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/cpu.h>
  31 #include <linux/cpumask.h>
  32 #include <linux/spinlock.h>
  33 #include <linux/page-flags.h>
  34 #include <linux/srcu.h>
  35 #include <linux/miscdevice.h>
  36 #include <linux/debugfs.h>
  37
  38 #include <asm/reg.h>
  39 #include <asm/cputable.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 <asm/hmi.h>
  56 #include <asm/pnv-pci.h>
  57 #include <linux/gfp.h>
  58 #include <linux/vmalloc.h>
  59 #include <linux/highmem.h>
  60 #include <linux/hugetlb.h>
  61 #include <linux/kvm_irqfd.h>
  62 #include <linux/irqbypass.h>
  63 #include <linux/module.h>
  64 #include <linux/compiler.h>
  65
  66 #include "book3s.h"
  67
  68 #define CREATE_TRACE_POINTS
  69 #include "trace_hv.h"
  70
  71 /* #define EXIT_DEBUG */
  72 /* #define EXIT_DEBUG_SIMPLE */
  73 /* #define EXIT_DEBUG_INT */
  74
  75 /* Used to indicate that a guest page fault needs to be handled */
  76 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  77 /* Used to indicate that a guest passthrough interrupt needs to be handled */
  78 #define RESUME_PASSTHROUGH      (RESUME_GUEST | RESUME_FLAG_ARCH2)
  79
  80 /* Used as a "null" value for timebase values */
  81 #define TB_NIL  (~(u64)0)
  82
  83 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
  84
  85 static int dynamic_mt_modes = 6;
  86 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
  87 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
  88 static int target_smt_mode;
  89 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
  90 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
  91
  92 #ifdef CONFIG_KVM_XICS
  93 static struct kernel_param_ops module_param_ops = {
  94         .set = param_set_int,
  95         .get = param_get_int,
  96 };
  97
  98 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass,
  99                                                         S_IRUGO | S_IWUSR);
 100 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
 101
 102 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect,
 103                                                         S_IRUGO | S_IWUSR);
 104 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
 105 #endif
 106
 107 /* Maximum halt poll interval defaults to KVM_HALT_POLL_NS_DEFAULT */
 108 static unsigned int halt_poll_max_ns = KVM_HALT_POLL_NS_DEFAULT;
 109 module_param(halt_poll_max_ns, uint, S_IRUGO | S_IWUSR);
 110 MODULE_PARM_DESC(halt_poll_max_ns, "Maximum halt poll time in ns");
 111
 112 /* Factor by which the vcore halt poll interval is grown, default is to double
 113  */
 114 static unsigned int halt_poll_ns_grow = 2;
 115 module_param(halt_poll_ns_grow, int, S_IRUGO);
 116 MODULE_PARM_DESC(halt_poll_ns_grow, "Factor halt poll time is grown by");
 117
 118 /* Factor by which the vcore halt poll interval is shrunk, default is to reset
 119  */
 120 static unsigned int halt_poll_ns_shrink;
 121 module_param(halt_poll_ns_shrink, int, S_IRUGO);
 122 MODULE_PARM_DESC(halt_poll_ns_shrink, "Factor halt poll time is shrunk by");
 123
 124 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
 125 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
 126
 127 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
 128                 int *ip)
 129 {
 130         int i = *ip;
 131         struct kvm_vcpu *vcpu;
 132
 133         while (++i < MAX_SMT_THREADS) {
 134                 vcpu = READ_ONCE(vc->runnable_threads[i]);
 135                 if (vcpu) {
 136                         *ip = i;
 137                         return vcpu;
 138                 }
 139         }
 140         return NULL;
 141 }
 142
 143 /* Used to traverse the list of runnable threads for a given vcore */
 144 #define for_each_runnable_thread(i, vcpu, vc) \
 145         for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
 146
 147 static bool kvmppc_ipi_thread(int cpu)
 148 {
 149         /* On POWER8 for IPIs to threads in the same core, use msgsnd */
 150         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
 151                 preempt_disable();
 152                 if (cpu_first_thread_sibling(cpu) ==
 153                     cpu_first_thread_sibling(smp_processor_id())) {
 154                         unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
 155                         msg |= cpu_thread_in_core(cpu);
 156                         smp_mb();
 157                         __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 158                         preempt_enable();
 159                         return true;
 160                 }
 161                 preempt_enable();
 162         }
 163
 164 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
 165         if (cpu >= 0 && cpu < nr_cpu_ids && paca[cpu].kvm_hstate.xics_phys) {
 166                 xics_wake_cpu(cpu);
 167                 return true;
 168         }
 169 #endif
 170
 171         return false;
 172 }
 173
 174 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
 175 {
 176         int cpu;
 177         struct swait_queue_head *wqp;
 178
 179         wqp = kvm_arch_vcpu_wq(vcpu);
 180         if (swait_active(wqp)) {
 181                 swake_up(wqp);
 182                 ++vcpu->stat.halt_wakeup;
 183         }
 184
 185         if (kvmppc_ipi_thread(vcpu->arch.thread_cpu))
 186                 return;
 187
 188         /* CPU points to the first thread of the core */
 189         cpu = vcpu->cpu;
 190         if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
 191                 smp_send_reschedule(cpu);
 192 }
 193
 194 /*
 195  * We use the vcpu_load/put functions to measure stolen time.
 196  * Stolen time is counted as time when either the vcpu is able to
 197  * run as part of a virtual core, but the task running the vcore
 198  * is preempted or sleeping, or when the vcpu needs something done
 199  * in the kernel by the task running the vcpu, but that task is
 200  * preempted or sleeping.  Those two things have to be counted
 201  * separately, since one of the vcpu tasks will take on the job
 202  * of running the core, and the other vcpu tasks in the vcore will
 203  * sleep waiting for it to do that, but that sleep shouldn't count
 204  * as stolen time.
 205  *
 206  * Hence we accumulate stolen time when the vcpu can run as part of
 207  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 208  * needs its task to do other things in the kernel (for example,
 209  * service a page fault) in busy_stolen.  We don't accumulate
 210  * stolen time for a vcore when it is inactive, or for a vcpu
 211  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 212  * a misnomer; it means that the vcpu task is not executing in
 213  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 214  * the kernel.  We don't have any way of dividing up that time
 215  * between time that the vcpu is genuinely stopped, time that
 216  * the task is actively working on behalf of the vcpu, and time
 217  * that the task is preempted, so we don't count any of it as
 218  * stolen.
 219  *
 220  * Updates to busy_stolen are protected by arch.tbacct_lock;
 221  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
 222  * lock.  The stolen times are measured in units of timebase ticks.
 223  * (Note that the != TB_NIL checks below are purely defensive;
 224  * they should never fail.)
 225  */
 226
 227 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
 228 {
 229         unsigned long flags;
 230
 231         spin_lock_irqsave(&vc->stoltb_lock, flags);
 232         vc->preempt_tb = mftb();
 233         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 234 }
 235
 236 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
 237 {
 238         unsigned long flags;
 239
 240         spin_lock_irqsave(&vc->stoltb_lock, flags);
 241         if (vc->preempt_tb != TB_NIL) {
 242                 vc->stolen_tb += mftb() - vc->preempt_tb;
 243                 vc->preempt_tb = TB_NIL;
 244         }
 245         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 246 }
 247
 248 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 249 {
 250         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 251         unsigned long flags;
 252
 253         /*
 254          * We can test vc->runner without taking the vcore lock,
 255          * because only this task ever sets vc->runner to this
 256          * vcpu, and once it is set to this vcpu, only this task
 257          * ever sets it to NULL.
 258          */
 259         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 260                 kvmppc_core_end_stolen(vc);
 261
 262         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 263         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 264             vcpu->arch.busy_preempt != TB_NIL) {
 265                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 266                 vcpu->arch.busy_preempt = TB_NIL;
 267         }
 268         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 269 }
 270
 271 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 272 {
 273         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 274         unsigned long flags;
 275
 276         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 277                 kvmppc_core_start_stolen(vc);
 278
 279         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 280         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 281                 vcpu->arch.busy_preempt = mftb();
 282         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 283 }
 284
 285 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
 286 {
 287         /*
 288          * Check for illegal transactional state bit combination
 289          * and if we find it, force the TS field to a safe state.
 290          */
 291         if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
 292                 msr &= ~MSR_TS_MASK;
 293         vcpu->arch.shregs.msr = msr;
 294         kvmppc_end_cede(vcpu);
 295 }
 296
 297 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 298 {
 299         vcpu->arch.pvr = pvr;
 300 }
 301
 302 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 303 {
 304         unsigned long pcr = 0;
 305         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 306
 307         if (arch_compat) {
 308                 switch (arch_compat) {
 309                 case PVR_ARCH_205:
 310                         /*
 311                          * If an arch bit is set in PCR, all the defined
 312                          * higher-order arch bits also have to be set.
 313                          */
 314                         pcr = PCR_ARCH_206 | PCR_ARCH_205;
 315                         break;
 316                 case PVR_ARCH_206:
 317                 case PVR_ARCH_206p:
 318                         pcr = PCR_ARCH_206;
 319                         break;
 320                 case PVR_ARCH_207:
 321                         break;
 322                 default:
 323                         return -EINVAL;
 324                 }
 325
 326                 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
 327                         /* POWER7 can't emulate POWER8 */
 328                         if (!(pcr & PCR_ARCH_206))
 329                                 return -EINVAL;
 330                         pcr &= ~PCR_ARCH_206;
 331                 }
 332         }
 333
 334         spin_lock(&vc->lock);
 335         vc->arch_compat = arch_compat;
 336         vc->pcr = pcr;
 337         spin_unlock(&vc->lock);
 338
 339         return 0;
 340 }
 341
 342 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 343 {
 344         int r;
 345
 346         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 347         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 348                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
 349         for (r = 0; r < 16; ++r)
 350                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 351                        r, kvmppc_get_gpr(vcpu, r),
 352                        r+16, kvmppc_get_gpr(vcpu, r+16));
 353         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 354                vcpu->arch.ctr, vcpu->arch.lr);
 355         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 356                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 357         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 358                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 359         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 360                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 361         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 362                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 363         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 364         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 365                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 366         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 367         for (r = 0; r < vcpu->arch.slb_max; ++r)
 368                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 369                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 370         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 371                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 372                vcpu->arch.last_inst);
 373 }
 374
 375 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 376 {
 377         return kvm_get_vcpu_by_id(kvm, id);
 378 }
 379
 380 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 381 {
 382         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 383         vpa->yield_count = cpu_to_be32(1);
 384 }
 385
 386 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 387                    unsigned long addr, unsigned long len)
 388 {
 389         /* check address is cacheline aligned */
 390         if (addr & (L1_CACHE_BYTES - 1))
 391                 return -EINVAL;
 392         spin_lock(&vcpu->arch.vpa_update_lock);
 393         if (v->next_gpa != addr || v->len != len) {
 394                 v->next_gpa = addr;
 395                 v->len = addr ? len : 0;
 396                 v->update_pending = 1;
 397         }
 398         spin_unlock(&vcpu->arch.vpa_update_lock);
 399         return 0;
 400 }
 401
 402 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 403 struct reg_vpa {
 404         u32 dummy;
 405         union {
 406                 __be16 hword;
 407                 __be32 word;
 408         } length;
 409 };
 410
 411 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 412 {
 413         if (vpap->update_pending)
 414                 return vpap->next_gpa != 0;
 415         return vpap->pinned_addr != NULL;
 416 }
 417
 418 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 419                                        unsigned long flags,
 420                                        unsigned long vcpuid, unsigned long vpa)
 421 {
 422         struct kvm *kvm = vcpu->kvm;
 423         unsigned long len, nb;
 424         void *va;
 425         struct kvm_vcpu *tvcpu;
 426         int err;
 427         int subfunc;
 428         struct kvmppc_vpa *vpap;
 429
 430         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 431         if (!tvcpu)
 432                 return H_PARAMETER;
 433
 434         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 435         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 436             subfunc == H_VPA_REG_SLB) {
 437                 /* Registering new area - address must be cache-line aligned */
 438                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 439                         return H_PARAMETER;
 440
 441                 /* convert logical addr to kernel addr and read length */
 442                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 443                 if (va == NULL)
 444                         return H_PARAMETER;
 445                 if (subfunc == H_VPA_REG_VPA)
 446                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 447                 else
 448                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 449                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 450
 451                 /* Check length */
 452                 if (len > nb || len < sizeof(struct reg_vpa))
 453                         return H_PARAMETER;
 454         } else {
 455                 vpa = 0;
 456                 len = 0;
 457         }
 458
 459         err = H_PARAMETER;
 460         vpap = NULL;
 461         spin_lock(&tvcpu->arch.vpa_update_lock);
 462
 463         switch (subfunc) {
 464         case H_VPA_REG_VPA:             /* register VPA */
 465                 if (len < sizeof(struct lppaca))
 466                         break;
 467                 vpap = &tvcpu->arch.vpa;
 468                 err = 0;
 469                 break;
 470
 471         case H_VPA_REG_DTL:             /* register DTL */
 472                 if (len < sizeof(struct dtl_entry))
 473                         break;
 474                 len -= len % sizeof(struct dtl_entry);
 475
 476                 /* Check that they have previously registered a VPA */
 477                 err = H_RESOURCE;
 478                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 479                         break;
 480
 481                 vpap = &tvcpu->arch.dtl;
 482                 err = 0;
 483                 break;
 484
 485         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 486                 /* Check that they have previously registered a VPA */
 487                 err = H_RESOURCE;
 488                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 489                         break;
 490
 491                 vpap = &tvcpu->arch.slb_shadow;
 492                 err = 0;
 493                 break;
 494
 495         case H_VPA_DEREG_VPA:           /* deregister VPA */
 496                 /* Check they don't still have a DTL or SLB buf registered */
 497                 err = H_RESOURCE;
 498                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 499                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 500                         break;
 501
 502                 vpap = &tvcpu->arch.vpa;
 503                 err = 0;
 504                 break;
 505
 506         case H_VPA_DEREG_DTL:           /* deregister DTL */
 507                 vpap = &tvcpu->arch.dtl;
 508                 err = 0;
 509                 break;
 510
 511         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 512                 vpap = &tvcpu->arch.slb_shadow;
 513                 err = 0;
 514                 break;
 515         }
 516
 517         if (vpap) {
 518                 vpap->next_gpa = vpa;
 519                 vpap->len = len;
 520                 vpap->update_pending = 1;
 521         }
 522
 523         spin_unlock(&tvcpu->arch.vpa_update_lock);
 524
 525         return err;
 526 }
 527
 528 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 529 {
 530         struct kvm *kvm = vcpu->kvm;
 531         void *va;
 532         unsigned long nb;
 533         unsigned long gpa;
 534
 535         /*
 536          * We need to pin the page pointed to by vpap->next_gpa,
 537          * but we can't call kvmppc_pin_guest_page under the lock
 538          * as it does get_user_pages() and down_read().  So we
 539          * have to drop the lock, pin the page, then get the lock
 540          * again and check that a new area didn't get registered
 541          * in the meantime.
