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 #include <linux/srcu.h>
  34 #include <linux/miscdevice.h>
  35 #include <linux/debugfs.h>
  36
  37 #include <asm/reg.h>
  38 #include <asm/cputable.h>
  39 #include <asm/cache.h>
  40 #include <asm/cacheflush.h>
  41 #include <asm/tlbflush.h>
  42 #include <asm/uaccess.h>
  43 #include <asm/io.h>
  44 #include <asm/kvm_ppc.h>
  45 #include <asm/kvm_book3s.h>
  46 #include <asm/mmu_context.h>
  47 #include <asm/lppaca.h>
  48 #include <asm/processor.h>
  49 #include <asm/cputhreads.h>
  50 #include <asm/page.h>
  51 #include <asm/hvcall.h>
  52 #include <asm/switch_to.h>
  53 #include <asm/smp.h>
  54 #include <asm/dbell.h>
  55 #include <linux/gfp.h>
  56 #include <linux/vmalloc.h>
  57 #include <linux/highmem.h>
  58 #include <linux/hugetlb.h>
  59 #include <linux/module.h>
  60
  61 #include "book3s.h"
  62
  63 #define CREATE_TRACE_POINTS
  64 #include "trace_hv.h"
  65
  66 /* #define EXIT_DEBUG */
  67 /* #define EXIT_DEBUG_SIMPLE */
  68 /* #define EXIT_DEBUG_INT */
  69
  70 /* Used to indicate that a guest page fault needs to be handled */
  71 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  72
  73 /* Used as a "null" value for timebase values */
  74 #define TB_NIL  (~(u64)0)
  75
  76 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
  77
  78 #if defined(CONFIG_PPC_64K_PAGES)
  79 #define MPP_BUFFER_ORDER        0
  80 #elif defined(CONFIG_PPC_4K_PAGES)
  81 #define MPP_BUFFER_ORDER        3
  82 #endif
  83
  84 static int dynamic_mt_modes = 6;
  85 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
  86 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
  87 static int target_smt_mode;
  88 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
  89 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
  90
  91 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
  92 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
  93
  94 static bool kvmppc_ipi_thread(int cpu)
  95 {
  96         /* On POWER8 for IPIs to threads in the same core, use msgsnd */
  97         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
  98                 preempt_disable();
  99                 if (cpu_first_thread_sibling(cpu) ==
 100                     cpu_first_thread_sibling(smp_processor_id())) {
 101                         unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
 102                         msg |= cpu_thread_in_core(cpu);
 103                         smp_mb();
 104                         __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 105                         preempt_enable();
 106                         return true;
 107                 }
 108                 preempt_enable();
 109         }
 110
 111 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
 112         if (cpu >= 0 && cpu < nr_cpu_ids && paca[cpu].kvm_hstate.xics_phys) {
 113                 xics_wake_cpu(cpu);
 114                 return true;
 115         }
 116 #endif
 117
 118         return false;
 119 }
 120
 121 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
 122 {
 123         int cpu;
 124         wait_queue_head_t *wqp;
 125
 126         wqp = kvm_arch_vcpu_wq(vcpu);
 127         if (waitqueue_active(wqp)) {
 128                 wake_up_interruptible(wqp);
 129                 ++vcpu->stat.halt_wakeup;
 130         }
 131
 132         if (kvmppc_ipi_thread(vcpu->arch.thread_cpu))
 133                 return;
 134
 135         /* CPU points to the first thread of the core */
 136         cpu = vcpu->cpu;
 137         if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
 138                 smp_send_reschedule(cpu);
 139 }
 140
 141 /*
 142  * We use the vcpu_load/put functions to measure stolen time.
 143  * Stolen time is counted as time when either the vcpu is able to
 144  * run as part of a virtual core, but the task running the vcore
 145  * is preempted or sleeping, or when the vcpu needs something done
 146  * in the kernel by the task running the vcpu, but that task is
 147  * preempted or sleeping.  Those two things have to be counted
 148  * separately, since one of the vcpu tasks will take on the job
 149  * of running the core, and the other vcpu tasks in the vcore will
 150  * sleep waiting for it to do that, but that sleep shouldn't count
 151  * as stolen time.
 152  *
 153  * Hence we accumulate stolen time when the vcpu can run as part of
 154  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 155  * needs its task to do other things in the kernel (for example,
 156  * service a page fault) in busy_stolen.  We don't accumulate
 157  * stolen time for a vcore when it is inactive, or for a vcpu
 158  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 159  * a misnomer; it means that the vcpu task is not executing in
 160  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 161  * the kernel.  We don't have any way of dividing up that time
 162  * between time that the vcpu is genuinely stopped, time that
 163  * the task is actively working on behalf of the vcpu, and time
 164  * that the task is preempted, so we don't count any of it as
 165  * stolen.
 166  *
 167  * Updates to busy_stolen are protected by arch.tbacct_lock;
 168  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
 169  * lock.  The stolen times are measured in units of timebase ticks.
 170  * (Note that the != TB_NIL checks below are purely defensive;
 171  * they should never fail.)
 172  */
 173
 174 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
 175 {
 176         unsigned long flags;
 177
 178         spin_lock_irqsave(&vc->stoltb_lock, flags);
 179         vc->preempt_tb = mftb();
 180         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 181 }
 182
 183 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
 184 {
 185         unsigned long flags;
 186
 187         spin_lock_irqsave(&vc->stoltb_lock, flags);
 188         if (vc->preempt_tb != TB_NIL) {
 189                 vc->stolen_tb += mftb() - vc->preempt_tb;
 190                 vc->preempt_tb = TB_NIL;
 191         }
 192         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 193 }
 194
 195 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 196 {
 197         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 198         unsigned long flags;
 199
 200         /*
 201          * We can test vc->runner without taking the vcore lock,
 202          * because only this task ever sets vc->runner to this
 203          * vcpu, and once it is set to this vcpu, only this task
 204          * ever sets it to NULL.
 205          */
 206         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 207                 kvmppc_core_end_stolen(vc);
 208
 209         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 210         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 211             vcpu->arch.busy_preempt != TB_NIL) {
 212                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 213                 vcpu->arch.busy_preempt = TB_NIL;
 214         }
 215         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 216 }
 217
 218 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 219 {
 220         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 221         unsigned long flags;
 222
 223         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 224                 kvmppc_core_start_stolen(vc);
 225
 226         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 227         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 228                 vcpu->arch.busy_preempt = mftb();
 229         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 230 }
 231
 232 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
 233 {
 234         vcpu->arch.shregs.msr = msr;
 235         kvmppc_end_cede(vcpu);
 236 }
 237
 238 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 239 {
 240         vcpu->arch.pvr = pvr;
 241 }
 242
 243 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 244 {
 245         unsigned long pcr = 0;
 246         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 247
 248         if (arch_compat) {
 249                 switch (arch_compat) {
 250                 case PVR_ARCH_205:
 251                         /*
 252                          * If an arch bit is set in PCR, all the defined
 253                          * higher-order arch bits also have to be set.
 254                          */
 255                         pcr = PCR_ARCH_206 | PCR_ARCH_205;
 256                         break;
 257                 case PVR_ARCH_206:
 258                 case PVR_ARCH_206p:
 259                         pcr = PCR_ARCH_206;
 260                         break;
 261                 case PVR_ARCH_207:
 262                         break;
 263                 default:
 264                         return -EINVAL;
 265                 }
 266
 267                 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
 268                         /* POWER7 can't emulate POWER8 */
 269                         if (!(pcr & PCR_ARCH_206))
 270                                 return -EINVAL;
 271                         pcr &= ~PCR_ARCH_206;
 272                 }
 273         }
 274
 275         spin_lock(&vc->lock);
 276         vc->arch_compat = arch_compat;
 277         vc->pcr = pcr;
 278         spin_unlock(&vc->lock);
 279
 280         return 0;
 281 }
 282
 283 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 284 {
 285         int r;
 286
 287         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 288         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 289                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
 290         for (r = 0; r < 16; ++r)
 291                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 292                        r, kvmppc_get_gpr(vcpu, r),
 293                        r+16, kvmppc_get_gpr(vcpu, r+16));
 294         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 295                vcpu->arch.ctr, vcpu->arch.lr);
 296         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 297                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 298         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 299                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 300         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 301                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 302         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 303                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 304         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 305         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 306                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 307         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 308         for (r = 0; r < vcpu->arch.slb_max; ++r)
 309                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 310                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 311         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 312                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 313                vcpu->arch.last_inst);
 314 }
 315
 316 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 317 {
 318         int r;
 319         struct kvm_vcpu *v, *ret = NULL;
 320
 321         mutex_lock(&kvm->lock);
 322         kvm_for_each_vcpu(r, v, kvm) {
 323                 if (v->vcpu_id == id) {
 324                         ret = v;
 325                         break;
 326                 }
 327         }
 328         mutex_unlock(&kvm->lock);
 329         return ret;
 330 }
 331
 332 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 333 {
 334         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 335         vpa->yield_count = cpu_to_be32(1);
 336 }
 337
 338 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 339                    unsigned long addr, unsigned long len)
 340 {
 341         /* check address is cacheline aligned */
 342         if (addr & (L1_CACHE_BYTES - 1))
 343                 return -EINVAL;
 344         spin_lock(&vcpu->arch.vpa_update_lock);
 345         if (v->next_gpa != addr || v->len != len) {
 346                 v->next_gpa = addr;
 347                 v->len = addr ? len : 0;
 348                 v->update_pending = 1;
 349         }
 350         spin_unlock(&vcpu->arch.vpa_update_lock);
 351         return 0;
 352 }
 353
 354 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 355 struct reg_vpa {
 356         u32 dummy;
 357         union {
 358                 __be16 hword;
 359                 __be32 word;
 360         } length;
 361 };
 362
 363 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 364 {
 365         if (vpap->update_pending)
 366                 return vpap->next_gpa != 0;
 367         return vpap->pinned_addr != NULL;
 368 }
 369
 370 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 371                                        unsigned long flags,
 372                                        unsigned long vcpuid, unsigned long vpa)
 373 {
 374         struct kvm *kvm = vcpu->kvm;
 375         unsigned long len, nb;
 376         void *va;
 377         struct kvm_vcpu *tvcpu;
 378         int err;
 379         int subfunc;
 380         struct kvmppc_vpa *vpap;
 381
 382         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 383         if (!tvcpu)
 384                 return H_PARAMETER;
 385
 386         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 387         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 388             subfunc == H_VPA_REG_SLB) {
 389                 /* Registering new area - address must be cache-line aligned */
 390                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 391                         return H_PARAMETER;
 392
 393                 /* convert logical addr to kernel addr and read length */
 394                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 395                 if (va == NULL)
 396                         return H_PARAMETER;
 397                 if (subfunc == H_VPA_REG_VPA)
 398                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 399                 else
 400                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 401                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 402
 403                 /* Check length */
 404                 if (len > nb || len < sizeof(struct reg_vpa))
 405                         return H_PARAMETER;
 406         } else {
 407                 vpa = 0;
 408                 len = 0;
 409         }
 410
 411         err = H_PARAMETER;
 412         vpap = NULL;
 413         spin_lock(&tvcpu->arch.vpa_update_lock);
 414
 415         switch (subfunc) {
 416         case H_VPA_REG_VPA:             /* register VPA */
 417                 if (len < sizeof(struct lppaca))
 418                         break;
 419                 vpap = &tvcpu->arch.vpa;
 420                 err = 0;
 421                 break;
 422
 423         case H_VPA_REG_DTL:             /* register DTL */
 424                 if (len < sizeof(struct dtl_entry))
 425                         break;
 426                 len -= len % sizeof(struct dtl_entry);
 427
 428                 /* Check that they have previously registered a VPA */
 429                 err = H_RESOURCE;
 430                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 431                         break;
 432
 433                 vpap = &tvcpu->arch.dtl;
 434                 err = 0;
 435                 break;
 436
 437         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 438                 /* Check that they have previously registered a VPA */
 439                 err = H_RESOURCE;
 440                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 441                         break;
 442
 443                 vpap = &tvcpu->arch.slb_shadow;
 444                 err = 0;
 445                 break;
 446
 447         case H_VPA_DEREG_VPA:           /* deregister VPA */
 448                 /* Check they don't still have a DTL or SLB buf registered */
 449                 err = H_RESOURCE;
 450                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 451                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 452                         break;
 453
 454                 vpap = &tvcpu->arch.vpa;
 455                 err = 0;
 456                 break;
 457
 458         case H_VPA_DEREG_DTL:           /* deregister DTL */
 459                 vpap = &tvcpu->arch.dtl;
 460                 err = 0;
 461                 break;
 462
 463         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 464                 vpap = &tvcpu->arch.slb_shadow;
 465                 err = 0;
 466                 break;
 467         }
 468
 469         if (vpap) {
 470                 vpap->next_gpa = vpa;
 471                 vpap->len = len;
 472                 vpap->update_pending = 1;
 473         }
 474
 475         spin_unlock(&tvcpu->arch.vpa_update_lock);
 476
 477         return err;
 478 }
 479
 480 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 481 {
 482         struct kvm *kvm = vcpu->kvm;
 483         void *va;
 484         unsigned long nb;
 485         unsigned long gpa;
 486
 487         /*
 488          * We need to pin the page pointed to by vpap->next_gpa,
 489          * but we can't call kvmppc_pin_guest_page under the lock
 490          * as it does get_user_pages() and down_read().  So we
 491          * have to drop the lock, pin the page, then get the lock
 492          * again and check that a new area didn't get registered
 493          * in the meantime.
