);
1122
1123                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1124                     !(vcpu->arch.shregs.msr & MSR_PR)) {
1125                         r = kvmppc_pseries_do_hcall(vcpu);
1126                         kvmppc_core_prepare_to_enter(vcpu);
1127                 }
1128         } while (r == RESUME_GUEST);
1129         return r;
1130 }
1131
1132
1133 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1134    Assumes POWER7 or PPC970. */
1135 static inline int lpcr_rmls(unsigned long rma_size)
1136 {
1137         switch (rma_size) {
1138         case 32ul << 20:        /* 32 MB */
1139                 if (cpu_has_feature(CPU_FTR_ARCH_206))
1140                         return 8;       /* only supported on POWER7 */
1141                 return -1;
1142         case 64ul << 20:        /* 64 MB */
1143                 return 3;
1144         case 128ul << 20:       /* 128 MB */
1145                 return 7;
1146         case 256ul << 20:       /* 256 MB */
1147                 return 4;
1148         case 1ul << 30:         /* 1 GB */
1149                 return 2;
1150         case 16ul << 30:        /* 16 GB */
1151                 return 1;
1152         case 256ul << 30:       /* 256 GB */
1153                 return 0;
1154         default:
1155                 return -1;
1156         }
1157 }
1158
1159 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1160 {
1161         struct kvmppc_linear_info *ri = vma->vm_file->private_data;
1162         struct page *page;
1163
1164         if (vmf->pgoff >= ri->npages)
1165                 return VM_FAULT_SIGBUS;
1166
1167         page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1168         get_page(page);
1169         vmf->page = page;
1170         return 0;
1171 }
1172
1173 static const struct vm_operations_struct kvm_rma_vm_ops = {
1174         .fault = kvm_rma_fault,
1175 };
1176
1177 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1178 {
1179         vma->vm_flags |= VM_RESERVED;
1180         vma->vm_ops = &kvm_rma_vm_ops;
1181         return 0;
1182 }
1183
1184 static int kvm_rma_release(struct inode *inode, struct file *filp)
1185 {
1186         struct kvmppc_linear_info *ri = filp->private_data;
1187
1188         kvm_release_rma(ri);
1189         return 0;
1190 }
1191
1192 static struct file_operations kvm_rma_fops = {
1193         .mmap           = kvm_rma_mmap,
1194         .release        = kvm_rma_release,
1195 };
1196
1197 long kvm_vm_ioctl_allocate_rma(struct kvm *kvm, struct kvm_allocate_rma *ret)
1198 {
1199         struct kvmppc_linear_info *ri;
1200         long fd;
1201
1202         ri = kvm_alloc_rma();
1203         if (!ri)
1204                 return -ENOMEM;
1205
1206         fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR);
1207         if (fd < 0)
1208                 kvm_release_rma(ri);
1209
1210         ret->rma_size = ri->npages << PAGE_SHIFT;
1211         return fd;
1212 }
1213
1214 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1215                                      int linux_psize)
1216 {
1217         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1218
1219         if (!def->shift)
1220                 return;
1221         (*sps)->page_shift = def->shift;
1222         (*sps)->slb_enc = def->sllp;
1223         (*sps)->enc[0].page_shift = def->shift;
1224         (*sps)->enc[0].pte_enc = def->penc;
1225         (*sps)++;
1226 }
1227
1228 int kvm_vm_ioctl_get_smmu_info(struct kvm *kvm, struct kvm_ppc_smmu_info *info)
1229 {
1230         struct kvm_ppc_one_seg_page_size *sps;
1231
1232         info->flags = KVM_PPC_PAGE_SIZES_REAL;
1233         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1234                 info->flags |= KVM_PPC_1T_SEGMENTS;
1235         info->slb_size = mmu_slb_size;
1236
1237         /* We only support these sizes for now, and no muti-size segments */
1238         sps = &info->sps[0];
1239         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1240         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1241         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1242
1243         return 0;
1244 }
1245
1246 /*
1247  * Get (and clear) the dirty memory log for a memory slot.