 542          */
 543         for (;;) {
 544                 gpa = vpap->next_gpa;
 545                 spin_unlock(&vcpu->arch.vpa_update_lock);
 546                 va = NULL;
 547                 nb = 0;
 548                 if (gpa)
 549                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 550                 spin_lock(&vcpu->arch.vpa_update_lock);
 551                 if (gpa == vpap->next_gpa)
 552                         break;
 553                 /* sigh... unpin that one and try again */
 554                 if (va)
 555                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 556         }
 557
 558         vpap->update_pending = 0;
 559         if (va && nb < vpap->len) {
 560                 /*
 561                  * If it's now too short, it must be that userspace
 562                  * has changed the mappings underlying guest memory,
 563                  * so unregister the region.
 564                  */
 565                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 566                 va = NULL;
 567         }
 568         if (vpap->pinned_addr)
 569                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 570                                         vpap->dirty);
 571         vpap->gpa = gpa;
 572         vpap->pinned_addr = va;
 573         vpap->dirty = false;
 574         if (va)
 575                 vpap->pinned_end = va + vpap->len;
 576 }
 577
 578 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 579 {
 580         if (!(vcpu->arch.vpa.update_pending ||
 581               vcpu->arch.slb_shadow.update_pending ||
 582               vcpu->arch.dtl.update_pending))
 583                 return;
 584
 585         spin_lock(&vcpu->arch.vpa_update_lock);
 586         if (vcpu->arch.vpa.update_pending) {
 587                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 588                 if (vcpu->arch.vpa.pinned_addr)
 589                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 590         }
 591         if (vcpu->arch.dtl.update_pending) {
 592                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 593                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 594                 vcpu->arch.dtl_index = 0;
 595         }
 596         if (vcpu->arch.slb_shadow.update_pending)
 597                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 598         spin_unlock(&vcpu->arch.vpa_update_lock);
 599 }
 600
 601 /*
 602  * Return the accumulated stolen time for the vcore up until `now'.
 603  * The caller should hold the vcore lock.
 604  */
 605 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 606 {
 607         u64 p;
 608         unsigned long flags;
 609
 610         spin_lock_irqsave(&vc->stoltb_lock, flags);
 611         p = vc->stolen_tb;
 612         if (vc->vcore_state != VCORE_INACTIVE &&
 613             vc->preempt_tb != TB_NIL)
 614                 p += now - vc->preempt_tb;
 615         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 616         return p;
 617 }
 618
 619 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 620                                     struct kvmppc_vcore *vc)
 621 {
 622         struct dtl_entry *dt;
 623         struct lppaca *vpa;
 624         unsigned long stolen;
 625         unsigned long core_stolen;
 626         u64 now;
 627
 628         dt = vcpu->arch.dtl_ptr;
 629         vpa = vcpu->arch.vpa.pinned_addr;
 630         now = mftb();
 631         core_stolen = vcore_stolen_time(vc, now);
 632         stolen = core_stolen - vcpu->arch.stolen_logged;
 633         vcpu->arch.stolen_logged = core_stolen;
 634         spin_lock_irq(&vcpu->arch.tbacct_lock);
 635         stolen += vcpu->arch.busy_stolen;
 636         vcpu->arch.busy_stolen = 0;
 637         spin_unlock_irq(&vcpu->arch.tbacct_lock);
 638         if (!dt || !vpa)
 639                 return;
 640         memset(dt, 0, sizeof(struct dtl_entry));
 641         dt->dispatch_reason = 7;
 642         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 643         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 644         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 645         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 646         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 647         ++dt;
 648         if (dt == vcpu->arch.dtl.pinned_end)
 649                 dt = vcpu->arch.dtl.pinned_addr;
 650         vcpu->arch.dtl_ptr = dt;
 651         /* order writing *dt vs. writing vpa->dtl_idx */
 652         smp_wmb();
 653         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 654         vcpu->arch.dtl.dirty = true;
 655 }
 656
 657 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 658 {
 659         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 660                 return true;
 661         if ((!vcpu->arch.vcore->arch_compat) &&
 662             cpu_has_feature(CPU_FTR_ARCH_207S))
 663                 return true;
 664         return false;
 665 }
 666
 667 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 668                              unsigned long resource, unsigned long value1,
 669                              unsigned long value2)
 670 {
 671         switch (resource) {
 672         case H_SET_MODE_RESOURCE_SET_CIABR:
 673                 if (!kvmppc_power8_compatible(vcpu))
 674                         return H_P2;
 675                 if (value2)
 676                         return H_P4;
 677                 if (mflags)
 678                         return H_UNSUPPORTED_FLAG_START;
 679                 /* Guests can't breakpoint the hypervisor */
 680                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 681                         return H_P3;
 682                 vcpu->arch.ciabr  = value1;
 683                 return H_SUCCESS;
 684         case H_SET_MODE_RESOURCE_SET_DAWR:
 685                 if (!kvmppc_power8_compatible(vcpu))
 686                         return H_P2;
 687                 if (mflags)
 688                         return H_UNSUPPORTED_FLAG_START;
 689                 if (value2 & DABRX_HYP)
 690                         return H_P4;
 691                 vcpu->arch.dawr  = value1;
 692                 vcpu->arch.dawrx = value2;
 693                 return H_SUCCESS;
 694         default:
 695                 return H_TOO_HARD;
 696         }
 697 }
 698
 699 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 700 {
 701         struct kvmppc_vcore *vcore = target->arch.vcore;
 702
 703         /*
 704          * We expect to have been called by the real mode handler
 705          * (kvmppc_rm_h_confer()) which would have directly returned
 706          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 707          * have useful work to do and should not confer) so we don't
 708          * recheck that here.
 709          */
 710
 711         spin_lock(&vcore->lock);
 712         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 713             vcore->vcore_state != VCORE_INACTIVE &&
 714             vcore->runner)
 715                 target = vcore->runner;
 716         spin_unlock(&vcore->lock);
 717
 718         return kvm_vcpu_yield_to(target);
 719 }
 720
 721 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 722 {
 723         int yield_count = 0;
 724         struct lppaca *lppaca;
 725
 726         spin_lock(&vcpu->arch.vpa_update_lock);
 727         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 728         if (lppaca)
 729                 yield_count = be32_to_cpu(lppaca->yield_count);
 730         spin_unlock(&vcpu->arch.vpa_update_lock);
 731         return yield_count;
 732 }
 733
 734 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 735 {
 736         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 737         unsigned long target, ret = H_SUCCESS;
 738         int yield_count;
 739         struct kvm_vcpu *tvcpu;
 740         int idx, rc;
 741
 742         if (req <= MAX_HCALL_OPCODE &&
 743             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 744                 return RESUME_HOST;
 745
 746         switch (req) {
 747         case H_CEDE:
 748                 break;
 749         case H_PROD:
 750                 target = kvmppc_get_gpr(vcpu, 4);
 751                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 752                 if (!tvcpu) {
 753                         ret = H_PARAMETER;
 754                         break;
 755                 }
 756                 tvcpu->arch.prodded = 1;
 757                 smp_mb();
 758                 if (vcpu->arch.ceded) {
 759                         if (swait_active(&vcpu->wq)) {
 760                                 swake_up(&vcpu->wq);
 761                                 vcpu->stat.halt_wakeup++;
 762                         }
 763                 }
 764                 break;
 765         case H_CONFER:
 766                 target = kvmppc_get_gpr(vcpu, 4);
 767                 if (target == -1)
 768                         break;
 769                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 770                 if (!tvcpu) {
 771                         ret = H_PARAMETER;
 772                         break;
 773                 }
 774                 yield_count = kvmppc_get_gpr(vcpu, 5);
 775                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 776                         break;
 777                 kvm_arch_vcpu_yield_to(tvcpu);
 778                 break;
 779         case H_REGISTER_VPA:
 780                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 781                                         kvmppc_get_gpr(vcpu, 5),
 782                                         kvmppc_get_gpr(vcpu, 6));
 783                 break;
 784         case H_RTAS:
 785                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 786                         return RESUME_HOST;
 787
 788                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 789                 rc = kvmppc_rtas_hcall(vcpu);
 790                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 791
 792                 if (rc == -ENOENT)
 793                         return RESUME_HOST;
 794                 else if (rc == 0)
 795                         break;
 796
 797                 /* Send the error out to userspace via KVM_RUN */
 798                 return rc;
 799         case H_LOGICAL_CI_LOAD:
 800                 ret = kvmppc_h_logical_ci_load(vcpu);
 801                 if (ret == H_TOO_HARD)
 802                         return RESUME_HOST;
 803                 break;
 804         case H_LOGICAL_CI_STORE:
 805                 ret = kvmppc_h_logical_ci_store(vcpu);
 806                 if (ret == H_TOO_HARD)
 807                         return RESUME_HOST;
 808                 break;
 809         case H_SET_MODE:
 810                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 811                                         kvmppc_get_gpr(vcpu, 5),
 812                                         kvmppc_get_gpr(vcpu, 6),
 813                                         kvmppc_get_gpr(vcpu, 7));
 814                 if (ret == H_TOO_HARD)
 815                         return RESUME_HOST;
 816                 break;
 817         case H_XIRR:
 818         case H_CPPR:
 819         case H_EOI:
 820         case H_IPI:
 821         case H_IPOLL:
 822         case H_XIRR_X:
 823                 if (kvmppc_xics_enabled(vcpu)) {
 824                         ret = kvmppc_xics_hcall(vcpu, req);
 825                         break;
 826                 }
 827                 return RESUME_HOST;
 828         case H_PUT_TCE:
 829                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 830                                                 kvmppc_get_gpr(vcpu, 5),
 831                                                 kvmppc_get_gpr(vcpu, 6));
 832                 if (ret == H_TOO_HARD)
 833                         return RESUME_HOST;
 834                 break;
 835         case H_PUT_TCE_INDIRECT:
 836                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
 837                                                 kvmppc_get_gpr(vcpu, 5),
 838                                                 kvmppc_get_gpr(vcpu, 6),
 839                                                 kvmppc_get_gpr(vcpu, 7));
 840                 if (ret == H_TOO_HARD)
 841                         return RESUME_HOST;
 842                 break;
 843         case H_STUFF_TCE:
 844                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 845                                                 kvmppc_get_gpr(vcpu, 5),
 846                                                 kvmppc_get_gpr(vcpu, 6),
 847                                                 kvmppc_get_gpr(vcpu, 7));
 848                 if (ret == H_TOO_HARD)
 849                         return RESUME_HOST;
 850                 break;
 851         default:
 852                 return RESUME_HOST;
 853         }
 854         kvmppc_set_gpr(vcpu, 3, ret);
 855         vcpu->arch.hcall_needed = 0;
 856         return RESUME_GUEST;
 857 }
 858
 859 static int kvmppc_hcall_impl_hv(unsigned long cmd)
 860 {
 861         switch (cmd) {
 862         case H_CEDE:
 863         case H_PROD:
 864         case H_CONFER:
 865         case H_REGISTER_VPA:
 866         case H_SET_MODE:
 867         case H_LOGICAL_CI_LOAD:
 868         case H_LOGICAL_CI_STORE:
 869 #ifdef CONFIG_KVM_XICS
 870         case H_XIRR:
 871         case H_CPPR:
 872         case H_EOI:
 873         case H_IPI:
 874         case H_IPOLL:
 875         case H_XIRR_X:
 876 #endif
 877                 return 1;
 878         }
 879
 880         /* See if it's in the real-mode table */
 881         return kvmppc_hcall_impl_hv_realmode(cmd);
 882 }
 883
 884 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
 885                                         struct kvm_vcpu *vcpu)
 886 {
 887         u32 last_inst;
 888
 889         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
 890                                         EMULATE_DONE) {
 891                 /*
 892                  * Fetch failed, so return to guest and
 893                  * try executing it again.
 894                  */
 895                 return RESUME_GUEST;
 896         }
 897
 898         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
 899                 run->exit_reason = KVM_EXIT_DEBUG;
 900                 run->debug.arch.address = kvmppc_get_pc(vcpu);
 901                 return RESUME_HOST;
 902         } else {
 903                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 904                 return RESUME_GUEST;
 905         }
 906 }
 907
 908 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
 909                                  struct task_struct *tsk)
 910 {
 911         int r = RESUME_HOST;
 912
 913         vcpu->stat.sum_exits++;
 914
 915         /*
 916          * This can happen if an interrupt occurs in the last stages
 917          * of guest entry or the first stages of guest exit (i.e. after
 918          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
 919          * and before setting it to KVM_GUEST_MODE_HOST_HV).
 920          * That can happen due to a bug, or due to a machine check
 921          * occurring at just the wrong time.
 922          */
 923         if (vcpu->arch.shregs.msr & MSR_HV) {
 924                 printk(KERN_EMERG "KVM trap in HV mode!\n");
 925                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 926                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 927                         vcpu->arch.shregs.msr);
 928                 kvmppc_dump_regs(vcpu);
 929                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
 930                 run->hw.hardware_exit_reason = vcpu->arch.trap;
 931                 return RESUME_HOST;
 932         }
 933         run->exit_reason = KVM_EXIT_UNKNOWN;
 934         run->ready_for_interrupt_injection = 1;
 935         switch (vcpu->arch.trap) {
 936         /* We're good on these - the host merely wanted to get our attention */
 937         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 938                 vcpu->stat.dec_exits++;
 939                 r = RESUME_GUEST;
 940                 break;
 941         case BOOK3S_INTERRUPT_EXTERNAL:
 942         case BOOK3S_INTERRUPT_H_DOORBELL:
 943                 vcpu->stat.ext_intr_exits++;
 944                 r = RESUME_GUEST;
 945                 break;
 946         /* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
 947         case BOOK3S_INTERRUPT_HMI:
 948         case BOOK3S_INTERRUPT_PERFMON:
 949                 r = RESUME_GUEST;
 950                 break;
 951         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 952                 /*
 953                  * Deliver a machine check interrupt to the guest.
 954                  * We have to do this, even if the host has handled the
 955                  * machine check, because machine checks use SRR0/1 and
 956                  * the interrupt might have trashed guest state in them.
 957                  */
 958                 kvmppc_book3s_queue_irqprio(vcpu,
 959                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 960                 r = RESUME_GUEST;
 961                 break;
 962         case BOOK3S_INTERRUPT_PROGRAM:
 963         {
 964                 ulong flags;
 965                 /*
 966                  * Normally program interrupts are delivered directly
 967                  * to the guest by the hardware, but we can get here
 968                  * as a result of a hypervisor emulation interrupt
 969                  * (e40) getting turned into a 700 by BML RTAS.
 970                  */
 971                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 972                 kvmppc_core_queue_program(vcpu, flags);
 973                 r = RESUME_GUEST;
 974                 break;
 975         }
 976         case BOOK3S_INTERRUPT_SYSCALL:
 977         {
 978                 /* hcall - punt to userspace */
 979                 int i;
 980
 981                 /* hypercall with MSR_PR has already been handled in rmode,
 982                  * and never reaches here.
