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/kernel.h>
  23 #include <linux/err.h>
  24 #include <linux/slab.h>
  25 #include <linux/preempt.h>
  26 #include <linux/sched/signal.h>
  27 #include <linux/sched/stat.h>
  28 #include <linux/delay.h>
  29 #include <linux/export.h>
  30 #include <linux/fs.h>
  31 #include <linux/anon_inodes.h>
  32 #include <linux/cpu.h>
  33 #include <linux/cpumask.h>
  34 #include <linux/spinlock.h>
  35 #include <linux/page-flags.h>
  36 #include <linux/srcu.h>
  37 #include <linux/miscdevice.h>
  38 #include <linux/debugfs.h>
  39 #include <linux/gfp.h>
  40 #include <linux/vmalloc.h>
  41 #include <linux/highmem.h>
  42 #include <linux/hugetlb.h>
  43 #include <linux/kvm_irqfd.h>
  44 #include <linux/irqbypass.h>
  45 #include <linux/module.h>
  46 #include <linux/compiler.h>
  47 #include <linux/of.h>
  48
  49 #include <asm/ftrace.h>
  50 #include <asm/reg.h>
  51 #include <asm/ppc-opcode.h>
  52 #include <asm/asm-prototypes.h>
  53 #include <asm/debug.h>
  54 #include <asm/disassemble.h>
  55 #include <asm/cputable.h>
  56 #include <asm/cacheflush.h>
  57 #include <linux/uaccess.h>
  58 #include <asm/io.h>
  59 #include <asm/kvm_ppc.h>
  60 #include <asm/kvm_book3s.h>
  61 #include <asm/mmu_context.h>
  62 #include <asm/lppaca.h>
  63 #include <asm/processor.h>
  64 #include <asm/cputhreads.h>
  65 #include <asm/page.h>
  66 #include <asm/hvcall.h>
  67 #include <asm/switch_to.h>
  68 #include <asm/smp.h>
  69 #include <asm/dbell.h>
  70 #include <asm/hmi.h>
  71 #include <asm/pnv-pci.h>
  72 #include <asm/mmu.h>
  73 #include <asm/opal.h>
  74 #include <asm/xics.h>
  75 #include <asm/xive.h>
  76
  77 #include "book3s.h"
  78
  79 #define CREATE_TRACE_POINTS
  80 #include "trace_hv.h"
  81
  82 /* #define EXIT_DEBUG */
  83 /* #define EXIT_DEBUG_SIMPLE */
  84 /* #define EXIT_DEBUG_INT */
  85
  86 /* Used to indicate that a guest page fault needs to be handled */
  87 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  88 /* Used to indicate that a guest passthrough interrupt needs to be handled */
  89 #define RESUME_PASSTHROUGH      (RESUME_GUEST | RESUME_FLAG_ARCH2)
  90
  91 /* Used as a "null" value for timebase values */
  92 #define TB_NIL  (~(u64)0)
  93
  94 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
  95
  96 static int dynamic_mt_modes = 6;
  97 module_param(dynamic_mt_modes, int, 0644);
  98 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
  99 static int target_smt_mode;
 100 module_param(target_smt_mode, int, 0644);
 101 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
 102
 103 static bool indep_threads_mode = true;
 104 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
 105 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
 106
 107 #ifdef CONFIG_KVM_XICS
 108 static struct kernel_param_ops module_param_ops = {
 109         .set = param_set_int,
 110         .get = param_get_int,
 111 };
 112
 113 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
 114 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
 115
 116 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
 117 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
 118 #endif
 119
 120 /* If set, the threads on each CPU core have to be in the same MMU mode */
 121 static bool no_mixing_hpt_and_radix;
 122
 123 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
 124 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
 125
 126 /*
 127  * RWMR values for POWER8.  These control the rate at which PURR
 128  * and SPURR count and should be set according to the number of
 129  * online threads in the vcore being run.
 130  */
 131 #define RWMR_RPA_P8_1THREAD     0x164520C62609AECAUL
 132 #define RWMR_RPA_P8_2THREAD     0x7FFF2908450D8DA9UL
 133 #define RWMR_RPA_P8_3THREAD     0x164520C62609AECAUL
 134 #define RWMR_RPA_P8_4THREAD     0x199A421245058DA9UL
 135 #define RWMR_RPA_P8_5THREAD     0x164520C62609AECAUL
 136 #define RWMR_RPA_P8_6THREAD     0x164520C62609AECAUL
 137 #define RWMR_RPA_P8_7THREAD     0x164520C62609AECAUL
 138 #define RWMR_RPA_P8_8THREAD     0x164520C62609AECAUL
 139
 140 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
 141         RWMR_RPA_P8_1THREAD,
 142         RWMR_RPA_P8_1THREAD,
 143         RWMR_RPA_P8_2THREAD,
 144         RWMR_RPA_P8_3THREAD,
 145         RWMR_RPA_P8_4THREAD,
 146         RWMR_RPA_P8_5THREAD,
 147         RWMR_RPA_P8_6THREAD,
 148         RWMR_RPA_P8_7THREAD,
 149         RWMR_RPA_P8_8THREAD,
 150 };
 151
 152 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
 153                 int *ip)
 154 {
 155         int i = *ip;
 156         struct kvm_vcpu *vcpu;
 157
 158         while (++i < MAX_SMT_THREADS) {
 159                 vcpu = READ_ONCE(vc->runnable_threads[i]);
 160                 if (vcpu) {
 161                         *ip = i;
 162                         return vcpu;
 163                 }
 164         }
 165         return NULL;
 166 }
 167
 168 /* Used to traverse the list of runnable threads for a given vcore */
 169 #define for_each_runnable_thread(i, vcpu, vc) \
 170         for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
 171
 172 static bool kvmppc_ipi_thread(int cpu)
 173 {
 174         unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
 175
 176         /* On POWER9 we can use msgsnd to IPI any cpu */
 177         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
 178                 msg |= get_hard_smp_processor_id(cpu);
 179                 smp_mb();
 180                 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 181                 return true;
 182         }
 183
 184         /* On POWER8 for IPIs to threads in the same core, use msgsnd */
 185         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
 186                 preempt_disable();
 187                 if (cpu_first_thread_sibling(cpu) ==
 188                     cpu_first_thread_sibling(smp_processor_id())) {
 189                         msg |= cpu_thread_in_core(cpu);
 190                         smp_mb();
 191                         __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 192                         preempt_enable();
 193                         return true;
 194                 }
 195                 preempt_enable();
 196         }
 197
 198 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
 199         if (cpu >= 0 && cpu < nr_cpu_ids) {
 200                 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
 201                         xics_wake_cpu(cpu);
 202                         return true;
 203                 }
 204                 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
 205                 return true;
 206         }
 207 #endif
 208
 209         return false;
 210 }
 211
 212 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
 213 {
 214         int cpu;
 215         struct swait_queue_head *wqp;
 216
 217         wqp = kvm_arch_vcpu_wq(vcpu);
 218         if (swq_has_sleeper(wqp)) {
 219                 swake_up_one(wqp);
 220                 ++vcpu->stat.halt_wakeup;
 221         }
 222
 223         cpu = READ_ONCE(vcpu->arch.thread_cpu);
 224         if (cpu >= 0 && kvmppc_ipi_thread(cpu))
 225                 return;
 226
 227         /* CPU points to the first thread of the core */
 228         cpu = vcpu->cpu;
 229         if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
 230                 smp_send_reschedule(cpu);
 231 }
 232
 233 /*
 234  * We use the vcpu_load/put functions to measure stolen time.
 235  * Stolen time is counted as time when either the vcpu is able to
 236  * run as part of a virtual core, but the task running the vcore
 237  * is preempted or sleeping, or when the vcpu needs something done
 238  * in the kernel by the task running the vcpu, but that task is
 239  * preempted or sleeping.  Those two things have to be counted
 240  * separately, since one of the vcpu tasks will take on the job
 241  * of running the core, and the other vcpu tasks in the vcore will
 242  * sleep waiting for it to do that, but that sleep shouldn't count
 243  * as stolen time.
 244  *
 245  * Hence we accumulate stolen time when the vcpu can run as part of
 246  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 247  * needs its task to do other things in the kernel (for example,
 248  * service a page fault) in busy_stolen.  We don't accumulate
 249  * stolen time for a vcore when it is inactive, or for a vcpu
 250  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 251  * a misnomer; it means that the vcpu task is not executing in
 252  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 253  * the kernel.  We don't have any way of dividing up that time
 254  * between time that the vcpu is genuinely stopped, time that
 255  * the task is actively working on behalf of the vcpu, and time
 256  * that the task is preempted, so we don't count any of it as
 257  * stolen.
 258  *
 259  * Updates to busy_stolen are protected by arch.tbacct_lock;
 260  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
 261  * lock.  The stolen times are measured in units of timebase ticks.
 262  * (Note that the != TB_NIL checks below are purely defensive;
 263  * they should never fail.)
 264  */
 265
 266 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
 267 {
 268         unsigned long flags;
 269
 270         spin_lock_irqsave(&vc->stoltb_lock, flags);
 271         vc->preempt_tb = mftb();
 272         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 273 }
 274
 275 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
 276 {
 277         unsigned long flags;
 278
 279         spin_lock_irqsave(&vc->stoltb_lock, flags);
 280         if (vc->preempt_tb != TB_NIL) {
 281                 vc->stolen_tb += mftb() - vc->preempt_tb;
 282                 vc->preempt_tb = TB_NIL;
 283         }
 284         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 285 }
 286
 287 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 288 {
 289         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 290         unsigned long flags;
 291
 292         /*
 293          * We can test vc->runner without taking the vcore lock,
 294          * because only this task ever sets vc->runner to this
 295          * vcpu, and once it is set to this vcpu, only this task
 296          * ever sets it to NULL.
 297          */
 298         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 299                 kvmppc_core_end_stolen(vc);
 300
 301         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 302         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 303             vcpu->arch.busy_preempt != TB_NIL) {
 304                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 305                 vcpu->arch.busy_preempt = TB_NIL;
 306         }
 307         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 308 }
 309
 310 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 311 {
 312         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 313         unsigned long flags;
 314
 315         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 316                 kvmppc_core_start_stolen(vc);
 317
 318         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 319         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 320                 vcpu->arch.busy_preempt = mftb();
 321         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 322 }
 323
 324 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
 325 {
 326         /*
 327          * Check for illegal transactional state bit combination
 328          * and if we find it, force the TS field to a safe state.
 329          */
 330         if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
 331                 msr &= ~MSR_TS_MASK;
 332         vcpu->arch.shregs.msr = msr;
 333         kvmppc_end_cede(vcpu);
 334 }
 335
 336 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 337 {
 338         vcpu->arch.pvr = pvr;
 339 }
 340
 341 /* Dummy value used in computing PCR value below */
 342 #define PCR_ARCH_300    (PCR_ARCH_207 << 1)
 343
 344 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 345 {
 346         unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
 347         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 348
 349         /* We can (emulate) our own architecture version and anything older */
 350         if (cpu_has_feature(CPU_FTR_ARCH_300))
 351                 host_pcr_bit = PCR_ARCH_300;
 352         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
 353                 host_pcr_bit = PCR_ARCH_207;
 354         else if (cpu_has_feature(CPU_FTR_ARCH_206))
 355                 host_pcr_bit = PCR_ARCH_206;
 356         else
 357                 host_pcr_bit = PCR_ARCH_205;
 358
 359         /* Determine lowest PCR bit needed to run guest in given PVR level */
 360         guest_pcr_bit = host_pcr_bit;
 361         if (arch_compat) {
 362                 switch (arch_compat) {
 363                 case PVR_ARCH_205:
 364                         guest_pcr_bit = PCR_ARCH_205;
 365                         break;
 366                 case PVR_ARCH_206:
 367                 case PVR_ARCH_206p:
 368                         guest_pcr_bit = PCR_ARCH_206;
 369                         break;
 370                 case PVR_ARCH_207:
 371                         guest_pcr_bit = PCR_ARCH_207;
 372                         break;
 373                 case PVR_ARCH_300:
 374                         guest_pcr_bit = PCR_ARCH_300;
 375                         break;
 376                 default:
 377                         return -EINVAL;
 378                 }
 379         }
 380
 381         /* Check requested PCR bits don't exceed our capabilities */
 382         if (guest_pcr_bit > host_pcr_bit)
 383                 return -EINVAL;
 384
 385         spin_lock(&vc->lock);
 386         vc->arch_compat = arch_compat;
 387         /* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
 388         vc->pcr = host_pcr_bit - guest_pcr_bit;
 389         spin_unlock(&vc->lock);
 390
 391         return 0;
 392 }
 393
 394 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 395 {
 396         int r;
 397
 398         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 399         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 400                vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
 401         for (r = 0; r < 16; ++r)
 402                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 403                        r, kvmppc_get_gpr(vcpu, r),
 404                        r+16, kvmppc_get_gpr(vcpu, r+16));
 405         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 406                vcpu->arch.regs.ctr, vcpu->arch.regs.link);
 407         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 408                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 409         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 410                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 411         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 412                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 413         pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
 414                vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
 415         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 416         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 417                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 418         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 419         for (r = 0; r < vcpu->arch.slb_max; ++r)
 420                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 421                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 422         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 423                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 424                vcpu->arch.last_inst);
 425 }
 426
 427 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 428 {
 429         return kvm_get_vcpu_by_id(kvm, id);
 430 }
 431
 432 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 433 {
 434         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 435         vpa->yield_count = cpu_to_be32(1);
 436 }
 437
 438 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 439                    unsigned long addr, unsigned long len)
 440 {
 441         /* check address is cacheline aligned */
 442         if (addr & (L1_CACHE_BYTES - 1))
 443                 return -EINVAL;
 444         spin_lock(&vcpu->arch.vpa_update_lock);
 445         if (v->next_gpa != addr || v->len != len) {
 446                 v->next_gpa = addr;
 447                 v->len = addr ? len : 0;
 448                 v->update_pending = 1;
 449         }
 450         spin_unlock(&vcpu->arch.vpa_update_lock);
 451         return 0;
 452 }
 453
 454 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 455 struct reg_vpa {
 456         u32 dummy;
 457         union {
 458                 __be16 hword;
 459                 __be32 word;
 460         } length;
 461 };
 462
 463 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 464 {
 465         if (vpap->update_pending)
 466                 return vpap->next_gpa != 0;
 467         return vpap->pinned_addr != NULL;
 468 }
 469
 470 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 471                                        unsigned long flags,
 472                                        unsigned long vcpuid, unsigned long vpa)
 473 {
 474         struct kvm *kvm = vcpu->kvm;
 475         unsigned long len, nb;
 476         void *va;
 477         struct kvm_vcpu *tvcpu;
 478         int err;
 479         int subfunc;
 480         struct kvmppc_vpa *vpap;
 481
 482         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 483         if (!tvcpu)
 484                 return H_PARAMETER;
 485
 486         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 487         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 488             subfunc == H_VPA_REG_SLB) {
 489                 /* Registering new area - address must be cache-line aligned */
 490                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 491                         return H_PARAMETER;
 492
 493                 /* convert logical addr to kernel addr and read length */
 494                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 495                 if (va == NULL)
 496                         return H_PARAMETER;
 497                 if (subfunc == H_VPA_REG_VPA)
 498                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 499                 else
 500                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 501                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 502
 503                 /* Check length */
 504                 if (len > nb || len < sizeof(struct reg_vpa))
 505                         return H_PARAMETER;
 506         } else {
 507                 vpa = 0;
 508                 len = 0;
 509         }
 510
 511         err = H_PARAMETER;
 512         vpap = NULL;
 513         spin_lock(&tvcpu->arch.vpa_update_lock);
 514
 515         switch (subfunc) {
 516         case H_VPA_REG_VPA:             /* register VPA */
 517                 /*
 518                  * The size of our lppaca is 1kB because of the way we align
 519                  * it for the guest to avoid crossing a 4kB boundary. We only
 520                  * use 640 bytes of the structure though, so we should accept
 521                  * clients that set a size of 640.
