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         struct kvm_vcpu *ret;
 378
 379         mutex_lock(&kvm->lock);
 380         ret = kvm_get_vcpu_by_id(kvm, id);
 381         mutex_unlock(&kvm->lock);
 382         return ret;
 383 }
 384
 385 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 386 {
 387         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 388         vpa->yield_count = cpu_to_be32(1);
 389 }
 390
 391 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 392                    unsigned long addr, unsigned long len)
 393 {
 394         /* check address is cacheline aligned */
 395         if (addr & (L1_CACHE_BYTES - 1))
 396                 return -EINVAL;
 397         spin_lock(&vcpu->arch.vpa_update_lock);
 398         if (v->next_gpa != addr || v->len != len) {
 399                 v->next_gpa = addr;
 400                 v->len = addr ? len : 0;
 401                 v->update_pending = 1;
 402         }
 403         spin_unlock(&vcpu->arch.vpa_update_lock);
 404         return 0;
 405 }
 406
 407 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 408 struct reg_vpa {
 409         u32 dummy;
 410         union {
 411                 __be16 hword;
 412                 __be32 word;
 413         } length;
 414 };
 415
 416 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 417 {
 418         if (vpap->update_pending)
 419                 return vpap->next_gpa != 0;
 420         return vpap->pinned_addr != NULL;
 421 }
 422
 423 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 424                                        unsigned long flags,
 425                                        unsigned long vcpuid, unsigned long vpa)
 426 {
 427         struct kvm *kvm = vcpu->kvm;
 428         unsigned long len, nb;
 429         void *va;
 430         struct kvm_vcpu *tvcpu;
 431         int err;
 432         int subfunc;
 433         struct kvmppc_vpa *vpap;
 434
 435         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 436         if (!tvcpu)
 437                 return H_PARAMETER;
 438
 439         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 440         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 441             subfunc == H_VPA_REG_SLB) {
 442                 /* Registering new area - address must be cache-line aligned */
 443                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 444                         return H_PARAMETER;
 445
 446                 /* convert logical addr to kernel addr and read length */
 447                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 448                 if (va == NULL)
 449                         return H_PARAMETER;
 450                 if (subfunc == H_VPA_REG_VPA)
 451                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 452                 else
 453                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 454                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 455
 456                 /* Check length */
 457                 if (len > nb || len < sizeof(struct reg_vpa))
 458                         return H_PARAMETER;
 459         } else {
 460                 vpa = 0;
 461                 len = 0;
 462         }
 463
 464         err = H_PARAMETER;
 465         vpap = NULL;
 466         spin_lock(&tvcpu->arch.vpa_update_lock);
 467
 468         switch (subfunc) {
 469         case H_VPA_REG_VPA:             /* register VPA */
 470                 if (len < sizeof(struct lppaca))
 471                         break;
 472                 vpap = &tvcpu->arch.vpa;
 473                 err = 0;
 474                 break;
 475
 476         case H_VPA_REG_DTL:             /* register DTL */
 477                 if (len < sizeof(struct dtl_entry))
 478                         break;
 479                 len -= len % sizeof(struct dtl_entry);
 480
 481                 /* Check that they have previously registered a VPA */
 482                 err = H_RESOURCE;
 483                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 484                         break;
 485
 486                 vpap = &tvcpu->arch.dtl;
 487                 err = 0;
 488                 break;
 489
 490         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 491                 /* Check that they have previously registered a VPA */
 492                 err = H_RESOURCE;
 493                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 494                         break;
 495
 496                 vpap = &tvcpu->arch.slb_shadow;
 497                 err = 0;
 498                 break;
 499
 500         case H_VPA_DEREG_VPA:           /* deregister VPA */
 501                 /* Check they don't still have a DTL or SLB buf registered */
 502                 err = H_RESOURCE;
 503                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 504                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 505                         break;
 506
 507                 vpap = &tvcpu->arch.vpa;
 508                 err = 0;
 509                 break;
 510
 511         case H_VPA_DEREG_DTL:           /* deregister DTL */
 512                 vpap = &tvcpu->arch.dtl;
 513                 err = 0;
 514                 break;
 515
 516         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 517                 vpap = &tvcpu->arch.slb_shadow;
 518                 err = 0;
 519                 break;
 520         }
 521
 522         if (vpap) {
 523                 vpap->next_gpa = vpa;
 524                 vpap->len = len;
 525                 vpap->update_pending = 1;
 526         }
 527
 528         spin_unlock(&tvcpu->arch.vpa_update_lock);
 529
 530         return err;
 531 }
 532
 533 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 534 {
 535         struct kvm *kvm = vcpu->kvm;
 536         void *va;
 537         unsigned long nb;
 538         unsigned long gpa;
 539
 540         /*
 541          * We need to pin the page pointed to by vpap->next_gpa,
 542          * but we can't call kvmppc_pin_guest_page under the lock
 543          * as it does get_user_pages() and down_read().  So we
 544          * have to drop the lock, pin the page, then get the lock
 545          * again and check that a new area didn't get registered
 546          * in the meantime.
 547          */
 548         for (;;) {
 549                 gpa = vpap->next_gpa;
 550                 spin_unlock(&vcpu->arch.vpa_update_lock);
 551                 va = NULL;
 552                 nb = 0;
 553                 if (gpa)
 554                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 555                 spin_lock(&vcpu->arch.vpa_update_lock);
 556                 if (gpa == vpap->next_gpa)
 557                         break;
 558                 /* sigh... unpin that one and try again */
 559                 if (va)
 560                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 561         }
 562
 563         vpap->update_pending = 0;
 564         if (va && nb < vpap->len) {
 565                 /*
 566                  * If it's now too short, it must be that userspace
 567                  * has changed the mappings underlying guest memory,
 568                  * so unregister the region.
 569                  */
 570                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 571                 va = NULL;
 572         }
 573         if (vpap->pinned_addr)
 574                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 575                                         vpap->dirty);
 576         vpap->gpa = gpa;
 577         vpap->pinned_addr = va;
 578         vpap->dirty = false;
 579         if (va)
 580                 vpap->pinned_end = va + vpap->len;
 581 }
 582
 583 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 584 {
 585         if (!(vcpu->arch.vpa.update_pending ||
 586               vcpu->arch.slb_shadow.update_pending ||
 587               vcpu->arch.dtl.update_pending))
 588                 return;
 589
 590         spin_lock(&vcpu->arch.vpa_update_lock);
 591         if (vcpu->arch.vpa.update_pending) {
 592                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 593                 if (vcpu->arch.vpa.pinned_addr)
 594                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 595         }
 596         if (vcpu->arch.dtl.update_pending) {
 597                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 598                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 599                 vcpu->arch.dtl_index = 0;
 600         }
 601         if (vcpu->arch.slb_shadow.update_pending)
 602                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 603         spin_unlock(&vcpu->arch.vpa_update_lock);
 604 }
 605
 606 /*
 607  * Return the accumulated stolen time for the vcore up until `now'.
 608  * The caller should hold the vcore lock.
 609  */
 610 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 611 {
 612         u64 p;
 613         unsigned long flags;
 614
 615         spin_lock_irqsave(&vc->stoltb_lock, flags);
 616         p = vc->stolen_tb;
 617         if (vc->vcore_state != VCORE_INACTIVE &&
 618             vc->preempt_tb != TB_NIL)
 619                 p += now - vc->preempt_tb;
 620         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 621         return p;
 622 }
 623
 624 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 625                                     struct kvmppc_vcore *vc)
 626 {
 627         struct dtl_entry *dt;
 628         struct lppaca *vpa;
 629         unsigned long stolen;
 630         unsigned long core_stolen;
 631         u64 now;
 632
 633         dt = vcpu->arch.dtl_ptr;
 634         vpa = vcpu->arch.vpa.pinned_addr;
 635         now = mftb();
 636         core_stolen = vcore_stolen_time(vc, now);
 637         stolen = core_stolen - vcpu->arch.stolen_logged;
 638         vcpu->arch.stolen_logged = core_stolen;
 639         spin_lock_irq(&vcpu->arch.tbacct_lock);
 640         stolen += vcpu->arch.busy_stolen;
 641         vcpu->arch.busy_stolen = 0;
 642         spin_unlock_irq(&vcpu->arch.tbacct_lock);
 643         if (!dt || !vpa)
 644                 return;
 645         memset(dt, 0, sizeof(struct dtl_entry));
 646         dt->dispatch_reason = 7;
 647         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 648         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 649         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 650         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 651         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 652         ++dt;
 653         if (dt == vcpu->arch.dtl.pinned_end)
 654                 dt = vcpu->arch.dtl.pinned_addr;
 655         vcpu->arch.dtl_ptr = dt;
 656         /* order writing *dt vs. writing vpa->dtl_idx */
 657         smp_wmb();
 658         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 659         vcpu->arch.dtl.dirty = true;
 660 }
 661
 662 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 663 {
 664         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 665                 return true;
 666         if ((!vcpu->arch.vcore->arch_compat) &&
 667             cpu_has_feature(CPU_FTR_ARCH_207S))
 668                 return true;
 669         return false;
 670 }
 671
 672 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 673                              unsigned long resource, unsigned long value1,
 674                              unsigned long value2)
 675 {
 676         switch (resource) {
 677         case H_SET_MODE_RESOURCE_SET_CIABR:
 678                 if (!kvmppc_power8_compatible(vcpu))
 679                         return H_P2;
 680                 if (value2)
 681                         return H_P4;
 682                 if (mflags)
 683                         return H_UNSUPPORTED_FLAG_START;
 684                 /* Guests can't breakpoint the hypervisor */
 685                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 686                         return H_P3;
 687                 vcpu->arch.ciabr  = value1;
 688                 return H_SUCCESS;
 689         case H_SET_MODE_RESOURCE_SET_DAWR:
 690                 if (!kvmppc_power8_compatible(vcpu))
 691                         return H_P2;
 692                 if (mflags)
 693                         return H_UNSUPPORTED_FLAG_START;
 694                 if (value2 & DABRX_HYP)
 695                         return H_P4;
 696                 vcpu->arch.dawr  = value1;
 697                 vcpu->arch.dawrx = value2;
 698                 return H_SUCCESS;
 699         default:
 700                 return H_TOO_HARD;
 701         }
 702 }
 703
 704 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 705 {
 706         struct kvmppc_vcore *vcore = target->arch.vcore;
 707
 708         /*
 709          * We expect to have been called by the real mode handler
 710          * (kvmppc_rm_h_confer()) which would have directly returned
 711          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 712          * have useful work to do and should not confer) so we don't
 713          * recheck that here.
 714          */
 715
 716         spin_lock(&vcore->lock);
 717         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 718             vcore->vcore_state != VCORE_INACTIVE &&
 719             vcore->runner)
 720                 target = vcore->runner;
 721         spin_unlock(&vcore->lock);
 722
 723         return kvm_vcpu_yield_to(target);
 724 }
 725
 726 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 727 {
 728         int yield_count = 0;
 729         struct lppaca *lppaca;
 730
 731         spin_lock(&vcpu->arch.vpa_update_lock);
 732         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 733         if (lppaca)
 734                 yield_count = be32_to_cpu(lppaca->yield_count);
 735         spin_unlock(&vcpu->arch.vpa_update_lock);
 736         return yield_count;
 737 }
 738
 739 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 740 {
 741         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 742         unsigned long target, ret = H_SUCCESS;
 743         int yield_count;
 744         struct kvm_vcpu *tvcpu;
 745         int idx, rc;
 746
 747         if (req <= MAX_HCALL_OPCODE &&
 748             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 749                 return RESUME_HOST;
 750
 751         switch (req) {
 752         case H_CEDE:
 753                 break;
 754         case H_PROD:
 755                 target = kvmppc_get_gpr(vcpu, 4);
 756                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 757                 if (!tvcpu) {
 758                         ret = H_PARAMETER;
 759                         break;
 760                 }
 761                 tvcpu->arch.prodded = 1;
 762                 smp_mb();
 763                 if (vcpu->arch.ceded) {
 764                         if (swait_active(&vcpu->wq)) {
 765                                 swake_up(&vcpu->wq);
 766                                 vcpu->stat.halt_wakeup++;
 767                         }
 768                 }
 769                 break;
 770         case H_CONFER:
 771                 target = kvmppc_get_gpr(vcpu, 4);
 772                 if (target == -1)
 773                         break;
 774                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 775                 if (!tvcpu) {
 776                         ret = H_PARAMETER;
 777                         break;
 778                 }
 779                 yield_count = kvmppc_get_gpr(vcpu, 5);
 780                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 781                         break;
 782                 kvm_arch_vcpu_yield_to(tvcpu);
 783                 break;
 784         case H_REGISTER_VPA:
 785                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 786                                         kvmppc_get_gpr(vcpu, 5),
 787                                         kvmppc_get_gpr(vcpu, 6));
 788                 break;
 789         case H_RTAS:
 790                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 791                         return RESUME_HOST;
 792
 793                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 794                 rc = kvmppc_rtas_hcall(vcpu);
 795                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 796
 797                 if (rc == -ENOENT)
 798                         return RESUME_HOST;
 799                 else if (rc == 0)
 800                         break;
 801
 802                 /* Send the error out to userspace via KVM_RUN */
 803                 return rc;
 804         case H_LOGICAL_CI_LOAD:
 805                 ret = kvmppc_h_logical_ci_load(vcpu);
 806                 if (ret == H_TOO_HARD)
 807                         return RESUME_HOST;
 808                 break;
 809         case H_LOGICAL_CI_STORE:
 810                 ret = kvmppc_h_logical_ci_store(vcpu);
 811                 if (ret == H_TOO_HARD)
 812                         return RESUME_HOST;
 813                 break;
 814         case H_SET_MODE:
 815                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 816                                         kvmppc_get_gpr(vcpu, 5),
 817                                         kvmppc_get_gpr(vcpu, 6),
 818                                         kvmppc_get_gpr(vcpu, 7));
 819                 if (ret == H_TOO_HARD)
 820                         return RESUME_HOST;
 821                 break;
 822         case H_XIRR:
 823         case H_CPPR:
 824         case H_EOI:
 825         case H_IPI:
 826         case H_IPOLL:
 827         case H_XIRR_X:
 828                 if (kvmppc_xics_enabled(vcpu)) {
 829                         ret = kvmppc_xics_hcall(vcpu, req);
 830                         break;
 831                 }
 832                 return RESUME_HOST;
 833         case H_PUT_TCE:
 834                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 835                                                 kvmppc_get_gpr(vcpu, 5),
 836                                                 kvmppc_get_gpr(vcpu, 6));
 837                 if (ret == H_TOO_HARD)
 838                         return RESUME_HOST;
 839                 break;
 840         case H_PUT_TCE_INDIRECT:
 841                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
 842                                                 kvmppc_get_gpr(vcpu, 5),
 843                                                 kvmppc_get_gpr(vcpu, 6),
 844                                                 kvmppc_get_gpr(vcpu, 7));
 845                 if (ret == H_TOO_HARD)
 846                         return RESUME_HOST;
 847                 break;
 848         case H_STUFF_TCE:
 849                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 850                                                 kvmppc_get_gpr(vcpu, 5),
 851                                                 kvmppc_get_gpr(vcpu, 6),
 852                                                 kvmppc_get_gpr(vcpu, 7));
 853                 if (ret == H_TOO_HARD)
 854                         return RESUME_HOST;
 855                 break;
 856         default:
 857                 return RESUME_HOST;
 858         }
 859         kvmppc_set_gpr(vcpu, 3, ret);
 860         vcpu->arch.hcall_needed = 0;
 861         return RESUME_GUEST;
 862 }
 863
 864 static int kvmppc_hcall_impl_hv(unsigned long cmd)
 865 {
 866         switch (cmd) {
 867         case H_CEDE:
 868         case H_PROD:
 869         case H_CONFER:
 870         case H_REGISTER_VPA:
 871         case H_SET_MODE:
 872         case H_LOGICAL_CI_LOAD:
 873         case H_LOGICAL_CI_STORE:
 874 #ifdef CONFIG_KVM_XICS
 875         case H_XIRR:
 876         case H_CPPR:
 877         case H_EOI:
 878         case H_IPI:
 879         case H_IPOLL:
 880         case H_XIRR_X:
 881 #endif
 882                 return 1;
 883         }
 884
 885         /* See if it's in the real-mode table */
 886         return kvmppc_hcall_impl_hv_realmode(cmd);
 887 }
 888
 889 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
 890                                         struct kvm_vcpu *vcpu)
 891 {
 892         u32 last_inst;
 893
 894         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
 895                                         EMULATE_DONE) {
 896                 /*
 897                  * Fetch failed, so return to guest and
 898                  * try executing it again.