 983                  */
 984
 985                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 986                 for (i = 0; i < 9; ++i)
 987                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 988                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 989                 vcpu->arch.hcall_needed = 1;
 990                 r = RESUME_HOST;
 991                 break;
 992         }
 993         /*
 994          * We get these next two if the guest accesses a page which it thinks
 995          * it has mapped but which is not actually present, either because
 996          * it is for an emulated I/O device or because the corresonding
 997          * host page has been paged out.  Any other HDSI/HISI interrupts
 998          * have been handled already.
 999          */
1000         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1001                 r = RESUME_PAGE_FAULT;
1002                 break;
1003         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1004                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1005                 vcpu->arch.fault_dsisr = 0;
1006                 r = RESUME_PAGE_FAULT;
1007                 break;
1008         /*
1009          * This occurs if the guest executes an illegal instruction.
1010          * If the guest debug is disabled, generate a program interrupt
1011          * to the guest. If guest debug is enabled, we need to check
1012          * whether the instruction is a software breakpoint instruction.
1013          * Accordingly return to Guest or Host.
1014          */
1015         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1016                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1017                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1018                                 swab32(vcpu->arch.emul_inst) :
1019                                 vcpu->arch.emul_inst;
1020                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1021                         r = kvmppc_emulate_debug_inst(run, vcpu);
1022                 } else {
1023                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1024                         r = RESUME_GUEST;
1025                 }
1026                 break;
1027         /*
1028          * This occurs if the guest (kernel or userspace), does something that
1029          * is prohibited by HFSCR.  We just generate a program interrupt to
1030          * the guest.
1031          */
1032         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1033                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1034                 r = RESUME_GUEST;
1035                 break;
1036         case BOOK3S_INTERRUPT_HV_RM_HARD:
1037                 r = RESUME_PASSTHROUGH;
1038                 break;
1039         default:
1040                 kvmppc_dump_regs(vcpu);
1041                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1042                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1043                         vcpu->arch.shregs.msr);
1044                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1045                 r = RESUME_HOST;
1046                 break;
1047         }
1048
1049         return r;
1050 }
1051
1052 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1053                                             struct kvm_sregs *sregs)
1054 {
1055         int i;
1056
1057         memset(sregs, 0, sizeof(struct kvm_sregs));
1058         sregs->pvr = vcpu->arch.pvr;
1059         for (i = 0; i < vcpu->arch.slb_max; i++) {
1060                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1061                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1062         }
1063
1064         return 0;
1065 }
1066
1067 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1068                                             struct kvm_sregs *sregs)
1069 {
1070         int i, j;
1071
1072         /* Only accept the same PVR as the host's, since we can't spoof it */
1073         if (sregs->pvr != vcpu->arch.pvr)
1074                 return -EINVAL;
1075
1076         j = 0;
1077         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1078                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1079                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1080                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1081                         ++j;
1082                 }
1083         }
1084         vcpu->arch.slb_max = j;
1085
1086         return 0;
1087 }
1088
1089 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1090                 bool preserve_top32)
1091 {
1092         struct kvm *kvm = vcpu->kvm;
1093         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1094         u64 mask;
1095
1096         spin_lock(&vc->lock);
1097         /*
1098          * If ILE (interrupt little-endian) has changed, update the
1099          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1100          */
1101         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1102                 struct kvm_vcpu *vcpu;
1103                 int i;
1104
1105                 kvm_for_each_vcpu(i, vcpu, kvm) {
1106                         if (vcpu->arch.vcore != vc)
1107                                 continue;
1108                         if (new_lpcr & LPCR_ILE)
1109                                 vcpu->arch.intr_msr |= MSR_LE;
1110                         else
1111                                 vcpu->arch.intr_msr &= ~MSR_LE;
1112                 }
1113         }
1114
1115         /*
1116          * Userspace can only modify DPFD (default prefetch depth),
1117          * ILE (interrupt little-endian) and TC (translation control).
1118          * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
1119          */
1120         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1121         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1122                 mask |= LPCR_AIL;
1123
1124         /* Broken 32-bit version of LPCR must not clear top bits */
1125         if (preserve_top32)
1126                 mask &= 0xFFFFFFFF;
1127         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1128         spin_unlock(&vc->lock);
1129 }
1130
1131 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1132                                  union kvmppc_one_reg *val)
1133 {
1134         int r = 0;
1135         long int i;
1136
1137         switch (id) {
1138         case KVM_REG_PPC_DEBUG_INST:
1139                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1140                 break;
1141         case KVM_REG_PPC_HIOR:
1142                 *val = get_reg_val(id, 0);
1143                 break;
1144         case KVM_REG_PPC_DABR:
1145                 *val = get_reg_val(id, vcpu->arch.dabr);
1146                 break;
1147         case KVM_REG_PPC_DABRX:
1148                 *val = get_reg_val(id, vcpu->arch.dabrx);
1149                 break;
1150         case KVM_REG_PPC_DSCR:
1151                 *val = get_reg_val(id, vcpu->arch.dscr);
1152                 break;
1153         case KVM_REG_PPC_PURR:
1154                 *val = get_reg_val(id, vcpu->arch.purr);
1155                 break;
1156         case KVM_REG_PPC_SPURR:
1157                 *val = get_reg_val(id, vcpu->arch.spurr);
1158                 break;
1159         case KVM_REG_PPC_AMR:
1160                 *val = get_reg_val(id, vcpu->arch.amr);
1161                 break;
1162         case KVM_REG_PPC_UAMOR:
1163                 *val = get_reg_val(id, vcpu->arch.uamor);
1164                 break;
1165         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1166                 i = id - KVM_REG_PPC_MMCR0;
1167                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1168                 break;
1169         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1170                 i = id - KVM_REG_PPC_PMC1;
1171                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1172                 break;
1173         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1174                 i = id - KVM_REG_PPC_SPMC1;
1175                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1176                 break;
1177         case KVM_REG_PPC_SIAR:
1178                 *val = get_reg_val(id, vcpu->arch.siar);
1179                 break;
1180         case KVM_REG_PPC_SDAR:
1181                 *val = get_reg_val(id, vcpu->arch.sdar);
1182                 break;
1183         case KVM_REG_PPC_SIER:
1184                 *val = get_reg_val(id, vcpu->arch.sier);
1185                 break;
1186         case KVM_REG_PPC_IAMR:
1187                 *val = get_reg_val(id, vcpu->arch.iamr);
1188                 break;
1189         case KVM_REG_PPC_PSPB:
1190                 *val = get_reg_val(id, vcpu->arch.pspb);
1191                 break;
1192         case KVM_REG_PPC_DPDES:
1193                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1194                 break;
1195         case KVM_REG_PPC_VTB:
1196                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1197                 break;
1198         case KVM_REG_PPC_DAWR:
1199                 *val = get_reg_val(id, vcpu->arch.dawr);
1200                 break;
1201         case KVM_REG_PPC_DAWRX:
1202                 *val = get_reg_val(id, vcpu->arch.dawrx);
1203                 break;
1204         case KVM_REG_PPC_CIABR:
1205                 *val = get_reg_val(id, vcpu->arch.ciabr);
1206                 break;
1207         case KVM_REG_PPC_CSIGR:
1208                 *val = get_reg_val(id, vcpu->arch.csigr);
1209                 break;
1210         case KVM_REG_PPC_TACR:
1211                 *val = get_reg_val(id, vcpu->arch.tacr);
1212                 break;
1213         case KVM_REG_PPC_TCSCR:
1214                 *val = get_reg_val(id, vcpu->arch.tcscr);
1215                 break;
1216         case KVM_REG_PPC_PID:
1217                 *val = get_reg_val(id, vcpu->arch.pid);
1218                 break;
1219         case KVM_REG_PPC_ACOP:
1220                 *val = get_reg_val(id, vcpu->arch.acop);
1221                 break;
1222         case KVM_REG_PPC_WORT:
1223                 *val = get_reg_val(id, vcpu->arch.wort);
1224                 break;
1225         case KVM_REG_PPC_VPA_ADDR:
1226                 spin_lock(&vcpu->arch.vpa_update_lock);
1227                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1228                 spin_unlock(&vcpu->arch.vpa_update_lock);
1229                 break;
1230         case KVM_REG_PPC_VPA_SLB:
1231                 spin_lock(&vcpu->arch.vpa_update_lock);
1232                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1233                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1234                 spin_unlock(&vcpu->arch.vpa_update_lock);
1235                 break;
1236         case KVM_REG_PPC_VPA_DTL:
1237                 spin_lock(&vcpu->arch.vpa_update_lock);
1238                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1239                 val->vpaval.length = vcpu->arch.dtl.len;
1240                 spin_unlock(&vcpu->arch.vpa_update_lock);
1241                 break;
1242         case KVM_REG_PPC_TB_OFFSET:
1243                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1244                 break;
1245         case KVM_REG_PPC_LPCR:
1246         case KVM_REG_PPC_LPCR_64:
1247                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1248                 break;
1249         case KVM_REG_PPC_PPR:
1250                 *val = get_reg_val(id, vcpu->arch.ppr);
1251                 break;
1252 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1253         case KVM_REG_PPC_TFHAR:
1254                 *val = get_reg_val(id, vcpu->arch.tfhar);
1255                 break;
1256         case KVM_REG_PPC_TFIAR:
1257                 *val = get_reg_val(id, vcpu->arch.tfiar);
1258                 break;
1259         case KVM_REG_PPC_TEXASR:
1260                 *val = get_reg_val(id, vcpu->arch.texasr);
1261                 break;
1262         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1263                 i = id - KVM_REG_PPC_TM_GPR0;
1264                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1265                 break;
1266         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1267         {
1268                 int j;
1269                 i = id - KVM_REG_PPC_TM_VSR0;
1270                 if (i < 32)
1271                         for (j = 0; j < TS_FPRWIDTH; j++)
1272                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1273                 else {
1274                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1275                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1276                         else
1277                                 r = -ENXIO;
1278                 }
1279                 break;
1280         }
1281         case KVM_REG_PPC_TM_CR:
1282                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1283                 break;
1284         case KVM_REG_PPC_TM_XER:
1285                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1286                 break;
1287         case KVM_REG_PPC_TM_LR:
1288                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1289                 break;
1290         case KVM_REG_PPC_TM_CTR:
1291                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1292                 break;
1293         case KVM_REG_PPC_TM_FPSCR:
1294                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1295                 break;
1296         case KVM_REG_PPC_TM_AMR:
1297                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1298                 break;
1299         case KVM_REG_PPC_TM_PPR:
1300                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1301                 break;
1302         case KVM_REG_PPC_TM_VRSAVE:
1303                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1304                 break;
1305         case KVM_REG_PPC_TM_VSCR:
1306                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1307                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1308                 else
1309                         r = -ENXIO;
1310                 break;
1311         case KVM_REG_PPC_TM_DSCR:
1312                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1313                 break;
1314         case KVM_REG_PPC_TM_TAR:
1315                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1316                 break;
1317 #endif
1318         case KVM_REG_PPC_ARCH_COMPAT:
1319                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1320                 break;
1321         default:
1322                 r = -EINVAL;
1323                 break;
1324         }
1325
1326         return r;
1327 }
1328
1329 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1330                                  union kvmppc_one_reg *val)
1331 {
1332         int r = 0;
1333         long int i;
1334         unsigned long addr, len;
1335
1336         switch (id) {
1337         case KVM_REG_PPC_HIOR:
1338                 /* Only allow this to be set to zero */
1339                 if (set_reg_val(id, *val))
1340                         r = -EINVAL;
1341                 break;
1342         case KVM_REG_PPC_DABR:
1343                 vcpu->arch.dabr = set_reg_val(id, *val);
1344                 break;
1345         case KVM_REG_PPC_DABRX:
1346                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1347                 break;
1348         case KVM_REG_PPC_DSCR:
1349                 vcpu->arch.dscr = set_reg_val(id, *val);
1350                 break;
1351         case KVM_REG_PPC_PURR:
1352                 vcpu->arch.purr = set_reg_val(id, *val);
1353                 break;
1354         case KVM_REG_PPC_SPURR:
1355                 vcpu->arch.spurr = set_reg_val(id, *val);
1356                 break;
1357         case KVM_REG_PPC_AMR:
1358                 vcpu->arch.amr = set_reg_val(id, *val);
1359                 break;
1360         case KVM_REG_PPC_UAMOR:
1361                 vcpu->arch.uamor = set_reg_val(id, *val);
1362                 break;
1363         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1364                 i = id - KVM_REG_PPC_MMCR0;
1365                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1366                 break;
1367         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1368                 i = id - KVM_REG_PPC_PMC1;
1369                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1370                 break;
1371         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1372                 i = id - KVM_REG_PPC_SPMC1;
1373                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1374                 break;
1375         case KVM_REG_PPC_SIAR:
1376                 vcpu->arch.siar = set_reg_val(id, *val);
1377                 break;
1378         case KVM_REG_PPC_SDAR:
1379                 vcpu->arch.sdar = set_reg_val(id, *val);
1380                 break;
1381         case KVM_REG_PPC_SIER:
1382                 vcpu->arch.sier = set_reg_val(id, *val);
1383                 break;
1384         case KVM_REG_PPC_IAMR:
1385                 vcpu->arch.iamr = set_reg_val(id, *val);
1386                 break;
1387         case KVM_REG_PPC_PSPB:
1388                 vcpu->arch.pspb = set_reg_val(id, *val);
1389                 break;
1390         case KVM_REG_PPC_DPDES:
1391                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1392                 break;
1393         case KVM_REG_PPC_VTB:
1394                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1395                 break;
1396         case KVM_REG_PPC_DAWR:
1397                 vcpu->arch.dawr = set_reg_val(id, *val);
1398                 break;
1399         case KVM_REG_PPC_DAWRX:
1400                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1401                 break;
1402         case KVM_REG_PPC_CIABR:
1403                 vcpu->arch.ciabr = set_reg_val(id, *val);
1404                 /* Don't allow setting breakpoints in hypervisor code */
1405                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1406                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1407                 break;
1408         case KVM_REG_PPC_CSIGR:
1409                 vcpu->arch.csigr = set_reg_val(id, *val);
1410                 break;
1411         case KVM_REG_PPC_TACR:
1412                 vcpu->arch.tacr = set_reg_val(id, *val);
1413                 break;
1414         case KVM_REG_PPC_TCSCR:
1415                 vcpu->arch.tcscr = set_reg_val(id, *val);
1416                 break;
1417         case KVM_REG_PPC_PID:
1418                 vcpu->arch.pid = set_reg_val(id, *val);
1419                 break;
1420         case KVM_REG_PPC_ACOP:
1421                 vcpu->arch.acop = set_reg_val(id, *val);
1422                 break;
1423         case KVM_REG_PPC_WORT:
1424                 vcpu->arch.wort = set_reg_val(id, *val);
1425                 break;
1426         case KVM_REG_PPC_VPA_ADDR:
1427                 addr = set_reg_val(id, *val);
1428                 r = -EINVAL;
1429                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1430                               vcpu->arch.dtl.next_gpa))
1431                         break;
1432                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1433                 break;
1434         case KVM_REG_PPC_VPA_SLB:
1435                 addr = val->vpaval.addr;
1436                 len = val->vpaval.length;
1437                 r = -EINVAL;
1438                 if (addr && !vcpu->arch.vpa.next_gpa)
1439                         break;
1440                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1441                 break;
1442         case KVM_REG_PPC_VPA_DTL:
1443                 addr = val->vpaval.addr;
1444                 len = val->vpaval.length;
1445                 r = -EINVAL;
1446                 if (addr && (len < sizeof(struct dtl_entry) ||
1447                              !vcpu->arch.vpa.next_gpa))
1448                         break;
1449                 len -= len % sizeof(struct dtl_entry);
1450                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1451                 break;
1452         case KVM_REG_PPC_TB_OFFSET:
1453                 /* round up to multiple of 2^24 */
1454                 vcpu->arch.vcore->tb_offset =
1455                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1456                 break;
1457         case KVM_REG_PPC_LPCR:
1458                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1459                 break;
1460         case KVM_REG_PPC_LPCR_64:
1461                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1462                 break;
1463         case KVM_REG_PPC_PPR:
1464                 vcpu->arch.ppr = set_reg_val(id, *val);
1465                 break;
1466 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1467         case KVM_REG_PPC_TFHAR:
1468                 vcpu->arch.tfhar = set_reg_val(id, *val);
1469                 break;
1470         case KVM_REG_PPC_TFIAR:
1471                 vcpu->arch.tfiar = set_reg_val(id, *val);
1472                 break;
1473         case KVM_REG_PPC_TEXASR:
1474                 vcpu->arch.texasr = set_reg_val(id, *val);
1475                 break;
1476         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1477                 i = id - KVM_REG_PPC_TM_GPR0;
1478                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1479                 break;
1480         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1481         {
1482                 int j;
1483                 i = id - KVM_REG_PPC_TM_VSR0;
1484                 if (i < 32)
1485                         for (j = 0; j < TS_FPRWIDTH; j++)
1486                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1487                 else
1488                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1489                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1490                         else
1491                                 r = -ENXIO;
1492                 break;
1493         }
1494         case KVM_REG_PPC_TM_CR:
1495                 vcpu->arch.cr_tm = set_reg_val(id, *val);
1496                 break;
1497         case KVM_REG_PPC_TM_XER:
1498                 vcpu->arch.xer_tm = set_reg_val(id, *val);
1499                 break;
1500         case KVM_REG_PPC_TM_LR:
1501                 vcpu->arch.lr_tm = set_reg_val(id, *val);
1502                 break;
1503         case KVM_REG_PPC_TM_CTR:
1504                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1505                 break;
1506         case KVM_REG_PPC_TM_FPSCR:
1507                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1508                 break;
1509         case KVM_REG_PPC_TM_AMR:
1510                 vcpu->arch.amr_tm = set_reg_val(id, *val);
1511                 break;
1512         case KVM_REG_PPC_TM_PPR:
1513                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1514                 break;
1515         case KVM_REG_PPC_TM_VRSAVE:
1516                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1517                 break;
1518         case KVM_REG_PPC_TM_VSCR:
1519                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1520                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1521                 else
1522                         r = - ENXIO;
1523                 break;
1524         case KVM_REG_PPC_TM_DSCR:
1525                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1526                 break;
1527         case KVM_REG_PPC_TM_TAR:
1528                 vcpu->arch.tar_tm = set_reg_val(id, *val);
1529                 break;
1530 #endif
1531         case KVM_REG_PPC_ARCH_COMPAT:
1532                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1533                 break;
1534         default:
1535                 r = -EINVAL;
1536                 break;
1537         }
1538
1539         return r;
1540 }
1541
1542 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1543 {
1544         struct kvmppc_vcore *vcore;
1545
1546         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1547
1548         if (vcore == NULL)
1549                 return NULL;
1550
1551         spin_lock_init(&vcore->lock);
1552         spin_lock_init(&vcore->stoltb_lock);
1553         init_swait_queue_head(&vcore->wq);
1554         vcore->preempt_tb = TB_NIL;
1555         vcore->lpcr = kvm->arch.lpcr;
1556         vcore->first_vcpuid = core * threads_per_subcore;
1557         vcore->kvm = kvm;
1558         INIT_LIST_HEAD(&vcore->preempt_list);
1559
1560         return vcore;
1561 }
1562
1563 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1564 static struct debugfs_timings_element {
1565         const char *name;
1566         size_t offset;
1567 } timings[] = {
1568         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
1569         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
1570         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
1571         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
1572         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
1573 };
1574
1575 #define N_TIMINGS       (sizeof(timings) / sizeof(timings[0]))
1576
1577 struct debugfs_timings_state {
1578         struct kvm_vcpu *vcpu;
1579         unsigned int    buflen;
1580         char            buf[N_TIMINGS * 100];
1581 };
1582
1583 static int debugfs_timings_open(struct inode *inode, struct file *file)
1584 {
1585         struct kvm_vcpu *vcpu = inode->i_private;
1586         struct debugfs_timings_state *p;
1587
1588         p = kzalloc(sizeof(*p), GFP_KERNEL);
1589         if (!p)
1590                 return -ENOMEM;
1591
1592         kvm_get_kvm(vcpu->kvm);
1593         p->vcpu = vcpu;
1594         file->private_data = p;
1595
1596         return nonseekable_open(inode, file);
1597 }
1598
1599 static int debugfs_timings_release(struct inode *inode, struct file *file)
1600 {
1601         struct debugfs_timings_state *p = file->private_data;
1602
1603         kvm_put_kvm(p->vcpu->kvm);
1604         kfree(p);
1605         return 0;
1606 }
1607
1608 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1609                                     size_t len, loff_t *ppos)
1610 {
1611         struct debugfs_timings_state *p = file->private_data;
1612         struct kvm_vcpu *vcpu = p->vcpu;
1613         char *s, *buf_end;
1614         struct kvmhv_tb_accumulator tb;
1615         u64 count;
1616         loff_t pos;
1617         ssize_t n;
1618         int i, loops;
1619         bool ok;
1620
1621         if (!p->buflen) {
1622                 s = p->buf;
1623                 buf_end = s + sizeof(p->buf);
1624                 for (i = 0; i < N_TIMINGS; ++i) {
1625                         struct kvmhv_tb_accumulator *acc;
1626
1627                         acc = (struct kvmhv_tb_accumulator *)
1628                                 ((unsigned long)vcpu + timings[i].offset);
1629                         ok = false;
1630                         for (loops = 0; loops < 1000; ++loops) {
1631                                 count = acc->seqcount;
1632                                 if (!(count & 1)) {
1633                                         smp_rmb();
1634                                         tb = *acc;
1635                                         smp_rmb();
1636                                         if (count == acc->seqcount) {
1637                                                 ok = true;
1638                                                 break;
1639                                         }
1640                                 }
1641                                 udelay(1);
1642                         }
1643                         if (!ok)
1644                                 snprintf(s, buf_end - s, "%s: stuck\n",
1645                                         timings[i].name);
1646                         else
1647                                 snprintf(s, buf_end - s,
1648                                         "%s: %llu %llu %llu %llu\n",
1649                                         timings[i].name, count / 2,
1650                                         tb_to_ns(tb.tb_total),
1651                                         tb_to_ns(tb.tb_min),
1652                                         tb_to_ns(tb.tb_max));
1653                         s += strlen(s);
1654                 }
1655                 p->buflen = s - p->buf;
1656         }
1657
1658         pos = *ppos;
1659         if (pos >= p->buflen)
1660                 return 0;
1661         if (len > p->buflen - pos)
1662                 len = p->buflen - pos;
1663         n = copy_to_user(buf, p->buf + pos, len);
1664         if (n) {
1665                 if (n == len)
1666                         return -EFAULT;
1667                 len -= n;
1668         }
1669         *ppos = pos + len;
1670         return len;
1671 }
1672
1673 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1674                                      size_t len, loff_t *ppos)
1675 {
1676         return -EACCES;
1677 }
1678
1679 static const struct file_operations debugfs_timings_ops = {
1680         .owner   = THIS_MODULE,
1681         .open    = debugfs_timings_open,
1682         .release = debugfs_timings_release,
1683         .read    = debugfs_timings_read,
1684         .write   = debugfs_timings_write,
1685         .llseek  = generic_file_llseek,
1686 };
1687
1688 /* Create a debugfs directory for the vcpu */
1689 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1690 {
1691         char buf[16];
1692         struct kvm *kvm = vcpu->kvm;
1693
1694         snprintf(buf, sizeof(buf), "vcpu%u", id);
1695         if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1696                 return;
1697         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1698         if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1699                 return;
1700         vcpu->arch.debugfs_timings =
1701                 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1702                                     vcpu, &debugfs_timings_ops);
1703 }
1704
1705 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1706 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1707 {
1708 }
1709 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1710
1711 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1712                                                    unsigned int id)
1713 {
1714         struct kvm_vcpu *vcpu;
1715         int err = -EINVAL;
1716         int core;
1717         struct kvmppc_vcore *vcore;
1718
1719         core = id / threads_per_subcore;
1720         if (core >= KVM_MAX_VCORES)
1721                 goto out;
1722
1723         err = -ENOMEM;
1724         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1725         if (!vcpu)
1726                 goto out;
1727
1728         err = kvm_vcpu_init(vcpu, kvm, id);
1729         if (err)
1730                 goto free_vcpu;
1731
1732         vcpu->arch.shared = &vcpu->arch.shregs;
1733 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1734         /*
1735          * The shared struct is never shared on HV,
1736          * so we can always use host endianness
1737          */
1738 #ifdef __BIG_ENDIAN__
1739         vcpu->arch.shared_big_endian = true;
1740 #else
1741         vcpu->arch.shared_big_endian = false;
1742 #endif
1743 #endif
1744         vcpu->arch.mmcr[0] = MMCR0_FC;
1745         vcpu->arch.ctrl = CTRL_RUNLATCH;
1746         /* default to host PVR, since we can't spoof it */
1747         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1748         spin_lock_init(&vcpu->arch.vpa_update_lock);
1749         spin_lock_init(&vcpu->arch.tbacct_lock);
1750         vcpu->arch.busy_preempt = TB_NIL;
1751         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1752
1753         kvmppc_mmu_book3s_hv_init(vcpu);
1754
1755         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1756
1757         init_waitqueue_head(&vcpu->arch.cpu_run);
1758
1759         mutex_lock(&kvm->lock);
1760         vcore = kvm->arch.vcores[core];
1761         if (!vcore) {
1762                 vcore = kvmppc_vcore_create(kvm, core);
1763                 kvm->arch.vcores[core] = vcore;
1764                 kvm->arch.online_vcores++;
1765         }
1766         mutex_unlock(&kvm->lock);
1767
1768         if (!vcore)
1769                 goto uninit_vcpu;
1770
1771         spin_lock(&vcore->lock);
1772         ++vcore->num_threads;
1773         spin_unlock(&vcore->lock);
1774         vcpu->arch.vcore = vcore;
1775         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1776         vcpu->arch.thread_cpu = -1;
1777
1778         vcpu->arch.cpu_type = KVM_CPU_3S_64;
1779         kvmppc_sanity_check(vcpu);
1780
1781         debugfs_vcpu_init(vcpu, id);
1782
1783         return vcpu;
1784
1785 uninit_vcpu:
1786         kvm_vcpu_uninit(vcpu);
1787 free_vcpu:
1788         kmem_cache_free(kvm_vcpu_cache, vcpu);
1789 out:
1790         return ERR_PTR(err);
1791 }
1792
1793 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1794 {
1795         if (vpa->pinned_addr)
1796                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1797                                         vpa->dirty);
1798 }
1799
1800 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1801 {
1802         spin_lock(&vcpu->arch.vpa_update_lock);
1803         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1804         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1805         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1806         spin_unlock(&vcpu->arch.vpa_update_lock);
1807         kvm_vcpu_uninit(vcpu);
1808         kmem_cache_free(kvm_vcpu_cache, vcpu);
1809 }
1810
1811 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1812 {
1813         /* Indicate we want to get back into the guest */
1814         return 1;
1815 }
1816
1817 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1818 {
1819         unsigned long dec_nsec, now;
1820
1821         now = get_tb();
1822         if (now > vcpu->arch.dec_expires) {
1823                 /* decrementer has already gone negative */
1824                 kvmppc_core_queue_dec(vcpu);
1825                 kvmppc_core_prepare_to_enter(vcpu);
1826                 return;
1827         }
1828         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1829                    / tb_ticks_per_sec;
1830         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1831                       HRTIMER_MODE_REL);
1832         vcpu->arch.timer_running = 1;
1833 }
1834
1835 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1836 {
1837         vcpu->arch.ceded = 0;
1838         if (vcpu->arch.timer_running) {
1839                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1840                 vcpu->arch.timer_running = 0;
1841         }
1842 }
1843
1844 extern void __kvmppc_vcore_entry(void);
1845
1846 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1847                                    struct kvm_vcpu *vcpu)
1848 {
1849         u64 now;
1850
1851         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1852                 return;
1853         spin_lock_irq(&vcpu->arch.tbacct_lock);
1854         now = mftb();
1855         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1856                 vcpu->arch.stolen_logged;
1857         vcpu->arch.busy_preempt = now;
1858         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1859         spin_unlock_irq(&vcpu->arch.tbacct_lock);
1860         --vc->n_runnable;
1861         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
1862 }
1863
1864 static int kvmppc_grab_hwthread(int cpu)
1865 {
1866         struct paca_struct *tpaca;
1867         long timeout = 10000;
1868
1869         tpaca = &paca[cpu];
1870
1871         /* Ensure the thread won't go into the kernel if it wakes */
1872         tpaca->kvm_hstate.kvm_vcpu = NULL;
1873         tpaca->kvm_hstate.kvm_vcore = NULL;
1874         tpaca->kvm_hstate.napping = 0;
1875         smp_wmb();
1876         tpaca->kvm_hstate.hwthread_req = 1;
1877
1878         /*
1879          * If the thread is already executing in the kernel (e.g. handling
1880          * a stray interrupt), wait for it to get back to nap mode.
1881          * The smp_mb() is to ensure that our setting of hwthread_req
1882          * is visible before we look at hwthread_state, so if this
1883          * races with the code at system_reset_pSeries and the thread
1884          * misses our setting of hwthread_req, we are sure to see its
1885          * setting of hwthread_state, and vice versa.