1248  */
1249 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1250 {
1251         struct kvm_memory_slot *memslot;
1252         int r;
1253         unsigned long n;
1254
1255         mutex_lock(&kvm->slots_lock);
1256
1257         r = -EINVAL;
1258         if (log->slot >= KVM_MEMORY_SLOTS)
1259                 goto out;
1260
1261         memslot = id_to_memslot(kvm->memslots, log->slot);
1262         r = -ENOENT;
1263         if (!memslot->dirty_bitmap)
1264                 goto out;
1265
1266         n = kvm_dirty_bitmap_bytes(memslot);
1267         memset(memslot->dirty_bitmap, 0, n);
1268
1269         r = kvmppc_hv_get_dirty_log(kvm, memslot);
1270         if (r)
1271                 goto out;
1272
1273         r = -EFAULT;
1274         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1275                 goto out;
1276
1277         r = 0;
1278 out:
1279         mutex_unlock(&kvm->slots_lock);
1280         return r;
1281 }
1282
1283 static unsigned long slb_pgsize_encoding(unsigned long psize)
1284 {
1285         unsigned long senc = 0;
1286
1287         if (psize > 0x1000) {
1288                 senc = SLB_VSID_L;
1289                 if (psize == 0x10000)
1290                         senc |= SLB_VSID_LP_01;
1291         }
1292         return senc;
1293 }
1294
1295 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1296                                 struct kvm_userspace_memory_region *mem)
1297 {
1298         unsigned long npages;
1299         unsigned long *phys;
1300
1301         /* Allocate a slot_phys array */
1302         phys = kvm->arch.slot_phys[mem->slot];
1303         if (!kvm->arch.using_mmu_notifiers && !phys) {
1304                 npages = mem->memory_size >> PAGE_SHIFT;
1305                 phys = vzalloc(npages * sizeof(unsigned long));
1306                 if (!phys)
1307                         return -ENOMEM;
1308                 kvm->arch.slot_phys[mem->slot] = phys;
1309                 kvm->arch.slot_npages[mem->slot] = npages;
1310         }
1311
1312         return 0;
1313 }
1314
1315 static void unpin_slot(struct kvm *kvm, int slot_id)
1316 {
1317         unsigned long *physp;
1318         unsigned long j, npages, pfn;
1319         struct page *page;
1320
1321         physp = kvm->arch.slot_phys[slot_id];
1322         npages = kvm->arch.slot_npages[slot_id];
1323         if (physp) {
1324                 spin_lock(&kvm->arch.slot_phys_lock);
1325                 for (j = 0; j < npages; j++) {
1326                         if (!(physp[j] & KVMPPC_GOT_PAGE))
1327                                 continue;
1328                         pfn = physp[j] >> PAGE_SHIFT;
1329                         page = pfn_to_page(pfn);
1330                         SetPageDirty(page);
1331                         put_page(page);
1332                 }
1333                 kvm->arch.slot_phys[slot_id] = NULL;
1334                 spin_unlock(&kvm->arch.slot_phys_lock);
1335                 vfree(physp);
1336         }
1337 }
1338
1339 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1340                                 struct kvm_userspace_memory_region *mem)
1341 {
1342 }
1343
1344 static int kvmppc_hv_setup_rma(struct kvm_vcpu *vcpu)
1345 {
1346         int err = 0;
1347         struct kvm *kvm = vcpu->kvm;
1348         struct kvmppc_linear_info *ri = NULL;
1349         unsigned long hva;
1350         struct kvm_memory_slot *memslot;
1351         struct vm_area_struct *vma;
1352         unsigned long lpcr, senc;
1353         unsigned long psize, porder;
1354         unsigned long rma_size;
1355         unsigned long rmls;
1356         unsigned long *physp;
1357         unsigned long i, npages;
1358
1359         mutex_lock(&kvm->lock);
1360         if (kvm->arch.rma_setup_done)
1361                 goto out;       /* another vcpu beat us to it */
1362
1363         /* Look up the memslot for guest physical address 0 */
1364         memslot = gfn_to_memslot(kvm, 0);
1365
1366         /* We must have some memory at 0 by now */
1367         err = -EINVAL;
1368         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1369                 goto out;
1370
1371         /* Look up the VMA for the start of this memory slot */
1372         hva = memslot->userspace_addr;
1373         down_read(&current->mm->mmap_sem);
1374         vma = find_vma(current->mm, hva);
1375         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
1376                 goto up_out;
1377
1378         psize = vma_kernel_pagesize(vma);
1379         porder = __ilog2(psize);
1380
1381         /* Is this one of our preallocated RMAs? */
1382         if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
1383             hva == vma->vm_start)
1384                 ri = vma->vm_file->private_data;
1385
1386         up_read(&current->mm->mmap_sem);
1387
1388         if (!ri) {
1389                 /* On POWER7, use VRMA; on PPC970, give up */
1390                 err = -EPERM;
1391                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1392                         pr_err("KVM: CPU requires an RMO\n");
1393                         goto out;
1394                 }
1395
1396                 /* We can handle 4k, 64k or 16M pages in the VRMA */
1397                 err = -EINVAL;
1398                 if (!(psize == 0x1000 || psize == 0x10000 ||
1399                       psize == 0x1000000))
1400                         goto out;
1401
1402                 /* Update VRMASD field in the LPCR */
1403                 senc = slb_pgsize_encoding(psize);
1404                 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1405                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
1406                 lpcr = kvm->arch.lpcr & ~LPCR_VRMASD;
1407                 lpcr |= senc << (LPCR_VRMASD_SH - 4);
1408                 kvm->arch.lpcr = lpcr;
1409
1410                 /* Create HPTEs in the hash page table for the VRMA */
1411                 kvmppc_map_vrma(vcpu, memslot, porder);
1412
1413         } else {