 522                  */
 523                 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
 524                 if (len < sizeof(struct lppaca))
 525                         break;
 526                 vpap = &tvcpu->arch.vpa;
 527                 err = 0;
 528                 break;
 529
 530         case H_VPA_REG_DTL:             /* register DTL */
 531                 if (len < sizeof(struct dtl_entry))
 532                         break;
 533                 len -= len % sizeof(struct dtl_entry);
 534
 535                 /* Check that they have previously registered a VPA */
 536                 err = H_RESOURCE;
 537                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 538                         break;
 539
 540                 vpap = &tvcpu->arch.dtl;
 541                 err = 0;
 542                 break;
 543
 544         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 545                 /* Check that they have previously registered a VPA */
 546                 err = H_RESOURCE;
 547                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 548                         break;
 549
 550                 vpap = &tvcpu->arch.slb_shadow;
 551                 err = 0;
 552                 break;
 553
 554         case H_VPA_DEREG_VPA:           /* deregister VPA */
 555                 /* Check they don't still have a DTL or SLB buf registered */
 556                 err = H_RESOURCE;
 557                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 558                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 559                         break;
 560
 561                 vpap = &tvcpu->arch.vpa;
 562                 err = 0;
 563                 break;
 564
 565         case H_VPA_DEREG_DTL:           /* deregister DTL */
 566                 vpap = &tvcpu->arch.dtl;
 567                 err = 0;
 568                 break;
 569
 570         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 571                 vpap = &tvcpu->arch.slb_shadow;
 572                 err = 0;
 573                 break;
 574         }
 575
 576         if (vpap) {
 577                 vpap->next_gpa = vpa;
 578                 vpap->len = len;
 579                 vpap->update_pending = 1;
 580         }
 581
 582         spin_unlock(&tvcpu->arch.vpa_update_lock);
 583
 584         return err;
 585 }
 586
 587 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 588 {
 589         struct kvm *kvm = vcpu->kvm;
 590         void *va;
 591         unsigned long nb;
 592         unsigned long gpa;
 593
 594         /*
 595          * We need to pin the page pointed to by vpap->next_gpa,
 596          * but we can't call kvmppc_pin_guest_page under the lock
 597          * as it does get_user_pages() and down_read().  So we
 598          * have to drop the lock, pin the page, then get the lock
 599          * again and check that a new area didn't get registered
 600          * in the meantime.
 601          */
 602         for (;;) {
 603                 gpa = vpap->next_gpa;
 604                 spin_unlock(&vcpu->arch.vpa_update_lock);
 605                 va = NULL;
 606                 nb = 0;
 607                 if (gpa)
 608                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 609                 spin_lock(&vcpu->arch.vpa_update_lock);
 610                 if (gpa == vpap->next_gpa)
 611                         break;
 612                 /* sigh... unpin that one and try again */
 613                 if (va)
 614                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 615         }
 616
 617         vpap->update_pending = 0;
 618         if (va && nb < vpap->len) {
 619                 /*
 620                  * If it's now too short, it must be that userspace
 621                  * has changed the mappings underlying guest memory,
 622                  * so unregister the region.
 623                  */
 624                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 625                 va = NULL;
 626         }
 627         if (vpap->pinned_addr)
 628                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 629                                         vpap->dirty);
 630         vpap->gpa = gpa;
 631         vpap->pinned_addr = va;
 632         vpap->dirty = false;
 633         if (va)
 634                 vpap->pinned_end = va + vpap->len;
 635 }
 636
 637 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 638 {
 639         if (!(vcpu->arch.vpa.update_pending ||
 640               vcpu->arch.slb_shadow.update_pending ||
 641               vcpu->arch.dtl.update_pending))
 642                 return;
 643
 644         spin_lock(&vcpu->arch.vpa_update_lock);
 645         if (vcpu->arch.vpa.update_pending) {
 646                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 647                 if (vcpu->arch.vpa.pinned_addr)
 648                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 649         }
 650         if (vcpu->arch.dtl.update_pending) {
 651                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 652                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 653                 vcpu->arch.dtl_index = 0;
 654         }
 655         if (vcpu->arch.slb_shadow.update_pending)
 656                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 657         spin_unlock(&vcpu->arch.vpa_update_lock);
 658 }
 659
 660 /*
 661  * Return the accumulated stolen time for the vcore up until `now'.
 662  * The caller should hold the vcore lock.
 663  */
 664 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 665 {
 666         u64 p;
 667         unsigned long flags;
 668
 669         spin_lock_irqsave(&vc->stoltb_lock, flags);
 670         p = vc->stolen_tb;
 671         if (vc->vcore_state != VCORE_INACTIVE &&
 672             vc->preempt_tb != TB_NIL)
 673                 p += now - vc->preempt_tb;
 674         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 675         return p;
 676 }
 677
 678 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 679                                     struct kvmppc_vcore *vc)
 680 {
 681         struct dtl_entry *dt;
 682         struct lppaca *vpa;
 683         unsigned long stolen;
 684         unsigned long core_stolen;
 685         u64 now;
 686         unsigned long flags;
 687
 688         dt = vcpu->arch.dtl_ptr;
 689         vpa = vcpu->arch.vpa.pinned_addr;
 690         now = mftb();
 691         core_stolen = vcore_stolen_time(vc, now);
 692         stolen = core_stolen - vcpu->arch.stolen_logged;
 693         vcpu->arch.stolen_logged = core_stolen;
 694         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 695         stolen += vcpu->arch.busy_stolen;
 696         vcpu->arch.busy_stolen = 0;
 697         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 698         if (!dt || !vpa)
 699                 return;
 700         memset(dt, 0, sizeof(struct dtl_entry));
 701         dt->dispatch_reason = 7;
 702         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 703         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 704         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 705         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 706         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 707         ++dt;
 708         if (dt == vcpu->arch.dtl.pinned_end)
 709                 dt = vcpu->arch.dtl.pinned_addr;
 710         vcpu->arch.dtl_ptr = dt;
 711         /* order writing *dt vs. writing vpa->dtl_idx */
 712         smp_wmb();
 713         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 714         vcpu->arch.dtl.dirty = true;
 715 }
 716
 717 /* See if there is a doorbell interrupt pending for a vcpu */
 718 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
 719 {
 720         int thr;
 721         struct kvmppc_vcore *vc;
 722
 723         if (vcpu->arch.doorbell_request)
 724                 return true;
 725         /*
 726          * Ensure that the read of vcore->dpdes comes after the read
 727          * of vcpu->doorbell_request.  This barrier matches the
 728          * lwsync in book3s_hv_rmhandlers.S just before the
 729          * fast_guest_return label.
 730          */
 731         smp_rmb();
 732         vc = vcpu->arch.vcore;
 733         thr = vcpu->vcpu_id - vc->first_vcpuid;
 734         return !!(vc->dpdes & (1 << thr));
 735 }
 736
 737 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 738 {
 739         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 740                 return true;
 741         if ((!vcpu->arch.vcore->arch_compat) &&
 742             cpu_has_feature(CPU_FTR_ARCH_207S))
 743                 return true;
 744         return false;
 745 }
 746
 747 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 748                              unsigned long resource, unsigned long value1,
 749                              unsigned long value2)
 750 {
 751         switch (resource) {
 752         case H_SET_MODE_RESOURCE_SET_CIABR:
 753                 if (!kvmppc_power8_compatible(vcpu))
 754                         return H_P2;
 755                 if (value2)
 756                         return H_P4;
 757                 if (mflags)
 758                         return H_UNSUPPORTED_FLAG_START;
 759                 /* Guests can't breakpoint the hypervisor */
 760                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 761                         return H_P3;
 762                 vcpu->arch.ciabr  = value1;
 763                 return H_SUCCESS;
 764         case H_SET_MODE_RESOURCE_SET_DAWR:
 765                 if (!kvmppc_power8_compatible(vcpu))
 766                         return H_P2;
 767                 if (!ppc_breakpoint_available())
 768                         return H_P2;
 769                 if (mflags)
 770                         return H_UNSUPPORTED_FLAG_START;
 771                 if (value2 & DABRX_HYP)
 772                         return H_P4;
 773                 vcpu->arch.dawr  = value1;
 774                 vcpu->arch.dawrx = value2;
 775                 return H_SUCCESS;
 776         default:
 777                 return H_TOO_HARD;
 778         }
 779 }
 780
 781 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 782 {
 783         struct kvmppc_vcore *vcore = target->arch.vcore;
 784
 785         /*
 786          * We expect to have been called by the real mode handler
 787          * (kvmppc_rm_h_confer()) which would have directly returned
 788          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 789          * have useful work to do and should not confer) so we don't
 790          * recheck that here.