1886          */
1887         smp_mb();
1888         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1889                 if (--timeout <= 0) {
1890                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1891                         return -EBUSY;
1892                 }
1893                 udelay(1);
1894         }
1895         return 0;
1896 }
1897
1898 static void kvmppc_release_hwthread(int cpu)
1899 {
1900         struct paca_struct *tpaca;
1901
1902         tpaca = &paca[cpu];
1903         tpaca->kvm_hstate.hwthread_req = 0;
1904         tpaca->kvm_hstate.kvm_vcpu = NULL;
1905         tpaca->kvm_hstate.kvm_vcore = NULL;
1906         tpaca->kvm_hstate.kvm_split_mode = NULL;
1907 }
1908
1909 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
1910 {
1911         int cpu;
1912         struct paca_struct *tpaca;
1913         struct kvmppc_vcore *mvc = vc->master_vcore;
1914
1915         cpu = vc->pcpu;
1916         if (vcpu) {
1917                 if (vcpu->arch.timer_running) {
1918                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1919                         vcpu->arch.timer_running = 0;
1920                 }
1921                 cpu += vcpu->arch.ptid;
1922                 vcpu->cpu = mvc->pcpu;
1923                 vcpu->arch.thread_cpu = cpu;
1924         }
1925         tpaca = &paca[cpu];
1926         tpaca->kvm_hstate.kvm_vcpu = vcpu;
1927         tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
1928         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
1929         smp_wmb();
1930         tpaca->kvm_hstate.kvm_vcore = mvc;
1931         if (cpu != smp_processor_id())
1932                 kvmppc_ipi_thread(cpu);
1933 }
1934
1935 static void kvmppc_wait_for_nap(void)
1936 {
1937         int cpu = smp_processor_id();
1938         int i, loops;
1939
1940         for (loops = 0; loops < 1000000; ++loops) {
1941                 /*
1942                  * Check if all threads are finished.
1943                  * We set the vcore pointer when starting a thread
1944                  * and the thread clears it when finished, so we look
1945                  * for any threads that still have a non-NULL vcore ptr.
1946                  */
1947                 for (i = 1; i < threads_per_subcore; ++i)
1948                         if (paca[cpu + i].kvm_hstate.kvm_vcore)
1949                                 break;
1950                 if (i == threads_per_subcore) {
1951                         HMT_medium();
1952                         return;
1953                 }
1954                 HMT_low();
1955         }
1956         HMT_medium();
1957         for (i = 1; i < threads_per_subcore; ++i)
1958                 if (paca[cpu + i].kvm_hstate.kvm_vcore)
1959                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
1960 }
1961
1962 /*
1963  * Check that we are on thread 0 and that any other threads in
1964  * this core are off-line.  Then grab the threads so they can't
1965  * enter the kernel.
1966  */
1967 static int on_primary_thread(void)
1968 {
1969         int cpu = smp_processor_id();
1970         int thr;
1971
1972         /* Are we on a primary subcore? */
1973         if (cpu_thread_in_subcore(cpu))
1974                 return 0;
1975
1976         thr = 0;
1977         while (++thr < threads_per_subcore)
1978                 if (cpu_online(cpu + thr))
1979                         return 0;
1980
1981         /* Grab all hw threads so they can't go into the kernel */
1982         for (thr = 1; thr < threads_per_subcore; ++thr) {
1983                 if (kvmppc_grab_hwthread(cpu + thr)) {
1984                         /* Couldn't grab one; let the others go */
1985                         do {
1986                                 kvmppc_release_hwthread(cpu + thr);
1987                         } while (--thr > 0);
1988                         return 0;
1989                 }
1990         }
1991         return 1;
1992 }
1993
1994 /*
1995  * A list of virtual cores for each physical CPU.
1996  * These are vcores that could run but their runner VCPU tasks are
1997  * (or may be) preempted.
1998  */
1999 struct preempted_vcore_list {
2000         struct list_head        list;
2001         spinlock_t              lock;
2002 };
2003
2004 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2005
2006 static void init_vcore_lists(void)
2007 {
2008         int cpu;
2009
2010         for_each_possible_cpu(cpu) {
2011                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2012                 spin_lock_init(&lp->lock);
2013                 INIT_LIST_HEAD(&lp->list);
2014         }
2015 }
2016
2017 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2018 {
2019         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2020
2021         vc->vcore_state = VCORE_PREEMPT;
2022         vc->pcpu = smp_processor_id();
2023         if (vc->num_threads < threads_per_subcore) {
2024                 spin_lock(&lp->lock);
2025                 list_add_tail(&vc->preempt_list, &lp->list);
2026                 spin_unlock(&lp->lock);
2027         }
2028
2029         /* Start accumulating stolen time */
2030         kvmppc_core_start_stolen(vc);
2031 }
2032
2033 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2034 {
2035         struct preempted_vcore_list *lp;
2036
2037         kvmppc_core_end_stolen(vc);
2038         if (!list_empty(&vc->preempt_list)) {
2039                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2040                 spin_lock(&lp->lock);
2041                 list_del_init(&vc->preempt_list);
2042                 spin_unlock(&lp->lock);
2043         }
2044         vc->vcore_state = VCORE_INACTIVE;
2045 }
2046
2047 /*
2048  * This stores information about the virtual cores currently
2049  * assigned to a physical core.
2050  */
2051 struct core_info {
2052         int             n_subcores;
2053         int             max_subcore_threads;
2054         int             total_threads;
2055         int             subcore_threads[MAX_SUBCORES];
2056         struct kvm      *subcore_vm[MAX_SUBCORES];
2057         struct list_head vcs[MAX_SUBCORES];
2058 };
2059
2060 /*
2061  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2062  * respectively in 2-way micro-threading (split-core) mode.
2063  */
2064 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2065
2066 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2067 {
2068         int sub;
2069
2070         memset(cip, 0, sizeof(*cip));
2071         cip->n_subcores = 1;
2072         cip->max_subcore_threads = vc->num_threads;
2073         cip->total_threads = vc->num_threads;
2074         cip->subcore_threads[0] = vc->num_threads;
2075         cip->subcore_vm[0] = vc->kvm;
2076         for (sub = 0; sub < MAX_SUBCORES; ++sub)
2077                 INIT_LIST_HEAD(&cip->vcs[sub]);
2078         list_add_tail(&vc->preempt_list, &cip->vcs[0]);
2079 }
2080
2081 static bool subcore_config_ok(int n_subcores, int n_threads)
2082 {
2083         /* Can only dynamically split if unsplit to begin with */
2084         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2085                 return false;
2086         if (n_subcores > MAX_SUBCORES)
2087                 return false;
2088         if (n_subcores > 1) {
2089                 if (!(dynamic_mt_modes & 2))
2090                         n_subcores = 4;
2091                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2092                         return false;
2093         }
2094
2095         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2096 }
2097
2098 static void init_master_vcore(struct kvmppc_vcore *vc)
2099 {
2100         vc->master_vcore = vc;
2101         vc->entry_exit_map = 0;
2102         vc->in_guest = 0;
2103         vc->napping_threads = 0;
2104         vc->conferring_threads = 0;
2105 }
2106
2107 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2108 {
2109         int n_threads = vc->num_threads;
2110         int sub;
2111
2112         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2113                 return false;
2114
2115         if (n_threads < cip->max_subcore_threads)
2116                 n_threads = cip->max_subcore_threads;
2117         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2118                 return false;
2119         cip->max_subcore_threads = n_threads;
2120
2121         sub = cip->n_subcores;
2122         ++cip->n_subcores;
2123         cip->total_threads += vc->num_threads;
2124         cip->subcore_threads[sub] = vc->num_threads;
2125         cip->subcore_vm[sub] = vc->kvm;
2126         init_master_vcore(vc);
2127         list_del(&vc->preempt_list);
2128         list_add_tail(&vc->preempt_list, &cip->vcs[sub]);
2129
2130         return true;
2131 }
2132
2133 /*
2134  * Work out whether it is possible to piggyback the execution of
2135  * vcore *pvc onto the execution of the other vcores described in *cip.
2136  */
2137 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2138                           int target_threads)
2139 {
2140         if (cip->total_threads + pvc->num_threads > target_threads)
2141                 return false;
2142
2143         return can_dynamic_split(pvc, cip);
2144 }
2145
2146 static void prepare_threads(struct kvmppc_vcore *vc)
2147 {
2148         int i;
2149         struct kvm_vcpu *vcpu;
2150
2151         for_each_runnable_thread(i, vcpu, vc) {
2152                 if (signal_pending(vcpu->arch.run_task))
2153                         vcpu->arch.ret = -EINTR;
2154                 else if (vcpu->arch.vpa.update_pending ||
2155                          vcpu->arch.slb_shadow.update_pending ||
2156                          vcpu->arch.dtl.update_pending)
2157                         vcpu->arch.ret = RESUME_GUEST;
2158                 else
2159                         continue;
2160                 kvmppc_remove_runnable(vc, vcpu);
2161                 wake_up(&vcpu->arch.cpu_run);
2162         }
2163 }
2164
2165 static void collect_piggybacks(struct core_info *cip, int target_threads)
2166 {
2167         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2168         struct kvmppc_vcore *pvc, *vcnext;
2169
2170         spin_lock(&lp->lock);
2171         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2172                 if (!spin_trylock(&pvc->lock))
2173                         continue;
2174                 prepare_threads(pvc);
2175                 if (!pvc->n_runnable) {
2176                         list_del_init(&pvc->preempt_list);
2177                         if (pvc->runner == NULL) {
2178                                 pvc->vcore_state = VCORE_INACTIVE;
2179                                 kvmppc_core_end_stolen(pvc);
2180                         }
2181                         spin_unlock(&pvc->lock);
2182                         continue;
2183                 }
2184                 if (!can_piggyback(pvc, cip, target_threads)) {
2185                         spin_unlock(&pvc->lock);
2186                         continue;
2187                 }
2188                 kvmppc_core_end_stolen(pvc);
2189                 pvc->vcore_state = VCORE_PIGGYBACK;
2190                 if (cip->total_threads >= target_threads)
2191                         break;
2192         }
2193         spin_unlock(&lp->lock);
2194 }
2195
2196 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2197 {
2198         int still_running = 0, i;
2199         u64 now;
2200         long ret;
2201         struct kvm_vcpu *vcpu;
2202
2203         spin_lock(&vc->lock);
2204         now = get_tb();
2205         for_each_runnable_thread(i, vcpu, vc) {
2206                 /* cancel pending dec exception if dec is positive */
2207                 if (now < vcpu->arch.dec_expires &&
2208                     kvmppc_core_pending_dec(vcpu))
2209                         kvmppc_core_dequeue_dec(vcpu);
2210
2211                 trace_kvm_guest_exit(vcpu);
2212
2213                 ret = RESUME_GUEST;
2214                 if (vcpu->arch.trap)
2215                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2216                                                     vcpu->arch.run_task);
2217
2218                 vcpu->arch.ret = ret;
2219                 vcpu->arch.trap = 0;
2220
2221                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2222                         if (vcpu->arch.pending_exceptions)
2223                                 kvmppc_core_prepare_to_enter(vcpu);
2224                         if (vcpu->arch.ceded)
2225                                 kvmppc_set_timer(vcpu);
2226                         else
2227                                 ++still_running;
2228                 } else {
2229                         kvmppc_remove_runnable(vc, vcpu);
2230                         wake_up(&vcpu->arch.cpu_run);
2231                 }
2232         }
2233         list_del_init(&vc->preempt_list);
2234         if (!is_master) {
2235                 if (still_running > 0) {
2236                         kvmppc_vcore_preempt(vc);
2237                 } else if (vc->runner) {
2238                         vc->vcore_state = VCORE_PREEMPT;
2239                         kvmppc_core_start_stolen(vc);
2240                 } else {
2241                         vc->vcore_state = VCORE_INACTIVE;
2242                 }
2243                 if (vc->n_runnable > 0 && vc->runner == NULL) {
2244                         /* make sure there's a candidate runner awake */
2245                         i = -1;
2246                         vcpu = next_runnable_thread(vc, &i);
2247                         wake_up(&vcpu->arch.cpu_run);
2248                 }
2249         }
2250         spin_unlock(&vc->lock);
2251 }
2252
2253 /*
2254  * Clear core from the list of active host cores as we are about to
2255  * enter the guest. Only do this if it is the primary thread of the
2256  * core (not if a subcore) that is entering the guest.
2257  */
2258 static inline void kvmppc_clear_host_core(int cpu)
2259 {
2260         int core;
2261
2262         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2263                 return;
2264         /*
2265          * Memory barrier can be omitted here as we will do a smp_wmb()
2266          * later in kvmppc_start_thread and we need ensure that state is
2267          * visible to other CPUs only after we enter guest.
2268          */
2269         core = cpu >> threads_shift;
2270         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2271 }
2272
2273 /*
2274  * Advertise this core as an active host core since we exited the guest
2275  * Only need to do this if it is the primary thread of the core that is
2276  * exiting.
2277  */
2278 static inline void kvmppc_set_host_core(int cpu)
2279 {
2280         int core;
2281
2282         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2283                 return;
2284
2285         /*
2286          * Memory barrier can be omitted here because we do a spin_unlock
2287          * immediately after this which provides the memory barrier.
2288          */
2289         core = cpu >> threads_shift;
2290         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2291 }
2292
2293 /*
2294  * Run a set of guest threads on a physical core.
2295  * Called with vc->lock held.
2296  */
2297 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2298 {
2299         struct kvm_vcpu *vcpu;
2300         int i;
2301         int srcu_idx;
2302         struct core_info core_info;
2303         struct kvmppc_vcore *pvc, *vcnext;
2304         struct kvm_split_mode split_info, *sip;
2305         int split, subcore_size, active;
2306         int sub;
2307         bool thr0_done;
2308         unsigned long cmd_bit, stat_bit;
2309         int pcpu, thr;
2310         int target_threads;
2311
2312         /*
2313          * Remove from the list any threads that have a signal pending
2314          * or need a VPA update done
2315          */
2316         prepare_threads(vc);
2317
2318         /* if the runner is no longer runnable, let the caller pick a new one */
2319         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2320                 return;
2321
2322         /*
2323          * Initialize *vc.
2324          */
2325         init_master_vcore(vc);
2326         vc->preempt_tb = TB_NIL;
2327
2328         /*
2329          * Make sure we are running on primary threads, and that secondary
2330          * threads are offline.  Also check if the number of threads in this
2331          * guest are greater than the current system threads per guest.
2332          */
2333         if ((threads_per_core > 1) &&
2334             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2335                 for_each_runnable_thread(i, vcpu, vc) {
2336                         vcpu->arch.ret = -EBUSY;
2337                         kvmppc_remove_runnable(vc, vcpu);
2338                         wake_up(&vcpu->arch.cpu_run);
2339                 }
2340                 goto out;
2341         }
2342
2343         /*
2344          * See if we could run any other vcores on the physical core
2345          * along with this one.