 494          */
 495         for (;;) {
 496                 gpa = vpap->next_gpa;
 497                 spin_unlock(&vcpu->arch.vpa_update_lock);
 498                 va = NULL;
 499                 nb = 0;
 500                 if (gpa)
 501                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 502                 spin_lock(&vcpu->arch.vpa_update_lock);
 503                 if (gpa == vpap->next_gpa)
 504                         break;
 505                 /* sigh... unpin that one and try again */
 506                 if (va)
 507                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 508         }
 509
 510         vpap->update_pending = 0;
 511         if (va && nb < vpap->len) {
 512                 /*
 513                  * If it's now too short, it must be that userspace
 514                  * has changed the mappings underlying guest memory,
 515                  * so unregister the region.
 516                  */
 517                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 518                 va = NULL;
 519         }
 520         if (vpap->pinned_addr)
 521                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 522                                         vpap->dirty);
 523         vpap->gpa = gpa;
 524         vpap->pinned_addr = va;
 525         vpap->dirty = false;
 526         if (va)
 527                 vpap->pinned_end = va + vpap->len;
 528 }
 529
 530 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 531 {
 532         if (!(vcpu->arch.vpa.update_pending ||
 533               vcpu->arch.slb_shadow.update_pending ||
 534               vcpu->arch.dtl.update_pending))
 535                 return;
 536
 537         spin_lock(&vcpu->arch.vpa_update_lock);
 538         if (vcpu->arch.vpa.update_pending) {
 539                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 540                 if (vcpu->arch.vpa.pinned_addr)
 541                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 542         }
 543         if (vcpu->arch.dtl.update_pending) {
 544                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 545                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 546                 vcpu->arch.dtl_index = 0;
 547         }
 548         if (vcpu->arch.slb_shadow.update_pending)
 549                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 550         spin_unlock(&vcpu->arch.vpa_update_lock);
 551 }
 552
 553 /*
 554  * Return the accumulated stolen time for the vcore up until `now'.
 555  * The caller should hold the vcore lock.
 556  */
 557 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 558 {
 559         u64 p;
 560         unsigned long flags;
 561
 562         spin_lock_irqsave(&vc->stoltb_lock, flags);
 563         p = vc->stolen_tb;
 564         if (vc->vcore_state != VCORE_INACTIVE &&
 565             vc->preempt_tb != TB_NIL)
 566                 p += now - vc->preempt_tb;
 567         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 568         return p;
 569 }
 570
 571 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 572                                     struct kvmppc_vcore *vc)
 573 {
 574         struct dtl_entry *dt;
 575         struct lppaca *vpa;
 576         unsigned long stolen;
 577         unsigned long core_stolen;
 578         u64 now;
 579
 580         dt = vcpu->arch.dtl_ptr;
 581         vpa = vcpu->arch.vpa.pinned_addr;
 582         now = mftb();
 583         core_stolen = vcore_stolen_time(vc, now);
 584         stolen = core_stolen - vcpu->arch.stolen_logged;
 585         vcpu->arch.stolen_logged = core_stolen;
 586         spin_lock_irq(&vcpu->arch.tbacct_lock);
 587         stolen += vcpu->arch.busy_stolen;
 588         vcpu->arch.busy_stolen = 0;
 589         spin_unlock_irq(&vcpu->arch.tbacct_lock);
 590         if (!dt || !vpa)
 591                 return;
 592         memset(dt, 0, sizeof(struct dtl_entry));
 593         dt->dispatch_reason = 7;
 594         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 595         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 596         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 597         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 598         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 599         ++dt;
 600         if (dt == vcpu->arch.dtl.pinned_end)
 601                 dt = vcpu->arch.dtl.pinned_addr;
 602         vcpu->arch.dtl_ptr = dt;
 603         /* order writing *dt vs. writing vpa->dtl_idx */
 604         smp_wmb();
 605         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 606         vcpu->arch.dtl.dirty = true;
 607 }
 608
 609 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 610 {
 611         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 612                 return true;
 613         if ((!vcpu->arch.vcore->arch_compat) &&
 614             cpu_has_feature(CPU_FTR_ARCH_207S))
 615                 return true;
 616         return false;
 617 }
 618
 619 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 620                              unsigned long resource, unsigned long value1,
 621                              unsigned long value2)
 622 {
 623         switch (resource) {
 624         case H_SET_MODE_RESOURCE_SET_CIABR:
 625                 if (!kvmppc_power8_compatible(vcpu))
 626                         return H_P2;
 627                 if (value2)
 628                         return H_P4;
 629                 if (mflags)
 630                         return H_UNSUPPORTED_FLAG_START;
 631                 /* Guests can't breakpoint the hypervisor */
 632                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 633                         return H_P3;
 634                 vcpu->arch.ciabr  = value1;
 635                 return H_SUCCESS;
 636         case H_SET_MODE_RESOURCE_SET_DAWR:
 637                 if (!kvmppc_power8_compatible(vcpu))
 638                         return H_P2;
 639                 if (mflags)
 640                         return H_UNSUPPORTED_FLAG_START;
 641                 if (value2 & DABRX_HYP)
 642                         return H_P4;
 643                 vcpu->arch.dawr  = value1;
 644                 vcpu->arch.dawrx = value2;
 645                 return H_SUCCESS;
 646         default:
 647                 return H_TOO_HARD;
 648         }
 649 }
 650
 651 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 652 {
 653         struct kvmppc_vcore *vcore = target->arch.vcore;
 654
 655         /*
 656          * We expect to have been called by the real mode handler
 657          * (kvmppc_rm_h_confer()) which would have directly returned
 658          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 659          * have useful work to do and should not confer) so we don't
 660          * recheck that here.
 661          */
 662
 663         spin_lock(&vcore->lock);
 664         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 665             vcore->vcore_state != VCORE_INACTIVE &&
 666             vcore->runner)
 667                 target = vcore->runner;
 668         spin_unlock(&vcore->lock);
 669
 670         return kvm_vcpu_yield_to(target);
 671 }
 672
 673 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 674 {
 675         int yield_count = 0;
 676         struct lppaca *lppaca;
 677
 678         spin_lock(&vcpu->arch.vpa_update_lock);
 679         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 680         if (lppaca)
 681                 yield_count = be32_to_cpu(lppaca->yield_count);
 682         spin_unlock(&vcpu->arch.vpa_update_lock);
 683         return yield_count;
 684 }
 685
 686 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 687 {
 688         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 689         unsigned long target, ret = H_SUCCESS;
 690         int yield_count;
 691         struct kvm_vcpu *tvcpu;
 692         int idx, rc;
 693
 694         if (req <= MAX_HCALL_OPCODE &&
 695             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 696                 return RESUME_HOST;
 697
 698         switch (req) {
 699         case H_CEDE:
 700                 break;
 701         case H_PROD:
 702                 target = kvmppc_get_gpr(vcpu, 4);
 703                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 704                 if (!tvcpu) {
 705                         ret = H_PARAMETER;
 706                         break;
 707                 }
 708                 tvcpu->arch.prodded = 1;
 709                 smp_mb();
 710                 if (vcpu->arch.ceded) {
 711                         if (waitqueue_active(&vcpu->wq)) {
 712                                 wake_up_interruptible(&vcpu->wq);
 713                                 vcpu->stat.halt_wakeup++;
 714                         }
 715                 }
 716                 break;
 717         case H_CONFER:
 718                 target = kvmppc_get_gpr(vcpu, 4);
 719                 if (target == -1)
 720                         break;
 721                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 722                 if (!tvcpu) {
 723                         ret = H_PARAMETER;
 724                         break;
 725                 }
 726                 yield_count = kvmppc_get_gpr(vcpu, 5);
 727                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 728                         break;
 729                 kvm_arch_vcpu_yield_to(tvcpu);
 730                 break;
 731         case H_REGISTER_VPA:
 732                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 733                                         kvmppc_get_gpr(vcpu, 5),
 734                                         kvmppc_get_gpr(vcpu, 6));
 735                 break;
 736         case H_RTAS:
 737                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 738                         return RESUME_HOST;
 739
 740                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 741                 rc = kvmppc_rtas_hcall(vcpu);
 742                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 743
 744                 if (rc == -ENOENT)
 745                         return RESUME_HOST;
 746                 else if (rc == 0)
 747                         break;
 748
 749                 /* Send the error out to userspace via KVM_RUN */
 750                 return rc;
 751         case H_LOGICAL_CI_LOAD:
 752                 ret = kvmppc_h_logical_ci_load(vcpu);
 753                 if (ret == H_TOO_HARD)
 754                         return RESUME_HOST;
 755                 break;
 756         case H_LOGICAL_CI_STORE:
 757                 ret = kvmppc_h_logical_ci_store(vcpu);
 758                 if (ret == H_TOO_HARD)
 759                         return RESUME_HOST;
 760                 break;
 761         case H_SET_MODE:
 762                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 763                                         kvmppc_get_gpr(vcpu, 5),
 764                                         kvmppc_get_gpr(vcpu, 6),
 765                                         kvmppc_get_gpr(vcpu, 7));
 766                 if (ret == H_TOO_HARD)
 767                         return RESUME_HOST;
 768                 break;
 769         case H_XIRR:
 770         case H_CPPR:
 771         case H_EOI:
 772         case H_IPI:
 773         case H_IPOLL:
 774         case H_XIRR_X:
 775                 if (kvmppc_xics_enabled(vcpu)) {
 776                         ret = kvmppc_xics_hcall(vcpu, req);
 777                         break;
 778                 } /* fallthrough */
 779         default:
 780                 return RESUME_HOST;
 781         }
 782         kvmppc_set_gpr(vcpu, 3, ret);
 783         vcpu->arch.hcall_needed = 0;
 784         return RESUME_GUEST;
 785 }
 786
 787 static int kvmppc_hcall_impl_hv(unsigned long cmd)
 788 {
 789         switch (cmd) {
 790         case H_CEDE:
 791         case H_PROD:
 792         case H_CONFER:
 793         case H_REGISTER_VPA:
 794         case H_SET_MODE:
 795         case H_LOGICAL_CI_LOAD:
 796         case H_LOGICAL_CI_STORE:
 797 #ifdef CONFIG_KVM_XICS
 798         case H_XIRR:
 799         case H_CPPR:
 800         case H_EOI:
 801         case H_IPI:
 802         case H_IPOLL:
 803         case H_XIRR_X:
 804 #endif
 805                 return 1;
 806         }
 807
 808         /* See if it's in the real-mode table */
 809         return kvmppc_hcall_impl_hv_realmode(cmd);
 810 }
 811
 812 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
 813                                         struct kvm_vcpu *vcpu)
 814 {
 815         u32 last_inst;
 816
 817         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
 818                                         EMULATE_DONE) {
 819                 /*
 820                  * Fetch failed, so return to guest and
 821                  * try executing it again.
 822                  */
 823                 return RESUME_GUEST;
 824         }
 825
 826         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
 827                 run->exit_reason = KVM_EXIT_DEBUG;
 828                 run->debug.arch.address = kvmppc_get_pc(vcpu);
 829                 return RESUME_HOST;
 830         } else {
 831                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 832                 return RESUME_GUEST;
 833         }
 834 }
 835
 836 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
 837                                  struct task_struct *tsk)
 838 {
 839         int r = RESUME_HOST;
 840
 841         vcpu->stat.sum_exits++;
 842
 843         run->exit_reason = KVM_EXIT_UNKNOWN;
 844         run->ready_for_interrupt_injection = 1;
 845         switch (vcpu->arch.trap) {
 846         /* We're good on these - the host merely wanted to get our attention */
 847         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 848                 vcpu->stat.dec_exits++;
 849                 r = RESUME_GUEST;
 850                 break;
 851         case BOOK3S_INTERRUPT_EXTERNAL:
 852         case BOOK3S_INTERRUPT_H_DOORBELL:
 853                 vcpu->stat.ext_intr_exits++;
 854                 r = RESUME_GUEST;
 855                 break;
 856         /* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
 857         case BOOK3S_INTERRUPT_HMI:
 858         case BOOK3S_INTERRUPT_PERFMON:
 859                 r = RESUME_GUEST;
 860                 break;
 861         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 862                 /*
 863                  * Deliver a machine check interrupt to the guest.
 864                  * We have to do this, even if the host has handled the
 865                  * machine check, because machine checks use SRR0/1 and
 866                  * the interrupt might have trashed guest state in them.
 867                  */
 868                 kvmppc_book3s_queue_irqprio(vcpu,
 869                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 870                 r = RESUME_GUEST;
 871                 break;
 872         case BOOK3S_INTERRUPT_PROGRAM:
 873         {
 874                 ulong flags;
 875                 /*
 876                  * Normally program interrupts are delivered directly
 877                  * to the guest by the hardware, but we can get here
 878                  * as a result of a hypervisor emulation interrupt
 879                  * (e40) getting turned into a 700 by BML RTAS.
 880                  */
 881                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 882                 kvmppc_core_queue_program(vcpu, flags);
 883                 r = RESUME_GUEST;
 884                 break;
 885         }
 886         case BOOK3S_INTERRUPT_SYSCALL:
 887         {
 888                 /* hcall - punt to userspace */
 889                 int i;
 890
 891                 /* hypercall with MSR_PR has already been handled in rmode,
 892                  * and never reaches here.