 791          */
 792
 793         spin_lock(&vcore->lock);
 794         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 795             vcore->vcore_state != VCORE_INACTIVE &&
 796             vcore->runner)
 797                 target = vcore->runner;
 798         spin_unlock(&vcore->lock);
 799
 800         return kvm_vcpu_yield_to(target);
 801 }
 802
 803 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 804 {
 805         int yield_count = 0;
 806         struct lppaca *lppaca;
 807
 808         spin_lock(&vcpu->arch.vpa_update_lock);
 809         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 810         if (lppaca)
 811                 yield_count = be32_to_cpu(lppaca->yield_count);
 812         spin_unlock(&vcpu->arch.vpa_update_lock);
 813         return yield_count;
 814 }
 815
 816 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 817 {
 818         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 819         unsigned long target, ret = H_SUCCESS;
 820         int yield_count;
 821         struct kvm_vcpu *tvcpu;
 822         int idx, rc;
 823
 824         if (req <= MAX_HCALL_OPCODE &&
 825             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 826                 return RESUME_HOST;
 827
 828         switch (req) {
 829         case H_CEDE:
 830                 break;
 831         case H_PROD:
 832                 target = kvmppc_get_gpr(vcpu, 4);
 833                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 834                 if (!tvcpu) {
 835                         ret = H_PARAMETER;
 836                         break;
 837                 }
 838                 tvcpu->arch.prodded = 1;
 839                 smp_mb();
 840                 if (tvcpu->arch.ceded)
 841                         kvmppc_fast_vcpu_kick_hv(tvcpu);
 842                 break;
 843         case H_CONFER:
 844                 target = kvmppc_get_gpr(vcpu, 4);
 845                 if (target == -1)
 846                         break;
 847                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 848                 if (!tvcpu) {
 849                         ret = H_PARAMETER;
 850                         break;
 851                 }
 852                 yield_count = kvmppc_get_gpr(vcpu, 5);
 853                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 854                         break;
 855                 kvm_arch_vcpu_yield_to(tvcpu);
 856                 break;
 857         case H_REGISTER_VPA:
 858                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 859                                         kvmppc_get_gpr(vcpu, 5),
 860                                         kvmppc_get_gpr(vcpu, 6));
 861                 break;
 862         case H_RTAS:
 863                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 864                         return RESUME_HOST;
 865
 866                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 867                 rc = kvmppc_rtas_hcall(vcpu);
 868                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 869
 870                 if (rc == -ENOENT)
 871                         return RESUME_HOST;
 872                 else if (rc == 0)
 873                         break;
 874
 875                 /* Send the error out to userspace via KVM_RUN */
 876                 return rc;
 877         case H_LOGICAL_CI_LOAD:
 878                 ret = kvmppc_h_logical_ci_load(vcpu);
 879                 if (ret == H_TOO_HARD)
 880                         return RESUME_HOST;
 881                 break;
 882         case H_LOGICAL_CI_STORE:
 883                 ret = kvmppc_h_logical_ci_store(vcpu);
 884                 if (ret == H_TOO_HARD)
 885                         return RESUME_HOST;
 886                 break;
 887         case H_SET_MODE:
 888                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 889                                         kvmppc_get_gpr(vcpu, 5),
 890                                         kvmppc_get_gpr(vcpu, 6),
 891                                         kvmppc_get_gpr(vcpu, 7));
 892                 if (ret == H_TOO_HARD)
 893                         return RESUME_HOST;
 894                 break;
 895         case H_XIRR:
 896         case H_CPPR:
 897         case H_EOI:
 898         case H_IPI:
 899         case H_IPOLL:
 900         case H_XIRR_X:
 901                 if (kvmppc_xics_enabled(vcpu)) {
 902                         if (xive_enabled()) {
 903                                 ret = H_NOT_AVAILABLE;
 904                                 return RESUME_GUEST;
 905                         }
 906                         ret = kvmppc_xics_hcall(vcpu, req);
 907                         break;
 908                 }
 909                 return RESUME_HOST;
 910         case H_PUT_TCE:
 911                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 912                                                 kvmppc_get_gpr(vcpu, 5),
 913                                                 kvmppc_get_gpr(vcpu, 6));
 914                 if (ret == H_TOO_HARD)
 915                         return RESUME_HOST;
 916                 break;
 917         case H_PUT_TCE_INDIRECT:
 918                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
 919                                                 kvmppc_get_gpr(vcpu, 5),
 920                                                 kvmppc_get_gpr(vcpu, 6),
 921                                                 kvmppc_get_gpr(vcpu, 7));
 922                 if (ret == H_TOO_HARD)
 923                         return RESUME_HOST;
 924                 break;
 925         case H_STUFF_TCE:
 926                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 927                                                 kvmppc_get_gpr(vcpu, 5),
 928                                                 kvmppc_get_gpr(vcpu, 6),
 929                                                 kvmppc_get_gpr(vcpu, 7));
 930                 if (ret == H_TOO_HARD)
 931                         return RESUME_HOST;
 932                 break;
 933         default:
 934                 return RESUME_HOST;
 935         }
 936         kvmppc_set_gpr(vcpu, 3, ret);
 937         vcpu->arch.hcall_needed = 0;
 938         return RESUME_GUEST;
 939 }
 940
 941 static int kvmppc_hcall_impl_hv(unsigned long cmd)
 942 {
 943         switch (cmd) {
 944         case H_CEDE:
 945         case H_PROD:
 946         case H_CONFER:
 947         case H_REGISTER_VPA:
 948         case H_SET_MODE:
 949         case H_LOGICAL_CI_LOAD:
 950         case H_LOGICAL_CI_STORE:
 951 #ifdef CONFIG_KVM_XICS
 952         case H_XIRR:
 953         case H_CPPR:
 954         case H_EOI:
 955         case H_IPI:
 956         case H_IPOLL:
 957         case H_XIRR_X:
 958 #endif
 959                 return 1;
 960         }
 961
 962         /* See if it's in the real-mode table */
 963         return kvmppc_hcall_impl_hv_realmode(cmd);
 964 }
 965
 966 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
 967                                         struct kvm_vcpu *vcpu)
 968 {
 969         u32 last_inst;
 970
 971         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
 972                                         EMULATE_DONE) {
 973                 /*
 974                  * Fetch failed, so return to guest and
 975                  * try executing it again.
 976                  */
 977                 return RESUME_GUEST;
 978         }
 979
 980         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
 981                 run->exit_reason = KVM_EXIT_DEBUG;
 982                 run->debug.arch.address = kvmppc_get_pc(vcpu);
 983                 return RESUME_HOST;
 984         } else {
 985                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 986                 return RESUME_GUEST;
 987         }
 988 }
 989
 990 static void do_nothing(void *x)
 991 {
 992 }
 993
 994 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
 995 {
 996         int thr, cpu, pcpu, nthreads;
 997         struct kvm_vcpu *v;
 998         unsigned long dpdes;
 999
1000         nthreads = vcpu->kvm->arch.emul_smt_mode;
1001         dpdes = 0;
1002         cpu = vcpu->vcpu_id & ~(nthreads - 1);
1003         for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1004                 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1005                 if (!v)
1006                         continue;
1007                 /*
1008                  * If the vcpu is currently running on a physical cpu thread,
1009                  * interrupt it in order to pull it out of the guest briefly,
1010                  * which will update its vcore->dpdes value.
1011                  */
1012                 pcpu = READ_ONCE(v->cpu);
1013                 if (pcpu >= 0)
1014                         smp_call_function_single(pcpu, do_nothing, NULL, 1);
1015                 if (kvmppc_doorbell_pending(v))
1016                         dpdes |= 1 << thr;
1017         }
1018         return dpdes;
1019 }
1020
1021 /*
1022  * On POWER9, emulate doorbell-related instructions in order to
1023  * give the guest the illusion of running on a multi-threaded core.
1024  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1025  * and mfspr DPDES.
1026  */
1027 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1028 {
1029         u32 inst, rb, thr;
1030         unsigned long arg;
1031         struct kvm *kvm = vcpu->kvm;
1032         struct kvm_vcpu *tvcpu;
1033
1034         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1035                 return RESUME_GUEST;
1036         if (get_op(inst) != 31)
1037                 return EMULATE_FAIL;
1038         rb = get_rb(inst);
1039         thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1040         switch (get_xop(inst)) {
1041         case OP_31_XOP_MSGSNDP:
1042                 arg = kvmppc_get_gpr(vcpu, rb);
1043                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1044                         break;
1045                 arg &= 0x3f;
1046                 if (arg >= kvm->arch.emul_smt_mode)
1047                         break;
1048                 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1049                 if (!tvcpu)
1050                         break;
1051                 if (!tvcpu->arch.doorbell_request) {
1052                         tvcpu->arch.doorbell_request = 1;
1053                         kvmppc_fast_vcpu_kick_hv(tvcpu);
1054                 }
1055                 break;
1056         case OP_31_XOP_MSGCLRP:
1057                 arg = kvmppc_get_gpr(vcpu, rb);
1058                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1059                         break;
1060                 vcpu->arch.vcore->dpdes = 0;
1061                 vcpu->arch.doorbell_request = 0;
1062                 break;
1063         case OP_31_XOP_MFSPR:
1064                 switch (get_sprn(inst)) {
1065                 case SPRN_TIR:
1066                         arg = thr;
1067                         break;
1068                 case SPRN_DPDES:
1069                         arg = kvmppc_read_dpdes(vcpu);
1070                         break;
1071                 default:
1072                         return EMULATE_FAIL;
1073                 }
1074                 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1075                 break;
1076         default:
1077                 return EMULATE_FAIL;
1078         }
1079         kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1080         return RESUME_GUEST;
1081 }
1082
1083 /* Called with vcpu->arch.vcore->lock held */
1084 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1085                                  struct task_struct *tsk)
1086 {
1087         int r = RESUME_HOST;
1088
1089         vcpu->stat.sum_exits++;
1090
1091         /*
1092          * This can happen if an interrupt occurs in the last stages
1093          * of guest entry or the first stages of guest exit (i.e. after
1094          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1095          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1096          * That can happen due to a bug, or due to a machine check
1097          * occurring at just the wrong time.
1098          */
1099         if (vcpu->arch.shregs.msr & MSR_HV) {
1100                 printk(KERN_EMERG "KVM trap in HV mode!\n");
1101                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1102                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1103                         vcpu->arch.shregs.msr);
1104                 kvmppc_dump_regs(vcpu);
1105                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1106                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1107                 return RESUME_HOST;
1108         }
1109         run->exit_reason = KVM_EXIT_UNKNOWN;
1110         run->ready_for_interrupt_injection = 1;
1111         switch (vcpu->arch.trap) {
1112         /* We're good on these - the host merely wanted to get our attention */
1113         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1114                 vcpu->stat.dec_exits++;
1115                 r = RESUME_GUEST;
1116                 break;
1117         case BOOK3S_INTERRUPT_EXTERNAL:
1118         case BOOK3S_INTERRUPT_H_DOORBELL:
1119         case BOOK3S_INTERRUPT_H_VIRT:
1120                 vcpu->stat.ext_intr_exits++;
1121                 r = RESUME_GUEST;
1122                 break;
1123         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1124         case BOOK3S_INTERRUPT_HMI:
1125         case BOOK3S_INTERRUPT_PERFMON:
1126         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1127                 r = RESUME_GUEST;
1128                 break;
1129         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1130                 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1131                 run->exit_reason = KVM_EXIT_NMI;
1132                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1133                 /* Clear out the old NMI status from run->flags */
1134                 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1135                 /* Now set the NMI status */
1136                 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1137                         run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1138                 else
1139                         run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1140
1141                 r = RESUME_HOST;
1142                 /* Print the MCE event to host console. */
1143                 machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1144                 break;
1145         case BOOK3S_INTERRUPT_PROGRAM:
1146         {
1147                 ulong flags;
1148                 /*
1149                  * Normally program interrupts are delivered directly
1150                  * to the guest by the hardware, but we can get here
1151                  * as a result of a hypervisor emulation interrupt
1152                  * (e40) getting turned into a 700 by BML RTAS.
1153                  */
1154                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1155                 kvmppc_core_queue_program(vcpu, flags);
1156                 r = RESUME_GUEST;
1157                 break;
1158         }
1159         case BOOK3S_INTERRUPT_SYSCALL:
1160         {
1161                 /* hcall - punt to userspace */
1162                 int i;
1163
1164                 /* hypercall with MSR_PR has already been handled in rmode,
1165                  * and never reaches here.
1166                  */
1167
1168                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1169                 for (i = 0; i < 9; ++i)
1170                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1171                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1172                 vcpu->arch.hcall_needed = 1;
1173                 r = RESUME_HOST;
1174                 break;
1175         }
1176         /*
1177          * We get these next two if the guest accesses a page which it thinks
1178          * it has mapped but which is not actually present, either because
1179          * it is for an emulated I/O device or because the corresonding
1180          * host page has been paged out.  Any other HDSI/HISI interrupts
1181          * have been handled already.
1182          */
1183         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1184                 r = RESUME_PAGE_FAULT;
1185                 break;
1186         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1187                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1188                 vcpu->arch.fault_dsisr = 0;
1189                 r = RESUME_PAGE_FAULT;
1190                 break;
1191         /*
1192          * This occurs if the guest executes an illegal instruction.
1193          * If the guest debug is disabled, generate a program interrupt
1194          * to the guest. If guest debug is enabled, we need to check
1195          * whether the instruction is a software breakpoint instruction.
1196          * Accordingly return to Guest or Host.
1197          */
1198         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1199                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1200                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1201                                 swab32(vcpu->arch.emul_inst) :
1202                                 vcpu->arch.emul_inst;
1203                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1204                         /* Need vcore unlocked to call kvmppc_get_last_inst */
1205                         spin_unlock(&vcpu->arch.vcore->lock);
1206                         r = kvmppc_emulate_debug_inst(run, vcpu);
1207                         spin_lock(&vcpu->arch.vcore->lock);
1208                 } else {
1209                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1210                         r = RESUME_GUEST;
1211                 }
1212                 break;
1213         /*
1214          * This occurs if the guest (kernel or userspace), does something that
1215          * is prohibited by HFSCR.
1216          * On POWER9, this could be a doorbell instruction that we need
1217          * to emulate.
1218          * Otherwise, we just generate a program interrupt to the guest.
1219          */
1220         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1221                 r = EMULATE_FAIL;
1222                 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1223                     cpu_has_feature(CPU_FTR_ARCH_300)) {
1224                         /* Need vcore unlocked to call kvmppc_get_last_inst */
1225                         spin_unlock(&vcpu->arch.vcore->lock);
1226                         r = kvmppc_emulate_doorbell_instr(vcpu);
1227                         spin_lock(&vcpu->arch.vcore->lock);
1228                 }
1229                 if (r == EMULATE_FAIL) {
1230                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1231                         r = RESUME_GUEST;
1232                 }
1233                 break;
1234
1235 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1236         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1237                 /*
1238                  * This occurs for various TM-related instructions that
1239                  * we need to emulate on POWER9 DD2.2.  We have already
1240                  * handled the cases where the guest was in real-suspend
1241                  * mode and was transitioning to transactional state.