2346          */
2347         init_core_info(&core_info, vc);
2348         pcpu = smp_processor_id();
2349         target_threads = threads_per_subcore;
2350         if (target_smt_mode && target_smt_mode < target_threads)
2351                 target_threads = target_smt_mode;
2352         if (vc->num_threads < target_threads)
2353                 collect_piggybacks(&core_info, target_threads);
2354
2355         /* Decide on micro-threading (split-core) mode */
2356         subcore_size = threads_per_subcore;
2357         cmd_bit = stat_bit = 0;
2358         split = core_info.n_subcores;
2359         sip = NULL;
2360         if (split > 1) {
2361                 /* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2362                 if (split == 2 && (dynamic_mt_modes & 2)) {
2363                         cmd_bit = HID0_POWER8_1TO2LPAR;
2364                         stat_bit = HID0_POWER8_2LPARMODE;
2365                 } else {
2366                         split = 4;
2367                         cmd_bit = HID0_POWER8_1TO4LPAR;
2368                         stat_bit = HID0_POWER8_4LPARMODE;
2369                 }
2370                 subcore_size = MAX_SMT_THREADS / split;
2371                 sip = &split_info;
2372                 memset(&split_info, 0, sizeof(split_info));
2373                 split_info.rpr = mfspr(SPRN_RPR);
2374                 split_info.pmmar = mfspr(SPRN_PMMAR);
2375                 split_info.ldbar = mfspr(SPRN_LDBAR);
2376                 split_info.subcore_size = subcore_size;
2377                 for (sub = 0; sub < core_info.n_subcores; ++sub)
2378                         split_info.master_vcs[sub] =
2379                                 list_first_entry(&core_info.vcs[sub],
2380                                         struct kvmppc_vcore, preempt_list);
2381                 /* order writes to split_info before kvm_split_mode pointer */
2382                 smp_wmb();
2383         }
2384         pcpu = smp_processor_id();
2385         for (thr = 0; thr < threads_per_subcore; ++thr)
2386                 paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2387
2388         /* Initiate micro-threading (split-core) if required */
2389         if (cmd_bit) {
2390                 unsigned long hid0 = mfspr(SPRN_HID0);
2391
2392                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2393                 mb();
2394                 mtspr(SPRN_HID0, hid0);
2395                 isync();
2396                 for (;;) {
2397                         hid0 = mfspr(SPRN_HID0);
2398                         if (hid0 & stat_bit)
2399                                 break;
2400                         cpu_relax();
2401                 }
2402         }
2403
2404         kvmppc_clear_host_core(pcpu);
2405
2406         /* Start all the threads */
2407         active = 0;
2408         for (sub = 0; sub < core_info.n_subcores; ++sub) {
2409                 thr = subcore_thread_map[sub];
2410                 thr0_done = false;
2411                 active |= 1 << thr;
2412                 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2413                         pvc->pcpu = pcpu + thr;
2414                         for_each_runnable_thread(i, vcpu, pvc) {
2415                                 kvmppc_start_thread(vcpu, pvc);
2416                                 kvmppc_create_dtl_entry(vcpu, pvc);
2417                                 trace_kvm_guest_enter(vcpu);
2418                                 if (!vcpu->arch.ptid)
2419                                         thr0_done = true;
2420                                 active |= 1 << (thr + vcpu->arch.ptid);
2421                         }
2422                         /*
2423                          * We need to start the first thread of each subcore
2424                          * even if it doesn't have a vcpu.
2425                          */
2426                         if (pvc->master_vcore == pvc && !thr0_done)
2427                                 kvmppc_start_thread(NULL, pvc);
2428                         thr += pvc->num_threads;
2429                 }
2430         }
2431
2432         /*
2433          * Ensure that split_info.do_nap is set after setting
2434          * the vcore pointer in the PACA of the secondaries.
2435          */
2436         smp_mb();
2437         if (cmd_bit)
2438                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
2439
2440         /*
2441          * When doing micro-threading, poke the inactive threads as well.
2442          * This gets them to the nap instruction after kvm_do_nap,
2443          * which reduces the time taken to unsplit later.
2444          */
2445         if (split > 1)
2446                 for (thr = 1; thr < threads_per_subcore; ++thr)
2447                         if (!(active & (1 << thr)))
2448                                 kvmppc_ipi_thread(pcpu + thr);
2449
2450         vc->vcore_state = VCORE_RUNNING;
2451         preempt_disable();
2452
2453         trace_kvmppc_run_core(vc, 0);
2454
2455         for (sub = 0; sub < core_info.n_subcores; ++sub)
2456                 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2457                         spin_unlock(&pvc->lock);
2458
2459         guest_enter();
2460
2461         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2462
2463         __kvmppc_vcore_entry();
2464
2465         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2466
2467         spin_lock(&vc->lock);
2468         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2469         vc->vcore_state = VCORE_EXITING;
2470
2471         /* wait for secondary threads to finish writing their state to memory */
2472         kvmppc_wait_for_nap();
2473
2474         /* Return to whole-core mode if we split the core earlier */
2475         if (split > 1) {
2476                 unsigned long hid0 = mfspr(SPRN_HID0);
2477                 unsigned long loops = 0;
2478
2479                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2480                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2481                 mb();
2482                 mtspr(SPRN_HID0, hid0);
2483                 isync();
2484                 for (;;) {
2485                         hid0 = mfspr(SPRN_HID0);
2486                         if (!(hid0 & stat_bit))
2487                                 break;
2488                         cpu_relax();
2489                         ++loops;
2490                 }
2491                 split_info.do_nap = 0;
2492         }
2493
2494         /* Let secondaries go back to the offline loop */
2495         for (i = 0; i < threads_per_subcore; ++i) {
2496                 kvmppc_release_hwthread(pcpu + i);
2497                 if (sip && sip->napped[i])
2498                         kvmppc_ipi_thread(pcpu + i);
2499         }
2500
2501         kvmppc_set_host_core(pcpu);
2502
2503         spin_unlock(&vc->lock);
2504
2505         /* make sure updates to secondary vcpu structs are visible now */
2506         smp_mb();
2507         guest_exit();
2508
2509         for (sub = 0; sub < core_info.n_subcores; ++sub)
2510                 list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2511                                          preempt_list)
2512                         post_guest_process(pvc, pvc == vc);
2513
2514         spin_lock(&vc->lock);
2515         preempt_enable();
2516
2517  out:
2518         vc->vcore_state = VCORE_INACTIVE;
2519         trace_kvmppc_run_core(vc, 1);
2520 }
2521
2522 /*
2523  * Wait for some other vcpu thread to execute us, and
2524  * wake us up when we need to handle something in the host.
2525  */
2526 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2527                                  struct kvm_vcpu *vcpu, int wait_state)
2528 {
2529         DEFINE_WAIT(wait);
2530
2531         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2532         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2533                 spin_unlock(&vc->lock);
2534                 schedule();
2535                 spin_lock(&vc->lock);
2536         }
2537         finish_wait(&vcpu->arch.cpu_run, &wait);
2538 }
2539
2540 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
2541 {
2542         /* 10us base */
2543         if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
2544                 vc->halt_poll_ns = 10000;
2545         else
2546                 vc->halt_poll_ns *= halt_poll_ns_grow;
2547
2548         if (vc->halt_poll_ns > halt_poll_max_ns)
2549                 vc->halt_poll_ns = halt_poll_max_ns;
2550 }
2551
2552 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
2553 {
2554         if (halt_poll_ns_shrink == 0)
2555                 vc->halt_poll_ns = 0;
2556         else
2557                 vc->halt_poll_ns /= halt_poll_ns_shrink;
2558 }
2559
2560 /* Check to see if any of the runnable vcpus on the vcore have pending
2561  * exceptions or are no longer ceded
2562  */
2563 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
2564 {
2565         struct kvm_vcpu *vcpu;
2566         int i;
2567
2568         for_each_runnable_thread(i, vcpu, vc) {
2569                 if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded)
2570                         return 1;
2571         }
2572
2573         return 0;
2574 }
2575
2576 /*
2577  * All the vcpus in this vcore are idle, so wait for a decrementer
2578  * or external interrupt to one of the vcpus.  vc->lock is held.
2579  */
2580 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2581 {
2582         ktime_t cur, start_poll, start_wait;
2583         int do_sleep = 1;
2584         u64 block_ns;
2585         DECLARE_SWAITQUEUE(wait);
2586
2587         /* Poll for pending exceptions and ceded state */
2588         cur = start_poll = ktime_get();
2589         if (vc->halt_poll_ns) {
2590                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
2591                 ++vc->runner->stat.halt_attempted_poll;
2592
2593                 vc->vcore_state = VCORE_POLLING;
2594                 spin_unlock(&vc->lock);
2595
2596                 do {
2597                         if (kvmppc_vcore_check_block(vc)) {
2598                                 do_sleep = 0;
2599                                 break;
2600                         }
2601                         cur = ktime_get();
2602                 } while (single_task_running() && ktime_before(cur, stop));
2603
2604                 spin_lock(&vc->lock);
2605                 vc->vcore_state = VCORE_INACTIVE;
2606
2607                 if (!do_sleep) {
2608                         ++vc->runner->stat.halt_successful_poll;
2609                         goto out;
2610                 }
2611         }
2612
2613         prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2614
2615         if (kvmppc_vcore_check_block(vc)) {
2616                 finish_swait(&vc->wq, &wait);
2617                 do_sleep = 0;
2618                 /* If we polled, count this as a successful poll */
2619                 if (vc->halt_poll_ns)
2620                         ++vc->runner->stat.halt_successful_poll;
2621                 goto out;
2622         }
2623
2624         start_wait = ktime_get();
2625
2626         vc->vcore_state = VCORE_SLEEPING;
2627         trace_kvmppc_vcore_blocked(vc, 0);
2628         spin_unlock(&vc->lock);
2629         schedule();
2630         finish_swait(&vc->wq, &wait);
2631         spin_lock(&vc->lock);
2632         vc->vcore_state = VCORE_INACTIVE;
2633         trace_kvmppc_vcore_blocked(vc, 1);
2634         ++vc->runner->stat.halt_successful_wait;
2635
2636         cur = ktime_get();
2637
2638 out:
2639         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
2640
2641         /* Attribute wait time */
2642         if (do_sleep) {
2643                 vc->runner->stat.halt_wait_ns +=
2644                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
2645                 /* Attribute failed poll time */
2646                 if (vc->halt_poll_ns)
2647                         vc->runner->stat.halt_poll_fail_ns +=
2648                                 ktime_to_ns(start_wait) -
2649                                 ktime_to_ns(start_poll);
2650         } else {
2651                 /* Attribute successful poll time */
2652                 if (vc->halt_poll_ns)
2653                         vc->runner->stat.halt_poll_success_ns +=
2654                                 ktime_to_ns(cur) -
2655                                 ktime_to_ns(start_poll);
2656         }
2657
2658         /* Adjust poll time */
2659         if (halt_poll_max_ns) {
2660                 if (block_ns <= vc->halt_poll_ns)
2661                         ;
2662                 /* We slept and blocked for longer than the max halt time */
2663                 else if (vc->halt_poll_ns && block_ns > halt_poll_max_ns)
2664                         shrink_halt_poll_ns(vc);
2665                 /* We slept and our poll time is too small */
2666                 else if (vc->halt_poll_ns < halt_poll_max_ns &&
2667                                 block_ns < halt_poll_max_ns)
2668                         grow_halt_poll_ns(vc);
2669         } else
2670                 vc->halt_poll_ns = 0;
2671
2672         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
2673 }
2674
2675 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2676 {
2677         int n_ceded, i;
2678         struct kvmppc_vcore *vc;
2679         struct kvm_vcpu *v;
2680
2681         trace_kvmppc_run_vcpu_enter(vcpu);
2682
2683         kvm_run->exit_reason = 0;
2684         vcpu->arch.ret = RESUME_GUEST;
2685         vcpu->arch.trap = 0;
2686         kvmppc_update_vpas(vcpu);
2687
2688         /*
2689          * Synchronize with other threads in this virtual core
2690          */
2691         vc = vcpu->arch.vcore;
2692         spin_lock(&vc->lock);
2693         vcpu->arch.ceded = 0;
2694         vcpu->arch.run_task = current;
2695         vcpu->arch.kvm_run = kvm_run;
2696         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2697         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2698         vcpu->arch.busy_preempt = TB_NIL;
2699         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
2700         ++vc->n_runnable;
2701
2702         /*
2703          * This happens the first time this is called for a vcpu.
2704          * If the vcore is already running, we may be able to start
2705          * this thread straight away and have it join in.
2706          */
2707         if (!signal_pending(current)) {
2708                 if (vc->vcore_state == VCORE_PIGGYBACK) {
2709                         struct kvmppc_vcore *mvc = vc->master_vcore;
2710                         if (spin_trylock(&mvc->lock)) {
2711                                 if (mvc->vcore_state == VCORE_RUNNING &&
2712                                     !VCORE_IS_EXITING(mvc)) {
2713                                         kvmppc_create_dtl_entry(vcpu, vc);
2714                                         kvmppc_start_thread(vcpu, vc);
2715                                         trace_kvm_guest_enter(vcpu);
2716                                 }
2717                                 spin_unlock(&mvc->lock);
2718                         }
2719                 } else if (vc->vcore_state == VCORE_RUNNING &&
2720                            !VCORE_IS_EXITING(vc)) {
2721                         kvmppc_create_dtl_entry(vcpu, vc);
2722                         kvmppc_start_thread(vcpu, vc);
2723                         trace_kvm_guest_enter(vcpu);
2724                 } else if (vc->vcore_state == VCORE_SLEEPING) {
2725                         swake_up(&vc->wq);
2726                 }
2727
2728         }
2729
2730         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2731                !signal_pending(current)) {
2732                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2733                         kvmppc_vcore_end_preempt(vc);
2734
2735                 if (vc->vcore_state != VCORE_INACTIVE) {
2736                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
2737                         continue;
2738                 }
2739                 for_each_runnable_thread(i, v, vc) {
2740                         kvmppc_core_prepare_to_enter(v);
2741                         if (signal_pending(v->arch.run_task)) {
2742                                 kvmppc_remove_runnable(vc, v);
2743                                 v->stat.signal_exits++;
2744                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2745                                 v->arch.ret = -EINTR;
2746                                 wake_up(&v->arch.cpu_run);
2747                         }
2748                 }
2749                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2750                         break;
2751                 n_ceded = 0;
2752                 for_each_runnable_thread(i, v, vc) {
2753                         if (!v->arch.pending_exceptions)
2754                                 n_ceded += v->arch.ceded;
2755                         else
2756                                 v->arch.ceded = 0;
2757                 }
2758                 vc->runner = vcpu;
2759                 if (n_ceded == vc->n_runnable) {
2760                         kvmppc_vcore_blocked(vc);
2761                 } else if (need_resched()) {
2762                         kvmppc_vcore_preempt(vc);
2763                         /* Let something else run */
2764                         cond_resched_lock(&vc->lock);
2765                         if (vc->vcore_state == VCORE_PREEMPT)
2766                                 kvmppc_vcore_end_preempt(vc);
2767                 } else {
2768                         kvmppc_run_core(vc);
2769                 }
2770                 vc->runner = NULL;
2771         }
2772
2773         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2774                (vc->vcore_state == VCORE_RUNNING ||
2775                 vc->vcore_state == VCORE_EXITING ||
2776                 vc->vcore_state == VCORE_PIGGYBACK))
2777                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
2778
2779         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2780                 kvmppc_vcore_end_preempt(vc);
2781
2782         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2783                 kvmppc_remove_runnable(vc, vcpu);
2784                 vcpu->stat.signal_exits++;
2785                 kvm_run->exit_reason = KVM_EXIT_INTR;
2786                 vcpu->arch.ret = -EINTR;
2787         }
2788
2789         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2790                 /* Wake up some vcpu to run the core */
2791                 i = -1;
2792                 v = next_runnable_thread(vc, &i);
2793                 wake_up(&v->arch.cpu_run);
2794         }
2795
2796         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2797         spin_unlock(&vc->lock);
2798         return vcpu->arch.ret;
2799 }
2800
2801 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2802 {
2803         int r;
2804         int srcu_idx;
2805         unsigned long ebb_regs[3] = {}; /* shut up GCC */
2806         unsigned long user_tar = 0;
2807         unsigned int user_vrsave;
2808
2809         if (!vcpu->arch.sane) {
2810                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2811                 return -EINVAL;
2812         }
2813
2814         /*
2815          * Don't allow entry with a suspended transaction, because
2816          * the guest entry/exit code will lose it.