 893                  */
 894
 895                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 896                 for (i = 0; i < 9; ++i)
 897                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 898                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 899                 vcpu->arch.hcall_needed = 1;
 900                 r = RESUME_HOST;
 901                 break;
 902         }
 903         /*
 904          * We get these next two if the guest accesses a page which it thinks
 905          * it has mapped but which is not actually present, either because
 906          * it is for an emulated I/O device or because the corresonding
 907          * host page has been paged out.  Any other HDSI/HISI interrupts
 908          * have been handled already.
 909          */
 910         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
 911                 r = RESUME_PAGE_FAULT;
 912                 break;
 913         case BOOK3S_INTERRUPT_H_INST_STORAGE:
 914                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
 915                 vcpu->arch.fault_dsisr = 0;
 916                 r = RESUME_PAGE_FAULT;
 917                 break;
 918         /*
 919          * This occurs if the guest executes an illegal instruction.
 920          * If the guest debug is disabled, generate a program interrupt
 921          * to the guest. If guest debug is enabled, we need to check
 922          * whether the instruction is a software breakpoint instruction.
 923          * Accordingly return to Guest or Host.
 924          */
 925         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 926                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
 927                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
 928                                 swab32(vcpu->arch.emul_inst) :
 929                                 vcpu->arch.emul_inst;
 930                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
 931                         r = kvmppc_emulate_debug_inst(run, vcpu);
 932                 } else {
 933                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 934                         r = RESUME_GUEST;
 935                 }
 936                 break;
 937         /*
 938          * This occurs if the guest (kernel or userspace), does something that
 939          * is prohibited by HFSCR.  We just generate a program interrupt to
 940          * the guest.
 941          */
 942         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
 943                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 944                 r = RESUME_GUEST;
 945                 break;
 946         default:
 947                 kvmppc_dump_regs(vcpu);
 948                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 949                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 950                         vcpu->arch.shregs.msr);
 951                 run->hw.hardware_exit_reason = vcpu->arch.trap;
 952                 r = RESUME_HOST;
 953                 break;
 954         }
 955
 956         return r;
 957 }
 958
 959 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
 960                                             struct kvm_sregs *sregs)
 961 {
 962         int i;
 963
 964         memset(sregs, 0, sizeof(struct kvm_sregs));
 965         sregs->pvr = vcpu->arch.pvr;
 966         for (i = 0; i < vcpu->arch.slb_max; i++) {
 967                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
 968                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 969         }
 970
 971         return 0;
 972 }
 973
 974 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
 975                                             struct kvm_sregs *sregs)
 976 {
 977         int i, j;
 978
 979         /* Only accept the same PVR as the host's, since we can't spoof it */
 980         if (sregs->pvr != vcpu->arch.pvr)
 981                 return -EINVAL;
 982
 983         j = 0;
 984         for (i = 0; i < vcpu->arch.slb_nr; i++) {
 985                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
 986                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
 987                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
 988                         ++j;
 989                 }
 990         }
 991         vcpu->arch.slb_max = j;
 992
 993         return 0;
 994 }
 995
 996 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
 997                 bool preserve_top32)
 998 {
 999         struct kvm *kvm = vcpu->kvm;
1000         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1001         u64 mask;
1002
1003         mutex_lock(&kvm->lock);
1004         spin_lock(&vc->lock);
1005         /*
1006          * If ILE (interrupt little-endian) has changed, update the
1007          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1008          */
1009         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1010                 struct kvm_vcpu *vcpu;
1011                 int i;
1012
1013                 kvm_for_each_vcpu(i, vcpu, kvm) {
1014                         if (vcpu->arch.vcore != vc)
1015                                 continue;
1016                         if (new_lpcr & LPCR_ILE)
1017                                 vcpu->arch.intr_msr |= MSR_LE;
1018                         else
1019                                 vcpu->arch.intr_msr &= ~MSR_LE;
1020                 }
1021         }
1022
1023         /*
1024          * Userspace can only modify DPFD (default prefetch depth),
1025          * ILE (interrupt little-endian) and TC (translation control).
1026          * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
1027          */
1028         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1029         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1030                 mask |= LPCR_AIL;
1031
1032         /* Broken 32-bit version of LPCR must not clear top bits */
1033         if (preserve_top32)
1034                 mask &= 0xFFFFFFFF;
1035         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1036         spin_unlock(&vc->lock);
1037         mutex_unlock(&kvm->lock);
1038 }
1039
1040 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1041                                  union kvmppc_one_reg *val)
1042 {
1043         int r = 0;
1044         long int i;
1045
1046         switch (id) {
1047         case KVM_REG_PPC_DEBUG_INST:
1048                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1049                 break;
1050         case KVM_REG_PPC_HIOR:
1051                 *val = get_reg_val(id, 0);
1052                 break;
1053         case KVM_REG_PPC_DABR:
1054                 *val = get_reg_val(id, vcpu->arch.dabr);
1055                 break;
1056         case KVM_REG_PPC_DABRX:
1057                 *val = get_reg_val(id, vcpu->arch.dabrx);
1058                 break;
1059         case KVM_REG_PPC_DSCR:
1060                 *val = get_reg_val(id, vcpu->arch.dscr);
1061                 break;
1062         case KVM_REG_PPC_PURR:
1063                 *val = get_reg_val(id, vcpu->arch.purr);
1064                 break;
1065         case KVM_REG_PPC_SPURR:
1066                 *val = get_reg_val(id, vcpu->arch.spurr);
1067                 break;
1068         case KVM_REG_PPC_AMR:
1069                 *val = get_reg_val(id, vcpu->arch.amr);
1070                 break;
1071         case KVM_REG_PPC_UAMOR:
1072                 *val = get_reg_val(id, vcpu->arch.uamor);
1073                 break;
1074         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1075                 i = id - KVM_REG_PPC_MMCR0;
1076                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1077                 break;
1078         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1079                 i = id - KVM_REG_PPC_PMC1;
1080                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1081                 break;
1082         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1083                 i = id - KVM_REG_PPC_SPMC1;
1084                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1085                 break;
1086         case KVM_REG_PPC_SIAR:
1087                 *val = get_reg_val(id, vcpu->arch.siar);
1088                 break;
1089         case KVM_REG_PPC_SDAR:
1090                 *val = get_reg_val(id, vcpu->arch.sdar);
1091                 break;
1092         case KVM_REG_PPC_SIER:
1093                 *val = get_reg_val(id, vcpu->arch.sier);
1094                 break;
1095         case KVM_REG_PPC_IAMR:
1096                 *val = get_reg_val(id, vcpu->arch.iamr);
1097                 break;
1098         case KVM_REG_PPC_PSPB:
1099                 *val = get_reg_val(id, vcpu->arch.pspb);
1100                 break;
1101         case KVM_REG_PPC_DPDES:
1102                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1103                 break;
1104         case KVM_REG_PPC_DAWR:
1105                 *val = get_reg_val(id, vcpu->arch.dawr);
1106                 break;
1107         case KVM_REG_PPC_DAWRX:
1108                 *val = get_reg_val(id, vcpu->arch.dawrx);
1109                 break;
1110         case KVM_REG_PPC_CIABR:
1111                 *val = get_reg_val(id, vcpu->arch.ciabr);
1112                 break;
1113         case KVM_REG_PPC_CSIGR:
1114                 *val = get_reg_val(id, vcpu->arch.csigr);
1115                 break;
1116         case KVM_REG_PPC_TACR:
1117                 *val = get_reg_val(id, vcpu->arch.tacr);
1118                 break;
1119         case KVM_REG_PPC_TCSCR:
1120                 *val = get_reg_val(id, vcpu->arch.tcscr);
1121                 break;
1122         case KVM_REG_PPC_PID:
1123                 *val = get_reg_val(id, vcpu->arch.pid);
1124                 break;
1125         case KVM_REG_PPC_ACOP:
1126                 *val = get_reg_val(id, vcpu->arch.acop);
1127                 break;
1128         case KVM_REG_PPC_WORT:
1129                 *val = get_reg_val(id, vcpu->arch.wort);
1130                 break;
1131         case KVM_REG_PPC_VPA_ADDR:
1132                 spin_lock(&vcpu->arch.vpa_update_lock);
1133                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1134                 spin_unlock(&vcpu->arch.vpa_update_lock);
1135                 break;
1136         case KVM_REG_PPC_VPA_SLB:
1137                 spin_lock(&vcpu->arch.vpa_update_lock);
1138                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1139                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1140                 spin_unlock(&vcpu->arch.vpa_update_lock);
1141                 break;
1142         case KVM_REG_PPC_VPA_DTL:
1143                 spin_lock(&vcpu->arch.vpa_update_lock);
1144                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1145                 val->vpaval.length = vcpu->arch.dtl.len;
1146                 spin_unlock(&vcpu->arch.vpa_update_lock);
1147                 break;
1148         case KVM_REG_PPC_TB_OFFSET:
1149                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1150                 break;
1151         case KVM_REG_PPC_LPCR:
1152         case KVM_REG_PPC_LPCR_64:
1153                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1154                 break;
1155         case KVM_REG_PPC_PPR:
1156                 *val = get_reg_val(id, vcpu->arch.ppr);
1157                 break;
1158 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1159         case KVM_REG_PPC_TFHAR:
1160                 *val = get_reg_val(id, vcpu->arch.tfhar);
1161                 break;
1162         case KVM_REG_PPC_TFIAR:
1163                 *val = get_reg_val(id, vcpu->arch.tfiar);
1164                 break;
1165         case KVM_REG_PPC_TEXASR:
1166                 *val = get_reg_val(id, vcpu->arch.texasr);
1167                 break;
1168         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1169                 i = id - KVM_REG_PPC_TM_GPR0;
1170                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1171                 break;
1172         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1173         {
1174                 int j;
1175                 i = id - KVM_REG_PPC_TM_VSR0;
1176                 if (i < 32)
1177                         for (j = 0; j < TS_FPRWIDTH; j++)
1178                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1179                 else {
1180                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1181                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1182                         else
1183                                 r = -ENXIO;
1184                 }
1185                 break;
1186         }
1187         case KVM_REG_PPC_TM_CR:
1188                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1189                 break;
1190         case KVM_REG_PPC_TM_LR:
1191                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1192                 break;
1193         case KVM_REG_PPC_TM_CTR:
1194                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1195                 break;
1196         case KVM_REG_PPC_TM_FPSCR:
1197                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1198                 break;
1199         case KVM_REG_PPC_TM_AMR:
1200                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1201                 break;
1202         case KVM_REG_PPC_TM_PPR:
1203                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1204                 break;
1205         case KVM_REG_PPC_TM_VRSAVE:
1206                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1207                 break;
1208         case KVM_REG_PPC_TM_VSCR:
1209                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1210                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1211                 else
1212                         r = -ENXIO;
1213                 break;
1214         case KVM_REG_PPC_TM_DSCR:
1215                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1216                 break;
1217         case KVM_REG_PPC_TM_TAR:
1218                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1219                 break;
1220 #endif
1221         case KVM_REG_PPC_ARCH_COMPAT:
1222                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1223                 break;
1224         default:
1225                 r = -EINVAL;
1226                 break;
1227         }
1228
1229         return r;
1230 }
1231
1232 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1233                                  union kvmppc_one_reg *val)
1234 {
1235         int r = 0;
1236         long int i;
1237         unsigned long addr, len;
1238
1239         switch (id) {
1240         case KVM_REG_PPC_HIOR:
1241                 /* Only allow this to be set to zero */
1242                 if (set_reg_val(id, *val))
1243                         r = -EINVAL;
1244                 break;
1245         case KVM_REG_PPC_DABR:
1246                 vcpu->arch.dabr = set_reg_val(id, *val);
1247                 break;
1248         case KVM_REG_PPC_DABRX:
1249                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1250                 break;
1251         case KVM_REG_PPC_DSCR:
1252                 vcpu->arch.dscr = set_reg_val(id, *val);
1253                 break;
1254         case KVM_REG_PPC_PURR:
1255                 vcpu->arch.purr = set_reg_val(id, *val);
1256                 break;
1257         case KVM_REG_PPC_SPURR:
1258                 vcpu->arch.spurr = set_reg_val(id, *val);
1259                 break;
1260         case KVM_REG_PPC_AMR:
1261                 vcpu->arch.amr = set_reg_val(id, *val);
1262                 break;
1263         case KVM_REG_PPC_UAMOR:
1264                 vcpu->arch.uamor = set_reg_val(id, *val);
1265                 break;
1266         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1267                 i = id - KVM_REG_PPC_MMCR0;
1268                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1269                 break;
1270         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1271                 i = id - KVM_REG_PPC_PMC1;
1272                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1273                 break;
1274         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1275                 i = id - KVM_REG_PPC_SPMC1;
1276                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1277                 break;
1278         case KVM_REG_PPC_SIAR:
1279                 vcpu->arch.siar = set_reg_val(id, *val);
1280                 break;
1281         case KVM_REG_PPC_SDAR:
1282                 vcpu->arch.sdar = set_reg_val(id, *val);
1283                 break;
1284         case KVM_REG_PPC_SIER:
1285                 vcpu->arch.sier = set_reg_val(id, *val);
1286                 break;
1287         case KVM_REG_PPC_IAMR:
1288                 vcpu->arch.iamr = set_reg_val(id, *val);
1289                 break;
1290         case KVM_REG_PPC_PSPB:
1291                 vcpu->arch.pspb = set_reg_val(id, *val);
1292                 break;
1293         case KVM_REG_PPC_DPDES:
1294                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1295                 break;
1296         case KVM_REG_PPC_DAWR:
1297                 vcpu->arch.dawr = set_reg_val(id, *val);
1298                 break;
1299         case KVM_REG_PPC_DAWRX:
1300                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1301                 break;
1302         case KVM_REG_PPC_CIABR:
1303                 vcpu->arch.ciabr = set_reg_val(id, *val);
1304                 /* Don't allow setting breakpoints in hypervisor code */
1305                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1306                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1307                 break;
1308         case KVM_REG_PPC_CSIGR:
1309                 vcpu->arch.csigr = set_reg_val(id, *val);
1310                 break;
1311         case KVM_REG_PPC_TACR:
1312                 vcpu->arch.tacr = set_reg_val(id, *val);
1313                 break;
1314         case KVM_REG_PPC_TCSCR:
1315                 vcpu->arch.tcscr = set_reg_val(id, *val);
1316                 break;
1317         case KVM_REG_PPC_PID:
1318                 vcpu->arch.pid = set_reg_val(id, *val);
1319                 break;
1320         case KVM_REG_PPC_ACOP:
1321                 vcpu->arch.acop = set_reg_val(id, *val);
1322                 break;
1323         case KVM_REG_PPC_WORT:
1324                 vcpu->arch.wort = set_reg_val(id, *val);
1325                 break;
1326         case KVM_REG_PPC_VPA_ADDR:
1327                 addr = set_reg_val(id, *val);
1328                 r = -EINVAL;
1329                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1330                               vcpu->arch.dtl.next_gpa))
1331                         break;
1332                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1333                 break;
1334         case KVM_REG_PPC_VPA_SLB:
1335                 addr = val->vpaval.addr;
1336                 len = val->vpaval.length;
1337                 r = -EINVAL;
1338                 if (addr && !vcpu->arch.vpa.next_gpa)
1339                         break;
1340                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1341                 break;
1342         case KVM_REG_PPC_VPA_DTL:
1343                 addr = val->vpaval.addr;
1344                 len = val->vpaval.length;
1345                 r = -EINVAL;
1346                 if (addr && (len < sizeof(struct dtl_entry) ||
1347                              !vcpu->arch.vpa.next_gpa))
1348                         break;
1349                 len -= len % sizeof(struct dtl_entry);
1350                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1351                 break;
1352         case KVM_REG_PPC_TB_OFFSET:
1353                 /* round up to multiple of 2^24 */
1354                 vcpu->arch.vcore->tb_offset =
1355                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1356                 break;
1357         case KVM_REG_PPC_LPCR:
1358                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1359                 break;
1360         case KVM_REG_PPC_LPCR_64:
1361                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1362                 break;
1363         case KVM_REG_PPC_PPR:
1364                 vcpu->arch.ppr = set_reg_val(id, *val);
1365                 break;
1366 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1367         case KVM_REG_PPC_TFHAR:
1368                 vcpu->arch.tfhar = set_reg_val(id, *val);
1369                 break;
1370         case KVM_REG_PPC_TFIAR:
1371                 vcpu->arch.tfiar = set_reg_val(id, *val);
1372                 break;
1373         case KVM_REG_PPC_TEXASR:
1374                 vcpu->arch.texasr = set_reg_val(id, *val);
1375                 break;
1376         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1377                 i = id - KVM_REG_PPC_TM_GPR0;
1378                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1379                 break;
1380         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1381         {
1382                 int j;
1383                 i = id - KVM_REG_PPC_TM_VSR0;
1384                 if (i < 32)
1385                         for (j = 0; j < TS_FPRWIDTH; j++)
1386                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1387                 else
1388                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1389                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1390                         else
1391                                 r = -ENXIO;
1392                 break;
1393         }
1394         case KVM_REG_PPC_TM_CR:
1395                 vcpu->arch.cr_tm = set_reg_val(id, *val);
1396                 break;
1397         case KVM_REG_PPC_TM_LR:
1398                 vcpu->arch.lr_tm = set_reg_val(id, *val);
1399                 break;
1400         case KVM_REG_PPC_TM_CTR:
1401                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1402                 break;
1403         case KVM_REG_PPC_TM_FPSCR:
1404                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1405                 break;
1406         case KVM_REG_PPC_TM_AMR:
1407                 vcpu->arch.amr_tm = set_reg_val(id, *val);
1408                 break;
1409         case KVM_REG_PPC_TM_PPR:
1410                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1411                 break;
1412         case KVM_REG_PPC_TM_VRSAVE:
1413                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1414                 break;
1415         case KVM_REG_PPC_TM_VSCR:
1416                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1417                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1418                 else
1419                         r = - ENXIO;
1420                 break;
1421         case KVM_REG_PPC_TM_DSCR:
1422                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1423                 break;
1424         case KVM_REG_PPC_TM_TAR:
1425                 vcpu->arch.tar_tm = set_reg_val(id, *val);
1426                 break;
1427 #endif
1428         case KVM_REG_PPC_ARCH_COMPAT:
1429                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1430                 break;
1431         default:
1432                 r = -EINVAL;
1433                 break;
1434         }
1435
1436         return r;
1437 }
1438
1439 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1440 {
1441         struct kvmppc_vcore *vcore;
1442
1443         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1444
1445         if (vcore == NULL)
1446                 return NULL;
1447
1448         INIT_LIST_HEAD(&vcore->runnable_threads);
1449         spin_lock_init(&vcore->lock);
1450         spin_lock_init(&vcore->stoltb_lock);
1451         init_waitqueue_head(&vcore->wq);
1452         vcore->preempt_tb = TB_NIL;
1453         vcore->lpcr = kvm->arch.lpcr;
1454         vcore->first_vcpuid = core * threads_per_subcore;
1455         vcore->kvm = kvm;
1456         INIT_LIST_HEAD(&vcore->preempt_list);
1457
1458         vcore->mpp_buffer_is_valid = false;
1459
1460         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1461                 vcore->mpp_buffer = (void *)__get_free_pages(
1462                         GFP_KERNEL|__GFP_ZERO,
1463                         MPP_BUFFER_ORDER);
1464
1465         return vcore;
1466 }
1467
1468 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1469 static struct debugfs_timings_element {
1470         const char *name;
1471         size_t offset;
1472 } timings[] = {
1473         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
1474         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
1475         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
1476         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
1477         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
1478 };
1479
1480 #define N_TIMINGS       (sizeof(timings) / sizeof(timings[0]))
1481
1482 struct debugfs_timings_state {
1483         struct kvm_vcpu *vcpu;
1484         unsigned int    buflen;
1485         char            buf[N_TIMINGS * 100];
1486 };
1487
1488 static int debugfs_timings_open(struct inode *inode, struct file *file)
1489 {
1490         struct kvm_vcpu *vcpu = inode->i_private;
1491         struct debugfs_timings_state *p;
1492
1493         p = kzalloc(sizeof(*p), GFP_KERNEL);
1494         if (!p)
1495                 return -ENOMEM;
1496
1497         kvm_get_kvm(vcpu->kvm);
1498         p->vcpu = vcpu;
1499         file->private_data = p;
1500
1501         return nonseekable_open(inode, file);
1502 }
1503
1504 static int debugfs_timings_release(struct inode *inode, struct file *file)
1505 {
1506         struct debugfs_timings_state *p = file->private_data;
1507
1508         kvm_put_kvm(p->vcpu->kvm);
1509         kfree(p);
1510         return 0;
1511 }
1512
1513 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1514                                     size_t len, loff_t *ppos)
1515 {
1516         struct debugfs_timings_state *p = file->private_data;
1517         struct kvm_vcpu *vcpu = p->vcpu;
1518         char *s, *buf_end;
1519         struct kvmhv_tb_accumulator tb;
1520         u64 count;
1521         loff_t pos;
1522         ssize_t n;
1523         int i, loops;
1524         bool ok;
1525
1526         if (!p->buflen) {
1527                 s = p->buf;
1528                 buf_end = s + sizeof(p->buf);
1529                 for (i = 0; i < N_TIMINGS; ++i) {
1530                         struct kvmhv_tb_accumulator *acc;
1531
1532                         acc = (struct kvmhv_tb_accumulator *)
1533                                 ((unsigned long)vcpu + timings[i].offset);
1534                         ok = false;
1535                         for (loops = 0; loops < 1000; ++loops) {
1536                                 count = acc->seqcount;
1537                                 if (!(count & 1)) {
1538                                         smp_rmb();
1539                                         tb = *acc;
1540                                         smp_rmb();
1541                                         if (count == acc->seqcount) {
1542                                                 ok = true;
1543                                                 break;
1544                                         }
1545                                 }
1546                                 udelay(1);
1547                         }
1548                         if (!ok)
1549                                 snprintf(s, buf_end - s, "%s: stuck\n",
1550                                         timings[i].name);
1551                         else
1552                                 snprintf(s, buf_end - s,
1553                                         "%s: %llu %llu %llu %llu\n",
1554                                         timings[i].name, count / 2,
1555                                         tb_to_ns(tb.tb_total),
1556                                         tb_to_ns(tb.tb_min),
1557                                         tb_to_ns(tb.tb_max));
1558                         s += strlen(s);
1559                 }
1560                 p->buflen = s - p->buf;
1561         }
1562
1563         pos = *ppos;
1564         if (pos >= p->buflen)
1565                 return 0;
1566         if (len > p->buflen - pos)
1567                 len = p->buflen - pos;
1568         n = copy_to_user(buf, p->buf + pos, len);
1569         if (n) {
1570                 if (n == len)
1571                         return -EFAULT;
1572                 len -= n;
1573         }
1574         *ppos = pos + len;
1575         return len;
1576 }
1577
1578 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1579                                      size_t len, loff_t *ppos)
1580 {
1581         return -EACCES;
1582 }
1583
1584 static const struct file_operations debugfs_timings_ops = {
1585         .owner   = THIS_MODULE,
1586         .open    = debugfs_timings_open,
1587         .release = debugfs_timings_release,
1588         .read    = debugfs_timings_read,
1589         .write   = debugfs_timings_write,
1590         .llseek  = generic_file_llseek,
1591 };
1592
1593 /* Create a debugfs directory for the vcpu */
1594 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1595 {
1596         char buf[16];
1597         struct kvm *kvm = vcpu->kvm;
1598
1599         snprintf(buf, sizeof(buf), "vcpu%u", id);
1600         if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1601                 return;
1602         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1603         if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1604                 return;
1605         vcpu->arch.debugfs_timings =
1606                 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1607                                     vcpu, &debugfs_timings_ops);
1608 }
1609
1610 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1611 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1612 {
1613 }
1614 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1615
1616 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1617                                                    unsigned int id)
1618 {
1619         struct kvm_vcpu *vcpu;
1620         int err = -EINVAL;
1621         int core;
1622         struct kvmppc_vcore *vcore;
1623
1624         core = id / threads_per_subcore;
1625         if (core >= KVM_MAX_VCORES)
1626                 goto out;
1627
1628         err = -ENOMEM;
1629         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1630         if (!vcpu)
1631                 goto out;
1632
1633         err = kvm_vcpu_init(vcpu, kvm, id);
1634         if (err)
1635                 goto free_vcpu;
1636
1637         vcpu->arch.shared = &vcpu->arch.shregs;
1638 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1639         /*
1640          * The shared struct is never shared on HV,
1641          * so we can always use host endianness
1642          */
1643 #ifdef __BIG_ENDIAN__
1644         vcpu->arch.shared_big_endian = true;
1645 #else
1646         vcpu->arch.shared_big_endian = false;
1647 #endif
1648 #endif
1649         vcpu->arch.mmcr[0] = MMCR0_FC;
1650         vcpu->arch.ctrl = CTRL_RUNLATCH;
1651         /* default to host PVR, since we can't spoof it */
1652         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1653         spin_lock_init(&vcpu->arch.vpa_update_lock);
1654         spin_lock_init(&vcpu->arch.tbacct_lock);
1655         vcpu->arch.busy_preempt = TB_NIL;
1656         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1657
1658         kvmppc_mmu_book3s_hv_init(vcpu);
1659
1660         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1661
1662         init_waitqueue_head(&vcpu->arch.cpu_run);
1663
1664         mutex_lock(&kvm->lock);
1665         vcore = kvm->arch.vcores[core];
1666         if (!vcore) {
1667                 vcore = kvmppc_vcore_create(kvm, core);
1668                 kvm->arch.vcores[core] = vcore;
1669                 kvm->arch.online_vcores++;
1670         }
1671         mutex_unlock(&kvm->lock);
1672
1673         if (!vcore)
1674                 goto free_vcpu;
1675
1676         spin_lock(&vcore->lock);
1677         ++vcore->num_threads;
1678         spin_unlock(&vcore->lock);
1679         vcpu->arch.vcore = vcore;
1680         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1681         vcpu->arch.thread_cpu = -1;
1682
1683         vcpu->arch.cpu_type = KVM_CPU_3S_64;
1684         kvmppc_sanity_check(vcpu);
1685
1686         debugfs_vcpu_init(vcpu, id);
1687
1688         return vcpu;
1689
1690 free_vcpu:
1691         kmem_cache_free(kvm_vcpu_cache, vcpu);
1692 out:
1693         return ERR_PTR(err);
1694 }
1695
1696 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1697 {
1698         if (vpa->pinned_addr)
1699                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1700                                         vpa->dirty);
1701 }
1702
1703 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1704 {
1705         spin_lock(&vcpu->arch.vpa_update_lock);
1706         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1707         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1708         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1709         spin_unlock(&vcpu->arch.vpa_update_lock);
1710         kvm_vcpu_uninit(vcpu);
1711         kmem_cache_free(kvm_vcpu_cache, vcpu);
1712 }
1713
1714 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1715 {
1716         /* Indicate we want to get back into the guest */
1717         return 1;
1718 }
1719
1720 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1721 {
1722         unsigned long dec_nsec, now;
1723
1724         now = get_tb();
1725         if (now > vcpu->arch.dec_expires) {
1726                 /* decrementer has already gone negative */
1727                 kvmppc_core_queue_dec(vcpu);
1728                 kvmppc_core_prepare_to_enter(vcpu);
1729                 return;
1730         }
1731         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1732                    / tb_ticks_per_sec;
1733         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1734                       HRTIMER_MODE_REL);
1735         vcpu->arch.timer_running = 1;
1736 }
1737
1738 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1739 {
1740         vcpu->arch.ceded = 0;
1741         if (vcpu->arch.timer_running) {
1742                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1743                 vcpu->arch.timer_running = 0;
1744         }
1745 }
1746
1747 extern void __kvmppc_vcore_entry(void);
1748
1749 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1750                                    struct kvm_vcpu *vcpu)
1751 {
1752         u64 now;
1753
1754         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1755                 return;
1756         spin_lock_irq(&vcpu->arch.tbacct_lock);
1757         now = mftb();
1758         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1759                 vcpu->arch.stolen_logged;
1760         vcpu->arch.busy_preempt = now;
1761         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1762         spin_unlock_irq(&vcpu->arch.tbacct_lock);
1763         --vc->n_runnable;
1764         list_del(&vcpu->arch.run_list);
1765 }
1766
1767 static int kvmppc_grab_hwthread(int cpu)
1768 {
1769         struct paca_struct *tpaca;
1770         long timeout = 10000;
1771
1772         tpaca = &paca[cpu];
1773
1774         /* Ensure the thread won't go into the kernel if it wakes */
1775         tpaca->kvm_hstate.kvm_vcpu = NULL;
1776         tpaca->kvm_hstate.kvm_vcore = NULL;
1777         tpaca->kvm_hstate.napping = 0;
1778         smp_wmb();
1779         tpaca->kvm_hstate.hwthread_req = 1;
1780
1781         /*
1782          * If the thread is already executing in the kernel (e.g. handling
1783          * a stray interrupt), wait for it to get back to nap mode.