1242                  */
1243                 r = kvmhv_p9_tm_emulation(vcpu);
1244                 break;
1245 #endif
1246
1247         case BOOK3S_INTERRUPT_HV_RM_HARD:
1248                 r = RESUME_PASSTHROUGH;
1249                 break;
1250         default:
1251                 kvmppc_dump_regs(vcpu);
1252                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1253                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1254                         vcpu->arch.shregs.msr);
1255                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1256                 r = RESUME_HOST;
1257                 break;
1258         }
1259
1260         return r;
1261 }
1262
1263 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1264                                             struct kvm_sregs *sregs)
1265 {
1266         int i;
1267
1268         memset(sregs, 0, sizeof(struct kvm_sregs));
1269         sregs->pvr = vcpu->arch.pvr;
1270         for (i = 0; i < vcpu->arch.slb_max; i++) {
1271                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1272                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1273         }
1274
1275         return 0;
1276 }
1277
1278 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1279                                             struct kvm_sregs *sregs)
1280 {
1281         int i, j;
1282
1283         /* Only accept the same PVR as the host's, since we can't spoof it */
1284         if (sregs->pvr != vcpu->arch.pvr)
1285                 return -EINVAL;
1286
1287         j = 0;
1288         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1289                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1290                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1291                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1292                         ++j;
1293                 }
1294         }
1295         vcpu->arch.slb_max = j;
1296
1297         return 0;
1298 }
1299
1300 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1301                 bool preserve_top32)
1302 {
1303         struct kvm *kvm = vcpu->kvm;
1304         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1305         u64 mask;
1306
1307         spin_lock(&vc->lock);
1308         /*
1309          * If ILE (interrupt little-endian) has changed, update the
1310          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1311          */
1312         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1313                 struct kvm_vcpu *vcpu;
1314                 int i;
1315
1316                 kvm_for_each_vcpu(i, vcpu, kvm) {
1317                         if (vcpu->arch.vcore != vc)
1318                                 continue;
1319                         if (new_lpcr & LPCR_ILE)
1320                                 vcpu->arch.intr_msr |= MSR_LE;
1321                         else
1322                                 vcpu->arch.intr_msr &= ~MSR_LE;
1323                 }
1324         }
1325
1326         /*
1327          * Userspace can only modify DPFD (default prefetch depth),
1328          * ILE (interrupt little-endian) and TC (translation control).
1329          * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1330          */
1331         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1332         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1333                 mask |= LPCR_AIL;
1334         /*
1335          * On POWER9, allow userspace to enable large decrementer for the
1336          * guest, whether or not the host has it enabled.
1337          */
1338         if (cpu_has_feature(CPU_FTR_ARCH_300))
1339                 mask |= LPCR_LD;
1340
1341         /* Broken 32-bit version of LPCR must not clear top bits */
1342         if (preserve_top32)
1343                 mask &= 0xFFFFFFFF;
1344         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1345         spin_unlock(&vc->lock);
1346 }
1347
1348 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1349                                  union kvmppc_one_reg *val)
1350 {
1351         int r = 0;
1352         long int i;
1353
1354         switch (id) {
1355         case KVM_REG_PPC_DEBUG_INST:
1356                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1357                 break;
1358         case KVM_REG_PPC_HIOR:
1359                 *val = get_reg_val(id, 0);
1360                 break;
1361         case KVM_REG_PPC_DABR:
1362                 *val = get_reg_val(id, vcpu->arch.dabr);
1363                 break;
1364         case KVM_REG_PPC_DABRX:
1365                 *val = get_reg_val(id, vcpu->arch.dabrx);
1366                 break;
1367         case KVM_REG_PPC_DSCR:
1368                 *val = get_reg_val(id, vcpu->arch.dscr);
1369                 break;
1370         case KVM_REG_PPC_PURR:
1371                 *val = get_reg_val(id, vcpu->arch.purr);
1372                 break;
1373         case KVM_REG_PPC_SPURR:
1374                 *val = get_reg_val(id, vcpu->arch.spurr);
1375                 break;
1376         case KVM_REG_PPC_AMR:
1377                 *val = get_reg_val(id, vcpu->arch.amr);
1378                 break;
1379         case KVM_REG_PPC_UAMOR:
1380                 *val = get_reg_val(id, vcpu->arch.uamor);
1381                 break;
1382         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1383                 i = id - KVM_REG_PPC_MMCR0;
1384                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1385                 break;
1386         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1387                 i = id - KVM_REG_PPC_PMC1;
1388                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1389                 break;
1390         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1391                 i = id - KVM_REG_PPC_SPMC1;
1392                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1393                 break;
1394         case KVM_REG_PPC_SIAR:
1395                 *val = get_reg_val(id, vcpu->arch.siar);
1396                 break;
1397         case KVM_REG_PPC_SDAR:
1398                 *val = get_reg_val(id, vcpu->arch.sdar);
1399                 break;
1400         case KVM_REG_PPC_SIER:
1401                 *val = get_reg_val(id, vcpu->arch.sier);
1402                 break;
1403         case KVM_REG_PPC_IAMR:
1404                 *val = get_reg_val(id, vcpu->arch.iamr);
1405                 break;
1406         case KVM_REG_PPC_PSPB:
1407                 *val = get_reg_val(id, vcpu->arch.pspb);
1408                 break;
1409         case KVM_REG_PPC_DPDES:
1410                 /*
1411                  * On POWER9, where we are emulating msgsndp etc.,
1412                  * we return 1 bit for each vcpu, which can come from
1413                  * either vcore->dpdes or doorbell_request.
1414                  * On POWER8, doorbell_request is 0.
1415                  */
1416                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes |
1417                                    vcpu->arch.doorbell_request);
1418                 break;
1419         case KVM_REG_PPC_VTB:
1420                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1421                 break;
1422         case KVM_REG_PPC_DAWR:
1423                 *val = get_reg_val(id, vcpu->arch.dawr);
1424                 break;
1425         case KVM_REG_PPC_DAWRX:
1426                 *val = get_reg_val(id, vcpu->arch.dawrx);
1427                 break;
1428         case KVM_REG_PPC_CIABR:
1429                 *val = get_reg_val(id, vcpu->arch.ciabr);
1430                 break;
1431         case KVM_REG_PPC_CSIGR:
1432                 *val = get_reg_val(id, vcpu->arch.csigr);
1433                 break;
1434         case KVM_REG_PPC_TACR:
1435                 *val = get_reg_val(id, vcpu->arch.tacr);
1436                 break;
1437         case KVM_REG_PPC_TCSCR:
1438                 *val = get_reg_val(id, vcpu->arch.tcscr);
1439                 break;
1440         case KVM_REG_PPC_PID:
1441                 *val = get_reg_val(id, vcpu->arch.pid);
1442                 break;
1443         case KVM_REG_PPC_ACOP:
1444                 *val = get_reg_val(id, vcpu->arch.acop);
1445                 break;
1446         case KVM_REG_PPC_WORT:
1447                 *val = get_reg_val(id, vcpu->arch.wort);
1448                 break;
1449         case KVM_REG_PPC_TIDR:
1450                 *val = get_reg_val(id, vcpu->arch.tid);
1451                 break;
1452         case KVM_REG_PPC_PSSCR:
1453                 *val = get_reg_val(id, vcpu->arch.psscr);
1454                 break;
1455         case KVM_REG_PPC_VPA_ADDR:
1456                 spin_lock(&vcpu->arch.vpa_update_lock);
1457                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1458                 spin_unlock(&vcpu->arch.vpa_update_lock);
1459                 break;
1460         case KVM_REG_PPC_VPA_SLB:
1461                 spin_lock(&vcpu->arch.vpa_update_lock);
1462                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1463                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1464                 spin_unlock(&vcpu->arch.vpa_update_lock);
1465                 break;
1466         case KVM_REG_PPC_VPA_DTL:
1467                 spin_lock(&vcpu->arch.vpa_update_lock);
1468                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1469                 val->vpaval.length = vcpu->arch.dtl.len;
1470                 spin_unlock(&vcpu->arch.vpa_update_lock);
1471                 break;
1472         case KVM_REG_PPC_TB_OFFSET:
1473                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1474                 break;
1475         case KVM_REG_PPC_LPCR:
1476         case KVM_REG_PPC_LPCR_64:
1477                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1478                 break;
1479         case KVM_REG_PPC_PPR:
1480                 *val = get_reg_val(id, vcpu->arch.ppr);
1481                 break;
1482 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1483         case KVM_REG_PPC_TFHAR:
1484                 *val = get_reg_val(id, vcpu->arch.tfhar);
1485                 break;
1486         case KVM_REG_PPC_TFIAR:
1487                 *val = get_reg_val(id, vcpu->arch.tfiar);
1488                 break;
1489         case KVM_REG_PPC_TEXASR:
1490                 *val = get_reg_val(id, vcpu->arch.texasr);
1491                 break;
1492         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1493                 i = id - KVM_REG_PPC_TM_GPR0;
1494                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1495                 break;
1496         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1497         {
1498                 int j;
1499                 i = id - KVM_REG_PPC_TM_VSR0;
1500                 if (i < 32)
1501                         for (j = 0; j < TS_FPRWIDTH; j++)
1502                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1503                 else {
1504                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1505                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1506                         else
1507                                 r = -ENXIO;
1508                 }
1509                 break;
1510         }
1511         case KVM_REG_PPC_TM_CR:
1512                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1513                 break;
1514         case KVM_REG_PPC_TM_XER:
1515                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1516                 break;
1517         case KVM_REG_PPC_TM_LR:
1518                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1519                 break;
1520         case KVM_REG_PPC_TM_CTR:
1521                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1522                 break;
1523         case KVM_REG_PPC_TM_FPSCR:
1524                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1525                 break;
1526         case KVM_REG_PPC_TM_AMR:
1527                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1528                 break;
1529         case KVM_REG_PPC_TM_PPR:
1530                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1531                 break;
1532         case KVM_REG_PPC_TM_VRSAVE:
1533                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1534                 break;
1535         case KVM_REG_PPC_TM_VSCR:
1536                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1537                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1538                 else
1539                         r = -ENXIO;
1540                 break;
1541         case KVM_REG_PPC_TM_DSCR:
1542                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1543                 break;
1544         case KVM_REG_PPC_TM_TAR:
1545                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1546                 break;
1547 #endif
1548         case KVM_REG_PPC_ARCH_COMPAT:
1549                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1550                 break;
1551         case KVM_REG_PPC_DEC_EXPIRY:
1552                 *val = get_reg_val(id, vcpu->arch.dec_expires +
1553                                    vcpu->arch.vcore->tb_offset);
1554                 break;
1555         case KVM_REG_PPC_ONLINE:
1556                 *val = get_reg_val(id, vcpu->arch.online);
1557                 break;
1558         default:
1559                 r = -EINVAL;
1560                 break;
1561         }
1562
1563         return r;
1564 }
1565
1566 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1567                                  union kvmppc_one_reg *val)
1568 {
1569         int r = 0;
1570         long int i;
1571         unsigned long addr, len;
1572
1573         switch (id) {
1574         case KVM_REG_PPC_HIOR:
1575                 /* Only allow this to be set to zero */
1576                 if (set_reg_val(id, *val))
1577                         r = -EINVAL;
1578                 break;
1579         case KVM_REG_PPC_DABR:
1580                 vcpu->arch.dabr = set_reg_val(id, *val);
1581                 break;
1582         case KVM_REG_PPC_DABRX:
1583                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1584                 break;
1585         case KVM_REG_PPC_DSCR:
1586                 vcpu->arch.dscr = set_reg_val(id, *val);
1587                 break;
1588         case KVM_REG_PPC_PURR:
1589                 vcpu->arch.purr = set_reg_val(id, *val);
1590                 break;
1591         case KVM_REG_PPC_SPURR:
1592                 vcpu->arch.spurr = set_reg_val(id, *val);
1593                 break;
1594         case KVM_REG_PPC_AMR:
1595                 vcpu->arch.amr = set_reg_val(id, *val);
1596                 break;
1597         case KVM_REG_PPC_UAMOR:
1598                 vcpu->arch.uamor = set_reg_val(id, *val);
1599                 break;
1600         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1601                 i = id - KVM_REG_PPC_MMCR0;
1602                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1603                 break;
1604         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1605                 i = id - KVM_REG_PPC_PMC1;
1606                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1607                 break;
1608         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1609                 i = id - KVM_REG_PPC_SPMC1;
1610                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1611                 break;
1612         case KVM_REG_PPC_SIAR:
1613                 vcpu->arch.siar = set_reg_val(id, *val);
1614                 break;
1615         case KVM_REG_PPC_SDAR:
1616                 vcpu->arch.sdar = set_reg_val(id, *val);
1617                 break;
1618         case KVM_REG_PPC_SIER:
1619                 vcpu->arch.sier = set_reg_val(id, *val);
1620                 break;
1621         case KVM_REG_PPC_IAMR:
1622                 vcpu->arch.iamr = set_reg_val(id, *val);
1623                 break;
1624         case KVM_REG_PPC_PSPB:
1625                 vcpu->arch.pspb = set_reg_val(id, *val);