2817          * If the guest has TM enabled, save away their TM-related SPRs
2818          * (they will get restored by the TM unavailable interrupt).
2819          */
2820 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2821         if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
2822             (current->thread.regs->msr & MSR_TM)) {
2823                 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
2824                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2825                         run->fail_entry.hardware_entry_failure_reason = 0;
2826                         return -EINVAL;
2827                 }
2828                 /* Enable TM so we can read the TM SPRs */
2829                 mtmsr(mfmsr() | MSR_TM);
2830                 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
2831                 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
2832                 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
2833                 current->thread.regs->msr &= ~MSR_TM;
2834         }
2835 #endif
2836
2837         kvmppc_core_prepare_to_enter(vcpu);
2838
2839         /* No need to go into the guest when all we'll do is come back out */
2840         if (signal_pending(current)) {
2841                 run->exit_reason = KVM_EXIT_INTR;
2842                 return -EINTR;
2843         }
2844
2845         atomic_inc(&vcpu->kvm->arch.vcpus_running);
2846         /* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2847         smp_mb();
2848
2849         /* On the first time here, set up HTAB and VRMA */
2850         if (!vcpu->kvm->arch.hpte_setup_done) {
2851                 r = kvmppc_hv_setup_htab_rma(vcpu);
2852                 if (r)
2853                         goto out;
2854         }
2855
2856         flush_all_to_thread(current);
2857
2858         /* Save userspace EBB and other register values */
2859         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
2860                 ebb_regs[0] = mfspr(SPRN_EBBHR);
2861                 ebb_regs[1] = mfspr(SPRN_EBBRR);
2862                 ebb_regs[2] = mfspr(SPRN_BESCR);
2863                 user_tar = mfspr(SPRN_TAR);
2864         }
2865         user_vrsave = mfspr(SPRN_VRSAVE);
2866
2867         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2868         vcpu->arch.pgdir = current->mm->pgd;
2869         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2870
2871         do {
2872                 r = kvmppc_run_vcpu(run, vcpu);
2873
2874                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2875                     !(vcpu->arch.shregs.msr & MSR_PR)) {
2876                         trace_kvm_hcall_enter(vcpu);
2877                         r = kvmppc_pseries_do_hcall(vcpu);
2878                         trace_kvm_hcall_exit(vcpu, r);
2879                         kvmppc_core_prepare_to_enter(vcpu);
2880                 } else if (r == RESUME_PAGE_FAULT) {
2881                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2882                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
2883                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2884                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2885                 } else if (r == RESUME_PASSTHROUGH)
2886                         r = kvmppc_xics_rm_complete(vcpu, 0);
2887         } while (is_kvmppc_resume_guest(r));
2888
2889         /* Restore userspace EBB and other register values */
2890         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
2891                 mtspr(SPRN_EBBHR, ebb_regs[0]);
2892                 mtspr(SPRN_EBBRR, ebb_regs[1]);
2893                 mtspr(SPRN_BESCR, ebb_regs[2]);
2894                 mtspr(SPRN_TAR, user_tar);
2895                 mtspr(SPRN_FSCR, current->thread.fscr);
2896         }
2897         mtspr(SPRN_VRSAVE, user_vrsave);
2898
2899  out:
2900         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2901         atomic_dec(&vcpu->kvm->arch.vcpus_running);
2902         return r;
2903 }
2904
2905 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2906                                      int linux_psize)
2907 {
2908         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2909
2910         if (!def->shift)
2911                 return;
2912         (*sps)->page_shift = def->shift;
2913         (*sps)->slb_enc = def->sllp;
2914         (*sps)->enc[0].page_shift = def->shift;
2915         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
2916         /*
2917          * Add 16MB MPSS support if host supports it
2918          */
2919         if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2920                 (*sps)->enc[1].page_shift = 24;
2921                 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2922         }
2923         (*sps)++;
2924 }
2925
2926 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2927                                          struct kvm_ppc_smmu_info *info)
2928 {
2929         struct kvm_ppc_one_seg_page_size *sps;
2930
2931         info->flags = KVM_PPC_PAGE_SIZES_REAL;
2932         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2933                 info->flags |= KVM_PPC_1T_SEGMENTS;
2934         info->slb_size = mmu_slb_size;
2935
2936         /* We only support these sizes for now, and no muti-size segments */
2937         sps = &info->sps[0];
2938         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2939         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2940         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2941
2942         return 0;
2943 }
2944
2945 /*
2946  * Get (and clear) the dirty memory log for a memory slot.
2947  */
2948 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2949                                          struct kvm_dirty_log *log)
2950 {
2951         struct kvm_memslots *slots;
2952         struct kvm_memory_slot *memslot;
2953         int r;
2954         unsigned long n;
2955
2956         mutex_lock(&kvm->slots_lock);
2957
2958         r = -EINVAL;
2959         if (log->slot >= KVM_USER_MEM_SLOTS)
2960                 goto out;
2961
2962         slots = kvm_memslots(kvm);
2963         memslot = id_to_memslot(slots, log->slot);
2964         r = -ENOENT;
2965         if (!memslot->dirty_bitmap)
2966                 goto out;
2967
2968         n = kvm_dirty_bitmap_bytes(memslot);
2969         memset(memslot->dirty_bitmap, 0, n);
2970
2971         r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2972         if (r)
2973                 goto out;
2974
2975         r = -EFAULT;
2976         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2977                 goto out;
2978
2979         r = 0;
2980 out:
2981         mutex_unlock(&kvm->slots_lock);
2982         return r;
2983 }
2984
2985 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2986                                         struct kvm_memory_slot *dont)
2987 {
2988         if (!dont || free->arch.rmap != dont->arch.rmap) {
2989                 vfree(free->arch.rmap);
2990                 free->arch.rmap = NULL;
2991         }
2992 }
2993
2994 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2995                                          unsigned long npages)
2996 {
2997         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2998         if (!slot->arch.rmap)
2999                 return -ENOMEM;
3000
3001         return 0;
3002 }
3003
3004 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3005                                         struct kvm_memory_slot *memslot,
3006                                         const struct kvm_userspace_memory_region *mem)
3007 {
3008         return 0;
3009 }
3010
3011 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3012                                 const struct kvm_userspace_memory_region *mem,
3013                                 const struct kvm_memory_slot *old,
3014                                 const struct kvm_memory_slot *new)
3015 {
3016         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3017         struct kvm_memslots *slots;
3018         struct kvm_memory_slot *memslot;
3019
3020         if (npages && old->npages) {
3021                 /*
3022                  * If modifying a memslot, reset all the rmap dirty bits.
3023                  * If this is a new memslot, we don't need to do anything
3024                  * since the rmap array starts out as all zeroes,
3025                  * i.e. no pages are dirty.
3026                  */
3027                 slots = kvm_memslots(kvm);
3028                 memslot = id_to_memslot(slots, mem->slot);
3029                 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
3030         }
3031 }
3032
3033 /*
3034  * Update LPCR values in kvm->arch and in vcores.
3035  * Caller must hold kvm->lock.
3036  */
3037 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3038 {
3039         long int i;
3040         u32 cores_done = 0;
3041
3042         if ((kvm->arch.lpcr & mask) == lpcr)
3043                 return;
3044
3045         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3046
3047         for (i = 0; i < KVM_MAX_VCORES; ++i) {
3048                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3049                 if (!vc)
3050                         continue;
3051                 spin_lock(&vc->lock);
3052                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3053                 spin_unlock(&vc->lock);
3054                 if (++cores_done >= kvm->arch.online_vcores)
3055                         break;
3056         }
3057 }
3058
3059 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3060 {
3061         return;
3062 }
3063
3064 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3065 {
3066         int err = 0;
3067         struct kvm *kvm = vcpu->kvm;
3068         unsigned long hva;
3069         struct kvm_memory_slot *memslot;
3070         struct vm_area_struct *vma;
3071         unsigned long lpcr = 0, senc;
3072         unsigned long psize, porder;
3073         int srcu_idx;
3074
3075         mutex_lock(&kvm->lock);
3076         if (kvm->arch.hpte_setup_done)
3077                 goto out;       /* another vcpu beat us to it */
3078
3079         /* Allocate hashed page table (if not done already) and reset it */
3080         if (!kvm->arch.hpt_virt) {
3081                 err = kvmppc_alloc_hpt(kvm, NULL);
3082                 if (err) {
3083                         pr_err("KVM: Couldn't alloc HPT\n");
3084                         goto out;
3085                 }
3086         }
3087
3088         /* Look up the memslot for guest physical address 0 */
3089         srcu_idx = srcu_read_lock(&kvm->srcu);
3090         memslot = gfn_to_memslot(kvm, 0);
3091
3092         /* We must have some memory at 0 by now */
3093         err = -EINVAL;
3094         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3095                 goto out_srcu;
3096
3097         /* Look up the VMA for the start of this memory slot */
3098         hva = memslot->userspace_addr;
3099         down_read(&current->mm->mmap_sem);
3100         vma = find_vma(current->mm, hva);
3101         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3102                 goto up_out;
3103
3104         psize = vma_kernel_pagesize(vma);
3105
3106         up_read(&current->mm->mmap_sem);
3107
3108         /* We can handle 4k, 64k or 16M pages in the VRMA */
3109         if (psize >= 0x1000000)
3110                 psize = 0x1000000;
3111         else if (psize >= 0x10000)
3112                 psize = 0x10000;
3113         else
3114                 psize = 0x1000;
3115         porder = __ilog2(psize);
3116
3117         /* Update VRMASD field in the LPCR */
3118         senc = slb_pgsize_encoding(psize);
3119         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3120                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3121         /* the -4 is to account for senc values starting at 0x10 */
3122         lpcr = senc << (LPCR_VRMASD_SH - 4);
3123
3124         /* Create HPTEs in the hash page table for the VRMA */
3125         kvmppc_map_vrma(vcpu, memslot, porder);
3126
3127         kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3128
3129         /* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
3130         smp_wmb();
3131         kvm->arch.hpte_setup_done = 1;
3132         err = 0;
3133  out_srcu:
3134         srcu_read_unlock(&kvm->srcu, srcu_idx);
3135  out:
3136         mutex_unlock(&kvm->lock);
3137         return err;
3138
3139  up_out:
3140         up_read(&current->mm->mmap_sem);
3141         goto out_srcu;
3142 }
3143
3144 #ifdef CONFIG_KVM_XICS
3145 static int kvmppc_cpu_notify(struct notifier_block *self, unsigned long action,
3146                         void *hcpu)
3147 {
3148         unsigned long cpu = (long)hcpu;
3149
3150         switch (action) {
3151         case CPU_UP_PREPARE:
3152         case CPU_UP_PREPARE_FROZEN:
3153                 kvmppc_set_host_core(cpu);
3154                 break;
3155
3156 #ifdef CONFIG_HOTPLUG_CPU
3157         case CPU_DEAD:
3158         case CPU_DEAD_FROZEN:
3159         case CPU_UP_CANCELED:
3160         case CPU_UP_CANCELED_FROZEN:
3161                 kvmppc_clear_host_core(cpu);
3162                 break;
3163 #endif
3164         default:
3165                 break;
3166         }
3167
3168         return NOTIFY_OK;
3169 }
3170
3171 static struct notifier_block kvmppc_cpu_notifier = {
3172             .notifier_call = kvmppc_cpu_notify,
3173 };
3174
3175 /*
3176  * Allocate a per-core structure for managing state about which cores are
3177  * running in the host versus the guest and for exchanging data between
3178  * real mode KVM and CPU running in the host.
3179  * This is only done for the first VM.
3180  * The allocated structure stays even if all VMs have stopped.
3181  * It is only freed when the kvm-hv module is unloaded.
3182  * It's OK for this routine to fail, we just don't support host
3183  * core operations like redirecting H_IPI wakeups.
3184  */
3185 void kvmppc_alloc_host_rm_ops(void)
3186 {
3187         struct kvmppc_host_rm_ops *ops;
3188         unsigned long l_ops;
3189         int cpu, core;
3190         int size;
3191
3192         /* Not the first time here ? */
3193         if (kvmppc_host_rm_ops_hv != NULL)
3194                 return;
3195
3196         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3197         if (!ops)
3198                 return;
3199
3200         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3201         ops->rm_core = kzalloc(size, GFP_KERNEL);
3202
3203         if (!ops->rm_core) {
3204                 kfree(ops);
3205                 return;
3206         }
3207
3208         get_online_cpus();
3209
3210         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3211                 if (!cpu_online(cpu))
3212                         continue;
3213
3214                 core = cpu >> threads_shift;
3215                 ops->rm_core[core].rm_state.in_host = 1;
3216         }
3217
3218         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3219
3220         /*
3221          * Make the contents of the kvmppc_host_rm_ops structure visible
3222          * to other CPUs before we assign it to the global variable.
3223          * Do an atomic assignment (no locks used here), but if someone
3224          * beats us to it, just free our copy and return.
3225          */
3226         smp_wmb();
3227         l_ops = (unsigned long) ops;
3228
3229         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3230                 put_online_cpus();
3231                 kfree(ops->rm_core);
3232                 kfree(ops);
3233                 return;
3234         }
3235
3236         register_cpu_notifier(&kvmppc_cpu_notifier);
3237
3238         put_online_cpus();
3239 }
3240
3241 void kvmppc_free_host_rm_ops(void)
3242 {
3243         if (kvmppc_host_rm_ops_hv) {
3244                 unregister_cpu_notifier(&kvmppc_cpu_notifier);
3245                 kfree(kvmppc_host_rm_ops_hv->rm_core);
3246                 kfree(kvmppc_host_rm_ops_hv);
3247                 kvmppc_host_rm_ops_hv = NULL;
3248         }
3249 }
3250 #endif
3251
3252 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3253 {
3254         unsigned long lpcr, lpid;
3255         char buf[32];
3256
3257         /* Allocate the guest's logical partition ID */
3258
3259         lpid = kvmppc_alloc_lpid();
3260         if ((long)lpid < 0)
3261                 return -ENOMEM;
3262         kvm->arch.lpid = lpid;
3263
3264         kvmppc_alloc_host_rm_ops();
3265
3266         /*
3267          * Since we don't flush the TLB when tearing down a VM,
3268          * and this lpid might have previously been used,
3269          * make sure we flush on each core before running the new VM.