1784          * The smp_mb() is to ensure that our setting of hwthread_req
1785          * is visible before we look at hwthread_state, so if this
1786          * races with the code at system_reset_pSeries and the thread
1787          * misses our setting of hwthread_req, we are sure to see its
1788          * setting of hwthread_state, and vice versa.
1789          */
1790         smp_mb();
1791         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1792                 if (--timeout <= 0) {
1793                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1794                         return -EBUSY;
1795                 }
1796                 udelay(1);
1797         }
1798         return 0;
1799 }
1800
1801 static void kvmppc_release_hwthread(int cpu)
1802 {
1803         struct paca_struct *tpaca;
1804
1805         tpaca = &paca[cpu];
1806         tpaca->kvm_hstate.hwthread_req = 0;
1807         tpaca->kvm_hstate.kvm_vcpu = NULL;
1808         tpaca->kvm_hstate.kvm_vcore = NULL;
1809         tpaca->kvm_hstate.kvm_split_mode = NULL;
1810 }
1811
1812 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
1813 {
1814         int cpu;
1815         struct paca_struct *tpaca;
1816         struct kvmppc_vcore *mvc = vc->master_vcore;
1817
1818         cpu = vc->pcpu;
1819         if (vcpu) {
1820                 if (vcpu->arch.timer_running) {
1821                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1822                         vcpu->arch.timer_running = 0;
1823                 }
1824                 cpu += vcpu->arch.ptid;
1825                 vcpu->cpu = mvc->pcpu;
1826                 vcpu->arch.thread_cpu = cpu;
1827         }
1828         tpaca = &paca[cpu];
1829         tpaca->kvm_hstate.kvm_vcpu = vcpu;
1830         tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
1831         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
1832         smp_wmb();
1833         tpaca->kvm_hstate.kvm_vcore = mvc;
1834         if (cpu != smp_processor_id())
1835                 kvmppc_ipi_thread(cpu);
1836 }
1837
1838 static void kvmppc_wait_for_nap(void)
1839 {
1840         int cpu = smp_processor_id();
1841         int i, loops;
1842
1843         for (loops = 0; loops < 1000000; ++loops) {
1844                 /*
1845                  * Check if all threads are finished.
1846                  * We set the vcore pointer when starting a thread
1847                  * and the thread clears it when finished, so we look
1848                  * for any threads that still have a non-NULL vcore ptr.
1849                  */
1850                 for (i = 1; i < threads_per_subcore; ++i)
1851                         if (paca[cpu + i].kvm_hstate.kvm_vcore)
1852                                 break;
1853                 if (i == threads_per_subcore) {
1854                         HMT_medium();
1855                         return;
1856                 }
1857                 HMT_low();
1858         }
1859         HMT_medium();
1860         for (i = 1; i < threads_per_subcore; ++i)
1861                 if (paca[cpu + i].kvm_hstate.kvm_vcore)
1862                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
1863 }
1864
1865 /*
1866  * Check that we are on thread 0 and that any other threads in
1867  * this core are off-line.  Then grab the threads so they can't
1868  * enter the kernel.
1869  */
1870 static int on_primary_thread(void)
1871 {
1872         int cpu = smp_processor_id();
1873         int thr;
1874
1875         /* Are we on a primary subcore? */
1876         if (cpu_thread_in_subcore(cpu))
1877                 return 0;
1878
1879         thr = 0;
1880         while (++thr < threads_per_subcore)
1881                 if (cpu_online(cpu + thr))
1882                         return 0;
1883
1884         /* Grab all hw threads so they can't go into the kernel */
1885         for (thr = 1; thr < threads_per_subcore; ++thr) {
1886                 if (kvmppc_grab_hwthread(cpu + thr)) {
1887                         /* Couldn't grab one; let the others go */
1888                         do {
1889                                 kvmppc_release_hwthread(cpu + thr);
1890                         } while (--thr > 0);
1891                         return 0;
1892                 }
1893         }
1894         return 1;
1895 }
1896
1897 static void kvmppc_start_saving_l2_cache(struct kvmppc_vcore *vc)
1898 {
1899         phys_addr_t phy_addr, mpp_addr;
1900
1901         phy_addr = (phys_addr_t)virt_to_phys(vc->mpp_buffer);
1902         mpp_addr = phy_addr & PPC_MPPE_ADDRESS_MASK;
1903
1904         mtspr(SPRN_MPPR, mpp_addr | PPC_MPPR_FETCH_ABORT);
1905         logmpp(mpp_addr | PPC_LOGMPP_LOG_L2);
1906
1907         vc->mpp_buffer_is_valid = true;
1908 }
1909
1910 static void kvmppc_start_restoring_l2_cache(const struct kvmppc_vcore *vc)
1911 {
1912         phys_addr_t phy_addr, mpp_addr;
1913
1914         phy_addr = virt_to_phys(vc->mpp_buffer);
1915         mpp_addr = phy_addr & PPC_MPPE_ADDRESS_MASK;
1916
1917         /* We must abort any in-progress save operations to ensure
1918          * the table is valid so that prefetch engine knows when to
1919          * stop prefetching. */
1920         logmpp(mpp_addr | PPC_LOGMPP_LOG_ABORT);
1921         mtspr(SPRN_MPPR, mpp_addr | PPC_MPPR_FETCH_WHOLE_TABLE);
1922 }
1923
1924 /*
1925  * A list of virtual cores for each physical CPU.
1926  * These are vcores that could run but their runner VCPU tasks are
1927  * (or may be) preempted.
1928  */
1929 struct preempted_vcore_list {
1930         struct list_head        list;
1931         spinlock_t              lock;
1932 };
1933
1934 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
1935
1936 static void init_vcore_lists(void)
1937 {
1938         int cpu;
1939
1940         for_each_possible_cpu(cpu) {
1941                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
1942                 spin_lock_init(&lp->lock);
1943                 INIT_LIST_HEAD(&lp->list);
1944         }
1945 }
1946
1947 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
1948 {
1949         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
1950
1951         vc->vcore_state = VCORE_PREEMPT;
1952         vc->pcpu = smp_processor_id();
1953         if (vc->num_threads < threads_per_subcore) {
1954                 spin_lock(&lp->lock);
1955                 list_add_tail(&vc->preempt_list, &lp->list);
1956                 spin_unlock(&lp->lock);
1957         }
1958
1959         /* Start accumulating stolen time */
1960         kvmppc_core_start_stolen(vc);
1961 }
1962
1963 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
1964 {
1965         struct preempted_vcore_list *lp;
1966
1967         kvmppc_core_end_stolen(vc);
1968         if (!list_empty(&vc->preempt_list)) {
1969                 lp = &per_cpu(preempted_vcores, vc->pcpu);
1970                 spin_lock(&lp->lock);
1971                 list_del_init(&vc->preempt_list);
1972                 spin_unlock(&lp->lock);
1973         }
1974         vc->vcore_state = VCORE_INACTIVE;
1975 }
1976
1977 /*
1978  * This stores information about the virtual cores currently
1979  * assigned to a physical core.
1980  */
1981 struct core_info {
1982         int             n_subcores;
1983         int             max_subcore_threads;
1984         int             total_threads;
1985         int             subcore_threads[MAX_SUBCORES];
1986         struct kvm      *subcore_vm[MAX_SUBCORES];
1987         struct list_head vcs[MAX_SUBCORES];
1988 };
1989
1990 /*
1991  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
1992  * respectively in 2-way micro-threading (split-core) mode.
1993  */
1994 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
1995
1996 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
1997 {
1998         int sub;
1999
2000         memset(cip, 0, sizeof(*cip));
2001         cip->n_subcores = 1;
2002         cip->max_subcore_threads = vc->num_threads;
2003         cip->total_threads = vc->num_threads;
2004         cip->subcore_threads[0] = vc->num_threads;
2005         cip->subcore_vm[0] = vc->kvm;
2006         for (sub = 0; sub < MAX_SUBCORES; ++sub)
2007                 INIT_LIST_HEAD(&cip->vcs[sub]);
2008         list_add_tail(&vc->preempt_list, &cip->vcs[0]);
2009 }
2010
2011 static bool subcore_config_ok(int n_subcores, int n_threads)
2012 {
2013         /* Can only dynamically split if unsplit to begin with */
2014         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2015                 return false;
2016         if (n_subcores > MAX_SUBCORES)
2017                 return false;
2018         if (n_subcores > 1) {
2019                 if (!(dynamic_mt_modes & 2))
2020                         n_subcores = 4;
2021                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2022                         return false;
2023         }
2024
2025         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2026 }
2027
2028 static void init_master_vcore(struct kvmppc_vcore *vc)
2029 {
2030         vc->master_vcore = vc;
2031         vc->entry_exit_map = 0;
2032         vc->in_guest = 0;
2033         vc->napping_threads = 0;
2034         vc->conferring_threads = 0;
2035 }
2036
2037 /*
2038  * See if the existing subcores can be split into 3 (or fewer) subcores
2039  * of at most two threads each, so we can fit in another vcore.  This
2040  * assumes there are at most two subcores and at most 6 threads in total.
2041  */
2042 static bool can_split_piggybacked_subcores(struct core_info *cip)
2043 {
2044         int sub, new_sub;
2045         int large_sub = -1;
2046         int thr;
2047         int n_subcores = cip->n_subcores;
2048         struct kvmppc_vcore *vc, *vcnext;
2049         struct kvmppc_vcore *master_vc = NULL;
2050
2051         for (sub = 0; sub < cip->n_subcores; ++sub) {
2052                 if (cip->subcore_threads[sub] <= 2)
2053                         continue;
2054                 if (large_sub >= 0)
2055                         return false;
2056                 large_sub = sub;
2057                 vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2058                                       preempt_list);
2059                 if (vc->num_threads > 2)
2060                         return false;
2061                 n_subcores += (cip->subcore_threads[sub] - 1) >> 1;
2062         }
2063         if (n_subcores > 3 || large_sub < 0)
2064                 return false;
2065
2066         /*
2067          * Seems feasible, so go through and move vcores to new subcores.
2068          * Note that when we have two or more vcores in one subcore,
2069          * all those vcores must have only one thread each.