1626                 break;
1627         case KVM_REG_PPC_DPDES:
1628                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1629                 break;
1630         case KVM_REG_PPC_VTB:
1631                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1632                 break;
1633         case KVM_REG_PPC_DAWR:
1634                 vcpu->arch.dawr = set_reg_val(id, *val);
1635                 break;
1636         case KVM_REG_PPC_DAWRX:
1637                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1638                 break;
1639         case KVM_REG_PPC_CIABR:
1640                 vcpu->arch.ciabr = set_reg_val(id, *val);
1641                 /* Don't allow setting breakpoints in hypervisor code */
1642                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1643                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1644                 break;
1645         case KVM_REG_PPC_CSIGR:
1646                 vcpu->arch.csigr = set_reg_val(id, *val);
1647                 break;
1648         case KVM_REG_PPC_TACR:
1649                 vcpu->arch.tacr = set_reg_val(id, *val);
1650                 break;
1651         case KVM_REG_PPC_TCSCR:
1652                 vcpu->arch.tcscr = set_reg_val(id, *val);
1653                 break;
1654         case KVM_REG_PPC_PID:
1655                 vcpu->arch.pid = set_reg_val(id, *val);
1656                 break;
1657         case KVM_REG_PPC_ACOP:
1658                 vcpu->arch.acop = set_reg_val(id, *val);
1659                 break;
1660         case KVM_REG_PPC_WORT:
1661                 vcpu->arch.wort = set_reg_val(id, *val);
1662                 break;
1663         case KVM_REG_PPC_TIDR:
1664                 vcpu->arch.tid = set_reg_val(id, *val);
1665                 break;
1666         case KVM_REG_PPC_PSSCR:
1667                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1668                 break;
1669         case KVM_REG_PPC_VPA_ADDR:
1670                 addr = set_reg_val(id, *val);
1671                 r = -EINVAL;
1672                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1673                               vcpu->arch.dtl.next_gpa))
1674                         break;
1675                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1676                 break;
1677         case KVM_REG_PPC_VPA_SLB:
1678                 addr = val->vpaval.addr;
1679                 len = val->vpaval.length;
1680                 r = -EINVAL;
1681                 if (addr && !vcpu->arch.vpa.next_gpa)
1682                         break;
1683                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1684                 break;
1685         case KVM_REG_PPC_VPA_DTL:
1686                 addr = val->vpaval.addr;
1687                 len = val->vpaval.length;
1688                 r = -EINVAL;
1689                 if (addr && (len < sizeof(struct dtl_entry) ||
1690                              !vcpu->arch.vpa.next_gpa))
1691                         break;
1692                 len -= len % sizeof(struct dtl_entry);
1693                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1694                 break;
1695         case KVM_REG_PPC_TB_OFFSET:
1696                 /* round up to multiple of 2^24 */
1697                 vcpu->arch.vcore->tb_offset =
1698                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1699                 break;
1700         case KVM_REG_PPC_LPCR:
1701                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1702                 break;
1703         case KVM_REG_PPC_LPCR_64:
1704                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1705                 break;
1706         case KVM_REG_PPC_PPR:
1707                 vcpu->arch.ppr = set_reg_val(id, *val);
1708                 break;
1709 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1710         case KVM_REG_PPC_TFHAR:
1711                 vcpu->arch.tfhar = set_reg_val(id, *val);
1712                 break;
1713         case KVM_REG_PPC_TFIAR:
1714                 vcpu->arch.tfiar = set_reg_val(id, *val);
1715                 break;
1716         case KVM_REG_PPC_TEXASR:
1717                 vcpu->arch.texasr = set_reg_val(id, *val);
1718                 break;
1719         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1720                 i = id - KVM_REG_PPC_TM_GPR0;
1721                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1722                 break;
1723         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1724         {
1725                 int j;
1726                 i = id - KVM_REG_PPC_TM_VSR0;
1727                 if (i < 32)
1728                         for (j = 0; j < TS_FPRWIDTH; j++)
1729                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1730                 else
1731                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1732                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1733                         else
1734                                 r = -ENXIO;
1735                 break;
1736         }
1737         case KVM_REG_PPC_TM_CR:
1738                 vcpu->arch.cr_tm = set_reg_val(id, *val);
1739                 break;
1740         case KVM_REG_PPC_TM_XER:
1741                 vcpu->arch.xer_tm = set_reg_val(id, *val);
1742                 break;
1743         case KVM_REG_PPC_TM_LR:
1744                 vcpu->arch.lr_tm = set_reg_val(id, *val);
1745                 break;
1746         case KVM_REG_PPC_TM_CTR:
1747                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1748                 break;
1749         case KVM_REG_PPC_TM_FPSCR:
1750                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1751                 break;
1752         case KVM_REG_PPC_TM_AMR:
1753                 vcpu->arch.amr_tm = set_reg_val(id, *val);
1754                 break;
1755         case KVM_REG_PPC_TM_PPR:
1756                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1757                 break;
1758         case KVM_REG_PPC_TM_VRSAVE:
1759                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1760                 break;
1761         case KVM_REG_PPC_TM_VSCR:
1762                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1763                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1764                 else
1765                         r = - ENXIO;
1766                 break;
1767         case KVM_REG_PPC_TM_DSCR:
1768                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1769                 break;
1770         case KVM_REG_PPC_TM_TAR:
1771                 vcpu->arch.tar_tm = set_reg_val(id, *val);
1772                 break;
1773 #endif
1774         case KVM_REG_PPC_ARCH_COMPAT:
1775                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1776                 break;
1777         case KVM_REG_PPC_DEC_EXPIRY:
1778                 vcpu->arch.dec_expires = set_reg_val(id, *val) -
1779                         vcpu->arch.vcore->tb_offset;
1780                 break;
1781         case KVM_REG_PPC_ONLINE:
1782                 i = set_reg_val(id, *val);
1783                 if (i && !vcpu->arch.online)
1784                         atomic_inc(&vcpu->arch.vcore->online_count);
1785                 else if (!i && vcpu->arch.online)
1786                         atomic_dec(&vcpu->arch.vcore->online_count);
1787                 vcpu->arch.online = i;
1788                 break;
1789         default:
1790                 r = -EINVAL;
1791                 break;
1792         }
1793
1794         return r;
1795 }
1796
1797 /*
1798  * On POWER9, threads are independent and can be in different partitions.
1799  * Therefore we consider each thread to be a subcore.
1800  * There is a restriction that all threads have to be in the same
1801  * MMU mode (radix or HPT), unfortunately, but since we only support
1802  * HPT guests on a HPT host so far, that isn't an impediment yet.
1803  */
1804 static int threads_per_vcore(struct kvm *kvm)
1805 {
1806         if (kvm->arch.threads_indep)
1807                 return 1;
1808         return threads_per_subcore;
1809 }
1810
1811 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
1812 {
1813         struct kvmppc_vcore *vcore;
1814
1815         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1816
1817         if (vcore == NULL)
1818                 return NULL;
1819
1820         spin_lock_init(&vcore->lock);
1821         spin_lock_init(&vcore->stoltb_lock);
1822         init_swait_queue_head(&vcore->wq);
1823         vcore->preempt_tb = TB_NIL;
1824         vcore->lpcr = kvm->arch.lpcr;
1825         vcore->first_vcpuid = id;
1826         vcore->kvm = kvm;
1827         INIT_LIST_HEAD(&vcore->preempt_list);
1828
1829         return vcore;
1830 }
1831
1832 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1833 static struct debugfs_timings_element {
1834         const char *name;
1835         size_t offset;
1836 } timings[] = {
1837         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
1838         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
1839         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
1840         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
1841         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
1842 };
1843
1844 #define N_TIMINGS       (ARRAY_SIZE(timings))
1845
1846 struct debugfs_timings_state {
1847         struct kvm_vcpu *vcpu;
1848         unsigned int    buflen;
1849         char            buf[N_TIMINGS * 100];
1850 };
1851
1852 static int debugfs_timings_open(struct inode *inode, struct file *file)
1853 {
1854         struct kvm_vcpu *vcpu = inode->i_private;
1855         struct debugfs_timings_state *p;
1856
1857         p = kzalloc(sizeof(*p), GFP_KERNEL);
1858         if (!p)
1859                 return -ENOMEM;
1860
1861         kvm_get_kvm(vcpu->kvm);
1862         p->vcpu = vcpu;
1863         file->private_data = p;
1864
1865         return nonseekable_open(inode, file);
1866 }
1867
1868 static int debugfs_timings_release(struct inode *inode, struct file *file)
1869 {
1870         struct debugfs_timings_state *p = file->private_data;
1871
1872         kvm_put_kvm(p->vcpu->kvm);
1873         kfree(p);
1874         return 0;
1875 }
1876
1877 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1878                                     size_t len, loff_t *ppos)
1879 {
1880         struct debugfs_timings_state *p = file->private_data;
1881         struct kvm_vcpu *vcpu = p->vcpu;
1882         char *s, *buf_end;
1883         struct kvmhv_tb_accumulator tb;
1884         u64 count;
1885         loff_t pos;
1886         ssize_t n;
1887         int i, loops;
1888         bool ok;
1889
1890         if (!p->buflen) {
1891                 s = p->buf;
1892                 buf_end = s + sizeof(p->buf);
1893                 for (i = 0; i < N_TIMINGS; ++i) {
1894                         struct kvmhv_tb_accumulator *acc;
1895
1896                         acc = (struct kvmhv_tb_accumulator *)
1897                                 ((unsigned long)vcpu + timings[i].offset);
1898                         ok = false;
1899                         for (loops = 0; loops < 1000; ++loops) {
1900                                 count = acc->seqcount;
1901                                 if (!(count & 1)) {
1902                                         smp_rmb();
1903                                         tb = *acc;
1904                                         smp_rmb();
1905                                         if (count == acc->seqcount) {
1906                                                 ok = true;
1907                                                 break;
1908                                         }
1909                                 }
1910                                 udelay(1);
1911                         }
1912                         if (!ok)
1913                                 snprintf(s, buf_end - s, "%s: stuck\n",
1914                                         timings[i].name);
1915                         else
1916                                 snprintf(s, buf_end - s,
1917                                         "%s: %llu %llu %llu %llu\n",
1918                                         timings[i].name, count / 2,
1919                                         tb_to_ns(tb.tb_total),
1920                                         tb_to_ns(tb.tb_min),
1921                                         tb_to_ns(tb.tb_max));
1922                         s += strlen(s);
1923                 }
1924                 p->buflen = s - p->buf;
1925         }
1926
1927         pos = *ppos;
1928         if (pos >= p->buflen)
1929                 return 0;
1930         if (len > p->buflen - pos)
1931                 len = p->buflen - pos;
1932         n = copy_to_user(buf, p->buf + pos, len);
1933         if (n) {
1934                 if (n == len)
1935                         return -EFAULT;
1936                 len -= n;
1937         }
1938         *ppos = pos + len;
1939         return len;
1940 }
1941
1942 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1943                                      size_t len, loff_t *ppos)
1944 {
1945         return -EACCES;
1946 }
1947
1948 static const struct file_operations debugfs_timings_ops = {
1949         .owner   = THIS_MODULE,
1950         .open    = debugfs_timings_open,
1951         .release = debugfs_timings_release,
1952         .read    = debugfs_timings_read,
1953         .write   = debugfs_timings_write,
1954         .llseek  = generic_file_llseek,
1955 };
1956
1957 /* Create a debugfs directory for the vcpu */
1958 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1959 {
1960         char buf[16];
1961         struct kvm *kvm = vcpu->kvm;
1962
1963         snprintf(buf, sizeof(buf), "vcpu%u", id);
1964         if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1965                 return;
1966         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1967         if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1968                 return;
1969         vcpu->arch.debugfs_timings =
1970                 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1971                                     vcpu, &debugfs_timings_ops);
1972 }
1973
1974 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1975 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1976 {
1977 }
1978 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1979
1980 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1981                                                    unsigned int id)
1982 {
1983         struct kvm_vcpu *vcpu;
1984         int err;
1985         int core;
1986         struct kvmppc_vcore *vcore;
1987
1988         err = -ENOMEM;
1989         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1990         if (!vcpu)
1991                 goto out;
1992
1993         err = kvm_vcpu_init(vcpu, kvm, id);
1994         if (err)
1995                 goto free_vcpu;
1996
1997         vcpu->arch.shared = &vcpu->arch.shregs;
1998 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1999         /*
2000          * The shared struct is never shared on HV,
2001          * so we can always use host endianness
2002          */
2003 #ifdef __BIG_ENDIAN__
2004         vcpu->arch.shared_big_endian = true;
2005 #else
2006         vcpu->arch.shared_big_endian = false;
2007 #endif
2008 #endif
2009         vcpu->arch.mmcr[0] = MMCR0_FC;
2010         vcpu->arch.ctrl = CTRL_RUNLATCH;
2011         /* default to host PVR, since we can't spoof it */
2012         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2013         spin_lock_init(&vcpu->arch.vpa_update_lock);
2014         spin_lock_init(&vcpu->arch.tbacct_lock);
2015         vcpu->arch.busy_preempt = TB_NIL;
2016         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2017
2018         /*
2019          * Set the default HFSCR for the guest from the host value.
2020          * This value is only used on POWER9.
2021          * On POWER9, we want to virtualize the doorbell facility, so we
2022          * turn off the HFSCR bit, which causes those instructions to trap.