3270          */
3271         cpumask_setall(&kvm->arch.need_tlb_flush);
3272
3273         /* Start out with the default set of hcalls enabled */
3274         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3275                sizeof(kvm->arch.enabled_hcalls));
3276
3277         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3278
3279         /* Init LPCR for virtual RMA mode */
3280         kvm->arch.host_lpid = mfspr(SPRN_LPID);
3281         kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3282         lpcr &= LPCR_PECE | LPCR_LPES;
3283         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3284                 LPCR_VPM0 | LPCR_VPM1;
3285         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3286                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3287         /* On POWER8 turn on online bit to enable PURR/SPURR */
3288         if (cpu_has_feature(CPU_FTR_ARCH_207S))
3289                 lpcr |= LPCR_ONL;
3290         kvm->arch.lpcr = lpcr;
3291
3292         /*
3293          * Track that we now have a HV mode VM active. This blocks secondary
3294          * CPU threads from coming online.
3295          */
3296         kvm_hv_vm_activated();
3297
3298         /*
3299          * Create a debugfs directory for the VM
3300          */
3301         snprintf(buf, sizeof(buf), "vm%d", current->pid);
3302         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3303         if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3304                 kvmppc_mmu_debugfs_init(kvm);
3305
3306         return 0;
3307 }
3308
3309 static void kvmppc_free_vcores(struct kvm *kvm)
3310 {
3311         long int i;
3312
3313         for (i = 0; i < KVM_MAX_VCORES; ++i)
3314                 kfree(kvm->arch.vcores[i]);
3315         kvm->arch.online_vcores = 0;
3316 }
3317
3318 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3319 {
3320         debugfs_remove_recursive(kvm->arch.debugfs_dir);
3321
3322         kvm_hv_vm_deactivated();
3323
3324         kvmppc_free_vcores(kvm);
3325
3326         kvmppc_free_hpt(kvm);
3327
3328         kvmppc_free_pimap(kvm);
3329 }
3330
3331 /* We don't need to emulate any privileged instructions or dcbz */
3332 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3333                                      unsigned int inst, int *advance)
3334 {
3335         return EMULATE_FAIL;
3336 }
3337
3338 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3339                                         ulong spr_val)
3340 {
3341         return EMULATE_FAIL;
3342 }
3343
3344 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3345                                         ulong *spr_val)
3346 {
3347         return EMULATE_FAIL;
3348 }
3349
3350 static int kvmppc_core_check_processor_compat_hv(void)
3351 {
3352         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3353             !cpu_has_feature(CPU_FTR_ARCH_206))
3354                 return -EIO;
3355         /*
3356          * Disable KVM for Power9, untill the required bits merged.
3357          */
3358         if (cpu_has_feature(CPU_FTR_ARCH_300))
3359                 return -EIO;
3360
3361         return 0;
3362 }
3363
3364 #ifdef CONFIG_KVM_XICS
3365
3366 void kvmppc_free_pimap(struct kvm *kvm)
3367 {
3368         kfree(kvm->arch.pimap);
3369 }
3370
3371 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
3372 {
3373         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
3374 }
3375
3376 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3377 {
3378         struct irq_desc *desc;
3379         struct kvmppc_irq_map *irq_map;
3380         struct kvmppc_passthru_irqmap *pimap;
3381         struct irq_chip *chip;
3382         int i;
3383
3384         if (!kvm_irq_bypass)
3385                 return 1;
3386
3387         desc = irq_to_desc(host_irq);
3388         if (!desc)
3389                 return -EIO;
3390
3391         mutex_lock(&kvm->lock);
3392
3393         pimap = kvm->arch.pimap;
3394         if (pimap == NULL) {
3395                 /* First call, allocate structure to hold IRQ map */
3396                 pimap = kvmppc_alloc_pimap();
3397                 if (pimap == NULL) {
3398                         mutex_unlock(&kvm->lock);
3399                         return -ENOMEM;
3400                 }
3401                 kvm->arch.pimap = pimap;
3402         }
3403
3404         /*
3405          * For now, we only support interrupts for which the EOI operation
3406          * is an OPAL call followed by a write to XIRR, since that's
3407          * what our real-mode EOI code does.
3408          */
3409         chip = irq_data_get_irq_chip(&desc->irq_data);
3410         if (!chip || !is_pnv_opal_msi(chip)) {
3411                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
3412                         host_irq, guest_gsi);
3413                 mutex_unlock(&kvm->lock);
3414                 return -ENOENT;
3415         }
3416
3417         /*
3418          * See if we already have an entry for this guest IRQ number.
3419          * If it's mapped to a hardware IRQ number, that's an error,
3420          * otherwise re-use this entry.
3421          */
3422         for (i = 0; i < pimap->n_mapped; i++) {
3423                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
3424                         if (pimap->mapped[i].r_hwirq) {
3425                                 mutex_unlock(&kvm->lock);
3426                                 return -EINVAL;
3427                         }
3428                         break;
3429                 }
3430         }
3431
3432         if (i == KVMPPC_PIRQ_MAPPED) {
3433                 mutex_unlock(&kvm->lock);
3434                 return -EAGAIN;         /* table is full */
3435         }
3436
3437         irq_map = &pimap->mapped[i];
3438
3439         irq_map->v_hwirq = guest_gsi;
3440         irq_map->desc = desc;
3441
3442         /*
3443          * Order the above two stores before the next to serialize with
3444          * the KVM real mode handler.
3445          */
3446         smp_wmb();
3447         irq_map->r_hwirq = desc->irq_data.hwirq;
3448
3449         if (i == pimap->n_mapped)
3450                 pimap->n_mapped++;
3451
3452         kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
3453
3454         mutex_unlock(&kvm->lock);
3455
3456         return 0;
3457 }
3458
3459 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3460 {
3461         struct irq_desc *desc;
3462         struct kvmppc_passthru_irqmap *pimap;
3463         int i;
3464
3465         if (!kvm_irq_bypass)
3466                 return 0;
3467
3468         desc = irq_to_desc(host_irq);
3469         if (!desc)
3470                 return -EIO;
3471
3472         mutex_lock(&kvm->lock);
3473
3474         if (kvm->arch.pimap == NULL) {
3475                 mutex_unlock(&kvm->lock);
3476                 return 0;
3477         }
3478         pimap = kvm->arch.pimap;
3479
3480         for (i = 0; i < pimap->n_mapped; i++) {
3481                 if (guest_gsi == pimap->mapped[i].v_hwirq)
3482                         break;
3483         }
3484
3485         if (i == pimap->n_mapped) {
3486                 mutex_unlock(&kvm->lock);
3487                 return -ENODEV;
3488         }
3489
3490         kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
3491
3492         /* invalidate the entry */
3493         pimap->mapped[i].r_hwirq = 0;
3494
3495         /*
3496          * We don't free this structure even when the count goes to
3497          * zero. The structure is freed when we destroy the VM.
3498          */
3499
3500         mutex_unlock(&kvm->lock);
3501         return 0;
3502 }
3503
3504 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
3505                                              struct irq_bypass_producer *prod)
3506 {
3507         int ret = 0;
3508         struct kvm_kernel_irqfd *irqfd =
3509                 container_of(cons, struct kvm_kernel_irqfd, consumer);
3510
3511         irqfd->producer = prod;
3512
3513         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
3514         if (ret)
3515                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
3516                         prod->irq, irqfd->gsi, ret);
3517
3518         return ret;
3519 }
3520
3521 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
3522                                               struct irq_bypass_producer *prod)
3523 {
3524         int ret;
3525         struct kvm_kernel_irqfd *irqfd =
3526                 container_of(cons, struct kvm_kernel_irqfd, consumer);
3527
3528         irqfd->producer = NULL;
3529
3530         /*
3531          * When producer of consumer is unregistered, we change back to
3532          * default external interrupt handling mode - KVM real mode
3533          * will switch back to host.
3534          */
3535         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
3536         if (ret)
3537                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
3538                         prod->irq, irqfd->gsi, ret);
3539 }
3540 #endif
3541
3542 static long kvm_arch_vm_ioctl_hv(struct file *filp,
3543                                  unsigned int ioctl, unsigned long arg)
3544 {
3545         struct kvm *kvm __maybe_unused = filp->private_data;
3546         void __user *argp = (void __user *)arg;
3547         long r;
3548
3549         switch (ioctl) {
3550
3551         case KVM_PPC_ALLOCATE_HTAB: {
3552                 u32 htab_order;
3553
3554                 r = -EFAULT;
3555                 if (get_user(htab_order, (u32 __user *)argp))
3556                         break;
3557                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
3558                 if (r)
3559                         break;
3560                 r = -EFAULT;
3561                 if (put_user(htab_order, (u32 __user *)argp))
3562                         break;
3563                 r = 0;
3564                 break;
3565         }
3566
3567         case KVM_PPC_GET_HTAB_FD: {
3568                 struct kvm_get_htab_fd ghf;
3569
3570                 r = -EFAULT;
3571                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
3572                         break;
3573                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3574                 break;
3575         }
3576
3577         default:
3578                 r = -ENOTTY;
3579         }
3580
3581         return r;
3582 }
3583
3584 /*
3585  * List of hcall numbers to enable by default.
3586  * For compatibility with old userspace, we enable by default
3587  * all hcalls that were implemented before the hcall-enabling
3588  * facility was added.  Note this list should not include H_RTAS.
3589  */
3590 static unsigned int default_hcall_list[] = {
3591         H_REMOVE,
3592         H_ENTER,
3593         H_READ,
3594         H_PROTECT,
3595         H_BULK_REMOVE,
3596         H_GET_TCE,
3597         H_PUT_TCE,
3598         H_SET_DABR,
3599         H_SET_XDABR,
3600         H_CEDE,
3601         H_PROD,
3602         H_CONFER,
3603         H_REGISTER_VPA,
3604 #ifdef CONFIG_KVM_XICS
3605         H_EOI,
3606         H_CPPR,
3607         H_IPI,
3608         H_IPOLL,
3609         H_XIRR,
3610         H_XIRR_X,
3611 #endif
3612         0
3613 };
3614
3615 static void init_default_hcalls(void)
3616 {
3617         int i;
3618         unsigned int hcall;
3619
3620         for (i = 0; default_hcall_list[i]; ++i) {
3621                 hcall = default_hcall_list[i];
3622                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3623                 __set_bit(hcall / 4, default_enabled_hcalls);
3624         }
3625 }
3626
3627 static struct kvmppc_ops kvm_ops_hv = {
3628         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3629         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3630         .get_one_reg = kvmppc_get_one_reg_hv,
3631         .set_one_reg = kvmppc_set_one_reg_hv,
3632         .vcpu_load   = kvmppc_core_vcpu_load_hv,
3633         .vcpu_put    = kvmppc_core_vcpu_put_hv,
3634         .set_msr     = kvmppc_set_msr_hv,
3635         .vcpu_run    = kvmppc_vcpu_run_hv,
3636         .vcpu_create = kvmppc_core_vcpu_create_hv,
3637         .vcpu_free   = kvmppc_core_vcpu_free_hv,
3638         .check_requests = kvmppc_core_check_requests_hv,
3639         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
3640         .flush_memslot  = kvmppc_core_flush_memslot_hv,
3641         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3642         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
3643         .unmap_hva = kvm_unmap_hva_hv,
3644         .unmap_hva_range = kvm_unmap_hva_range_hv,
3645         .age_hva  = kvm_age_hva_hv,
3646         .test_age_hva = kvm_test_age_hva_hv,
3647         .set_spte_hva = kvm_set_spte_hva_hv,
3648         .mmu_destroy  = kvmppc_mmu_destroy_hv,
3649         .free_memslot = kvmppc_core_free_memslot_hv,
3650         .create_memslot = kvmppc_core_create_memslot_hv,
3651         .init_vm =  kvmppc_core_init_vm_hv,
3652         .destroy_vm = kvmppc_core_destroy_vm_hv,
3653         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3654         .emulate_op = kvmppc_core_emulate_op_hv,
3655         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3656         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3657         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3658         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
3659         .hcall_implemented = kvmppc_hcall_impl_hv,
3660 #ifdef CONFIG_KVM_XICS
3661         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
3662         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
3663 #endif
3664 };
3665
3666 static int kvm_init_subcore_bitmap(void)
3667 {
3668         int i, j;
3669         int nr_cores = cpu_nr_cores();
3670         struct sibling_subcore_state *sibling_subcore_state;
3671
3672         for (i = 0; i < nr_cores; i++) {
3673                 int first_cpu = i * threads_per_core;
3674                 int node = cpu_to_node(first_cpu);
3675
3676                 /* Ignore if it is already allocated. */
3677                 if (paca[first_cpu].sibling_subcore_state)
3678                         continue;
3679
3680                 sibling_subcore_state =
3681                         kmalloc_node(sizeof(struct sibling_subcore_state),
3682                                                         GFP_KERNEL, node);
3683                 if (!sibling_subcore_state)
3684                         return -ENOMEM;
3685
3686                 memset(sibling_subcore_state, 0,
3687                                 sizeof(struct sibling_subcore_state));
3688
3689                 for (j = 0; j < threads_per_core; j++) {
3690                         int cpu = first_cpu + j;
3691
3692                         paca[cpu].sibling_subcore_state = sibling_subcore_state;
3693                 }
3694         }
3695         return 0;
3696 }
3697
3698 static int kvmppc_book3s_init_hv(void)
3699 {
3700         int r;
3701         /*
3702          * FIXME!! Do we need to check on all cpus ?
3703          */
3704         r = kvmppc_core_check_processor_compat_hv();
3705         if (r < 0)
3706                 return -ENODEV;
3707
3708         r = kvm_init_subcore_bitmap();
3709         if (r)
3710                 return r;
3711
3712         kvm_ops_hv.owner = THIS_MODULE;
3713         kvmppc_hv_ops = &kvm_ops_hv;
3714
3715         init_default_hcalls();
3716
3717         init_vcore_lists();
3718
3719         r = kvmppc_mmu_hv_init();
3720         return r;
3721 }
3722
3723 static void kvmppc_book3s_exit_hv(void)
3724 {
3725         kvmppc_free_host_rm_ops();
3726         kvmppc_hv_ops = NULL;
3727 }
3728
3729 module_init(kvmppc_book3s_init_hv);
3730 module_exit(kvmppc_book3s_exit_hv);
3731 MODULE_LICENSE("GPL");
3732 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3733 MODULE_ALIAS("devname:kvm");