2070          */
2071         new_sub = cip->n_subcores;
2072         thr = 0;
2073         sub = large_sub;
2074         list_for_each_entry_safe(vc, vcnext, &cip->vcs[sub], preempt_list) {
2075                 if (thr >= 2) {
2076                         list_del(&vc->preempt_list);
2077                         list_add_tail(&vc->preempt_list, &cip->vcs[new_sub]);
2078                         /* vc->num_threads must be 1 */
2079                         if (++cip->subcore_threads[new_sub] == 1) {
2080                                 cip->subcore_vm[new_sub] = vc->kvm;
2081                                 init_master_vcore(vc);
2082                                 master_vc = vc;
2083                                 ++cip->n_subcores;
2084                         } else {
2085                                 vc->master_vcore = master_vc;
2086                                 ++new_sub;
2087                         }
2088                 }
2089                 thr += vc->num_threads;
2090         }
2091         cip->subcore_threads[large_sub] = 2;
2092         cip->max_subcore_threads = 2;
2093
2094         return true;
2095 }
2096
2097 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2098 {
2099         int n_threads = vc->num_threads;
2100         int sub;
2101
2102         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2103                 return false;
2104
2105         if (n_threads < cip->max_subcore_threads)
2106                 n_threads = cip->max_subcore_threads;
2107         if (subcore_config_ok(cip->n_subcores + 1, n_threads)) {
2108                 cip->max_subcore_threads = n_threads;
2109         } else if (cip->n_subcores <= 2 && cip->total_threads <= 6 &&
2110                    vc->num_threads <= 2) {
2111                 /*
2112                  * We may be able to fit another subcore in by
2113                  * splitting an existing subcore with 3 or 4
2114                  * threads into two 2-thread subcores, or one
2115                  * with 5 or 6 threads into three subcores.
2116                  * We can only do this if those subcores have
2117                  * piggybacked virtual cores.
2118                  */
2119                 if (!can_split_piggybacked_subcores(cip))
2120                         return false;
2121         } else {
2122                 return false;
2123         }
2124
2125         sub = cip->n_subcores;
2126         ++cip->n_subcores;
2127         cip->total_threads += vc->num_threads;
2128         cip->subcore_threads[sub] = vc->num_threads;
2129         cip->subcore_vm[sub] = vc->kvm;
2130         init_master_vcore(vc);
2131         list_del(&vc->preempt_list);
2132         list_add_tail(&vc->preempt_list, &cip->vcs[sub]);
2133
2134         return true;
2135 }
2136
2137 static bool can_piggyback_subcore(struct kvmppc_vcore *pvc,
2138                                   struct core_info *cip, int sub)
2139 {
2140         struct kvmppc_vcore *vc;
2141         int n_thr;
2142
2143         vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2144                               preempt_list);
2145
2146         /* require same VM and same per-core reg values */
2147         if (pvc->kvm != vc->kvm ||
2148             pvc->tb_offset != vc->tb_offset ||
2149             pvc->pcr != vc->pcr ||
2150             pvc->lpcr != vc->lpcr)
2151                 return false;
2152
2153         /* P8 guest with > 1 thread per core would see wrong TIR value */
2154         if (cpu_has_feature(CPU_FTR_ARCH_207S) &&
2155             (vc->num_threads > 1 || pvc->num_threads > 1))
2156                 return false;
2157
2158         n_thr = cip->subcore_threads[sub] + pvc->num_threads;
2159         if (n_thr > cip->max_subcore_threads) {
2160                 if (!subcore_config_ok(cip->n_subcores, n_thr))
2161                         return false;
2162                 cip->max_subcore_threads = n_thr;
2163         }
2164
2165         cip->total_threads += pvc->num_threads;
2166         cip->subcore_threads[sub] = n_thr;
2167         pvc->master_vcore = vc;
2168         list_del(&pvc->preempt_list);
2169         list_add_tail(&pvc->preempt_list, &cip->vcs[sub]);
2170
2171         return true;
2172 }
2173
2174 /*
2175  * Work out whether it is possible to piggyback the execution of
2176  * vcore *pvc onto the execution of the other vcores described in *cip.
2177  */
2178 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2179                           int target_threads)
2180 {
2181         int sub;
2182
2183         if (cip->total_threads + pvc->num_threads > target_threads)
2184                 return false;
2185         for (sub = 0; sub < cip->n_subcores; ++sub)
2186                 if (cip->subcore_threads[sub] &&
2187                     can_piggyback_subcore(pvc, cip, sub))
2188                         return true;
2189
2190         if (can_dynamic_split(pvc, cip))
2191                 return true;
2192
2193         return false;
2194 }
2195
2196 static void prepare_threads(struct kvmppc_vcore *vc)
2197 {
2198         struct kvm_vcpu *vcpu, *vnext;
2199
2200         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2201                                  arch.run_list) {
2202                 if (signal_pending(vcpu->arch.run_task))
2203                         vcpu->arch.ret = -EINTR;
2204                 else if (vcpu->arch.vpa.update_pending ||
2205                          vcpu->arch.slb_shadow.update_pending ||
2206                          vcpu->arch.dtl.update_pending)
2207                         vcpu->arch.ret = RESUME_GUEST;
2208                 else
2209                         continue;
2210                 kvmppc_remove_runnable(vc, vcpu);
2211                 wake_up(&vcpu->arch.cpu_run);
2212         }
2213 }
2214
2215 static void collect_piggybacks(struct core_info *cip, int target_threads)
2216 {
2217         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2218         struct kvmppc_vcore *pvc, *vcnext;
2219
2220         spin_lock(&lp->lock);
2221         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2222                 if (!spin_trylock(&pvc->lock))
2223                         continue;
2224                 prepare_threads(pvc);
2225                 if (!pvc->n_runnable) {
2226                         list_del_init(&pvc->preempt_list);
2227                         if (pvc->runner == NULL) {
2228                                 pvc->vcore_state = VCORE_INACTIVE;
2229                                 kvmppc_core_end_stolen(pvc);
2230                         }
2231                         spin_unlock(&pvc->lock);
2232                         continue;
2233                 }
2234                 if (!can_piggyback(pvc, cip, target_threads)) {
2235                         spin_unlock(&pvc->lock);
2236                         continue;
2237                 }
2238                 kvmppc_core_end_stolen(pvc);
2239                 pvc->vcore_state = VCORE_PIGGYBACK;
2240                 if (cip->total_threads >= target_threads)
2241                         break;
2242         }
2243         spin_unlock(&lp->lock);
2244 }
2245
2246 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2247 {
2248         int still_running = 0;
2249         u64 now;
2250         long ret;
2251         struct kvm_vcpu *vcpu, *vnext;
2252
2253         spin_lock(&vc->lock);
2254         now = get_tb();
2255         list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2256                                  arch.run_list) {
2257                 /* cancel pending dec exception if dec is positive */
2258                 if (now < vcpu->arch.dec_expires &&
2259                     kvmppc_core_pending_dec(vcpu))
2260                         kvmppc_core_dequeue_dec(vcpu);
2261
2262                 trace_kvm_guest_exit(vcpu);
2263
2264                 ret = RESUME_GUEST;
2265                 if (vcpu->arch.trap)
2266                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2267                                                     vcpu->arch.run_task);
2268
2269                 vcpu->arch.ret = ret;
2270                 vcpu->arch.trap = 0;
2271
2272                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2273                         if (vcpu->arch.pending_exceptions)
2274                                 kvmppc_core_prepare_to_enter(vcpu);
2275                         if (vcpu->arch.ceded)
2276                                 kvmppc_set_timer(vcpu);
2277                         else
2278                                 ++still_running;
2279                 } else {
2280                         kvmppc_remove_runnable(vc, vcpu);
2281                         wake_up(&vcpu->arch.cpu_run);
2282                 }
2283         }
2284         list_del_init(&vc->preempt_list);
2285         if (!is_master) {
2286                 if (still_running > 0) {
2287                         kvmppc_vcore_preempt(vc);
2288                 } else if (vc->runner) {
2289                         vc->vcore_state = VCORE_PREEMPT;
2290                         kvmppc_core_start_stolen(vc);
2291                 } else {
2292                         vc->vcore_state = VCORE_INACTIVE;
2293                 }
2294                 if (vc->n_runnable > 0 && vc->runner == NULL) {
2295                         /* make sure there's a candidate runner awake */
2296                         vcpu = list_first_entry(&vc->runnable_threads,
2297                                                 struct kvm_vcpu, arch.run_list);
2298                         wake_up(&vcpu->arch.cpu_run);
2299                 }
2300         }
2301         spin_unlock(&vc->lock);
2302 }
2303
2304 /*
2305  * Run a set of guest threads on a physical core.
2306  * Called with vc->lock held.
2307  */
2308 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2309 {
2310         struct kvm_vcpu *vcpu, *vnext;
2311         int i;
2312         int srcu_idx;
2313         struct core_info core_info;
2314         struct kvmppc_vcore *pvc, *vcnext;
2315         struct kvm_split_mode split_info, *sip;
2316         int split, subcore_size, active;
2317         int sub;
2318         bool thr0_done;
2319         unsigned long cmd_bit, stat_bit;
2320         int pcpu, thr;
2321         int target_threads;
2322
2323         /*
2324          * Remove from the list any threads that have a signal pending
2325          * or need a VPA update done
2326          */
2327         prepare_threads(vc);
2328
2329         /* if the runner is no longer runnable, let the caller pick a new one */
2330         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2331                 return;
2332
2333         /*
2334          * Initialize *vc.
2335          */
2336         init_master_vcore(vc);
2337         vc->preempt_tb = TB_NIL;
2338
2339         /*
2340          * Make sure we are running on primary threads, and that secondary
2341          * threads are offline.  Also check if the number of threads in this
2342          * guest are greater than the current system threads per guest.
2343          */
2344         if ((threads_per_core > 1) &&
2345             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2346                 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2347                                          arch.run_list) {
2348                         vcpu->arch.ret = -EBUSY;
2349                         kvmppc_remove_runnable(vc, vcpu);
2350                         wake_up(&vcpu->arch.cpu_run);
2351                 }
2352                 goto out;
2353         }
2354
2355         /*
2356          * See if we could run any other vcores on the physical core
2357          * along with this one.
2358          */
2359         init_core_info(&core_info, vc);
2360         pcpu = smp_processor_id();
2361         target_threads = threads_per_subcore;
2362         if (target_smt_mode && target_smt_mode < target_threads)
2363                 target_threads = target_smt_mode;
2364         if (vc->num_threads < target_threads)
2365                 collect_piggybacks(&core_info, target_threads);
2366
2367         /* Decide on micro-threading (split-core) mode */
2368         subcore_size = threads_per_subcore;
2369         cmd_bit = stat_bit = 0;
2370         split = core_info.n_subcores;
2371         sip = NULL;
2372         if (split > 1) {
2373                 /* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2374                 if (split == 2 && (dynamic_mt_modes & 2)) {
2375                         cmd_bit = HID0_POWER8_1TO2LPAR;
2376                         stat_bit = HID0_POWER8_2LPARMODE;
2377                 } else {
2378                         split = 4;
2379                         cmd_bit = HID0_POWER8_1TO4LPAR;
2380                         stat_bit = HID0_POWER8_4LPARMODE;
2381                 }
2382                 subcore_size = MAX_SMT_THREADS / split;
2383                 sip = &split_info;
2384                 memset(&split_info, 0, sizeof(split_info));
2385                 split_info.rpr = mfspr(SPRN_RPR);
2386                 split_info.pmmar = mfspr(SPRN_PMMAR);
2387                 split_info.ldbar = mfspr(SPRN_LDBAR);
2388                 split_info.subcore_size = subcore_size;
2389                 for (sub = 0; sub < core_info.n_subcores; ++sub)
2390                         split_info.master_vcs[sub] =
2391                                 list_first_entry(&core_info.vcs[sub],
2392                                         struct kvmppc_vcore, preempt_list);
2393                 /* order writes to split_info before kvm_split_mode pointer */
2394                 smp_wmb();
2395         }
2396         pcpu = smp_processor_id();
2397         for (thr = 0; thr < threads_per_subcore; ++thr)
2398                 paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2399
2400         /* Initiate micro-threading (split-core) if required */
2401         if (cmd_bit) {
2402                 unsigned long hid0 = mfspr(SPRN_HID0);
2403
2404                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2405                 mb();
2406                 mtspr(SPRN_HID0, hid0);
2407                 isync();
2408                 for (;;) {
2409                         hid0 = mfspr(SPRN_HID0);
2410                         if (hid0 & stat_bit)
2411                                 break;
2412                         cpu_relax();
2413                 }
2414         }
2415
2416         /* Start all the threads */
2417         active = 0;
2418         for (sub = 0; sub < core_info.n_subcores; ++sub) {
2419                 thr = subcore_thread_map[sub];
2420                 thr0_done = false;
2421                 active |= 1 << thr;
2422                 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2423                         pvc->pcpu = pcpu + thr;
2424                         list_for_each_entry(vcpu, &pvc->runnable_threads,
2425                                             arch.run_list) {
2426                                 kvmppc_start_thread(vcpu, pvc);
2427                                 kvmppc_create_dtl_entry(vcpu, pvc);
2428                                 trace_kvm_guest_enter(vcpu);
2429                                 if (!vcpu->arch.ptid)
2430                                         thr0_done = true;
2431                                 active |= 1 << (thr + vcpu->arch.ptid);
2432                         }
2433                         /*
2434                          * We need to start the first thread of each subcore
2435                          * even if it doesn't have a vcpu.