1414                 /* Set up to use an RMO region */
1415                 rma_size = ri->npages;
1416                 if (rma_size > memslot->npages)
1417                         rma_size = memslot->npages;
1418                 rma_size <<= PAGE_SHIFT;
1419                 rmls = lpcr_rmls(rma_size);
1420                 err = -EINVAL;
1421                 if (rmls < 0) {
1422                         pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
1423                         goto out;
1424                 }
1425                 atomic_inc(&ri->use_count);
1426                 kvm->arch.rma = ri;
1427
1428                 /* Update LPCR and RMOR */
1429                 lpcr = kvm->arch.lpcr;
1430                 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1431                         /* PPC970; insert RMLS value (split field) in HID4 */
1432                         lpcr &= ~((1ul << HID4_RMLS0_SH) |
1433                                   (3ul << HID4_RMLS2_SH));
1434                         lpcr |= ((rmls >> 2) << HID4_RMLS0_SH) |
1435                                 ((rmls & 3) << HID4_RMLS2_SH);
1436                         /* RMOR is also in HID4 */
1437                         lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
1438                                 << HID4_RMOR_SH;
1439                 } else {
1440                         /* POWER7 */
1441                         lpcr &= ~(LPCR_VPM0 | LPCR_VRMA_L);
1442                         lpcr |= rmls << LPCR_RMLS_SH;
1443                         kvm->arch.rmor = kvm->arch.rma->base_pfn << PAGE_SHIFT;
1444                 }
1445                 kvm->arch.lpcr = lpcr;
1446                 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
1447                         ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
1448
1449                 /* Initialize phys addrs of pages in RMO */
1450                 npages = ri->npages;
1451                 porder = __ilog2(npages);
1452                 physp = kvm->arch.slot_phys[memslot->id];
1453                 spin_lock(&kvm->arch.slot_phys_lock);
1454                 for (i = 0; i < npages; ++i)
1455                         physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) + porder;
1456                 spin_unlock(&kvm->arch.slot_phys_lock);
1457         }
1458
1459         /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
1460         smp_wmb();
1461         kvm->arch.rma_setup_done = 1;
1462         err = 0;
1463  out:
1464         mutex_unlock(&kvm->lock);
1465         return err;
1466
1467  up_out:
1468         up_read(&current->mm->mmap_sem);
1469         goto out;
1470 }
1471
1472 int kvmppc_core_init_vm(struct kvm *kvm)
1473 {
1474         long r;
1475         unsigned long lpcr;
1476
1477         /* Allocate hashed page table */
1478         r = kvmppc_alloc_hpt(kvm);
1479         if (r)
1480                 return r;
1481
1482         INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
1483
1484         kvm->arch.rma = NULL;
1485
1486         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
1487
1488         if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1489                 /* PPC970; HID4 is effectively the LPCR */
1490                 unsigned long lpid = kvm->arch.lpid;
1491                 kvm->arch.host_lpid = 0;
1492                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
1493                 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
1494                 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
1495                         ((lpid & 0xf) << HID4_LPID5_SH);
1496         } else {
1497                 /* POWER7; init LPCR for virtual RMA mode */
1498                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
1499                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
1500                 lpcr &= LPCR_PECE | LPCR_LPES;
1501                 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
1502                         LPCR_VPM0 | LPCR_VPM1;
1503                 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
1504                         (VRMA_VSID << SLB_VSID_SHIFT_1T);
1505         }
1506         kvm->arch.lpcr = lpcr;
1507
1508         kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
1509         spin_lock_init(&kvm->arch.slot_phys_lock);
1510         return 0;
1511 }
1512
1513 void kvmppc_core_destroy_vm(struct kvm *kvm)
1514 {
1515         unsigned long i;
1516
1517         if (!kvm->arch.using_mmu_notifiers)
1518                 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
1519                         unpin_slot(kvm, i);
1520
1521         if (kvm->arch.rma) {
1522                 kvm_release_rma(kvm->arch.rma);
1523                 kvm->arch.rma = NULL;
1524         }
1525
1526         kvmppc_free_hpt(kvm);
1527         WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1528 }
1529
1530 /* These are stubs for now */
1531 void kvmppc_mmu_pte_pflush(struct kvm_vcpu *vcpu, ulong pa_start, ulong pa_end)
1532 {
1533 }
1534
1535 /* We don't need to emulate any privileged instructions or dcbz */
1536 int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu,
1537                            unsigned int inst, int *advance)
1538 {
1539         return EMULATE_FAIL;
1540 }
1541
1542 int kvmppc_core_emulate_mtspr(struct kvm_vcpu *vcpu, int sprn, ulong spr_val)
1543 {
1544         return EMULATE_FAIL;
1545 }
1546
1547 int kvmppc_core_emulate_mfspr(struct kvm_vcpu *vcpu, int sprn, ulong *spr_val)
1548 {
1549         return EMULATE_FAIL;
1550 }
1551
1552 static int kvmppc_book3s_hv_init(void)
1553 {
1554         int r;
1555
1556         r = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1557
1558         if (r)
1559                 return r;
1560
1561         r = kvmppc_mmu_hv_init();
1562
1563         return r;
1564 }
1565
1566 static void kvmppc_book3s_hv_exit(void)
1567 {
1568         kvm_exit();
1569 }
1570
1571 module_init(kvmppc_book3s_hv_init);
1572 module_exit(kvmppc_book3s_hv_exit);