 899                  */
 900                 return RESUME_GUEST;
 901         }
 902
 903         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
 904                 run->exit_reason = KVM_EXIT_DEBUG;
 905                 run->debug.arch.address = kvmppc_get_pc(vcpu);
 906                 return RESUME_HOST;
 907         } else {
 908                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 909                 return RESUME_GUEST;
 910         }
 911 }
 912
 913 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
 914                                  struct task_struct *tsk)
 915 {
 916         int r = RESUME_HOST;
 917
 918         vcpu->stat.sum_exits++;
 919
 920         /*
 921          * This can happen if an interrupt occurs in the last stages
 922          * of guest entry or the first stages of guest exit (i.e. after
 923          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
 924          * and before setting it to KVM_GUEST_MODE_HOST_HV).
 925          * That can happen due to a bug, or due to a machine check
 926          * occurring at just the wrong time.
 927          */
 928         if (vcpu->arch.shregs.msr & MSR_HV) {
 929                 printk(KERN_EMERG "KVM trap in HV mode!\n");
 930                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 931                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 932                         vcpu->arch.shregs.msr);
 933                 kvmppc_dump_regs(vcpu);
 934                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
 935                 run->hw.hardware_exit_reason = vcpu->arch.trap;
 936                 return RESUME_HOST;
 937         }
 938         run->exit_reason = KVM_EXIT_UNKNOWN;
 939         run->ready_for_interrupt_injection = 1;
 940         switch (vcpu->arch.trap) {
 941         /* We're good on these - the host merely wanted to get our attention */
 942         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 943                 vcpu->stat.dec_exits++;
 944                 r = RESUME_GUEST;
 945                 break;
 946         case BOOK3S_INTERRUPT_EXTERNAL:
 947         case BOOK3S_INTERRUPT_H_DOORBELL:
 948                 vcpu->stat.ext_intr_exits++;
 949                 r = RESUME_GUEST;
 950                 break;
 951         /* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
 952         case BOOK3S_INTERRUPT_HMI:
 953         case BOOK3S_INTERRUPT_PERFMON:
 954                 r = RESUME_GUEST;
 955                 break;
 956         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 957                 /*
 958                  * Deliver a machine check interrupt to the guest.
 959                  * We have to do this, even if the host has handled the
 960                  * machine check, because machine checks use SRR0/1 and
 961                  * the interrupt might have trashed guest state in them.
 962                  */
 963                 kvmppc_book3s_queue_irqprio(vcpu,
 964                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 965                 r = RESUME_GUEST;
 966                 break;
 967         case BOOK3S_INTERRUPT_PROGRAM:
 968         {
 969                 ulong flags;
 970                 /*
 971                  * Normally program interrupts are delivered directly
 972                  * to the guest by the hardware, but we can get here
 973                  * as a result of a hypervisor emulation interrupt
 974                  * (e40) getting turned into a 700 by BML RTAS.
 975                  */
 976                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 977                 kvmppc_core_queue_program(vcpu, flags);
 978                 r = RESUME_GUEST;
 979                 break;
 980         }
 981         case BOOK3S_INTERRUPT_SYSCALL:
 982         {
 983                 /* hcall - punt to userspace */
 984                 int i;
 985
 986                 /* hypercall with MSR_PR has already been handled in rmode,
 987                  * and never reaches here.
 988                  */
 989
 990                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
 991                 for (i = 0; i < 9; ++i)
 992                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
 993                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
 994                 vcpu->arch.hcall_needed = 1;
 995                 r = RESUME_HOST;
 996                 break;
 997         }
 998         /*
 999          * We get these next two if the guest accesses a page which it thinks
1000          * it has mapped but which is not actually present, either because
1001          * it is for an emulated I/O device or because the corresonding
1002          * host page has been paged out.  Any other HDSI/HISI interrupts
1003          * have been handled already.
1004          */
1005         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1006                 r = RESUME_PAGE_FAULT;
1007                 break;
1008         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1009                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1010                 vcpu->arch.fault_dsisr = 0;
1011                 r = RESUME_PAGE_FAULT;
1012                 break;
1013         /*
1014          * This occurs if the guest executes an illegal instruction.
1015          * If the guest debug is disabled, generate a program interrupt
1016          * to the guest. If guest debug is enabled, we need to check
1017          * whether the instruction is a software breakpoint instruction.
1018          * Accordingly return to Guest or Host.
1019          */
1020         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1021                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1022                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1023                                 swab32(vcpu->arch.emul_inst) :
1024                                 vcpu->arch.emul_inst;
1025                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1026                         r = kvmppc_emulate_debug_inst(run, vcpu);
1027                 } else {
1028                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1029                         r = RESUME_GUEST;
1030                 }
1031                 break;
1032         /*
1033          * This occurs if the guest (kernel or userspace), does something that
1034          * is prohibited by HFSCR.  We just generate a program interrupt to
1035          * the guest.
1036          */
1037         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1038                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1039                 r = RESUME_GUEST;
1040                 break;
1041         case BOOK3S_INTERRUPT_HV_RM_HARD:
1042                 r = RESUME_PASSTHROUGH;
1043                 break;
1044         default:
1045                 kvmppc_dump_regs(vcpu);
1046                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1047                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1048                         vcpu->arch.shregs.msr);
1049                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1050                 r = RESUME_HOST;
1051                 break;
1052         }
1053
1054         return r;
1055 }
1056
1057 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1058                                             struct kvm_sregs *sregs)
1059 {
1060         int i;
1061
1062         memset(sregs, 0, sizeof(struct kvm_sregs));
1063         sregs->pvr = vcpu->arch.pvr;
1064         for (i = 0; i < vcpu->arch.slb_max; i++) {
1065                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1066                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1067         }
1068
1069         return 0;
1070 }
1071
1072 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1073                                             struct kvm_sregs *sregs)
1074 {
1075         int i, j;
1076
1077         /* Only accept the same PVR as the host's, since we can't spoof it */
1078         if (sregs->pvr != vcpu->arch.pvr)
1079                 return -EINVAL;
1080
1081         j = 0;
1082         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1083                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1084                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1085                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1086                         ++j;
1087                 }
1088         }
1089         vcpu->arch.slb_max = j;
1090
1091         return 0;
1092 }
1093
1094 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1095                 bool preserve_top32)
1096 {
1097         struct kvm *kvm = vcpu->kvm;
1098         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1099         u64 mask;
1100
1101         mutex_lock(&kvm->lock);
1102         spin_lock(&vc->lock);
1103         /*
1104          * If ILE (interrupt little-endian) has changed, update the
1105          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1106          */
1107         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1108                 struct kvm_vcpu *vcpu;
1109                 int i;
1110
1111                 kvm_for_each_vcpu(i, vcpu, kvm) {
1112                         if (vcpu->arch.vcore != vc)
1113                                 continue;
1114                         if (new_lpcr & LPCR_ILE)
1115                                 vcpu->arch.intr_msr |= MSR_LE;
1116                         else
1117                                 vcpu->arch.intr_msr &= ~MSR_LE;
1118                 }
1119         }
1120
1121         /*
1122          * Userspace can only modify DPFD (default prefetch depth),
1123          * ILE (interrupt little-endian) and TC (translation control).
1124          * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
1125          */
1126         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1127         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1128                 mask |= LPCR_AIL;
1129
1130         /* Broken 32-bit version of LPCR must not clear top bits */
1131         if (preserve_top32)
1132                 mask &= 0xFFFFFFFF;
1133         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1134         spin_unlock(&vc->lock);
1135         mutex_unlock(&kvm->lock);
1136 }
1137
1138 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1139                                  union kvmppc_one_reg *val)
1140 {
1141         int r = 0;
1142         long int i;
1143
1144         switch (id) {
1145         case KVM_REG_PPC_DEBUG_INST:
1146                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1147                 break;
1148         case KVM_REG_PPC_HIOR:
1149                 *val = get_reg_val(id, 0);
1150                 break;
1151         case KVM_REG_PPC_DABR:
1152                 *val = get_reg_val(id, vcpu->arch.dabr);
1153                 break;
1154         case KVM_REG_PPC_DABRX:
1155                 *val = get_reg_val(id, vcpu->arch.dabrx);
1156                 break;
1157         case KVM_REG_PPC_DSCR:
1158                 *val = get_reg_val(id, vcpu->arch.dscr);
1159                 break;
1160         case KVM_REG_PPC_PURR:
1161                 *val = get_reg_val(id, vcpu->arch.purr);
1162                 break;
1163         case KVM_REG_PPC_SPURR:
1164                 *val = get_reg_val(id, vcpu->arch.spurr);
1165                 break;
1166         case KVM_REG_PPC_AMR:
1167                 *val = get_reg_val(id, vcpu->arch.amr);
1168                 break;
1169         case KVM_REG_PPC_UAMOR:
1170                 *val = get_reg_val(id, vcpu->arch.uamor);
1171                 break;
1172         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1173                 i = id - KVM_REG_PPC_MMCR0;
1174                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1175                 break;
1176         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1177                 i = id - KVM_REG_PPC_PMC1;
1178                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1179                 break;
1180         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1181                 i = id - KVM_REG_PPC_SPMC1;
1182                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1183                 break;
1184         case KVM_REG_PPC_SIAR:
1185                 *val = get_reg_val(id, vcpu->arch.siar);
1186                 break;
1187         case KVM_REG_PPC_SDAR:
1188                 *val = get_reg_val(id, vcpu->arch.sdar);
1189                 break;
1190         case KVM_REG_PPC_SIER:
1191                 *val = get_reg_val(id, vcpu->arch.sier);
1192                 break;
1193         case KVM_REG_PPC_IAMR:
1194                 *val = get_reg_val(id, vcpu->arch.iamr);
1195                 break;
1196         case KVM_REG_PPC_PSPB:
1197                 *val = get_reg_val(id, vcpu->arch.pspb);
1198                 break;
1199         case KVM_REG_PPC_DPDES:
1200                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1201                 break;
1202         case KVM_REG_PPC_DAWR:
1203                 *val = get_reg_val(id, vcpu->arch.dawr);
1204                 break;
1205         case KVM_REG_PPC_DAWRX:
1206                 *val = get_reg_val(id, vcpu->arch.dawrx);
1207                 break;
1208         case KVM_REG_PPC_CIABR:
1209                 *val = get_reg_val(id, vcpu->arch.ciabr);
1210                 break;
1211         case KVM_REG_PPC_CSIGR:
1212                 *val = get_reg_val(id, vcpu->arch.csigr);
1213                 break;
1214         case KVM_REG_PPC_TACR:
1215                 *val = get_reg_val(id, vcpu->arch.tacr);
1216                 break;
1217         case KVM_REG_PPC_TCSCR:
1218                 *val = get_reg_val(id, vcpu->arch.tcscr);
1219                 break;
1220         case KVM_REG_PPC_PID:
1221                 *val = get_reg_val(id, vcpu->arch.pid);
1222                 break;
1223         case KVM_REG_PPC_ACOP:
1224                 *val = get_reg_val(id, vcpu->arch.acop);
1225                 break;
1226         case KVM_REG_PPC_WORT:
1227                 *val = get_reg_val(id, vcpu->arch.wort);
1228                 break;
1229         case KVM_REG_PPC_VPA_ADDR:
1230                 spin_lock(&vcpu->arch.vpa_update_lock);
1231                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1232                 spin_unlock(&vcpu->arch.vpa_update_lock);
1233                 break;
1234         case KVM_REG_PPC_VPA_SLB:
1235                 spin_lock(&vcpu->arch.vpa_update_lock);
1236                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1237                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1238                 spin_unlock(&vcpu->arch.vpa_update_lock);
1239                 break;
1240         case KVM_REG_PPC_VPA_DTL:
1241                 spin_lock(&vcpu->arch.vpa_update_lock);
1242                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1243                 val->vpaval.length = vcpu->arch.dtl.len;
1244                 spin_unlock(&vcpu->arch.vpa_update_lock);
1245                 break;
1246         case KVM_REG_PPC_TB_OFFSET:
1247                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1248                 break;
1249         case KVM_REG_PPC_LPCR:
1250         case KVM_REG_PPC_LPCR_64:
1251                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1252                 break;
1253         case KVM_REG_PPC_PPR:
1254                 *val = get_reg_val(id, vcpu->arch.ppr);
1255                 break;
1256 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1257         case KVM_REG_PPC_TFHAR:
1258                 *val = get_reg_val(id, vcpu->arch.tfhar);
1259                 break;
1260         case KVM_REG_PPC_TFIAR:
1261                 *val = get_reg_val(id, vcpu->arch.tfiar);
1262                 break;
1263         case KVM_REG_PPC_TEXASR:
1264                 *val = get_reg_val(id, vcpu->arch.texasr);
1265                 break;
1266         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1267                 i = id - KVM_REG_PPC_TM_GPR0;