2023          */
2024         vcpu->arch.hfscr = mfspr(SPRN_HFSCR);
2025         if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2026                 vcpu->arch.hfscr |= HFSCR_TM;
2027         else if (!cpu_has_feature(CPU_FTR_TM_COMP))
2028                 vcpu->arch.hfscr &= ~HFSCR_TM;
2029         if (cpu_has_feature(CPU_FTR_ARCH_300))
2030                 vcpu->arch.hfscr &= ~HFSCR_MSGP;
2031
2032         kvmppc_mmu_book3s_hv_init(vcpu);
2033
2034         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2035
2036         init_waitqueue_head(&vcpu->arch.cpu_run);
2037
2038         mutex_lock(&kvm->lock);
2039         vcore = NULL;
2040         err = -EINVAL;
2041         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2042                 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2043                         pr_devel("KVM: VCPU ID too high\n");
2044                         core = KVM_MAX_VCORES;
2045                 } else {
2046                         BUG_ON(kvm->arch.smt_mode != 1);
2047                         core = kvmppc_pack_vcpu_id(kvm, id);
2048                 }
2049         } else {
2050                 core = id / kvm->arch.smt_mode;
2051         }
2052         if (core < KVM_MAX_VCORES) {
2053                 vcore = kvm->arch.vcores[core];
2054                 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2055                         pr_devel("KVM: collision on id %u", id);
2056                         vcore = NULL;
2057                 } else if (!vcore) {
2058                         err = -ENOMEM;
2059                         vcore = kvmppc_vcore_create(kvm,
2060                                         id & ~(kvm->arch.smt_mode - 1));
2061                         kvm->arch.vcores[core] = vcore;
2062                         kvm->arch.online_vcores++;
2063                 }
2064         }
2065         mutex_unlock(&kvm->lock);
2066
2067         if (!vcore)
2068                 goto uninit_vcpu;
2069
2070         spin_lock(&vcore->lock);
2071         ++vcore->num_threads;
2072         spin_unlock(&vcore->lock);
2073         vcpu->arch.vcore = vcore;
2074         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2075         vcpu->arch.thread_cpu = -1;
2076         vcpu->arch.prev_cpu = -1;
2077
2078         vcpu->arch.cpu_type = KVM_CPU_3S_64;
2079         kvmppc_sanity_check(vcpu);
2080
2081         debugfs_vcpu_init(vcpu, id);
2082
2083         return vcpu;
2084
2085 uninit_vcpu:
2086         kvm_vcpu_uninit(vcpu);
2087 free_vcpu:
2088         kmem_cache_free(kvm_vcpu_cache, vcpu);
2089 out:
2090         return ERR_PTR(err);
2091 }
2092
2093 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2094                               unsigned long flags)
2095 {
2096         int err;
2097         int esmt = 0;
2098
2099         if (flags)
2100                 return -EINVAL;
2101         if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2102                 return -EINVAL;
2103         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2104                 /*
2105                  * On POWER8 (or POWER7), the threading mode is "strict",
2106                  * so we pack smt_mode vcpus per vcore.
2107                  */
2108                 if (smt_mode > threads_per_subcore)
2109                         return -EINVAL;
2110         } else {
2111                 /*
2112                  * On POWER9, the threading mode is "loose",
2113                  * so each vcpu gets its own vcore.
2114                  */
2115                 esmt = smt_mode;
2116                 smt_mode = 1;
2117         }
2118         mutex_lock(&kvm->lock);
2119         err = -EBUSY;
2120         if (!kvm->arch.online_vcores) {
2121                 kvm->arch.smt_mode = smt_mode;
2122                 kvm->arch.emul_smt_mode = esmt;
2123                 err = 0;
2124         }
2125         mutex_unlock(&kvm->lock);
2126
2127         return err;
2128 }
2129
2130 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2131 {
2132         if (vpa->pinned_addr)
2133                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2134                                         vpa->dirty);
2135 }
2136
2137 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2138 {
2139         spin_lock(&vcpu->arch.vpa_update_lock);
2140         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2141         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2142         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2143         spin_unlock(&vcpu->arch.vpa_update_lock);
2144         kvm_vcpu_uninit(vcpu);
2145         kmem_cache_free(kvm_vcpu_cache, vcpu);
2146 }
2147
2148 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2149 {
2150         /* Indicate we want to get back into the guest */
2151         return 1;
2152 }
2153
2154 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2155 {
2156         unsigned long dec_nsec, now;
2157
2158         now = get_tb();
2159         if (now > vcpu->arch.dec_expires) {
2160                 /* decrementer has already gone negative */
2161                 kvmppc_core_queue_dec(vcpu);
2162                 kvmppc_core_prepare_to_enter(vcpu);
2163                 return;
2164         }
2165         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
2166                    / tb_ticks_per_sec;
2167         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2168         vcpu->arch.timer_running = 1;
2169 }
2170
2171 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2172 {
2173         vcpu->arch.ceded = 0;
2174         if (vcpu->arch.timer_running) {
2175                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2176                 vcpu->arch.timer_running = 0;
2177         }
2178 }
2179
2180 extern int __kvmppc_vcore_entry(void);
2181
2182 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2183                                    struct kvm_vcpu *vcpu)
2184 {
2185         u64 now;
2186
2187         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2188                 return;
2189         spin_lock_irq(&vcpu->arch.tbacct_lock);
2190         now = mftb();
2191         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2192                 vcpu->arch.stolen_logged;
2193         vcpu->arch.busy_preempt = now;
2194         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2195         spin_unlock_irq(&vcpu->arch.tbacct_lock);
2196         --vc->n_runnable;
2197         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2198 }
2199
2200 static int kvmppc_grab_hwthread(int cpu)
2201 {
2202         struct paca_struct *tpaca;
2203         long timeout = 10000;
2204
2205         tpaca = paca_ptrs[cpu];
2206
2207         /* Ensure the thread won't go into the kernel if it wakes */
2208         tpaca->kvm_hstate.kvm_vcpu = NULL;
2209         tpaca->kvm_hstate.kvm_vcore = NULL;
2210         tpaca->kvm_hstate.napping = 0;
2211         smp_wmb();
2212         tpaca->kvm_hstate.hwthread_req = 1;
2213
2214         /*
2215          * If the thread is already executing in the kernel (e.g. handling
2216          * a stray interrupt), wait for it to get back to nap mode.
2217          * The smp_mb() is to ensure that our setting of hwthread_req
2218          * is visible before we look at hwthread_state, so if this
2219          * races with the code at system_reset_pSeries and the thread
2220          * misses our setting of hwthread_req, we are sure to see its
2221          * setting of hwthread_state, and vice versa.
2222          */
2223         smp_mb();
2224         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2225                 if (--timeout <= 0) {
2226                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
2227                         return -EBUSY;
2228                 }
2229                 udelay(1);
2230         }
2231         return 0;
2232 }
2233
2234 static void kvmppc_release_hwthread(int cpu)
2235 {
2236         struct paca_struct *tpaca;
2237
2238         tpaca = paca_ptrs[cpu];
2239         tpaca->kvm_hstate.hwthread_req = 0;
2240         tpaca->kvm_hstate.kvm_vcpu = NULL;
2241         tpaca->kvm_hstate.kvm_vcore = NULL;
2242         tpaca->kvm_hstate.kvm_split_mode = NULL;
2243 }
2244
2245 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2246 {
2247         int i;
2248
2249         cpu = cpu_first_thread_sibling(cpu);
2250         cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2251         /*
2252          * Make sure setting of bit in need_tlb_flush precedes
2253          * testing of cpu_in_guest bits.  The matching barrier on
2254          * the other side is the first smp_mb() in kvmppc_run_core().
2255          */
2256         smp_mb();
2257         for (i = 0; i < threads_per_core; ++i)
2258                 if (cpumask_test_cpu(cpu + i, &kvm->arch.cpu_in_guest))
2259                         smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2260 }
2261
2262 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2263 {
2264         struct kvm *kvm = vcpu->kvm;
2265
2266         /*
2267          * With radix, the guest can do TLB invalidations itself,
2268          * and it could choose to use the local form (tlbiel) if
2269          * it is invalidating a translation that has only ever been
2270          * used on one vcpu.  However, that doesn't mean it has
2271          * only ever been used on one physical cpu, since vcpus
2272          * can move around between pcpus.  To cope with this, when
2273          * a vcpu moves from one pcpu to another, we need to tell
2274          * any vcpus running on the same core as this vcpu previously
2275          * ran to flush the TLB.  The TLB is shared between threads,
2276          * so we use a single bit in .need_tlb_flush for all 4 threads.
2277          */
2278         if (vcpu->arch.prev_cpu != pcpu) {
2279                 if (vcpu->arch.prev_cpu >= 0 &&
2280                     cpu_first_thread_sibling(vcpu->arch.prev_cpu) !=
2281                     cpu_first_thread_sibling(pcpu))
2282                         radix_flush_cpu(kvm, vcpu->arch.prev_cpu, vcpu);
2283                 vcpu->arch.prev_cpu = pcpu;
2284         }
2285 }
2286
2287 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2288 {
2289         int cpu;
2290         struct paca_struct *tpaca;
2291         struct kvm *kvm = vc->kvm;
2292
2293         cpu = vc->pcpu;
2294         if (vcpu) {
2295                 if (vcpu->arch.timer_running) {
2296                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2297                         vcpu->arch.timer_running = 0;
2298                 }
2299                 cpu += vcpu->arch.ptid;
2300                 vcpu->cpu = vc->pcpu;
2301                 vcpu->arch.thread_cpu = cpu;
2302                 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2303         }
2304         tpaca = paca_ptrs[cpu];
2305         tpaca->kvm_hstate.kvm_vcpu = vcpu;
2306         tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2307         tpaca->kvm_hstate.fake_suspend = 0;
2308         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2309         smp_wmb();
2310         tpaca->kvm_hstate.kvm_vcore = vc;
2311         if (cpu != smp_processor_id())
2312                 kvmppc_ipi_thread(cpu);
2313 }
2314
2315 static void kvmppc_wait_for_nap(int n_threads)
2316 {
2317         int cpu = smp_processor_id();
2318         int i, loops;
2319
2320         if (n_threads <= 1)
2321                 return;
2322         for (loops = 0; loops < 1000000; ++loops) {
2323                 /*
2324                  * Check if all threads are finished.
2325                  * We set the vcore pointer when starting a thread
2326                  * and the thread clears it when finished, so we look
2327                  * for any threads that still have a non-NULL vcore ptr.
2328                  */
2329                 for (i = 1; i < n_threads; ++i)
2330                         if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2331                                 break;
2332                 if (i == n_threads) {
2333                         HMT_medium();
2334                         return;
2335                 }
2336                 HMT_low();
2337         }
2338         HMT_medium();
2339         for (i = 1; i < n_threads; ++i)
2340                 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2341                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2342 }
2343
2344 /*
2345  * Check that we are on thread 0 and that any other threads in
2346  * this core are off-line.  Then grab the threads so they can't
2347  * enter the kernel.
2348  */
2349 static int on_primary_thread(void)
2350 {
2351         int cpu = smp_processor_id();
2352         int thr;
2353
2354         /* Are we on a primary subcore? */
2355         if (cpu_thread_in_subcore(cpu))
2356                 return 0;
2357
2358         thr = 0;
2359         while (++thr < threads_per_subcore)
2360                 if (cpu_online(cpu + thr))
2361                         return 0;
2362
2363         /* Grab all hw threads so they can't go into the kernel */
2364         for (thr = 1; thr < threads_per_subcore; ++thr) {
2365                 if (kvmppc_grab_hwthread(cpu + thr)) {
2366                         /* Couldn't grab one; let the others go */
2367                         do {
2368                                 kvmppc_release_hwthread(cpu + thr);
2369                         } while (--thr > 0);
2370                         return 0;
2371                 }
2372         }
2373         return 1;
2374 }
2375
2376 /*
2377  * A list of virtual cores for each physical CPU.
2378  * These are vcores that could run but their runner VCPU tasks are
2379  * (or may be) preempted.
2380  */
2381 struct preempted_vcore_list {
2382         struct list_head        list;
2383         spinlock_t              lock;
2384 };
2385
2386 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2387
2388 static void init_vcore_lists(void)
2389 {
2390         int cpu;
2391
2392         for_each_possible_cpu(cpu) {
2393                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2394                 spin_lock_init(&lp->lock);
2395                 INIT_LIST_HEAD(&lp->list);
2396         }
2397 }
2398
2399 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2400 {
2401         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2402
2403         vc->vcore_state = VCORE_PREEMPT;
2404         vc->pcpu = smp_processor_id();
2405         if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2406                 spin_lock(&lp->lock);
2407                 list_add_tail(&vc->preempt_list, &lp->list);
2408                 spin_unlock(&lp->lock);
2409         }
2410
2411         /* Start accumulating stolen time */
2412         kvmppc_core_start_stolen(vc);
2413 }
2414
2415 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2416 {
2417         struct preempted_vcore_list *lp;
2418
2419         kvmppc_core_end_stolen(vc);
2420         if (!list_empty(&vc->preempt_list)) {
2421                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2422                 spin_lock(&lp->lock);
2423                 list_del_init(&vc->preempt_list);
2424                 spin_unlock(&lp->lock);
2425         }
2426         vc->vcore_state = VCORE_INACTIVE;
2427 }
2428
2429 /*
2430  * This stores information about the virtual cores currently
2431  * assigned to a physical core.