2436                          */
2437                         if (pvc->master_vcore == pvc && !thr0_done)
2438                                 kvmppc_start_thread(NULL, pvc);
2439                         thr += pvc->num_threads;
2440                 }
2441         }
2442
2443         /*
2444          * Ensure that split_info.do_nap is set after setting
2445          * the vcore pointer in the PACA of the secondaries.
2446          */
2447         smp_mb();
2448         if (cmd_bit)
2449                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
2450
2451         /*
2452          * When doing micro-threading, poke the inactive threads as well.
2453          * This gets them to the nap instruction after kvm_do_nap,
2454          * which reduces the time taken to unsplit later.
2455          */
2456         if (split > 1)
2457                 for (thr = 1; thr < threads_per_subcore; ++thr)
2458                         if (!(active & (1 << thr)))
2459                                 kvmppc_ipi_thread(pcpu + thr);
2460
2461         vc->vcore_state = VCORE_RUNNING;
2462         preempt_disable();
2463
2464         trace_kvmppc_run_core(vc, 0);
2465
2466         for (sub = 0; sub < core_info.n_subcores; ++sub)
2467                 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2468                         spin_unlock(&pvc->lock);
2469
2470         kvm_guest_enter();
2471
2472         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2473
2474         if (vc->mpp_buffer_is_valid)
2475                 kvmppc_start_restoring_l2_cache(vc);
2476
2477         __kvmppc_vcore_entry();
2478
2479         if (vc->mpp_buffer)
2480                 kvmppc_start_saving_l2_cache(vc);
2481
2482         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2483
2484         spin_lock(&vc->lock);
2485         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2486         vc->vcore_state = VCORE_EXITING;
2487
2488         /* wait for secondary threads to finish writing their state to memory */
2489         kvmppc_wait_for_nap();
2490
2491         /* Return to whole-core mode if we split the core earlier */
2492         if (split > 1) {
2493                 unsigned long hid0 = mfspr(SPRN_HID0);
2494                 unsigned long loops = 0;
2495
2496                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2497                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2498                 mb();
2499                 mtspr(SPRN_HID0, hid0);
2500                 isync();
2501                 for (;;) {
2502                         hid0 = mfspr(SPRN_HID0);
2503                         if (!(hid0 & stat_bit))
2504                                 break;
2505                         cpu_relax();
2506                         ++loops;
2507                 }
2508                 split_info.do_nap = 0;
2509         }
2510
2511         /* Let secondaries go back to the offline loop */
2512         for (i = 0; i < threads_per_subcore; ++i) {
2513                 kvmppc_release_hwthread(pcpu + i);
2514                 if (sip && sip->napped[i])
2515                         kvmppc_ipi_thread(pcpu + i);
2516         }
2517
2518         spin_unlock(&vc->lock);
2519
2520         /* make sure updates to secondary vcpu structs are visible now */
2521         smp_mb();
2522         kvm_guest_exit();
2523
2524         for (sub = 0; sub < core_info.n_subcores; ++sub)
2525                 list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2526                                          preempt_list)
2527                         post_guest_process(pvc, pvc == vc);
2528
2529         spin_lock(&vc->lock);
2530         preempt_enable();
2531
2532  out:
2533         vc->vcore_state = VCORE_INACTIVE;
2534         trace_kvmppc_run_core(vc, 1);
2535 }
2536
2537 /*
2538  * Wait for some other vcpu thread to execute us, and
2539  * wake us up when we need to handle something in the host.
2540  */
2541 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2542                                  struct kvm_vcpu *vcpu, int wait_state)
2543 {
2544         DEFINE_WAIT(wait);
2545
2546         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2547         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2548                 spin_unlock(&vc->lock);
2549                 schedule();
2550                 spin_lock(&vc->lock);
2551         }
2552         finish_wait(&vcpu->arch.cpu_run, &wait);
2553 }
2554
2555 /*
2556  * All the vcpus in this vcore are idle, so wait for a decrementer
2557  * or external interrupt to one of the vcpus.  vc->lock is held.
2558  */
2559 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2560 {
2561         struct kvm_vcpu *vcpu;
2562         int do_sleep = 1;
2563
2564         DEFINE_WAIT(wait);
2565
2566         prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2567
2568         /*
2569          * Check one last time for pending exceptions and ceded state after
2570          * we put ourselves on the wait queue
2571          */
2572         list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
2573                 if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded) {
2574                         do_sleep = 0;
2575                         break;
2576                 }
2577         }
2578
2579         if (!do_sleep) {
2580                 finish_wait(&vc->wq, &wait);
2581                 return;
2582         }
2583
2584         vc->vcore_state = VCORE_SLEEPING;
2585         trace_kvmppc_vcore_blocked(vc, 0);
2586         spin_unlock(&vc->lock);
2587         schedule();
2588         finish_wait(&vc->wq, &wait);
2589         spin_lock(&vc->lock);
2590         vc->vcore_state = VCORE_INACTIVE;
2591         trace_kvmppc_vcore_blocked(vc, 1);
2592 }
2593
2594 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2595 {
2596         int n_ceded;
2597         struct kvmppc_vcore *vc;
2598         struct kvm_vcpu *v, *vn;
2599
2600         trace_kvmppc_run_vcpu_enter(vcpu);
2601
2602         kvm_run->exit_reason = 0;
2603         vcpu->arch.ret = RESUME_GUEST;
2604         vcpu->arch.trap = 0;
2605         kvmppc_update_vpas(vcpu);
2606
2607         /*
2608          * Synchronize with other threads in this virtual core
2609          */
2610         vc = vcpu->arch.vcore;
2611         spin_lock(&vc->lock);
2612         vcpu->arch.ceded = 0;
2613         vcpu->arch.run_task = current;
2614         vcpu->arch.kvm_run = kvm_run;
2615         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2616         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2617         vcpu->arch.busy_preempt = TB_NIL;
2618         list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
2619         ++vc->n_runnable;
2620
2621         /*
2622          * This happens the first time this is called for a vcpu.
2623          * If the vcore is already running, we may be able to start
2624          * this thread straight away and have it join in.
2625          */
2626         if (!signal_pending(current)) {
2627                 if (vc->vcore_state == VCORE_PIGGYBACK) {
2628                         struct kvmppc_vcore *mvc = vc->master_vcore;
2629                         if (spin_trylock(&mvc->lock)) {
2630                                 if (mvc->vcore_state == VCORE_RUNNING &&
2631                                     !VCORE_IS_EXITING(mvc)) {
2632                                         kvmppc_create_dtl_entry(vcpu, vc);
2633                                         kvmppc_start_thread(vcpu, vc);
2634                                         trace_kvm_guest_enter(vcpu);
2635                                 }
2636                                 spin_unlock(&mvc->lock);
2637                         }
2638                 } else if (vc->vcore_state == VCORE_RUNNING &&
2639                            !VCORE_IS_EXITING(vc)) {
2640                         kvmppc_create_dtl_entry(vcpu, vc);
2641                         kvmppc_start_thread(vcpu, vc);
2642                         trace_kvm_guest_enter(vcpu);
2643                 } else if (vc->vcore_state == VCORE_SLEEPING) {
2644                         wake_up(&vc->wq);
2645                 }
2646
2647         }
2648
2649         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2650                !signal_pending(current)) {
2651                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2652                         kvmppc_vcore_end_preempt(vc);
2653
2654                 if (vc->vcore_state != VCORE_INACTIVE) {
2655                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
2656                         continue;
2657                 }
2658                 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
2659                                          arch.run_list) {
2660                         kvmppc_core_prepare_to_enter(v);
2661                         if (signal_pending(v->arch.run_task)) {
2662                                 kvmppc_remove_runnable(vc, v);
2663                                 v->stat.signal_exits++;
2664                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2665                                 v->arch.ret = -EINTR;
2666                                 wake_up(&v->arch.cpu_run);
2667                         }
2668                 }
2669                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2670                         break;
2671                 n_ceded = 0;
2672                 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
2673                         if (!v->arch.pending_exceptions)
2674                                 n_ceded += v->arch.ceded;
2675                         else
2676                                 v->arch.ceded = 0;
2677                 }
2678                 vc->runner = vcpu;
2679                 if (n_ceded == vc->n_runnable) {
2680                         kvmppc_vcore_blocked(vc);
2681                 } else if (need_resched()) {
2682                         kvmppc_vcore_preempt(vc);
2683                         /* Let something else run */
2684                         cond_resched_lock(&vc->lock);
2685                         if (vc->vcore_state == VCORE_PREEMPT)
2686                                 kvmppc_vcore_end_preempt(vc);
2687                 } else {
2688                         kvmppc_run_core(vc);
2689                 }
2690                 vc->runner = NULL;
2691         }
2692
2693         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2694                (vc->vcore_state == VCORE_RUNNING ||
2695                 vc->vcore_state == VCORE_EXITING ||
2696                 vc->vcore_state == VCORE_PIGGYBACK))
2697                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
2698
2699         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2700                 kvmppc_vcore_end_preempt(vc);
2701
2702         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2703                 kvmppc_remove_runnable(vc, vcpu);
2704                 vcpu->stat.signal_exits++;
2705                 kvm_run->exit_reason = KVM_EXIT_INTR;
2706                 vcpu->arch.ret = -EINTR;
2707         }
2708
2709         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2710                 /* Wake up some vcpu to run the core */
2711                 v = list_first_entry(&vc->runnable_threads,
2712                                      struct kvm_vcpu, arch.run_list);
2713                 wake_up(&v->arch.cpu_run);
2714         }
2715
2716         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2717         spin_unlock(&vc->lock);
2718         return vcpu->arch.ret;
2719 }
2720
2721 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2722 {
2723         int r;
2724         int srcu_idx;
2725
2726         if (!vcpu->arch.sane) {
2727                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2728                 return -EINVAL;
2729         }
2730
2731         kvmppc_core_prepare_to_enter(vcpu);
2732
2733         /* No need to go into the guest when all we'll do is come back out */
2734         if (signal_pending(current)) {
2735                 run->exit_reason = KVM_EXIT_INTR;
2736                 return -EINTR;
2737         }
2738
2739         atomic_inc(&vcpu->kvm->arch.vcpus_running);
2740         /* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2741         smp_mb();
2742
2743         /* On the first time here, set up HTAB and VRMA */
2744         if (!vcpu->kvm->arch.hpte_setup_done) {
2745                 r = kvmppc_hv_setup_htab_rma(vcpu);
2746                 if (r)
2747                         goto out;
2748         }
2749
2750         flush_fp_to_thread(current);
2751         flush_altivec_to_thread(current);
2752         flush_vsx_to_thread(current);
2753         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2754         vcpu->arch.pgdir = current->mm->pgd;
2755         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2756
2757         do {
2758                 r = kvmppc_run_vcpu(run, vcpu);
2759
2760                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2761                     !(vcpu->arch.shregs.msr & MSR_PR)) {
2762                         trace_kvm_hcall_enter(vcpu);
2763                         r = kvmppc_pseries_do_hcall(vcpu);
2764                         trace_kvm_hcall_exit(vcpu, r);
2765                         kvmppc_core_prepare_to_enter(vcpu);
2766                 } else if (r == RESUME_PAGE_FAULT) {
2767                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2768                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
2769                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2770                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2771                 }
2772         } while (is_kvmppc_resume_guest(r));
2773
2774  out:
2775         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2776         atomic_dec(&vcpu->kvm->arch.vcpus_running);
2777         return r;
2778 }
2779
2780 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2781                                      int linux_psize)
2782 {
2783         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2784
2785         if (!def->shift)
2786                 return;
2787         (*sps)->page_shift = def->shift;
2788         (*sps)->slb_enc = def->sllp;
2789         (*sps)->enc[0].page_shift = def->shift;
2790         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
2791         /*
2792          * Add 16MB MPSS support if host supports it
2793          */
2794         if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2795                 (*sps)->enc[1].page_shift = 24;
2796                 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2797         }
2798         (*sps)++;
2799 }
2800
2801 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2802                                          struct kvm_ppc_smmu_info *info)
2803 {
2804         struct kvm_ppc_one_seg_page_size *sps;
2805
2806         info->flags = KVM_PPC_PAGE_SIZES_REAL;
2807         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2808                 info->flags |= KVM_PPC_1T_SEGMENTS;
2809         info->slb_size = mmu_slb_size;
2810
2811         /* We only support these sizes for now, and no muti-size segments */
2812         sps = &info->sps[0];
2813         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2814         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2815         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2816
2817         return 0;
2818 }
2819
2820 /*
2821  * Get (and clear) the dirty memory log for a memory slot.