1268                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1269                 break;
1270         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1271         {
1272                 int j;
1273                 i = id - KVM_REG_PPC_TM_VSR0;
1274                 if (i < 32)
1275                         for (j = 0; j < TS_FPRWIDTH; j++)
1276                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1277                 else {
1278                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1279                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1280                         else
1281                                 r = -ENXIO;
1282                 }
1283                 break;
1284         }
1285         case KVM_REG_PPC_TM_CR:
1286                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1287                 break;
1288         case KVM_REG_PPC_TM_LR:
1289                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1290                 break;
1291         case KVM_REG_PPC_TM_CTR:
1292                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1293                 break;
1294         case KVM_REG_PPC_TM_FPSCR:
1295                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1296                 break;
1297         case KVM_REG_PPC_TM_AMR:
1298                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1299                 break;
1300         case KVM_REG_PPC_TM_PPR:
1301                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1302                 break;
1303         case KVM_REG_PPC_TM_VRSAVE:
1304                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1305                 break;
1306         case KVM_REG_PPC_TM_VSCR:
1307                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1308                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1309                 else
1310                         r = -ENXIO;
1311                 break;
1312         case KVM_REG_PPC_TM_DSCR:
1313                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1314                 break;
1315         case KVM_REG_PPC_TM_TAR:
1316                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1317                 break;
1318 #endif
1319         case KVM_REG_PPC_ARCH_COMPAT:
1320                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1321                 break;
1322         default:
1323                 r = -EINVAL;
1324                 break;
1325         }
1326
1327         return r;
1328 }
1329
1330 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1331                                  union kvmppc_one_reg *val)
1332 {
1333         int r = 0;
1334         long int i;
1335         unsigned long addr, len;
1336
1337         switch (id) {
1338         case KVM_REG_PPC_HIOR:
1339                 /* Only allow this to be set to zero */
1340                 if (set_reg_val(id, *val))
1341                         r = -EINVAL;
1342                 break;
1343         case KVM_REG_PPC_DABR:
1344                 vcpu->arch.dabr = set_reg_val(id, *val);
1345                 break;
1346         case KVM_REG_PPC_DABRX:
1347                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1348                 break;
1349         case KVM_REG_PPC_DSCR:
1350                 vcpu->arch.dscr = set_reg_val(id, *val);
1351                 break;
1352         case KVM_REG_PPC_PURR:
1353                 vcpu->arch.purr = set_reg_val(id, *val);
1354                 break;
1355         case KVM_REG_PPC_SPURR:
1356                 vcpu->arch.spurr = set_reg_val(id, *val);
1357                 break;
1358         case KVM_REG_PPC_AMR:
1359                 vcpu->arch.amr = set_reg_val(id, *val);
1360                 break;
1361         case KVM_REG_PPC_UAMOR:
1362                 vcpu->arch.uamor = set_reg_val(id, *val);
1363                 break;
1364         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1365                 i = id - KVM_REG_PPC_MMCR0;
1366                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1367                 break;
1368         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1369                 i = id - KVM_REG_PPC_PMC1;
1370                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1371                 break;
1372         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1373                 i = id - KVM_REG_PPC_SPMC1;
1374                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1375                 break;
1376         case KVM_REG_PPC_SIAR:
1377                 vcpu->arch.siar = set_reg_val(id, *val);
1378                 break;
1379         case KVM_REG_PPC_SDAR:
1380                 vcpu->arch.sdar = set_reg_val(id, *val);
1381                 break;
1382         case KVM_REG_PPC_SIER:
1383                 vcpu->arch.sier = set_reg_val(id, *val);
1384                 break;
1385         case KVM_REG_PPC_IAMR:
1386                 vcpu->arch.iamr = set_reg_val(id, *val);
1387                 break;
1388         case KVM_REG_PPC_PSPB:
1389                 vcpu->arch.pspb = set_reg_val(id, *val);
1390                 break;
1391         case KVM_REG_PPC_DPDES:
1392                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1393                 break;
1394         case KVM_REG_PPC_DAWR:
1395                 vcpu->arch.dawr = set_reg_val(id, *val);
1396                 break;
1397         case KVM_REG_PPC_DAWRX:
1398                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1399                 break;
1400         case KVM_REG_PPC_CIABR:
1401                 vcpu->arch.ciabr = set_reg_val(id, *val);
1402                 /* Don't allow setting breakpoints in hypervisor code */
1403                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1404                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1405                 break;
1406         case KVM_REG_PPC_CSIGR:
1407                 vcpu->arch.csigr = set_reg_val(id, *val);
1408                 break;
1409         case KVM_REG_PPC_TACR:
1410                 vcpu->arch.tacr = set_reg_val(id, *val);
1411                 break;
1412         case KVM_REG_PPC_TCSCR:
1413                 vcpu->arch.tcscr = set_reg_val(id, *val);
1414                 break;
1415         case KVM_REG_PPC_PID:
1416                 vcpu->arch.pid = set_reg_val(id, *val);
1417                 break;
1418         case KVM_REG_PPC_ACOP:
1419                 vcpu->arch.acop = set_reg_val(id, *val);
1420                 break;
1421         case KVM_REG_PPC_WORT:
1422                 vcpu->arch.wort = set_reg_val(id, *val);
1423                 break;
1424         case KVM_REG_PPC_VPA_ADDR:
1425                 addr = set_reg_val(id, *val);
1426                 r = -EINVAL;
1427                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1428                               vcpu->arch.dtl.next_gpa))
1429                         break;
1430                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1431                 break;
1432         case KVM_REG_PPC_VPA_SLB:
1433                 addr = val->vpaval.addr;
1434                 len = val->vpaval.length;
1435                 r = -EINVAL;
1436                 if (addr && !vcpu->arch.vpa.next_gpa)
1437                         break;
1438                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1439                 break;
1440         case KVM_REG_PPC_VPA_DTL:
1441                 addr = val->vpaval.addr;
1442                 len = val->vpaval.length;
1443                 r = -EINVAL;
1444                 if (addr && (len < sizeof(struct dtl_entry) ||
1445                              !vcpu->arch.vpa.next_gpa))
1446                         break;
1447                 len -= len % sizeof(struct dtl_entry);
1448                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1449                 break;
1450         case KVM_REG_PPC_TB_OFFSET:
1451                 /* round up to multiple of 2^24 */
1452                 vcpu->arch.vcore->tb_offset =
1453                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1454                 break;
1455         case KVM_REG_PPC_LPCR:
1456                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1457                 break;
1458         case KVM_REG_PPC_LPCR_64:
1459                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1460                 break;
1461         case KVM_REG_PPC_PPR:
1462                 vcpu->arch.ppr = set_reg_val(id, *val);
1463                 break;
1464 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1465         case KVM_REG_PPC_TFHAR:
1466                 vcpu->arch.tfhar = set_reg_val(id, *val);
1467                 break;
1468         case KVM_REG_PPC_TFIAR:
1469                 vcpu->arch.tfiar = set_reg_val(id, *val);
1470                 break;
1471         case KVM_REG_PPC_TEXASR:
1472                 vcpu->arch.texasr = set_reg_val(id, *val);
1473                 break;
1474         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1475                 i = id - KVM_REG_PPC_TM_GPR0;
1476                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1477                 break;
1478         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1479         {
1480                 int j;
1481                 i = id - KVM_REG_PPC_TM_VSR0;
1482                 if (i < 32)
1483                         for (j = 0; j < TS_FPRWIDTH; j++)
1484                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1485                 else
1486                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1487                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1488                         else
1489                                 r = -ENXIO;
1490                 break;
1491         }
1492         case KVM_REG_PPC_TM_CR:
1493                 vcpu->arch.cr_tm = set_reg_val(id, *val);
1494                 break;
1495         case KVM_REG_PPC_TM_LR:
1496                 vcpu->arch.lr_tm = set_reg_val(id, *val);
1497                 break;
1498         case KVM_REG_PPC_TM_CTR:
1499                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1500                 break;
1501         case KVM_REG_PPC_TM_FPSCR:
1502                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1503                 break;
1504         case KVM_REG_PPC_TM_AMR:
1505                 vcpu->arch.amr_tm = set_reg_val(id, *val);
1506                 break;
1507         case KVM_REG_PPC_TM_PPR:
1508                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1509                 break;
1510         case KVM_REG_PPC_TM_VRSAVE:
1511                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1512                 break;
1513         case KVM_REG_PPC_TM_VSCR:
1514                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1515                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1516                 else
1517                         r = - ENXIO;
1518                 break;
1519         case KVM_REG_PPC_TM_DSCR:
1520                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1521                 break;
1522         case KVM_REG_PPC_TM_TAR:
1523                 vcpu->arch.tar_tm = set_reg_val(id, *val);
1524                 break;
1525 #endif
1526         case KVM_REG_PPC_ARCH_COMPAT:
1527                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1528                 break;
1529         default:
1530                 r = -EINVAL;
1531                 break;
1532         }
1533
1534         return r;
1535 }
1536
1537 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1538 {
1539         struct kvmppc_vcore *vcore;
1540
1541         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1542
1543         if (vcore == NULL)
1544                 return NULL;
1545
1546         spin_lock_init(&vcore->lock);
1547         spin_lock_init(&vcore->stoltb_lock);
1548         init_swait_queue_head(&vcore->wq);
1549         vcore->preempt_tb = TB_NIL;
1550         vcore->lpcr = kvm->arch.lpcr;
1551         vcore->first_vcpuid = core * threads_per_subcore;
1552         vcore->kvm = kvm;
1553         INIT_LIST_HEAD(&vcore->preempt_list);
1554
1555         return vcore;
1556 }
1557
1558 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1559 static struct debugfs_timings_element {
1560         const char *name;
1561         size_t offset;
1562 } timings[] = {
1563         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
1564         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
1565         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
1566         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
1567         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
1568 };
1569
1570 #define N_TIMINGS       (sizeof(timings) / sizeof(timings[0]))
1571
1572 struct debugfs_timings_state {
1573         struct kvm_vcpu *vcpu;
1574         unsigned int    buflen;
1575         char            buf[N_TIMINGS * 100];
1576 };
1577
1578 static int debugfs_timings_open(struct inode *inode, struct file *file)
1579 {
1580         struct kvm_vcpu *vcpu = inode->i_private;
1581         struct debugfs_timings_state *p;
1582
1583         p = kzalloc(sizeof(*p), GFP_KERNEL);
1584         if (!p)
1585                 return -ENOMEM;
1586
1587         kvm_get_kvm(vcpu->kvm);
1588         p->vcpu = vcpu;
1589         file->private_data = p;
1590
1591         return nonseekable_open(inode, file);
1592 }
1593
1594 static int debugfs_timings_release(struct inode *inode, struct file *file)
1595 {
1596         struct debugfs_timings_state *p = file->private_data;
1597
1598         kvm_put_kvm(p->vcpu->kvm);
1599         kfree(p);
1600         return 0;
1601 }
1602
1603 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1604                                     size_t len, loff_t *ppos)
1605 {
1606         struct debugfs_timings_state *p = file->private_data;
1607         struct kvm_vcpu *vcpu = p->vcpu;
1608         char *s, *buf_end;
1609         struct kvmhv_tb_accumulator tb;
1610         u64 count;
1611         loff_t pos;
1612         ssize_t n;
1613         int i, loops;
1614         bool ok;
1615
1616         if (!p->buflen) {
1617                 s = p->buf;
1618                 buf_end = s + sizeof(p->buf);
1619                 for (i = 0; i < N_TIMINGS; ++i) {
1620                         struct kvmhv_tb_accumulator *acc;
1621
1622                         acc = (struct kvmhv_tb_accumulator *)
1623                                 ((unsigned long)vcpu + timings[i].offset);
1624                         ok = false;
1625                         for (loops = 0; loops < 1000; ++loops) {
1626                                 count = acc->seqcount;
1627                                 if (!(count & 1)) {
1628                                         smp_rmb();
1629                                         tb = *acc;
1630                                         smp_rmb();
1631                                         if (count == acc->seqcount) {
1632                                                 ok = true;
1633                                                 break;
1634                                         }
1635                                 }
1636                                 udelay(1);
1637                         }
1638                         if (!ok)
1639                                 snprintf(s, buf_end - s, "%s: stuck\n",
1640                                         timings[i].name);
1641                         else
1642                                 snprintf(s, buf_end - s,
1643                                         "%s: %llu %llu %llu %llu\n",
1644                                         timings[i].name, count / 2,
1645                                         tb_to_ns(tb.tb_total),
1646                                         tb_to_ns(tb.tb_min),
1647                                         tb_to_ns(tb.tb_max));
1648                         s += strlen(s);
1649                 }
1650                 p->buflen = s - p->buf;