2432  */
2433 struct core_info {
2434         int             n_subcores;
2435         int             max_subcore_threads;
2436         int             total_threads;
2437         int             subcore_threads[MAX_SUBCORES];
2438         struct kvmppc_vcore *vc[MAX_SUBCORES];
2439 };
2440
2441 /*
2442  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2443  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2444  */
2445 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2446
2447 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2448 {
2449         memset(cip, 0, sizeof(*cip));
2450         cip->n_subcores = 1;
2451         cip->max_subcore_threads = vc->num_threads;
2452         cip->total_threads = vc->num_threads;
2453         cip->subcore_threads[0] = vc->num_threads;
2454         cip->vc[0] = vc;
2455 }
2456
2457 static bool subcore_config_ok(int n_subcores, int n_threads)
2458 {
2459         /*
2460          * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2461          * split-core mode, with one thread per subcore.
2462          */
2463         if (cpu_has_feature(CPU_FTR_ARCH_300))
2464                 return n_subcores <= 4 && n_threads == 1;
2465
2466         /* On POWER8, can only dynamically split if unsplit to begin with */
2467         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2468                 return false;
2469         if (n_subcores > MAX_SUBCORES)
2470                 return false;
2471         if (n_subcores > 1) {
2472                 if (!(dynamic_mt_modes & 2))
2473                         n_subcores = 4;
2474                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2475                         return false;
2476         }
2477
2478         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2479 }
2480
2481 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2482 {
2483         vc->entry_exit_map = 0;
2484         vc->in_guest = 0;
2485         vc->napping_threads = 0;
2486         vc->conferring_threads = 0;
2487         vc->tb_offset_applied = 0;
2488 }
2489
2490 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2491 {
2492         int n_threads = vc->num_threads;
2493         int sub;
2494
2495         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2496                 return false;
2497
2498         /* Some POWER9 chips require all threads to be in the same MMU mode */
2499         if (no_mixing_hpt_and_radix &&
2500             kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2501                 return false;
2502
2503         if (n_threads < cip->max_subcore_threads)
2504                 n_threads = cip->max_subcore_threads;
2505         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2506                 return false;
2507         cip->max_subcore_threads = n_threads;
2508
2509         sub = cip->n_subcores;
2510         ++cip->n_subcores;
2511         cip->total_threads += vc->num_threads;
2512         cip->subcore_threads[sub] = vc->num_threads;
2513         cip->vc[sub] = vc;
2514         init_vcore_to_run(vc);
2515         list_del_init(&vc->preempt_list);
2516
2517         return true;
2518 }
2519
2520 /*
2521  * Work out whether it is possible to piggyback the execution of
2522  * vcore *pvc onto the execution of the other vcores described in *cip.
2523  */
2524 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2525                           int target_threads)
2526 {
2527         if (cip->total_threads + pvc->num_threads > target_threads)
2528                 return false;
2529
2530         return can_dynamic_split(pvc, cip);
2531 }
2532
2533 static void prepare_threads(struct kvmppc_vcore *vc)
2534 {
2535         int i;
2536         struct kvm_vcpu *vcpu;
2537
2538         for_each_runnable_thread(i, vcpu, vc) {
2539                 if (signal_pending(vcpu->arch.run_task))
2540                         vcpu->arch.ret = -EINTR;
2541                 else if (vcpu->arch.vpa.update_pending ||
2542                          vcpu->arch.slb_shadow.update_pending ||
2543                          vcpu->arch.dtl.update_pending)
2544                         vcpu->arch.ret = RESUME_GUEST;
2545                 else
2546                         continue;
2547                 kvmppc_remove_runnable(vc, vcpu);
2548                 wake_up(&vcpu->arch.cpu_run);
2549         }
2550 }
2551
2552 static void collect_piggybacks(struct core_info *cip, int target_threads)
2553 {
2554         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2555         struct kvmppc_vcore *pvc, *vcnext;
2556
2557         spin_lock(&lp->lock);
2558         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2559                 if (!spin_trylock(&pvc->lock))
2560                         continue;
2561                 prepare_threads(pvc);
2562                 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
2563                         list_del_init(&pvc->preempt_list);
2564                         if (pvc->runner == NULL) {
2565                                 pvc->vcore_state = VCORE_INACTIVE;
2566                                 kvmppc_core_end_stolen(pvc);
2567                         }
2568                         spin_unlock(&pvc->lock);
2569                         continue;
2570                 }
2571                 if (!can_piggyback(pvc, cip, target_threads)) {
2572                         spin_unlock(&pvc->lock);
2573                         continue;
2574                 }
2575                 kvmppc_core_end_stolen(pvc);
2576                 pvc->vcore_state = VCORE_PIGGYBACK;
2577                 if (cip->total_threads >= target_threads)
2578                         break;
2579         }
2580         spin_unlock(&lp->lock);
2581 }
2582
2583 static bool recheck_signals_and_mmu(struct core_info *cip)
2584 {
2585         int sub, i;
2586         struct kvm_vcpu *vcpu;
2587         struct kvmppc_vcore *vc;
2588
2589         for (sub = 0; sub < cip->n_subcores; ++sub) {
2590                 vc = cip->vc[sub];
2591                 if (!vc->kvm->arch.mmu_ready)
2592                         return true;
2593                 for_each_runnable_thread(i, vcpu, vc)
2594                         if (signal_pending(vcpu->arch.run_task))
2595                                 return true;
2596         }
2597         return false;
2598 }
2599
2600 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2601 {
2602         int still_running = 0, i;
2603         u64 now;
2604         long ret;
2605         struct kvm_vcpu *vcpu;
2606
2607         spin_lock(&vc->lock);
2608         now = get_tb();
2609         for_each_runnable_thread(i, vcpu, vc) {
2610                 /* cancel pending dec exception if dec is positive */
2611                 if (now < vcpu->arch.dec_expires &&
2612                     kvmppc_core_pending_dec(vcpu))
2613                         kvmppc_core_dequeue_dec(vcpu);
2614
2615                 trace_kvm_guest_exit(vcpu);
2616
2617                 ret = RESUME_GUEST;
2618                 if (vcpu->arch.trap)
2619                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2620                                                     vcpu->arch.run_task);
2621
2622                 vcpu->arch.ret = ret;
2623                 vcpu->arch.trap = 0;
2624
2625                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2626                         if (vcpu->arch.pending_exceptions)
2627                                 kvmppc_core_prepare_to_enter(vcpu);
2628                         if (vcpu->arch.ceded)
2629                                 kvmppc_set_timer(vcpu);
2630                         else
2631                                 ++still_running;
2632                 } else {
2633                         kvmppc_remove_runnable(vc, vcpu);
2634                         wake_up(&vcpu->arch.cpu_run);
2635                 }
2636         }
2637         if (!is_master) {
2638                 if (still_running > 0) {
2639                         kvmppc_vcore_preempt(vc);
2640                 } else if (vc->runner) {
2641                         vc->vcore_state = VCORE_PREEMPT;
2642                         kvmppc_core_start_stolen(vc);
2643                 } else {
2644                         vc->vcore_state = VCORE_INACTIVE;
2645                 }
2646                 if (vc->n_runnable > 0 && vc->runner == NULL) {
2647                         /* make sure there's a candidate runner awake */
2648                         i = -1;
2649                         vcpu = next_runnable_thread(vc, &i);
2650                         wake_up(&vcpu->arch.cpu_run);
2651                 }
2652         }
2653         spin_unlock(&vc->lock);
2654 }
2655
2656 /*
2657  * Clear core from the list of active host cores as we are about to
2658  * enter the guest. Only do this if it is the primary thread of the
2659  * core (not if a subcore) that is entering the guest.
2660  */
2661 static inline int kvmppc_clear_host_core(unsigned int cpu)
2662 {
2663         int core;
2664
2665         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2666                 return 0;
2667         /*
2668          * Memory barrier can be omitted here as we will do a smp_wmb()
2669          * later in kvmppc_start_thread and we need ensure that state is
2670          * visible to other CPUs only after we enter guest.
2671          */
2672         core = cpu >> threads_shift;
2673         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2674         return 0;
2675 }
2676
2677 /*
2678  * Advertise this core as an active host core since we exited the guest
2679  * Only need to do this if it is the primary thread of the core that is
2680  * exiting.
2681  */
2682 static inline int kvmppc_set_host_core(unsigned int cpu)
2683 {
2684         int core;
2685
2686         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2687                 return 0;
2688
2689         /*
2690          * Memory barrier can be omitted here because we do a spin_unlock
2691          * immediately after this which provides the memory barrier.
2692          */
2693         core = cpu >> threads_shift;
2694         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2695         return 0;
2696 }
2697
2698 static void set_irq_happened(int trap)
2699 {
2700         switch (trap) {
2701         case BOOK3S_INTERRUPT_EXTERNAL:
2702                 local_paca->irq_happened |= PACA_IRQ_EE;
2703                 break;
2704         case BOOK3S_INTERRUPT_H_DOORBELL:
2705                 local_paca->irq_happened |= PACA_IRQ_DBELL;
2706                 break;
2707         case BOOK3S_INTERRUPT_HMI:
2708                 local_paca->irq_happened |= PACA_IRQ_HMI;
2709                 break;
2710         case BOOK3S_INTERRUPT_SYSTEM_RESET:
2711                 replay_system_reset();
2712                 break;
2713         }
2714 }
2715
2716 /*
2717  * Run a set of guest threads on a physical core.
2718  * Called with vc->lock held.
2719  */
2720 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2721 {
2722         struct kvm_vcpu *vcpu;
2723         int i;
2724         int srcu_idx;
2725         struct core_info core_info;
2726         struct kvmppc_vcore *pvc;
2727         struct kvm_split_mode split_info, *sip;
2728         int split, subcore_size, active;
2729         int sub;
2730         bool thr0_done;
2731         unsigned long cmd_bit, stat_bit;
2732         int pcpu, thr;
2733         int target_threads;
2734         int controlled_threads;
2735         int trap;
2736         bool is_power8;
2737         bool hpt_on_radix;
2738
2739         /*
2740          * Remove from the list any threads that have a signal pending
2741          * or need a VPA update done
2742          */
2743         prepare_threads(vc);
2744
2745         /* if the runner is no longer runnable, let the caller pick a new one */
2746         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2747                 return;
2748
2749         /*
2750          * Initialize *vc.
2751          */
2752         init_vcore_to_run(vc);
2753         vc->preempt_tb = TB_NIL;
2754
2755         /*
2756          * Number of threads that we will be controlling: the same as
2757          * the number of threads per subcore, except on POWER9,
2758          * where it's 1 because the threads are (mostly) independent.
2759          */
2760         controlled_threads = threads_per_vcore(vc->kvm);
2761
2762         /*
2763          * Make sure we are running on primary threads, and that secondary
2764          * threads are offline.  Also check if the number of threads in this
2765          * guest are greater than the current system threads per guest.
2766          * On POWER9, we need to be not in independent-threads mode if
2767          * this is a HPT guest on a radix host machine where the
2768          * CPU threads may not be in different MMU modes.
2769          */
2770         hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
2771                 !kvm_is_radix(vc->kvm);
2772         if (((controlled_threads > 1) &&
2773              ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
2774             (hpt_on_radix && vc->kvm->arch.threads_indep)) {
2775                 for_each_runnable_thread(i, vcpu, vc) {
2776                         vcpu->arch.ret = -EBUSY;
2777                         kvmppc_remove_runnable(vc, vcpu);
2778                         wake_up(&vcpu->arch.cpu_run);
2779                 }
2780                 goto out;
2781         }
2782
2783         /*
2784          * See if we could run any other vcores on the physical core
2785          * along with this one.
2786          */
2787         init_core_info(&core_info, vc);
2788         pcpu = smp_processor_id();
2789         target_threads = controlled_threads;
2790         if (target_smt_mode && target_smt_mode < target_threads)
2791                 target_threads = target_smt_mode;
2792         if (vc->num_threads < target_threads)
2793                 collect_piggybacks(&core_info, target_threads);
2794
2795         /*
2796          * On radix, arrange for TLB flushing if necessary.
2797          * This has to be done before disabling interrupts since
2798          * it uses smp_call_function().
2799          */
2800         pcpu = smp_processor_id();
2801         if (kvm_is_radix(vc->kvm)) {
2802                 for (sub = 0; sub < core_info.n_subcores; ++sub)
2803                         for_each_runnable_thread(i, vcpu, core_info.vc[sub])
2804                                 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
2805         }
2806
2807         /*
2808          * Hard-disable interrupts, and check resched flag and signals.
2809          * If we need to reschedule or deliver a signal, clean up
2810          * and return without going into the guest(s).
2811          * If the mmu_ready flag has been cleared, don't go into the
2812          * guest because that means a HPT resize operation is in progress.