2822  */
2823 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2824                                          struct kvm_dirty_log *log)
2825 {
2826         struct kvm_memslots *slots;
2827         struct kvm_memory_slot *memslot;
2828         int r;
2829         unsigned long n;
2830
2831         mutex_lock(&kvm->slots_lock);
2832
2833         r = -EINVAL;
2834         if (log->slot >= KVM_USER_MEM_SLOTS)
2835                 goto out;
2836
2837         slots = kvm_memslots(kvm);
2838         memslot = id_to_memslot(slots, log->slot);
2839         r = -ENOENT;
2840         if (!memslot->dirty_bitmap)
2841                 goto out;
2842
2843         n = kvm_dirty_bitmap_bytes(memslot);
2844         memset(memslot->dirty_bitmap, 0, n);
2845
2846         r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2847         if (r)
2848                 goto out;
2849
2850         r = -EFAULT;
2851         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2852                 goto out;
2853
2854         r = 0;
2855 out:
2856         mutex_unlock(&kvm->slots_lock);
2857         return r;
2858 }
2859
2860 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2861                                         struct kvm_memory_slot *dont)
2862 {
2863         if (!dont || free->arch.rmap != dont->arch.rmap) {
2864                 vfree(free->arch.rmap);
2865                 free->arch.rmap = NULL;
2866         }
2867 }
2868
2869 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2870                                          unsigned long npages)
2871 {
2872         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2873         if (!slot->arch.rmap)
2874                 return -ENOMEM;
2875
2876         return 0;
2877 }
2878
2879 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2880                                         struct kvm_memory_slot *memslot,
2881                                         const struct kvm_userspace_memory_region *mem)
2882 {
2883         return 0;
2884 }
2885
2886 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2887                                 const struct kvm_userspace_memory_region *mem,
2888                                 const struct kvm_memory_slot *old,
2889                                 const struct kvm_memory_slot *new)
2890 {
2891         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2892         struct kvm_memslots *slots;
2893         struct kvm_memory_slot *memslot;
2894
2895         if (npages && old->npages) {
2896                 /*
2897                  * If modifying a memslot, reset all the rmap dirty bits.
2898                  * If this is a new memslot, we don't need to do anything
2899                  * since the rmap array starts out as all zeroes,
2900                  * i.e. no pages are dirty.
2901                  */
2902                 slots = kvm_memslots(kvm);
2903                 memslot = id_to_memslot(slots, mem->slot);
2904                 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2905         }
2906 }
2907
2908 /*
2909  * Update LPCR values in kvm->arch and in vcores.
2910  * Caller must hold kvm->lock.
2911  */
2912 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2913 {
2914         long int i;
2915         u32 cores_done = 0;
2916
2917         if ((kvm->arch.lpcr & mask) == lpcr)
2918                 return;
2919
2920         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2921
2922         for (i = 0; i < KVM_MAX_VCORES; ++i) {
2923                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2924                 if (!vc)
2925                         continue;
2926                 spin_lock(&vc->lock);
2927                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2928                 spin_unlock(&vc->lock);
2929                 if (++cores_done >= kvm->arch.online_vcores)
2930                         break;
2931         }
2932 }
2933
2934 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2935 {
2936         return;
2937 }
2938
2939 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2940 {
2941         int err = 0;
2942         struct kvm *kvm = vcpu->kvm;
2943         unsigned long hva;
2944         struct kvm_memory_slot *memslot;
2945         struct vm_area_struct *vma;
2946         unsigned long lpcr = 0, senc;
2947         unsigned long psize, porder;
2948         int srcu_idx;
2949
2950         mutex_lock(&kvm->lock);
2951         if (kvm->arch.hpte_setup_done)
2952                 goto out;       /* another vcpu beat us to it */
2953
2954         /* Allocate hashed page table (if not done already) and reset it */
2955         if (!kvm->arch.hpt_virt) {
2956                 err = kvmppc_alloc_hpt(kvm, NULL);
2957                 if (err) {
2958                         pr_err("KVM: Couldn't alloc HPT\n");
2959                         goto out;
2960                 }
2961         }
2962
2963         /* Look up the memslot for guest physical address 0 */
2964         srcu_idx = srcu_read_lock(&kvm->srcu);
2965         memslot = gfn_to_memslot(kvm, 0);
2966
2967         /* We must have some memory at 0 by now */
2968         err = -EINVAL;
2969         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2970                 goto out_srcu;
2971
2972         /* Look up the VMA for the start of this memory slot */
2973         hva = memslot->userspace_addr;
2974         down_read(&current->mm->mmap_sem);
2975         vma = find_vma(current->mm, hva);
2976         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2977                 goto up_out;
2978
2979         psize = vma_kernel_pagesize(vma);
2980         porder = __ilog2(psize);
2981
2982         up_read(&current->mm->mmap_sem);
2983
2984         /* We can handle 4k, 64k or 16M pages in the VRMA */
2985         err = -EINVAL;
2986         if (!(psize == 0x1000 || psize == 0x10000 ||
2987               psize == 0x1000000))
2988                 goto out_srcu;
2989
2990         /* Update VRMASD field in the LPCR */
2991         senc = slb_pgsize_encoding(psize);
2992         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2993                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
2994         /* the -4 is to account for senc values starting at 0x10 */
2995         lpcr = senc << (LPCR_VRMASD_SH - 4);
2996
2997         /* Create HPTEs in the hash page table for the VRMA */
2998         kvmppc_map_vrma(vcpu, memslot, porder);
2999
3000         kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3001
3002         /* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
3003         smp_wmb();
3004         kvm->arch.hpte_setup_done = 1;
3005         err = 0;
3006  out_srcu:
3007         srcu_read_unlock(&kvm->srcu, srcu_idx);
3008  out:
3009         mutex_unlock(&kvm->lock);
3010         return err;
3011
3012  up_out:
3013         up_read(&current->mm->mmap_sem);
3014         goto out_srcu;
3015 }
3016
3017 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3018 {
3019         unsigned long lpcr, lpid;
3020         char buf[32];
3021
3022         /* Allocate the guest's logical partition ID */
3023
3024         lpid = kvmppc_alloc_lpid();
3025         if ((long)lpid < 0)
3026                 return -ENOMEM;
3027         kvm->arch.lpid = lpid;
3028
3029         /*
3030          * Since we don't flush the TLB when tearing down a VM,
3031          * and this lpid might have previously been used,
3032          * make sure we flush on each core before running the new VM.
3033          */
3034         cpumask_setall(&kvm->arch.need_tlb_flush);
3035
3036         /* Start out with the default set of hcalls enabled */
3037         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3038                sizeof(kvm->arch.enabled_hcalls));
3039
3040         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3041
3042         /* Init LPCR for virtual RMA mode */
3043         kvm->arch.host_lpid = mfspr(SPRN_LPID);
3044         kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3045         lpcr &= LPCR_PECE | LPCR_LPES;
3046         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3047                 LPCR_VPM0 | LPCR_VPM1;
3048         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3049                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3050         /* On POWER8 turn on online bit to enable PURR/SPURR */
3051         if (cpu_has_feature(CPU_FTR_ARCH_207S))
3052                 lpcr |= LPCR_ONL;
3053         kvm->arch.lpcr = lpcr;
3054
3055         /*
3056          * Track that we now have a HV mode VM active. This blocks secondary
3057          * CPU threads from coming online.
3058          */
3059         kvm_hv_vm_activated();
3060
3061         /*
3062          * Create a debugfs directory for the VM
3063          */
3064         snprintf(buf, sizeof(buf), "vm%d", current->pid);
3065         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3066         if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3067                 kvmppc_mmu_debugfs_init(kvm);
3068
3069         return 0;
3070 }
3071
3072 static void kvmppc_free_vcores(struct kvm *kvm)
3073 {
3074         long int i;
3075
3076         for (i = 0; i < KVM_MAX_VCORES; ++i) {
3077                 if (kvm->arch.vcores[i] && kvm->arch.vcores[i]->mpp_buffer) {
3078                         struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3079                         free_pages((unsigned long)vc->mpp_buffer,
3080                                    MPP_BUFFER_ORDER);
3081                 }
3082                 kfree(kvm->arch.vcores[i]);
3083         }
3084         kvm->arch.online_vcores = 0;
3085 }
3086
3087 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3088 {
3089         debugfs_remove_recursive(kvm->arch.debugfs_dir);
3090
3091         kvm_hv_vm_deactivated();
3092
3093         kvmppc_free_vcores(kvm);
3094
3095         kvmppc_free_hpt(kvm);
3096 }
3097
3098 /* We don't need to emulate any privileged instructions or dcbz */
3099 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3100                                      unsigned int inst, int *advance)
3101 {
3102         return EMULATE_FAIL;
3103 }
3104
3105 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3106                                         ulong spr_val)
3107 {
3108         return EMULATE_FAIL;
3109 }
3110
3111 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3112                                         ulong *spr_val)
3113 {
3114         return EMULATE_FAIL;
3115 }
3116
3117 static int kvmppc_core_check_processor_compat_hv(void)
3118 {
3119         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3120             !cpu_has_feature(CPU_FTR_ARCH_206))
3121                 return -EIO;
3122         return 0;
3123 }
3124
3125 static long kvm_arch_vm_ioctl_hv(struct file *filp,
3126                                  unsigned int ioctl, unsigned long arg)
3127 {
3128         struct kvm *kvm __maybe_unused = filp->private_data;
3129         void __user *argp = (void __user *)arg;
3130         long r;
3131
3132         switch (ioctl) {
3133
3134         case KVM_PPC_ALLOCATE_HTAB: {
3135                 u32 htab_order;
3136
3137                 r = -EFAULT;
3138                 if (get_user(htab_order, (u32 __user *)argp))
3139                         break;
3140                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
3141                 if (r)
3142                         break;
3143                 r = -EFAULT;
3144                 if (put_user(htab_order, (u32 __user *)argp))
3145                         break;
3146                 r = 0;
3147                 break;
3148         }
3149
3150         case KVM_PPC_GET_HTAB_FD: {
3151                 struct kvm_get_htab_fd ghf;
3152
3153                 r = -EFAULT;
3154                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
3155                         break;
3156                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3157                 break;
3158         }
3159
3160         default:
3161                 r = -ENOTTY;
3162         }
3163
3164         return r;
3165 }
3166
3167 /*
3168  * List of hcall numbers to enable by default.
3169  * For compatibility with old userspace, we enable by default
3170  * all hcalls that were implemented before the hcall-enabling
3171  * facility was added.  Note this list should not include H_RTAS.
3172  */
3173 static unsigned int default_hcall_list[] = {
3174         H_REMOVE,
3175         H_ENTER,
3176         H_READ,
3177         H_PROTECT,
3178         H_BULK_REMOVE,
3179         H_GET_TCE,
3180         H_PUT_TCE,
3181         H_SET_DABR,
3182         H_SET_XDABR,
3183         H_CEDE,
3184         H_PROD,
3185         H_CONFER,
3186         H_REGISTER_VPA,
3187 #ifdef CONFIG_KVM_XICS
3188         H_EOI,
3189         H_CPPR,
3190         H_IPI,
3191         H_IPOLL,
3192         H_XIRR,
3193         H_XIRR_X,
3194 #endif
3195         0
3196 };
3197
3198 static void init_default_hcalls(void)
3199 {
3200         int i;
3201         unsigned int hcall;
3202
3203         for (i = 0; default_hcall_list[i]; ++i) {
3204                 hcall = default_hcall_list[i];
3205                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3206                 __set_bit(hcall / 4, default_enabled_hcalls);
3207         }
3208 }
3209
3210 static struct kvmppc_ops kvm_ops_hv = {
3211         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3212         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3213         .get_one_reg = kvmppc_get_one_reg_hv,
3214         .set_one_reg = kvmppc_set_one_reg_hv,
3215         .vcpu_load   = kvmppc_core_vcpu_load_hv,
3216         .vcpu_put    = kvmppc_core_vcpu_put_hv,
3217         .set_msr     = kvmppc_set_msr_hv,
3218         .vcpu_run    = kvmppc_vcpu_run_hv,
3219         .vcpu_create = kvmppc_core_vcpu_create_hv,
3220         .vcpu_free   = kvmppc_core_vcpu_free_hv,
3221         .check_requests = kvmppc_core_check_requests_hv,
3222         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
3223         .flush_memslot  = kvmppc_core_flush_memslot_hv,
3224         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3225         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
3226         .unmap_hva = kvm_unmap_hva_hv,
3227         .unmap_hva_range = kvm_unmap_hva_range_hv,
3228         .age_hva  = kvm_age_hva_hv,
3229         .test_age_hva = kvm_test_age_hva_hv,
3230         .set_spte_hva = kvm_set_spte_hva_hv,
3231         .mmu_destroy  = kvmppc_mmu_destroy_hv,
3232         .free_memslot = kvmppc_core_free_memslot_hv,
3233         .create_memslot = kvmppc_core_create_memslot_hv,
3234         .init_vm =  kvmppc_core_init_vm_hv,
3235         .destroy_vm = kvmppc_core_destroy_vm_hv,
3236         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3237         .emulate_op = kvmppc_core_emulate_op_hv,
3238         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3239         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3240         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3241         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
3242         .hcall_implemented = kvmppc_hcall_impl_hv,
3243 };
3244
3245 static int kvmppc_book3s_init_hv(void)
3246 {
3247         int r;
3248         /*
3249          * FIXME!! Do we need to check on all cpus ?
3250          */
3251         r = kvmppc_core_check_processor_compat_hv();
3252         if (r < 0)
3253                 return -ENODEV;
3254
3255         kvm_ops_hv.owner = THIS_MODULE;
3256         kvmppc_hv_ops = &kvm_ops_hv;
3257
3258         init_default_hcalls();
3259
3260         init_vcore_lists();
3261
3262         r = kvmppc_mmu_hv_init();
3263         return r;
3264 }
3265
3266 static void kvmppc_book3s_exit_hv(void)
3267 {
3268         kvmppc_hv_ops = NULL;
3269 }
3270
3271 module_init(kvmppc_book3s_init_hv);
3272 module_exit(kvmppc_book3s_exit_hv);
3273 MODULE_LICENSE("GPL");
3274 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3275 MODULE_ALIAS("devname:kvm");