1651         }
1652
1653         pos = *ppos;
1654         if (pos >= p->buflen)
1655                 return 0;
1656         if (len > p->buflen - pos)
1657                 len = p->buflen - pos;
1658         n = copy_to_user(buf, p->buf + pos, len);
1659         if (n) {
1660                 if (n == len)
1661                         return -EFAULT;
1662                 len -= n;
1663         }
1664         *ppos = pos + len;
1665         return len;
1666 }
1667
1668 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1669                                      size_t len, loff_t *ppos)
1670 {
1671         return -EACCES;
1672 }
1673
1674 static const struct file_operations debugfs_timings_ops = {
1675         .owner   = THIS_MODULE,
1676         .open    = debugfs_timings_open,
1677         .release = debugfs_timings_release,
1678         .read    = debugfs_timings_read,
1679         .write   = debugfs_timings_write,
1680         .llseek  = generic_file_llseek,
1681 };
1682
1683 /* Create a debugfs directory for the vcpu */
1684 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1685 {
1686         char buf[16];
1687         struct kvm *kvm = vcpu->kvm;
1688
1689         snprintf(buf, sizeof(buf), "vcpu%u", id);
1690         if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1691                 return;
1692         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1693         if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1694                 return;
1695         vcpu->arch.debugfs_timings =
1696                 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1697                                     vcpu, &debugfs_timings_ops);
1698 }
1699
1700 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1701 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1702 {
1703 }
1704 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1705
1706 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1707                                                    unsigned int id)
1708 {
1709         struct kvm_vcpu *vcpu;
1710         int err = -EINVAL;
1711         int core;
1712         struct kvmppc_vcore *vcore;
1713
1714         core = id / threads_per_subcore;
1715         if (core >= KVM_MAX_VCORES)
1716                 goto out;
1717
1718         err = -ENOMEM;
1719         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1720         if (!vcpu)
1721                 goto out;
1722
1723         err = kvm_vcpu_init(vcpu, kvm, id);
1724         if (err)
1725                 goto free_vcpu;
1726
1727         vcpu->arch.shared = &vcpu->arch.shregs;
1728 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1729         /*
1730          * The shared struct is never shared on HV,
1731          * so we can always use host endianness
1732          */
1733 #ifdef __BIG_ENDIAN__
1734         vcpu->arch.shared_big_endian = true;
1735 #else
1736         vcpu->arch.shared_big_endian = false;
1737 #endif
1738 #endif
1739         vcpu->arch.mmcr[0] = MMCR0_FC;
1740         vcpu->arch.ctrl = CTRL_RUNLATCH;
1741         /* default to host PVR, since we can't spoof it */
1742         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1743         spin_lock_init(&vcpu->arch.vpa_update_lock);
1744         spin_lock_init(&vcpu->arch.tbacct_lock);
1745         vcpu->arch.busy_preempt = TB_NIL;
1746         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1747
1748         kvmppc_mmu_book3s_hv_init(vcpu);
1749
1750         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1751
1752         init_waitqueue_head(&vcpu->arch.cpu_run);
1753
1754         mutex_lock(&kvm->lock);
1755         vcore = kvm->arch.vcores[core];
1756         if (!vcore) {
1757                 vcore = kvmppc_vcore_create(kvm, core);
1758                 kvm->arch.vcores[core] = vcore;
1759                 kvm->arch.online_vcores++;
1760         }
1761         mutex_unlock(&kvm->lock);
1762
1763         if (!vcore)
1764                 goto free_vcpu;
1765
1766         spin_lock(&vcore->lock);
1767         ++vcore->num_threads;
1768         spin_unlock(&vcore->lock);
1769         vcpu->arch.vcore = vcore;
1770         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1771         vcpu->arch.thread_cpu = -1;
1772
1773         vcpu->arch.cpu_type = KVM_CPU_3S_64;
1774         kvmppc_sanity_check(vcpu);
1775
1776         debugfs_vcpu_init(vcpu, id);
1777
1778         return vcpu;
1779
1780 free_vcpu:
1781         kmem_cache_free(kvm_vcpu_cache, vcpu);
1782 out:
1783         return ERR_PTR(err);
1784 }
1785
1786 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1787 {
1788         if (vpa->pinned_addr)
1789                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1790                                         vpa->dirty);
1791 }
1792
1793 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1794 {
1795         spin_lock(&vcpu->arch.vpa_update_lock);
1796         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1797         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1798         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1799         spin_unlock(&vcpu->arch.vpa_update_lock);
1800         kvm_vcpu_uninit(vcpu);
1801         kmem_cache_free(kvm_vcpu_cache, vcpu);
1802 }
1803
1804 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1805 {
1806         /* Indicate we want to get back into the guest */
1807         return 1;
1808 }
1809
1810 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1811 {
1812         unsigned long dec_nsec, now;
1813
1814         now = get_tb();
1815         if (now > vcpu->arch.dec_expires) {
1816                 /* decrementer has already gone negative */
1817                 kvmppc_core_queue_dec(vcpu);
1818                 kvmppc_core_prepare_to_enter(vcpu);
1819                 return;
1820         }
1821         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1822                    / tb_ticks_per_sec;
1823         hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1824                       HRTIMER_MODE_REL);
1825         vcpu->arch.timer_running = 1;
1826 }
1827
1828 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1829 {
1830         vcpu->arch.ceded = 0;
1831         if (vcpu->arch.timer_running) {
1832                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1833                 vcpu->arch.timer_running = 0;
1834         }
1835 }
1836
1837 extern void __kvmppc_vcore_entry(void);
1838
1839 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1840                                    struct kvm_vcpu *vcpu)
1841 {
1842         u64 now;
1843
1844         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1845                 return;
1846         spin_lock_irq(&vcpu->arch.tbacct_lock);
1847         now = mftb();
1848         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1849                 vcpu->arch.stolen_logged;
1850         vcpu->arch.busy_preempt = now;
1851         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1852         spin_unlock_irq(&vcpu->arch.tbacct_lock);
1853         --vc->n_runnable;
1854         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
1855 }
1856
1857 static int kvmppc_grab_hwthread(int cpu)
1858 {
1859         struct paca_struct *tpaca;
1860         long timeout = 10000;
1861
1862         tpaca = &paca[cpu];
1863
1864         /* Ensure the thread won't go into the kernel if it wakes */
1865         tpaca->kvm_hstate.kvm_vcpu = NULL;
1866         tpaca->kvm_hstate.kvm_vcore = NULL;
1867         tpaca->kvm_hstate.napping = 0;
1868         smp_wmb();
1869         tpaca->kvm_hstate.hwthread_req = 1;
1870
1871         /*
1872          * If the thread is already executing in the kernel (e.g. handling
1873          * a stray interrupt), wait for it to get back to nap mode.
1874          * The smp_mb() is to ensure that our setting of hwthread_req
1875          * is visible before we look at hwthread_state, so if this
1876          * races with the code at system_reset_pSeries and the thread
1877          * misses our setting of hwthread_req, we are sure to see its
1878          * setting of hwthread_state, and vice versa.
1879          */
1880         smp_mb();
1881         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1882                 if (--timeout <= 0) {
1883                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1884                         return -EBUSY;
1885                 }
1886                 udelay(1);
1887         }
1888         return 0;
1889 }
1890
1891 static void kvmppc_release_hwthread(int cpu)
1892 {
1893         struct paca_struct *tpaca;
1894
1895         tpaca = &paca[cpu];
1896         tpaca->kvm_hstate.hwthread_req = 0;
1897         tpaca->kvm_hstate.kvm_vcpu = NULL;
1898         tpaca->kvm_hstate.kvm_vcore = NULL;
1899         tpaca->kvm_hstate.kvm_split_mode = NULL;
1900 }
1901
1902 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
1903 {
1904         int cpu;
1905         struct paca_struct *tpaca;
1906         struct kvmppc_vcore *mvc = vc->master_vcore;
1907
1908         cpu = vc->pcpu;
1909         if (vcpu) {
1910                 if (vcpu->arch.timer_running) {
1911                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1912                         vcpu->arch.timer_running = 0;
1913                 }
1914                 cpu += vcpu->arch.ptid;
1915                 vcpu->cpu = mvc->pcpu;
1916                 vcpu->arch.thread_cpu = cpu;
1917         }
1918         tpaca = &paca[cpu];
1919         tpaca->kvm_hstate.kvm_vcpu = vcpu;
1920         tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
1921         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
1922         smp_wmb();
1923         tpaca->kvm_hstate.kvm_vcore = mvc;
1924         if (cpu != smp_processor_id())
1925                 kvmppc_ipi_thread(cpu);
1926 }
1927
1928 static void kvmppc_wait_for_nap(void)
1929 {
1930         int cpu = smp_processor_id();
1931         int i, loops;
1932
1933         for (loops = 0; loops < 1000000; ++loops) {
1934                 /*
1935                  * Check if all threads are finished.
1936                  * We set the vcore pointer when starting a thread
1937                  * and the thread clears it when finished, so we look
1938                  * for any threads that still have a non-NULL vcore ptr.
1939                  */
1940                 for (i = 1; i < threads_per_subcore; ++i)
1941                         if (paca[cpu + i].kvm_hstate.kvm_vcore)
1942                                 break;
1943                 if (i == threads_per_subcore) {
1944                         HMT_medium();
1945                         return;
1946                 }
1947                 HMT_low();
1948         }
1949         HMT_medium();
1950         for (i = 1; i < threads_per_subcore; ++i)
1951                 if (paca[cpu + i].kvm_hstate.kvm_vcore)
1952                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
1953 }
1954
1955 /*
1956  * Check that we are on thread 0 and that any other threads in
1957  * this core are off-line.  Then grab the threads so they can't
1958  * enter the kernel.
1959  */
1960 static int on_primary_thread(void)
1961 {
1962         int cpu = smp_processor_id();
1963         int thr;
1964
1965         /* Are we on a primary subcore? */
1966         if (cpu_thread_in_subcore(cpu))
1967                 return 0;
1968
1969         thr = 0;
1970         while (++thr < threads_per_subcore)
1971                 if (cpu_online(cpu + thr))
1972                         return 0;
1973
1974         /* Grab all hw threads so they can't go into the kernel */
1975         for (thr = 1; thr < threads_per_subcore; ++thr) {
1976                 if (kvmppc_grab_hwthread(cpu + thr)) {
1977                         /* Couldn't grab one; let the others go */
1978                         do {
1979                                 kvmppc_release_hwthread(cpu + thr);
1980                         } while (--thr > 0);
1981                         return 0;
1982                 }
1983         }
1984         return 1;
1985 }
1986
1987 /*
1988  * A list of virtual cores for each physical CPU.
1989  * These are vcores that could run but their runner VCPU tasks are
1990  * (or may be) preempted.
1991  */
1992 struct preempted_vcore_list {
1993         struct list_head        list;
1994         spinlock_t              lock;
1995 };
1996
1997 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
1998
1999 static void init_vcore_lists(void)
2000 {
2001         int cpu;
2002
2003         for_each_possible_cpu(cpu) {
2004                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2005                 spin_lock_init(&lp->lock);
2006                 INIT_LIST_HEAD(&lp->list);
2007         }
2008 }
2009
2010 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2011 {
2012         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2013
2014         vc->vcore_state = VCORE_PREEMPT;
2015         vc->pcpu = smp_processor_id();
2016         if (vc->num_threads < threads_per_subcore) {
2017                 spin_lock(&lp->lock);
2018                 list_add_tail(&vc->preempt_list, &lp->list);
2019                 spin_unlock(&lp->lock);
2020         }
2021
2022         /* Start accumulating stolen time */
2023         kvmppc_core_start_stolen(vc);
2024 }
2025
2026 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2027 {
2028         struct preempted_vcore_list *lp;
2029
2030         kvmppc_core_end_stolen(vc);
2031         if (!list_empty(&vc->preempt_list)) {
2032                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2033                 spin_lock(&lp->lock);
2034                 list_del_init(&vc->preempt_list);
2035                 spin_unlock(&lp->lock);
2036         }
2037         vc->vcore_state = VCORE_INACTIVE;
2038 }
2039
2040 /*
2041  * This stores information about the virtual cores currently
2042  * assigned to a physical core.
2043  */
2044 struct core_info {
2045         int             n_subcores;
2046         int             max_subcore_threads;
2047         int             total_threads;
2048         int             subcore_threads[MAX_SUBCORES];
2049         struct kvm      *subcore_vm[MAX_SUBCORES];
2050         struct list_head vcs[MAX_SUBCORES];
2051 };
2052
2053 /*
2054  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2055  * respectively in 2-way micro-threading (split-core) mode.
2056  */
2057 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2058
2059 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2060 {
2061         int sub;
2062
2063         memset(cip, 0, sizeof(*cip));
2064         cip->n_subcores = 1;
2065         cip->max_subcore_threads = vc->num_threads;
2066         cip->total_threads = vc->num_threads;
2067         cip->subcore_threads[0] = vc->num_threads;
2068         cip->subcore_vm[0] = vc->kvm;
2069         for (sub = 0; sub < MAX_SUBCORES; ++sub)
2070                 INIT_LIST_HEAD(&cip->vcs[sub]);
2071         list_add_tail(&vc->preempt_list, &cip->vcs[0]);
2072 }
2073
2074 static bool subcore_config_ok(int n_subcores, int n_threads)
2075 {
2076         /* Can only dynamically split if unsplit to begin with */
2077         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2078                 return false;
2079         if (n_subcores > MAX_SUBCORES)
2080                 return false;
2081         if (n_subcores > 1) {
2082                 if (!(dynamic_mt_modes & 2))
2083                         n_subcores = 4;
2084                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2085                         return false;
2086         }
2087
2088         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2089 }
2090
2091 static void init_master_vcore(struct kvmppc_vcore *vc)
2092 {
2093         vc->master_vcore = vc;
2094         vc->entry_exit_map = 0;
2095         vc->in_guest = 0;
2096         vc->napping_threads = 0;
2097         vc->conferring_threads = 0;
2098 }
2099
2100 /*
2101  * See if the existing subcores can be split into 3 (or fewer) subcores
2102  * of at most two threads each, so we can fit in another vcore.  This
2103  * assumes there are at most two subcores and at most 6 threads in total.
2104  */
2105 static bool can_split_piggybacked_subcores(struct core_info *cip)
2106 {
2107         int sub, new_sub;
2108         int large_sub = -1;
2109         int thr;
2110         int n_subcores = cip->n_subcores;
2111         struct kvmppc_vcore *vc, *vcnext;
2112         struct kvmppc_vcore *master_vc = NULL;
2113
2114         for (sub = 0; sub < cip->n_subcores; ++sub) {
2115                 if (cip->subcore_threads[sub] <= 2)
2116                         continue;
2117                 if (large_sub >= 0)
2118                         return false;
2119                 large_sub = sub;
2120                 vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2121                                       preempt_list);
2122                 if (vc->num_threads > 2)
2123                         return false;
2124                 n_subcores += (cip->subcore_threads[sub] - 1) >> 1;
2125         }
2126         if (large_sub < 0 || !subcore_config_ok(n_subcores + 1, 2))
2127                 return false;
2128
2129         /*
2130          * Seems feasible, so go through and move vcores to new subcores.
2131          * Note that when we have two or more vcores in one subcore,
2132          * all those vcores must have only one thread each.
2133          */
2134         new_sub = cip->n_subcores;
2135         thr = 0;
2136         sub = large_sub;
2137         list_for_each_entry_safe(vc, vcnext, &cip->vcs[sub], preempt_list) {
2138                 if (thr >= 2) {
2139                         list_del(&vc->preempt_list);
2140                         list_add_tail(&vc->preempt_list, &cip->vcs[new_sub]);
2141                         /* vc->num_threads must be 1 */
2142                         if (++cip->subcore_threads[new_sub] == 1) {
2143                                 cip->subcore_vm[new_sub] = vc->kvm;
2144                                 init_master_vcore(vc);
2145                                 master_vc = vc;
2146                                 ++cip->n_subcores;
2147                         } else {
2148                                 vc->master_vcore = master_vc;
2149                                 ++new_sub;
2150                         }
2151                 }
2152                 thr += vc->num_threads;
2153         }
2154         cip->subcore_threads[large_sub] = 2;
2155         cip->max_subcore_threads = 2;
2156
2157         return true;
2158 }
2159
2160 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2161 {
2162         int n_threads = vc->num_threads;
2163         int sub;
2164
2165         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2166                 return false;
2167
2168         if (n_threads < cip->max_subcore_threads)
2169                 n_threads = cip->max_subcore_threads;
2170         if (subcore_config_ok(cip->n_subcores + 1, n_threads)) {
2171                 cip->max_subcore_threads = n_threads;
2172         } else if (cip->n_subcores <= 2 && cip->total_threads <= 6 &&
2173                    vc->num_threads <= 2) {
2174                 /*
2175                  * We may be able to fit another subcore in by
2176                  * splitting an existing subcore with 3 or 4
2177                  * threads into two 2-thread subcores, or one
2178                  * with 5 or 6 threads into three subcores.
2179                  * We can only do this if those subcores have
2180                  * piggybacked virtual cores.
2181                  */
2182                 if (!can_split_piggybacked_subcores(cip))
2183                         return false;
2184         } else {
2185                 return false;
2186         }
2187
2188         sub = cip->n_subcores;
2189         ++cip->n_subcores;
2190         cip->total_threads += vc->num_threads;
2191         cip->subcore_threads[sub] = vc->num_threads;
2192         cip->subcore_vm[sub] = vc->kvm;
2193         init_master_vcore(vc);
2194         list_del(&vc->preempt_list);
2195         list_add_tail(&vc->preempt_list, &cip->vcs[sub]);
2196
2197         return true;
2198 }
2199
2200 static bool can_piggyback_subcore(struct kvmppc_vcore *pvc,
2201                                   struct core_info *cip, int sub)
2202 {
2203         struct kvmppc_vcore *vc;
2204         int n_thr;
2205
2206         vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2207                               preempt_list);
2208
2209         /* require same VM and same per-core reg values */
2210         if (pvc->kvm != vc->kvm ||
2211             pvc->tb_offset != vc->tb_offset ||
2212             pvc->pcr != vc->pcr ||
2213             pvc->lpcr != vc->lpcr)
2214                 return false;
2215
2216         /* P8 guest with > 1 thread per core would see wrong TIR value */
2217         if (cpu_has_feature(CPU_FTR_ARCH_207S) &&
2218             (vc->num_threads > 1 || pvc->num_threads > 1))
2219                 return false;
2220
2221         n_thr = cip->subcore_threads[sub] + pvc->num_threads;
2222         if (n_thr > cip->max_subcore_threads) {
2223                 if (!subcore_config_ok(cip->n_subcores, n_thr))
2224                         return false;
2225                 cip->max_subcore_threads = n_thr;
2226         }
2227
2228         cip->total_threads += pvc->num_threads;
2229         cip->subcore_threads[sub] = n_thr;
2230         pvc->master_vcore = vc;
2231         list_del(&pvc->preempt_list);
2232         list_add_tail(&pvc->preempt_list, &cip->vcs[sub]);
2233
2234         return true;
2235 }
2236
2237 /*
2238  * Work out whether it is possible to piggyback the execution of
2239  * vcore *pvc onto the execution of the other vcores described in *cip.
2240  */
2241 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2242                           int target_threads)
2243 {
2244         int sub;
2245
2246         if (cip->total_threads + pvc->num_threads > target_threads)
2247                 return false;
2248         for (sub = 0; sub < cip->n_subcores; ++sub)
2249                 if (cip->subcore_threads[sub] &&
2250                     can_piggyback_subcore(pvc, cip, sub))
2251                         return true;
2252
2253         if (can_dynamic_split(pvc, cip))
2254                 return true;
2255
2256         return false;
2257 }
2258
2259 static void prepare_threads(struct kvmppc_vcore *vc)
2260 {
2261         int i;
2262         struct kvm_vcpu *vcpu;
2263
2264         for_each_runnable_thread(i, vcpu, vc) {
2265                 if (signal_pending(vcpu->arch.run_task))
2266                         vcpu->arch.ret = -EINTR;
2267                 else if (vcpu->arch.vpa.update_pending ||
2268                          vcpu->arch.slb_shadow.update_pending ||
2269                          vcpu->arch.dtl.update_pending)
2270                         vcpu->arch.ret = RESUME_GUEST;
2271                 else
2272                         continue;
2273                 kvmppc_remove_runnable(vc, vcpu);
2274                 wake_up(&vcpu->arch.cpu_run);
2275         }
2276 }
2277
2278 static void collect_piggybacks(struct core_info *cip, int target_threads)
2279 {
2280         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2281         struct kvmppc_vcore *pvc, *vcnext;
2282
2283         spin_lock(&lp->lock);
2284         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2285                 if (!spin_trylock(&pvc->lock))
2286                         continue;
2287                 prepare_threads(pvc);
2288                 if (!pvc->n_runnable) {
2289                         list_del_init(&pvc->preempt_list);
2290                         if (pvc->runner == NULL) {
2291                                 pvc->vcore_state = VCORE_INACTIVE;
2292                                 kvmppc_core_end_stolen(pvc);
2293                         }
2294                         spin_unlock(&pvc->lock);
2295                         continue;
2296                 }
2297                 if (!can_piggyback(pvc, cip, target_threads)) {
2298                         spin_unlock(&pvc->lock);
2299                         continue;
2300                 }
2301                 kvmppc_core_end_stolen(pvc);
2302                 pvc->vcore_state = VCORE_PIGGYBACK;
2303                 if (cip->total_threads >= target_threads)
2304                         break;
2305         }
2306         spin_unlock(&lp->lock);
2307 }
2308
2309 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2310 {
2311         int still_running = 0, i;
2312         u64 now;
2313         long ret;
2314         struct kvm_vcpu *vcpu;
2315
2316         spin_lock(&vc->lock);
2317         now = get_tb();
2318         for_each_runnable_thread(i, vcpu, vc) {
2319                 /* cancel pending dec exception if dec is positive */
2320                 if (now < vcpu->arch.dec_expires &&
2321                     kvmppc_core_pending_dec(vcpu))
2322                         kvmppc_core_dequeue_dec(vcpu);
2323
2324                 trace_kvm_guest_exit(vcpu);
2325
2326                 ret = RESUME_GUEST;
2327                 if (vcpu->arch.trap)
2328                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2329                                                     vcpu->arch.run_task);
2330
2331                 vcpu->arch.ret = ret;
2332                 vcpu->arch.trap = 0;
2333
2334                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2335                         if (vcpu->arch.pending_exceptions)
2336                                 kvmppc_core_prepare_to_enter(vcpu);
2337                         if (vcpu->arch.ceded)
2338                                 kvmppc_set_timer(vcpu);
2339                         else
2340                                 ++still_running;
2341                 } else {
2342                         kvmppc_remove_runnable(vc, vcpu);
2343                         wake_up(&vcpu->arch.cpu_run);
2344                 }
2345         }
2346         list_del_init(&vc->preempt_list);
2347         if (!is_master) {
2348                 if (still_running > 0) {
2349                         kvmppc_vcore_preempt(vc);
2350                 } else if (vc->runner) {
2351                         vc->vcore_state = VCORE_PREEMPT;
2352                         kvmppc_core_start_stolen(vc);
2353                 } else {
2354                         vc->vcore_state = VCORE_INACTIVE;
2355                 }
2356                 if (vc->n_runnable > 0 && vc->runner == NULL) {
2357                         /* make sure there's a candidate runner awake */
2358                         i = -1;
2359                         vcpu = next_runnable_thread(vc, &i);
2360                         wake_up(&vcpu->arch.cpu_run);
2361                 }
2362         }
2363         spin_unlock(&vc->lock);
2364 }
2365
2366 /*
2367  * Clear core from the list of active host cores as we are about to
2368  * enter the guest. Only do this if it is the primary thread of the
2369  * core (not if a subcore) that is entering the guest.
2370  */
2371 static inline void kvmppc_clear_host_core(int cpu)
2372 {
2373         int core;
2374
2375         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2376                 return;
2377         /*
2378          * Memory barrier can be omitted here as we will do a smp_wmb()
2379          * later in kvmppc_start_thread and we need ensure that state is
2380          * visible to other CPUs only after we enter guest.
2381          */
2382         core = cpu >> threads_shift;
2383         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2384 }
2385
2386 /*
2387  * Advertise this core as an active host core since we exited the guest
2388  * Only need to do this if it is the primary thread of the core that is
2389  * exiting.
2390  */
2391 static inline void kvmppc_set_host_core(int cpu)
2392 {
2393         int core;
2394
2395         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2396                 return;
2397
2398         /*
2399          * Memory barrier can be omitted here because we do a spin_unlock
2400          * immediately after this which provides the memory barrier.
2401          */
2402         core = cpu >> threads_shift;
2403         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2404 }
2405
2406 /*
2407  * Run a set of guest threads on a physical core.
2408  * Called with vc->lock held.
2409  */
2410 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2411 {
2412         struct kvm_vcpu *vcpu;
2413         int i;
2414         int srcu_idx;
2415         struct core_info core_info;
2416         struct kvmppc_vcore *pvc, *vcnext;
2417         struct kvm_split_mode split_info, *sip;
2418         int split, subcore_size, active;
2419         int sub;
2420         bool thr0_done;
2421         unsigned long cmd_bit, stat_bit;
2422         int pcpu, thr;
2423         int target_threads;
2424
2425         /*
2426          * Remove from the list any threads that have a signal pending
2427          * or need a VPA update done
2428          */
2429         prepare_threads(vc);
2430
2431         /* if the runner is no longer runnable, let the caller pick a new one */
2432         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2433                 return;
2434
2435         /*
2436          * Initialize *vc.
2437          */
2438         init_master_vcore(vc);
2439         vc->preempt_tb = TB_NIL;
2440
2441         /*
2442          * Make sure we are running on primary threads, and that secondary
2443          * threads are offline.  Also check if the number of threads in this
2444          * guest are greater than the current system threads per guest.
2445          */
2446         if ((threads_per_core > 1) &&
2447             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2448                 for_each_runnable_thread(i, vcpu, vc) {
2449                         vcpu->arch.ret = -EBUSY;
2450                         kvmppc_remove_runnable(vc, vcpu);
2451                         wake_up(&vcpu->arch.cpu_run);
2452                 }
2453                 goto out;
2454         }
2455
2456         /*
2457          * See if we could run any other vcores on the physical core
2458          * along with this one.
2459          */
2460         init_core_info(&core_info, vc);
2461         pcpu = smp_processor_id();
2462         target_threads = threads_per_subcore;
2463         if (target_smt_mode && target_smt_mode < target_threads)
2464                 target_threads = target_smt_mode;
2465         if (vc->num_threads < target_threads)
2466                 collect_piggybacks(&core_info, target_threads);
2467
2468         /* Decide on micro-threading (split-core) mode */
2469         subcore_size = threads_per_subcore;
2470         cmd_bit = stat_bit = 0;
2471         split = core_info.n_subcores;
2472         sip = NULL;
2473         if (split > 1) {
2474                 /* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2475                 if (split == 2 && (dynamic_mt_modes & 2)) {
2476                         cmd_bit = HID0_POWER8_1TO2LPAR;
2477                         stat_bit = HID0_POWER8_2LPARMODE;
2478                 } else {
2479                         split = 4;
2480                         cmd_bit = HID0_POWER8_1TO4LPAR;
2481                         stat_bit = HID0_POWER8_4LPARMODE;
2482                 }
2483                 subcore_size = MAX_SMT_THREADS / split;
2484                 sip = &split_info;
2485                 memset(&split_info, 0, sizeof(split_info));
2486                 split_info.rpr = mfspr(SPRN_RPR);
2487                 split_info.pmmar = mfspr(SPRN_PMMAR);
2488                 split_info.ldbar = mfspr(SPRN_LDBAR);
2489                 split_info.subcore_size = subcore_size;
2490                 for (sub = 0; sub < core_info.n_subcores; ++sub)
2491                         split_info.master_vcs[sub] =
2492                                 list_first_entry(&core_info.vcs[sub],
2493                                         struct kvmppc_vcore, preempt_list);
2494                 /* order writes to split_info before kvm_split_mode pointer */
2495                 smp_wmb();
2496         }
2497         pcpu = smp_processor_id();
2498         for (thr = 0; thr < threads_per_subcore; ++thr)
2499                 paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2500
2501         /* Initiate micro-threading (split-core) if required */
2502         if (cmd_bit) {
2503                 unsigned long hid0 = mfspr(SPRN_HID0);
2504
2505                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2506                 mb();
2507                 mtspr(SPRN_HID0, hid0);
2508                 isync();
2509                 for (;;) {
2510                         hid0 = mfspr(SPRN_HID0);
2511                         if (hid0 & stat_bit)
2512                                 break;
2513                         cpu_relax();
2514                 }
2515         }
2516
2517         kvmppc_clear_host_core(pcpu);
2518
2519         /* Start all the threads */
2520         active = 0;
2521         for (sub = 0; sub < core_info.n_subcores; ++sub) {
2522                 thr = subcore_thread_map[sub];
2523                 thr0_done = false;
2524                 active |= 1 << thr;
2525                 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2526                         pvc->pcpu = pcpu + thr;
2527                         for_each_runnable_thread(i, vcpu, pvc) {
2528                                 kvmppc_start_thread(vcpu, pvc);
2529                                 kvmppc_create_dtl_entry(vcpu, pvc);
2530                                 trace_kvm_guest_enter(vcpu);
2531                                 if (!vcpu->arch.ptid)
2532                                         thr0_done = true;
2533                                 active |= 1 << (thr + vcpu->arch.ptid);
2534                         }
2535                         /*
2536                          * We need to start the first thread of each subcore
2537                          * even if it doesn't have a vcpu.
2538                          */
2539                         if (pvc->master_vcore == pvc && !thr0_done)
2540                                 kvmppc_start_thread(NULL, pvc);
2541                         thr += pvc->num_threads;
2542                 }
2543         }
2544
2545         /*
2546          * Ensure that split_info.do_nap is set after setting
2547          * the vcore pointer in the PACA of the secondaries.
2548          */
2549         smp_mb();
2550         if (cmd_bit)
2551                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
2552
2553         /*
2554          * When doing micro-threading, poke the inactive threads as well.
2555          * This gets them to the nap instruction after kvm_do_nap,
2556          * which reduces the time taken to unsplit later.
2557          */
2558         if (split > 1)
2559                 for (thr = 1; thr < threads_per_subcore; ++thr)
2560                         if (!(active & (1 << thr)))
2561                                 kvmppc_ipi_thread(pcpu + thr);
2562
2563         vc->vcore_state = VCORE_RUNNING;
2564         preempt_disable();
2565
2566         trace_kvmppc_run_core(vc, 0);
2567
2568         for (sub = 0; sub < core_info.n_subcores; ++sub)
2569                 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2570                         spin_unlock(&pvc->lock);
2571
2572         guest_enter();
2573
2574         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2575
2576         __kvmppc_vcore_entry();
2577
2578         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2579
2580         spin_lock(&vc->lock);
2581         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2582         vc->vcore_state = VCORE_EXITING;
2583
2584         /* wait for secondary threads to finish writing their state to memory */
2585         kvmppc_wait_for_nap();
2586
2587         /* Return to whole-core mode if we split the core earlier */
2588         if (split > 1) {
2589                 unsigned long hid0 = mfspr(SPRN_HID0);
2590                 unsigned long loops = 0;
2591
2592                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2593                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2594                 mb();
2595                 mtspr(SPRN_HID0, hid0);
2596                 isync();
2597                 for (;;) {
2598                         hid0 = mfspr(SPRN_HID0);
2599                         if (!(hid0 & stat_bit))
2600                                 break;
2601                         cpu_relax();
2602                         ++loops;
2603                 }
2604                 split_info.do_nap = 0;
2605         }
2606
2607         /* Let secondaries go back to the offline loop */
2608         for (i = 0; i < threads_per_subcore; ++i) {
2609                 kvmppc_release_hwthread(pcpu + i);
2610                 if (sip && sip->napped[i])
2611                         kvmppc_ipi_thread(pcpu + i);
2612         }
2613
2614         kvmppc_set_host_core(pcpu);
2615
2616         spin_unlock(&vc->lock);
2617
2618         /* make sure updates to secondary vcpu structs are visible now */
2619         smp_mb();
2620         guest_exit();
2621
2622         for (sub = 0; sub < core_info.n_subcores; ++sub)
2623                 list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2624                                          preempt_list)
2625                         post_guest_process(pvc, pvc == vc);
2626
2627         spin_lock(&vc->lock);
2628         preempt_enable();
2629
2630  out:
2631         vc->vcore_state = VCORE_INACTIVE;
2632         trace_kvmppc_run_core(vc, 1);
2633 }
2634
2635 /*
2636  * Wait for some other vcpu thread to execute us, and
2637  * wake us up when we need to handle something in the host.
2638  */
2639 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2640                                  struct kvm_vcpu *vcpu, int wait_state)
2641 {
2642         DEFINE_WAIT(wait);
2643
2644         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2645         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2646                 spin_unlock(&vc->lock);
2647                 schedule();
2648                 spin_lock(&vc->lock);
2649         }
2650         finish_wait(&vcpu->arch.cpu_run, &wait);
2651 }
2652
2653 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
2654 {
2655         /* 10us base */
2656         if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
2657                 vc->halt_poll_ns = 10000;
2658         else
2659                 vc->halt_poll_ns *= halt_poll_ns_grow;
2660
2661         if (vc->halt_poll_ns > halt_poll_max_ns)
2662                 vc->halt_poll_ns = halt_poll_max_ns;
2663 }
2664
2665 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
2666 {
2667         if (halt_poll_ns_shrink == 0)
2668                 vc->halt_poll_ns = 0;
2669         else
2670                 vc->halt_poll_ns /= halt_poll_ns_shrink;
2671 }
2672
2673 /* Check to see if any of the runnable vcpus on the vcore have pending
2674  * exceptions or are no longer ceded
2675  */
2676 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
2677 {
2678         struct kvm_vcpu *vcpu;
2679         int i;
2680
2681         for_each_runnable_thread(i, vcpu, vc) {
2682                 if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded)
2683                         return 1;
2684         }
2685
2686         return 0;
2687 }
2688
2689 /*
2690  * All the vcpus in this vcore are idle, so wait for a decrementer
2691  * or external interrupt to one of the vcpus.  vc->lock is held.
2692  */
2693 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2694 {
2695         ktime_t cur, start_poll, start_wait;
2696         int do_sleep = 1;
2697         u64 block_ns;
2698         DECLARE_SWAITQUEUE(wait);
2699
2700         /* Poll for pending exceptions and ceded state */
2701         cur = start_poll = ktime_get();
2702         if (vc->halt_poll_ns) {
2703                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
2704                 ++vc->runner->stat.halt_attempted_poll;
2705
2706                 vc->vcore_state = VCORE_POLLING;
2707                 spin_unlock(&vc->lock);
2708
2709                 do {
2710                         if (kvmppc_vcore_check_block(vc)) {
2711                                 do_sleep = 0;
2712                                 break;
2713                         }
2714                         cur = ktime_get();
2715                 } while (single_task_running() && ktime_before(cur, stop));
2716
2717                 spin_lock(&vc->lock);
2718                 vc->vcore_state = VCORE_INACTIVE;
2719
2720                 if (!do_sleep) {
2721                         ++vc->runner->stat.halt_successful_poll;
2722                         goto out;
2723                 }
2724         }
2725
2726         prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2727
2728         if (kvmppc_vcore_check_block(vc)) {
2729                 finish_swait(&vc->wq, &wait);
2730                 do_sleep = 0;
2731                 /* If we polled, count this as a successful poll */
2732                 if (vc->halt_poll_ns)
2733                         ++vc->runner->stat.halt_successful_poll;
2734                 goto out;
2735         }
2736
2737         start_wait = ktime_get();
2738
2739         vc->vcore_state = VCORE_SLEEPING;
2740         trace_kvmppc_vcore_blocked(vc, 0);
2741         spin_unlock(&vc->lock);
2742         schedule();
2743         finish_swait(&vc->wq, &wait);
2744         spin_lock(&vc->lock);
2745         vc->vcore_state = VCORE_INACTIVE;
2746         trace_kvmppc_vcore_blocked(vc, 1);
2747         ++vc->runner->stat.halt_successful_wait;
2748
2749         cur = ktime_get();
2750
2751 out:
2752         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
2753
2754         /* Attribute wait time */
2755         if (do_sleep) {
2756                 vc->runner->stat.halt_wait_ns +=
2757                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
2758                 /* Attribute failed poll time */
2759                 if (vc->halt_poll_ns)
2760                         vc->runner->stat.halt_poll_fail_ns +=
2761                                 ktime_to_ns(start_wait) -
2762                                 ktime_to_ns(start_poll);
2763         } else {
2764                 /* Attribute successful poll time */
2765                 if (vc->halt_poll_ns)
2766                         vc->runner->stat.halt_poll_success_ns +=
2767                                 ktime_to_ns(cur) -
2768                                 ktime_to_ns(start_poll);
2769         }
2770
2771         /* Adjust poll time */
2772         if (halt_poll_max_ns) {
2773                 if (block_ns <= vc->halt_poll_ns)
2774                         ;
2775                 /* We slept and blocked for longer than the max halt time */
2776                 else if (vc->halt_poll_ns && block_ns > halt_poll_max_ns)
2777                         shrink_halt_poll_ns(vc);
2778                 /* We slept and our poll time is too small */
2779                 else if (vc->halt_poll_ns < halt_poll_max_ns &&
2780                                 block_ns < halt_poll_max_ns)
2781                         grow_halt_poll_ns(vc);
2782         } else
2783                 vc->halt_poll_ns = 0;
2784
2785         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
2786 }
2787
2788 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2789 {
2790         int n_ceded, i;
2791         struct kvmppc_vcore *vc;
2792         struct kvm_vcpu *v;
2793
2794         trace_kvmppc_run_vcpu_enter(vcpu);
2795
2796         kvm_run->exit_reason = 0;
2797         vcpu->arch.ret = RESUME_GUEST;
2798         vcpu->arch.trap = 0;
2799         kvmppc_update_vpas(vcpu);
2800
2801         /*
2802          * Synchronize with other threads in this virtual core
2803          */
2804         vc = vcpu->arch.vcore;
2805         spin_lock(&vc->lock);
2806         vcpu->arch.ceded = 0;
2807         vcpu->arch.run_task = current;
2808         vcpu->arch.kvm_run = kvm_run;
2809         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2810         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2811         vcpu->arch.busy_preempt = TB_NIL;
2812         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
2813         ++vc->n_runnable;
2814
2815         /*
2816          * This happens the first time this is called for a vcpu.
2817          * If the vcore is already running, we may be able to start
2818          * this thread straight away and have it join in.
2819          */
2820         if (!signal_pending(current)) {
2821                 if (vc->vcore_state == VCORE_PIGGYBACK) {
2822                         struct kvmppc_vcore *mvc = vc->master_vcore;
2823                         if (spin_trylock(&mvc->lock)) {
2824                                 if (mvc->vcore_state == VCORE_RUNNING &&
2825                                     !VCORE_IS_EXITING(mvc)) {
2826                                         kvmppc_create_dtl_entry(vcpu, vc);
2827                                         kvmppc_start_thread(vcpu, vc);
2828                                         trace_kvm_guest_enter(vcpu);
2829                                 }
2830                                 spin_unlock(&mvc->lock);
2831                         }
2832                 } else if (vc->vcore_state == VCORE_RUNNING &&
2833                            !VCORE_IS_EXITING(vc)) {
2834                         kvmppc_create_dtl_entry(vcpu, vc);
2835                         kvmppc_start_thread(vcpu, vc);
2836                         trace_kvm_guest_enter(vcpu);
2837                 } else if (vc->vcore_state == VCORE_SLEEPING) {
2838                         swake_up(&vc->wq);
2839                 }
2840
2841         }
2842
2843         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2844                !signal_pending(current)) {
2845                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2846                         kvmppc_vcore_end_preempt(vc);
2847
2848                 if (vc->vcore_state != VCORE_INACTIVE) {
2849                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
2850                         continue;
2851                 }
2852                 for_each_runnable_thread(i, v, vc) {
2853                         kvmppc_core_prepare_to_enter(v);
2854                         if (signal_pending(v->arch.run_task)) {
2855                                 kvmppc_remove_runnable(vc, v);
2856                                 v->stat.signal_exits++;
2857                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2858                                 v->arch.ret = -EINTR;
2859                                 wake_up(&v->arch.cpu_run);
2860                         }
2861                 }
2862                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2863                         break;
2864                 n_ceded = 0;
2865                 for_each_runnable_thread(i, v, vc) {
2866                         if (!v->arch.pending_exceptions)
2867                                 n_ceded += v->arch.ceded;
2868                         else
2869                                 v->arch.ceded = 0;
2870                 }
2871                 vc->runner = vcpu;
2872                 if (n_ceded == vc->n_runnable) {
2873                         kvmppc_vcore_blocked(vc);
2874                 } else if (need_resched()) {
2875                         kvmppc_vcore_preempt(vc);
2876                         /* Let something else run */
2877                         cond_resched_lock(&vc->lock);
2878                         if (vc->vcore_state == VCORE_PREEMPT)
2879                                 kvmppc_vcore_end_preempt(vc);
2880                 } else {
2881                         kvmppc_run_core(vc);
2882                 }
2883                 vc->runner = NULL;
2884         }
2885
2886         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2887                (vc->vcore_state == VCORE_RUNNING ||
2888                 vc->vcore_state == VCORE_EXITING ||
2889                 vc->vcore_state == VCORE_PIGGYBACK))
2890                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
2891
2892         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2893                 kvmppc_vcore_end_preempt(vc);
2894
2895         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2896                 kvmppc_remove_runnable(vc, vcpu);
2897                 vcpu->stat.signal_exits++;
2898                 kvm_run->exit_reason = KVM_EXIT_INTR;
2899                 vcpu->arch.ret = -EINTR;
2900         }
2901
2902         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2903                 /* Wake up some vcpu to run the core */
2904                 i = -1;
2905                 v = next_runnable_thread(vc, &i);
2906                 wake_up(&v->arch.cpu_run);
2907         }
2908
2909         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2910         spin_unlock(&vc->lock);
2911         return vcpu->arch.ret;
2912 }
2913
2914 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2915 {
2916         int r;
2917         int srcu_idx;
2918
2919         if (!vcpu->arch.sane) {
2920                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2921                 return -EINVAL;
2922         }
2923
2924         kvmppc_core_prepare_to_enter(vcpu);
2925
2926         /* No need to go into the guest when all we'll do is come back out */
2927         if (signal_pending(current)) {
2928                 run->exit_reason = KVM_EXIT_INTR;
2929                 return -EINTR;
2930         }
2931
2932         atomic_inc(&vcpu->kvm->arch.vcpus_running);
2933         /* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2934         smp_mb();
2935
2936         /* On the first time here, set up HTAB and VRMA */
2937         if (!vcpu->kvm->arch.hpte_setup_done) {
2938                 r = kvmppc_hv_setup_htab_rma(vcpu);
2939                 if (r)
2940                         goto out;
2941         }
2942
2943         flush_all_to_thread(current);
2944
2945         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2946         vcpu->arch.pgdir = current->mm->pgd;
2947         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2948
2949         do {
2950                 r = kvmppc_run_vcpu(run, vcpu);
2951
2952                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2953                     !(vcpu->arch.shregs.msr & MSR_PR)) {
2954                         trace_kvm_hcall_enter(vcpu);
2955                         r = kvmppc_pseries_do_hcall(vcpu);
2956                         trace_kvm_hcall_exit(vcpu, r);
2957                         kvmppc_core_prepare_to_enter(vcpu);
2958                 } else if (r == RESUME_PAGE_FAULT) {
2959                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2960                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
2961                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2962                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2963                 } else if (r == RESUME_PASSTHROUGH)
2964                         r = kvmppc_xics_rm_complete(vcpu, 0);
2965         } while (is_kvmppc_resume_guest(r));
2966
2967  out:
2968         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2969         atomic_dec(&vcpu->kvm->arch.vcpus_running);
2970         return r;
2971 }
2972
2973 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2974                                      int linux_psize)
2975 {
2976         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2977
2978         if (!def->shift)
2979                 return;
2980         (*sps)->page_shift = def->shift;
2981         (*sps)->slb_enc = def->sllp;
2982         (*sps)->enc[0].page_shift = def->shift;
2983         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
2984         /*
2985          * Add 16MB MPSS support if host supports it
2986          */
2987         if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2988                 (*sps)->enc[1].page_shift = 24;
2989                 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2990         }
2991         (*sps)++;
2992 }
2993
2994 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2995                                          struct kvm_ppc_smmu_info *info)
2996 {
2997         struct kvm_ppc_one_seg_page_size *sps;
2998
2999         info->flags = KVM_PPC_PAGE_SIZES_REAL;
3000         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
3001                 info->flags |= KVM_PPC_1T_SEGMENTS;
3002         info->slb_size = mmu_slb_size;
3003
3004         /* We only support these sizes for now, and no muti-size segments */
3005         sps = &info->sps[0];
3006         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
3007         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
3008         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
3009
3010         return 0;
3011 }
3012
3013 /*
3014  * Get (and clear) the dirty memory log for a memory slot.
3015  */
3016 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
3017                                          struct kvm_dirty_log *log)
3018 {
3019         struct kvm_memslots *slots;
3020         struct kvm_memory_slot *memslot;
3021         int r;
3022         unsigned long n;
3023
3024         mutex_lock(&kvm->slots_lock);
3025
3026         r = -EINVAL;
3027         if (log->slot >= KVM_USER_MEM_SLOTS)
3028                 goto out;
3029
3030         slots = kvm_memslots(kvm);
3031         memslot = id_to_memslot(slots, log->slot);
3032         r = -ENOENT;
3033         if (!memslot->dirty_bitmap)
3034                 goto out;
3035
3036         n = kvm_dirty_bitmap_bytes(memslot);
3037         memset(memslot->dirty_bitmap, 0, n);
3038
3039         r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
3040         if (r)
3041                 goto out;
3042
3043         r = -EFAULT;
3044         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
3045                 goto out;
3046
3047         r = 0;
3048 out:
3049         mutex_unlock(&kvm->slots_lock);
3050         return r;
3051 }
3052
3053 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
3054                                         struct kvm_memory_slot *dont)
3055 {
3056         if (!dont || free->arch.rmap != dont->arch.rmap) {
3057                 vfree(free->arch.rmap);
3058                 free->arch.rmap = NULL;
3059         }
3060 }
3061
3062 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3063                                          unsigned long npages)
3064 {
3065         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
3066         if (!slot->arch.rmap)
3067                 return -ENOMEM;
3068
3069         return 0;
3070 }
3071
3072 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3073                                         struct kvm_memory_slot *memslot,
3074                                         const struct kvm_userspace_memory_region *mem)
3075 {
3076         return 0;
3077 }
3078
3079 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3080                                 const struct kvm_userspace_memory_region *mem,
3081                                 const struct kvm_memory_slot *old,
3082                                 const struct kvm_memory_slot *new)
3083 {
3084         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3085         struct kvm_memslots *slots;
3086         struct kvm_memory_slot *memslot;
3087
3088         if (npages && old->npages) {
3089                 /*
3090                  * If modifying a memslot, reset all the rmap dirty bits.
3091                  * If this is a new memslot, we don't need to do anything
3092                  * since the rmap array starts out as all zeroes,
3093                  * i.e. no pages are dirty.
3094                  */
3095                 slots = kvm_memslots(kvm);
3096                 memslot = id_to_memslot(slots, mem->slot);
3097                 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
3098         }
3099 }
3100
3101 /*
3102  * Update LPCR values in kvm->arch and in vcores.
3103  * Caller must hold kvm->lock.
3104  */
3105 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3106 {
3107         long int i;
3108         u32 cores_done = 0;
3109
3110         if ((kvm->arch.lpcr & mask) == lpcr)
3111                 return;
3112
3113         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3114
3115         for (i = 0; i < KVM_MAX_VCORES; ++i) {
3116                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3117                 if (!vc)
3118                         continue;
3119                 spin_lock(&vc->lock);
3120                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3121                 spin_unlock(&vc->lock);
3122                 if (++cores_done >= kvm->arch.online_vcores)
3123                         break;
3124         }
3125 }
3126
3127 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3128 {
3129         return;
3130 }
3131
3132 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3133 {
3134         int err = 0;
3135         struct kvm *kvm = vcpu->kvm;
3136         unsigned long hva;
3137         struct kvm_memory_slot *memslot;
3138         struct vm_area_struct *vma;
3139         unsigned long lpcr = 0, senc;
3140         unsigned long psize, porder;
3141         int srcu_idx;
3142
3143         mutex_lock(&kvm->lock);
3144         if (kvm->arch.hpte_setup_done)
3145                 goto out;       /* another vcpu beat us to it */
3146
3147         /* Allocate hashed page table (if not done already) and reset it */
3148         if (!kvm->arch.hpt_virt) {
3149                 err = kvmppc_alloc_hpt(kvm, NULL);
3150                 if (err) {
3151                         pr_err("KVM: Couldn't alloc HPT\n");
3152                         goto out;
3153                 }
3154         }
3155
3156         /* Look up the memslot for guest physical address 0 */
3157         srcu_idx = srcu_read_lock(&kvm->srcu);
3158         memslot = gfn_to_memslot(kvm, 0);
3159
3160         /* We must have some memory at 0 by now */
3161         err = -EINVAL;
3162         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3163                 goto out_srcu;
3164
3165         /* Look up the VMA for the start of this memory slot */
3166         hva = memslot->userspace_addr;
3167         down_read(&current->mm->mmap_sem);
3168         vma = find_vma(current->mm, hva);
3169         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3170                 goto up_out;
3171
3172         psize = vma_kernel_pagesize(vma);
3173         porder = __ilog2(psize);
3174
3175         up_read(&current->mm->mmap_sem);
3176
3177         /* We can handle 4k, 64k or 16M pages in the VRMA */
3178         err = -EINVAL;
3179         if (!(psize == 0x1000 || psize == 0x10000 ||
3180               psize == 0x1000000))
3181                 goto out_srcu;
3182
3183         /* Update VRMASD field in the LPCR */
3184         senc = slb_pgsize_encoding(psize);
3185         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3186                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3187         /* the -4 is to account for senc values starting at 0x10 */
3188         lpcr = senc << (LPCR_VRMASD_SH - 4);
3189
3190         /* Create HPTEs in the hash page table for the VRMA */
3191         kvmppc_map_vrma(vcpu, memslot, porder);
3192
3193         kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3194
3195         /* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
3196         smp_wmb();
3197         kvm->arch.hpte_setup_done = 1;
3198         err = 0;
3199  out_srcu:
3200         srcu_read_unlock(&kvm->srcu, srcu_idx);
3201  out:
3202         mutex_unlock(&kvm->lock);
3203         return err;
3204
3205  up_out:
3206         up_read(&current->mm->mmap_sem);
3207         goto out_srcu;
3208 }
3209
3210 #ifdef CONFIG_KVM_XICS
3211 static int kvmppc_cpu_notify(struct notifier_block *self, unsigned long action,
3212                         void *hcpu)
3213 {
3214         unsigned long cpu = (long)hcpu;
3215
3216         switch (action) {
3217         case CPU_UP_PREPARE:
3218         case CPU_UP_PREPARE_FROZEN:
3219                 kvmppc_set_host_core(cpu);
3220                 break;
3221
3222 #ifdef CONFIG_HOTPLUG_CPU
3223         case CPU_DEAD:
3224         case CPU_DEAD_FROZEN:
3225         case CPU_UP_CANCELED:
3226         case CPU_UP_CANCELED_FROZEN:
3227                 kvmppc_clear_host_core(cpu);
3228                 break;
3229 #endif
3230         default:
3231                 break;
3232         }
3233
3234         return NOTIFY_OK;
3235 }
3236
3237 static struct notifier_block kvmppc_cpu_notifier = {
3238             .notifier_call = kvmppc_cpu_notify,
3239 };
3240
3241 /*
3242  * Allocate a per-core structure for managing state about which cores are
3243  * running in the host versus the guest and for exchanging data between
3244  * real mode KVM and CPU running in the host.
3245  * This is only done for the first VM.
3246  * The allocated structure stays even if all VMs have stopped.
3247  * It is only freed when the kvm-hv module is unloaded.
3248  * It's OK for this routine to fail, we just don't support host
3249  * core operations like redirecting H_IPI wakeups.
3250  */
3251 void kvmppc_alloc_host_rm_ops(void)
3252 {
3253         struct kvmppc_host_rm_ops *ops;
3254         unsigned long l_ops;
3255         int cpu, core;
3256         int size;
3257
3258         /* Not the first time here ? */
3259         if (kvmppc_host_rm_ops_hv != NULL)
3260                 return;
3261
3262         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3263         if (!ops)
3264                 return;
3265
3266         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3267         ops->rm_core = kzalloc(size, GFP_KERNEL);
3268
3269         if (!ops->rm_core) {
3270                 kfree(ops);
3271                 return;
3272         }
3273
3274         get_online_cpus();
3275
3276         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3277                 if (!cpu_online(cpu))
3278                         continue;
3279
3280                 core = cpu >> threads_shift;
3281                 ops->rm_core[core].rm_state.in_host = 1;
3282         }
3283
3284         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3285
3286         /*
3287          * Make the contents of the kvmppc_host_rm_ops structure visible
3288          * to other CPUs before we assign it to the global variable.
3289          * Do an atomic assignment (no locks used here), but if someone
3290          * beats us to it, just free our copy and return.
3291          */
3292         smp_wmb();
3293         l_ops = (unsigned long) ops;
3294
3295         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3296                 put_online_cpus();
3297                 kfree(ops->rm_core);
3298                 kfree(ops);
3299                 return;
3300         }
3301
3302         register_cpu_notifier(&kvmppc_cpu_notifier);
3303
3304         put_online_cpus();
3305 }
3306
3307 void kvmppc_free_host_rm_ops(void)
3308 {
3309         if (kvmppc_host_rm_ops_hv) {
3310                 unregister_cpu_notifier(&kvmppc_cpu_notifier);
3311                 kfree(kvmppc_host_rm_ops_hv->rm_core);
3312                 kfree(kvmppc_host_rm_ops_hv);
3313                 kvmppc_host_rm_ops_hv = NULL;
3314         }
3315 }
3316 #endif
3317
3318 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3319 {
3320         unsigned long lpcr, lpid;
3321         char buf[32];
3322
3323         /* Allocate the guest's logical partition ID */
3324
3325         lpid = kvmppc_alloc_lpid();
3326         if ((long)lpid < 0)
3327                 return -ENOMEM;
3328         kvm->arch.lpid = lpid;
3329
3330         kvmppc_alloc_host_rm_ops();
3331
3332         /*
3333          * Since we don't flush the TLB when tearing down a VM,
3334          * and this lpid might have previously been used,
3335          * make sure we flush on each core before running the new VM.
3336          */
3337         cpumask_setall(&kvm->arch.need_tlb_flush);
3338
3339         /* Start out with the default set of hcalls enabled */
3340         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3341                sizeof(kvm->arch.enabled_hcalls));
3342
3343         kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3344
3345         /* Init LPCR for virtual RMA mode */
3346         kvm->arch.host_lpid = mfspr(SPRN_LPID);
3347         kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3348         lpcr &= LPCR_PECE | LPCR_LPES;
3349         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3350                 LPCR_VPM0 | LPCR_VPM1;
3351         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3352                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3353         /* On POWER8 turn on online bit to enable PURR/SPURR */
3354         if (cpu_has_feature(CPU_FTR_ARCH_207S))
3355                 lpcr |= LPCR_ONL;
3356         kvm->arch.lpcr = lpcr;
3357
3358         /*
3359          * Track that we now have a HV mode VM active. This blocks secondary
3360          * CPU threads from coming online.
3361          */
3362         kvm_hv_vm_activated();
3363
3364         /*
3365          * Create a debugfs directory for the VM
3366          */
3367         snprintf(buf, sizeof(buf), "vm%d", current->pid);
3368         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3369         if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3370                 kvmppc_mmu_debugfs_init(kvm);
3371
3372         return 0;
3373 }
3374
3375 static void kvmppc_free_vcores(struct kvm *kvm)
3376 {
3377         long int i;
3378
3379         for (i = 0; i < KVM_MAX_VCORES; ++i)
3380                 kfree(kvm->arch.vcores[i]);
3381         kvm->arch.online_vcores = 0;
3382 }
3383
3384 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3385 {
3386         debugfs_remove_recursive(kvm->arch.debugfs_dir);
3387
3388         kvm_hv_vm_deactivated();
3389
3390         kvmppc_free_vcores(kvm);
3391
3392         kvmppc_free_hpt(kvm);
3393
3394         kvmppc_free_pimap(kvm);
3395 }
3396
3397 /* We don't need to emulate any privileged instructions or dcbz */
3398 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3399                                      unsigned int inst, int *advance)
3400 {
3401         return EMULATE_FAIL;
3402 }
3403
3404 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3405                                         ulong spr_val)
3406 {
3407         return EMULATE_FAIL;
3408 }
3409
3410 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3411                                         ulong *spr_val)
3412 {
3413         return EMULATE_FAIL;
3414 }
3415
3416 static int kvmppc_core_check_processor_compat_hv(void)
3417 {
3418         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3419             !cpu_has_feature(CPU_FTR_ARCH_206))
3420                 return -EIO;
3421         /*
3422          * Disable KVM for Power9, untill the required bits merged.
3423          */
3424         if (cpu_has_feature(CPU_FTR_ARCH_300))
3425                 return -EIO;
3426
3427         return 0;
3428 }
3429
3430 #ifdef CONFIG_KVM_XICS
3431
3432 void kvmppc_free_pimap(struct kvm *kvm)
3433 {
3434         kfree(kvm->arch.pimap);
3435 }
3436
3437 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
3438 {
3439         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
3440 }
3441
3442 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3443 {
3444         struct irq_desc *desc;
3445         struct kvmppc_irq_map *irq_map;
3446         struct kvmppc_passthru_irqmap *pimap;
3447         struct irq_chip *chip;
3448         int i;
3449
3450         if (!kvm_irq_bypass)
3451                 return 1;
3452
3453         desc = irq_to_desc(host_irq);
3454         if (!desc)
3455                 return -EIO;
3456
3457         mutex_lock(&kvm->lock);
3458
3459         pimap = kvm->arch.pimap;
3460         if (pimap == NULL) {
3461                 /* First call, allocate structure to hold IRQ map */
3462                 pimap = kvmppc_alloc_pimap();
3463                 if (pimap == NULL) {
3464                         mutex_unlock(&kvm->lock);
3465                         return -ENOMEM;
3466                 }
3467                 kvm->arch.pimap = pimap;
3468         }
3469
3470         /*
3471          * For now, we only support interrupts for which the EOI operation
3472          * is an OPAL call followed by a write to XIRR, since that's
3473          * what our real-mode EOI code does.
3474          */
3475         chip = irq_data_get_irq_chip(&desc->irq_data);
3476         if (!chip || !is_pnv_opal_msi(chip)) {
3477                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
3478                         host_irq, guest_gsi);
3479                 mutex_unlock(&kvm->lock);
3480                 return -ENOENT;
3481         }
3482
3483         /*
3484          * See if we already have an entry for this guest IRQ number.
3485          * If it's mapped to a hardware IRQ number, that's an error,
3486          * otherwise re-use this entry.
3487          */
3488         for (i = 0; i < pimap->n_mapped; i++) {
3489                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
3490                         if (pimap->mapped[i].r_hwirq) {
3491                                 mutex_unlock(&kvm->lock);
3492                                 return -EINVAL;
3493                         }
3494                         break;
3495                 }
3496         }
3497
3498         if (i == KVMPPC_PIRQ_MAPPED) {
3499                 mutex_unlock(&kvm->lock);
3500                 return -EAGAIN;         /* table is full */
3501         }
3502
3503         irq_map = &pimap->mapped[i];
3504
3505         irq_map->v_hwirq = guest_gsi;
3506         irq_map->desc = desc;
3507
3508         /*
3509          * Order the above two stores before the next to serialize with
3510          * the KVM real mode handler.
3511          */
3512         smp_wmb();
3513         irq_map->r_hwirq = desc->irq_data.hwirq;
3514
3515         if (i == pimap->n_mapped)
3516                 pimap->n_mapped++;
3517
3518         kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
3519
3520         mutex_unlock(&kvm->lock);
3521
3522         return 0;
3523 }
3524
3525 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3526 {
3527         struct irq_desc *desc;
3528         struct kvmppc_passthru_irqmap *pimap;
3529         int i;
3530
3531         if (!kvm_irq_bypass)
3532                 return 0;
3533
3534         desc = irq_to_desc(host_irq);
3535         if (!desc)
3536                 return -EIO;
3537
3538         mutex_lock(&kvm->lock);
3539
3540         if (kvm->arch.pimap == NULL) {
3541                 mutex_unlock(&kvm->lock);
3542                 return 0;
3543         }
3544         pimap = kvm->arch.pimap;
3545
3546         for (i = 0; i < pimap->n_mapped; i++) {
3547                 if (guest_gsi == pimap->mapped[i].v_hwirq)
3548                         break;
3549         }
3550
3551         if (i == pimap->n_mapped) {
3552                 mutex_unlock(&kvm->lock);
3553                 return -ENODEV;
3554         }
3555
3556         kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
3557
3558         /* invalidate the entry */
3559         pimap->mapped[i].r_hwirq = 0;
3560
3561         /*
3562          * We don't free this structure even when the count goes to
3563          * zero. The structure is freed when we destroy the VM.
3564          */
3565
3566         mutex_unlock(&kvm->lock);
3567         return 0;
3568 }
3569
3570 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
3571                                              struct irq_bypass_producer *prod)
3572 {
3573         int ret = 0;
3574         struct kvm_kernel_irqfd *irqfd =
3575                 container_of(cons, struct kvm_kernel_irqfd, consumer);
3576
3577         irqfd->producer = prod;
3578
3579         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
3580         if (ret)
3581                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
3582                         prod->irq, irqfd->gsi, ret);
3583
3584         return ret;
3585 }
3586
3587 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
3588                                               struct irq_bypass_producer *prod)
3589 {
3590         int ret;
3591         struct kvm_kernel_irqfd *irqfd =
3592                 container_of(cons, struct kvm_kernel_irqfd, consumer);
3593
3594         irqfd->producer = NULL;
3595
3596         /*
3597          * When producer of consumer is unregistered, we change back to
3598          * default external interrupt handling mode - KVM real mode
3599          * will switch back to host.
3600          */
3601         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
3602         if (ret)
3603                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
3604                         prod->irq, irqfd->gsi, ret);
3605 }
3606 #endif
3607
3608 static long kvm_arch_vm_ioctl_hv(struct file *filp,
3609                                  unsigned int ioctl, unsigned long arg)
3610 {
3611         struct kvm *kvm __maybe_unused = filp->private_data;
3612         void __user *argp = (void __user *)arg;
3613         long r;
3614
3615         switch (ioctl) {
3616
3617         case KVM_PPC_ALLOCATE_HTAB: {
3618                 u32 htab_order;
3619
3620                 r = -EFAULT;
3621                 if (get_user(htab_order, (u32 __user *)argp))
3622                         break;
3623                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
3624                 if (r)
3625                         break;
3626                 r = -EFAULT;
3627                 if (put_user(htab_order, (u32 __user *)argp))
3628                         break;
3629                 r = 0;
3630                 break;
3631         }
3632
3633         case KVM_PPC_GET_HTAB_FD: {
3634                 struct kvm_get_htab_fd ghf;
3635
3636                 r = -EFAULT;
3637                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
3638                         break;
3639                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3640                 break;
3641         }
3642
3643         default:
3644                 r = -ENOTTY;
3645         }
3646
3647         return r;
3648 }
3649
3650 /*
3651  * List of hcall numbers to enable by default.
3652  * For compatibility with old userspace, we enable by default
3653  * all hcalls that were implemented before the hcall-enabling
3654  * facility was added.  Note this list should not include H_RTAS.
3655  */
3656 static unsigned int default_hcall_list[] = {
3657         H_REMOVE,
3658         H_ENTER,
3659         H_READ,
3660         H_PROTECT,
3661         H_BULK_REMOVE,
3662         H_GET_TCE,
3663         H_PUT_TCE,
3664         H_SET_DABR,
3665         H_SET_XDABR,
3666         H_CEDE,
3667         H_PROD,
3668         H_CONFER,
3669         H_REGISTER_VPA,
3670 #ifdef CONFIG_KVM_XICS
3671         H_EOI,
3672         H_CPPR,
3673         H_IPI,
3674         H_IPOLL,
3675         H_XIRR,
3676         H_XIRR_X,
3677 #endif
3678         0
3679 };
3680
3681 static void init_default_hcalls(void)
3682 {
3683         int i;
3684         unsigned int hcall;
3685
3686         for (i = 0; default_hcall_list[i]; ++i) {
3687                 hcall = default_hcall_list[i];
3688                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3689                 __set_bit(hcall / 4, default_enabled_hcalls);
3690         }
3691 }
3692
3693 static struct kvmppc_ops kvm_ops_hv = {
3694         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3695         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3696         .get_one_reg = kvmppc_get_one_reg_hv,
3697         .set_one_reg = kvmppc_set_one_reg_hv,
3698         .vcpu_load   = kvmppc_core_vcpu_load_hv,
3699         .vcpu_put    = kvmppc_core_vcpu_put_hv,
3700         .set_msr     = kvmppc_set_msr_hv,
3701         .vcpu_run    = kvmppc_vcpu_run_hv,
3702         .vcpu_create = kvmppc_core_vcpu_create_hv,
3703         .vcpu_free   = kvmppc_core_vcpu_free_hv,
3704         .check_requests = kvmppc_core_check_requests_hv,
3705         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
3706         .flush_memslot  = kvmppc_core_flush_memslot_hv,
3707         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3708         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
3709         .unmap_hva = kvm_unmap_hva_hv,
3710         .unmap_hva_range = kvm_unmap_hva_range_hv,
3711         .age_hva  = kvm_age_hva_hv,
3712         .test_age_hva = kvm_test_age_hva_hv,
3713         .set_spte_hva = kvm_set_spte_hva_hv,
3714         .mmu_destroy  = kvmppc_mmu_destroy_hv,
3715         .free_memslot = kvmppc_core_free_memslot_hv,
3716         .create_memslot = kvmppc_core_create_memslot_hv,
3717         .init_vm =  kvmppc_core_init_vm_hv,
3718         .destroy_vm = kvmppc_core_destroy_vm_hv,
3719         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3720         .emulate_op = kvmppc_core_emulate_op_hv,
3721         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3722         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3723         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3724         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
3725         .hcall_implemented = kvmppc_hcall_impl_hv,
3726 #ifdef CONFIG_KVM_XICS
3727         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
3728         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
3729 #endif
3730 };
3731
3732 static int kvm_init_subcore_bitmap(void)
3733 {
3734         int i, j;
3735         int nr_cores = cpu_nr_cores();
3736         struct sibling_subcore_state *sibling_subcore_state;
3737
3738         for (i = 0; i < nr_cores; i++) {
3739                 int first_cpu = i * threads_per_core;
3740                 int node = cpu_to_node(first_cpu);
3741
3742                 /* Ignore if it is already allocated. */
3743                 if (paca[first_cpu].sibling_subcore_state)
3744                         continue;
3745
3746                 sibling_subcore_state =
3747                         kmalloc_node(sizeof(struct sibling_subcore_state),
3748                                                         GFP_KERNEL, node);
3749                 if (!sibling_subcore_state)
3750                         return -ENOMEM;
3751
3752                 memset(sibling_subcore_state, 0,
3753                                 sizeof(struct sibling_subcore_state));
3754
3755                 for (j = 0; j < threads_per_core; j++) {
3756                         int cpu = first_cpu + j;
3757
3758                         paca[cpu].sibling_subcore_state = sibling_subcore_state;
3759                 }
3760         }
3761         return 0;
3762 }
3763
3764 static int kvmppc_book3s_init_hv(void)
3765 {
3766         int r;
3767         /*
3768          * FIXME!! Do we need to check on all cpus ?
3769          */
3770         r = kvmppc_core_check_processor_compat_hv();
3771         if (r < 0)
3772                 return -ENODEV;
3773
3774         r = kvm_init_subcore_bitmap();
3775         if (r)
3776                 return r;
3777
3778         kvm_ops_hv.owner = THIS_MODULE;
3779         kvmppc_hv_ops = &kvm_ops_hv;
3780
3781         init_default_hcalls();
3782
3783         init_vcore_lists();
3784
3785         r = kvmppc_mmu_hv_init();
3786         return r;
3787 }
3788
3789 static void kvmppc_book3s_exit_hv(void)
3790 {
3791         kvmppc_free_host_rm_ops();
3792         kvmppc_hv_ops = NULL;
3793 }
3794
3795 module_init(kvmppc_book3s_init_hv);
3796 module_exit(kvmppc_book3s_exit_hv);
3797 MODULE_LICENSE("GPL");
3798 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3799 MODULE_ALIAS("devname:kvm");