2813          */
2814         local_irq_disable();
2815         hard_irq_disable();
2816         if (lazy_irq_pending() || need_resched() ||
2817             recheck_signals_and_mmu(&core_info)) {
2818                 local_irq_enable();
2819                 vc->vcore_state = VCORE_INACTIVE;
2820                 /* Unlock all except the primary vcore */
2821                 for (sub = 1; sub < core_info.n_subcores; ++sub) {
2822                         pvc = core_info.vc[sub];
2823                         /* Put back on to the preempted vcores list */
2824                         kvmppc_vcore_preempt(pvc);
2825                         spin_unlock(&pvc->lock);
2826                 }
2827                 for (i = 0; i < controlled_threads; ++i)
2828                         kvmppc_release_hwthread(pcpu + i);
2829                 return;
2830         }
2831
2832         kvmppc_clear_host_core(pcpu);
2833
2834         /* Decide on micro-threading (split-core) mode */
2835         subcore_size = threads_per_subcore;
2836         cmd_bit = stat_bit = 0;
2837         split = core_info.n_subcores;
2838         sip = NULL;
2839         is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
2840                 && !cpu_has_feature(CPU_FTR_ARCH_300);
2841
2842         if (split > 1 || hpt_on_radix) {
2843                 sip = &split_info;
2844                 memset(&split_info, 0, sizeof(split_info));
2845                 for (sub = 0; sub < core_info.n_subcores; ++sub)
2846                         split_info.vc[sub] = core_info.vc[sub];
2847
2848                 if (is_power8) {
2849                         if (split == 2 && (dynamic_mt_modes & 2)) {
2850                                 cmd_bit = HID0_POWER8_1TO2LPAR;
2851                                 stat_bit = HID0_POWER8_2LPARMODE;
2852                         } else {
2853                                 split = 4;
2854                                 cmd_bit = HID0_POWER8_1TO4LPAR;
2855                                 stat_bit = HID0_POWER8_4LPARMODE;
2856                         }
2857                         subcore_size = MAX_SMT_THREADS / split;
2858                         split_info.rpr = mfspr(SPRN_RPR);
2859                         split_info.pmmar = mfspr(SPRN_PMMAR);
2860                         split_info.ldbar = mfspr(SPRN_LDBAR);
2861                         split_info.subcore_size = subcore_size;
2862                 } else {
2863                         split_info.subcore_size = 1;
2864                         if (hpt_on_radix) {
2865                                 /* Use the split_info for LPCR/LPIDR changes */
2866                                 split_info.lpcr_req = vc->lpcr;
2867                                 split_info.lpidr_req = vc->kvm->arch.lpid;
2868                                 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
2869                                 split_info.do_set = 1;
2870                         }
2871                 }
2872
2873                 /* order writes to split_info before kvm_split_mode pointer */
2874                 smp_wmb();
2875         }
2876
2877         for (thr = 0; thr < controlled_threads; ++thr) {
2878                 struct paca_struct *paca = paca_ptrs[pcpu + thr];
2879
2880                 paca->kvm_hstate.tid = thr;
2881                 paca->kvm_hstate.napping = 0;
2882                 paca->kvm_hstate.kvm_split_mode = sip;
2883         }
2884
2885         /* Initiate micro-threading (split-core) on POWER8 if required */
2886         if (cmd_bit) {
2887                 unsigned long hid0 = mfspr(SPRN_HID0);
2888
2889                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2890                 mb();
2891                 mtspr(SPRN_HID0, hid0);
2892                 isync();
2893                 for (;;) {
2894                         hid0 = mfspr(SPRN_HID0);
2895                         if (hid0 & stat_bit)
2896                                 break;
2897                         cpu_relax();
2898                 }
2899         }
2900
2901         /*
2902          * On POWER8, set RWMR register.
2903          * Since it only affects PURR and SPURR, it doesn't affect
2904          * the host, so we don't save/restore the host value.
2905          */
2906         if (is_power8) {
2907                 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
2908                 int n_online = atomic_read(&vc->online_count);
2909
2910                 /*
2911                  * Use the 8-thread value if we're doing split-core
2912                  * or if the vcore's online count looks bogus.
2913                  */
2914                 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
2915                     n_online >= 1 && n_online <= MAX_SMT_THREADS)
2916                         rwmr_val = p8_rwmr_values[n_online];
2917                 mtspr(SPRN_RWMR, rwmr_val);
2918         }
2919
2920         /* Start all the threads */
2921         active = 0;
2922         for (sub = 0; sub < core_info.n_subcores; ++sub) {
2923                 thr = is_power8 ? subcore_thread_map[sub] : sub;
2924                 thr0_done = false;
2925                 active |= 1 << thr;
2926                 pvc = core_info.vc[sub];
2927                 pvc->pcpu = pcpu + thr;
2928                 for_each_runnable_thread(i, vcpu, pvc) {
2929                         kvmppc_start_thread(vcpu, pvc);
2930                         kvmppc_create_dtl_entry(vcpu, pvc);
2931                         trace_kvm_guest_enter(vcpu);
2932                         if (!vcpu->arch.ptid)
2933                                 thr0_done = true;
2934                         active |= 1 << (thr + vcpu->arch.ptid);
2935                 }
2936                 /*
2937                  * We need to start the first thread of each subcore
2938                  * even if it doesn't have a vcpu.
2939                  */
2940                 if (!thr0_done)
2941                         kvmppc_start_thread(NULL, pvc);
2942         }
2943
2944         /*
2945          * Ensure that split_info.do_nap is set after setting
2946          * the vcore pointer in the PACA of the secondaries.
2947          */
2948         smp_mb();
2949
2950         /*
2951          * When doing micro-threading, poke the inactive threads as well.
2952          * This gets them to the nap instruction after kvm_do_nap,
2953          * which reduces the time taken to unsplit later.
2954          * For POWER9 HPT guest on radix host, we need all the secondary
2955          * threads woken up so they can do the LPCR/LPIDR change.
2956          */
2957         if (cmd_bit || hpt_on_radix) {
2958                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
2959                 for (thr = 1; thr < threads_per_subcore; ++thr)
2960                         if (!(active & (1 << thr)))
2961                                 kvmppc_ipi_thread(pcpu + thr);
2962         }
2963
2964         vc->vcore_state = VCORE_RUNNING;
2965         preempt_disable();
2966
2967         trace_kvmppc_run_core(vc, 0);
2968
2969         for (sub = 0; sub < core_info.n_subcores; ++sub)
2970                 spin_unlock(&core_info.vc[sub]->lock);
2971
2972         if (kvm_is_radix(vc->kvm)) {
2973                 int tmp = pcpu;
2974
2975                 /*
2976                  * Do we need to flush the process scoped TLB for the LPAR?
2977                  *
2978                  * On POWER9, individual threads can come in here, but the
2979                  * TLB is shared between the 4 threads in a core, hence
2980                  * invalidating on one thread invalidates for all.
2981                  * Thus we make all 4 threads use the same bit here.
2982                  *
2983                  * Hash must be flushed in realmode in order to use tlbiel.
2984                  */
2985                 mtspr(SPRN_LPID, vc->kvm->arch.lpid);
2986                 isync();
2987
2988                 if (cpu_has_feature(CPU_FTR_ARCH_300))
2989                         tmp &= ~0x3UL;
2990
2991                 if (cpumask_test_cpu(tmp, &vc->kvm->arch.need_tlb_flush)) {
2992                         radix__local_flush_tlb_lpid_guest(vc->kvm->arch.lpid);
2993                         /* Clear the bit after the TLB flush */
2994                         cpumask_clear_cpu(tmp, &vc->kvm->arch.need_tlb_flush);
2995                 }
2996         }
2997
2998         guest_enter_irqoff();
2999
3000         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3001
3002         this_cpu_disable_ftrace();
3003
3004         /*
3005          * Interrupts will be enabled once we get into the guest,
3006          * so tell lockdep that we're about to enable interrupts.
3007          */
3008         trace_hardirqs_on();
3009
3010         trap = __kvmppc_vcore_entry();
3011
3012         trace_hardirqs_off();
3013
3014         this_cpu_enable_ftrace();
3015
3016         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3017
3018         set_irq_happened(trap);
3019
3020         spin_lock(&vc->lock);
3021         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3022         vc->vcore_state = VCORE_EXITING;
3023
3024         /* wait for secondary threads to finish writing their state to memory */
3025         kvmppc_wait_for_nap(controlled_threads);
3026
3027         /* Return to whole-core mode if we split the core earlier */
3028         if (cmd_bit) {
3029                 unsigned long hid0 = mfspr(SPRN_HID0);
3030                 unsigned long loops = 0;
3031
3032                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3033                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3034                 mb();
3035                 mtspr(SPRN_HID0, hid0);
3036                 isync();
3037                 for (;;) {
3038                         hid0 = mfspr(SPRN_HID0);
3039                         if (!(hid0 & stat_bit))
3040                                 break;
3041                         cpu_relax();
3042                         ++loops;
3043                 }
3044         } else if (hpt_on_radix) {
3045                 /* Wait for all threads to have seen final sync */
3046                 for (thr = 1; thr < controlled_threads; ++thr) {
3047                         struct paca_struct *paca = paca_ptrs[pcpu + thr];
3048
3049                         while (paca->kvm_hstate.kvm_split_mode) {
3050                                 HMT_low();
3051                                 barrier();
3052                         }
3053                         HMT_medium();
3054                 }
3055         }
3056         split_info.do_nap = 0;
3057
3058         kvmppc_set_host_core(pcpu);
3059
3060         local_irq_enable();
3061         guest_exit();
3062
3063         /* Let secondaries go back to the offline loop */
3064         for (i = 0; i < controlled_threads; ++i) {
3065                 kvmppc_release_hwthread(pcpu + i);
3066                 if (sip && sip->napped[i])
3067                         kvmppc_ipi_thread(pcpu + i);
3068                 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3069         }
3070
3071         spin_unlock(&vc->lock);
3072
3073         /* make sure updates to secondary vcpu structs are visible now */
3074         smp_mb();
3075
3076         preempt_enable();
3077
3078         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3079                 pvc = core_info.vc[sub];
3080                 post_guest_process(pvc, pvc == vc);
3081         }
3082
3083         spin_lock(&vc->lock);
3084
3085  out:
3086         vc->vcore_state = VCORE_INACTIVE;
3087         trace_kvmppc_run_core(vc, 1);
3088 }
3089
3090 /*
3091  * Wait for some other vcpu thread to execute us, and
3092  * wake us up when we need to handle something in the host.
3093  */
3094 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3095                                  struct kvm_vcpu *vcpu, int wait_state)
3096 {
3097         DEFINE_WAIT(wait);
3098
3099         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3100         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3101                 spin_unlock(&vc->lock);
3102                 schedule();
3103                 spin_lock(&vc->lock);
3104         }
3105         finish_wait(&vcpu->arch.cpu_run, &wait);
3106 }
3107
3108 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3109 {
3110         /* 10us base */
3111         if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
3112                 vc->halt_poll_ns = 10000;
3113         else
3114                 vc->halt_poll_ns *= halt_poll_ns_grow;
3115 }
3116
3117 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3118 {
3119         if (halt_poll_ns_shrink == 0)
3120                 vc->halt_poll_ns = 0;
3121         else
3122                 vc->halt_poll_ns /= halt_poll_ns_shrink;
3123 }
3124
3125 #ifdef CONFIG_KVM_XICS
3126 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3127 {
3128         if (!xive_enabled())
3129                 return false;
3130         return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3131                 vcpu->arch.xive_saved_state.cppr;
3132 }
3133 #else
3134 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3135 {
3136         return false;
3137 }
3138 #endif /* CONFIG_KVM_XICS */
3139
3140 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3141 {
3142         if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3143             kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3144                 return true;
3145
3146         return false;
3147 }
3148
3149 /*
3150  * Check to see if any of the runnable vcpus on the vcore have pending
3151  * exceptions or are no longer ceded
3152  */
3153 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3154 {
3155         struct kvm_vcpu *vcpu;
3156         int i;
3157
3158         for_each_runnable_thread(i, vcpu, vc) {
3159                 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3160                         return 1;
3161         }
3162
3163         return 0;
3164 }
3165
3166 /*
3167  * All the vcpus in this vcore are idle, so wait for a decrementer
3168  * or external interrupt to one of the vcpus.  vc->lock is held.
3169  */
3170 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3171 {
3172         ktime_t cur, start_poll, start_wait;
3173         int do_sleep = 1;
3174         u64 block_ns;
3175         DECLARE_SWAITQUEUE(wait);
3176
3177         /* Poll for pending exceptions and ceded state */
3178         cur = start_poll = ktime_get();
3179         if (vc->halt_poll_ns) {
3180                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3181                 ++vc->runner->stat.halt_attempted_poll;
3182
3183                 vc->vcore_state = VCORE_POLLING;
3184                 spin_unlock(&vc->lock);
3185
3186                 do {
3187                         if (kvmppc_vcore_check_block(vc)) {
3188                                 do_sleep = 0;
3189                                 break;
3190                         }
3191                         cur = ktime_get();
3192                 } while (single_task_running() && ktime_before(cur, stop));
3193
3194                 spin_lock(&vc->lock);
3195                 vc->vcore_state = VCORE_INACTIVE;
3196
3197                 if (!do_sleep) {
3198                         ++vc->runner->stat.halt_successful_poll;
3199                         goto out;
3200                 }
3201         }
3202
3203         prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3204
3205         if (kvmppc_vcore_check_block(vc)) {
3206                 finish_swait(&vc->wq, &wait);
3207                 do_sleep = 0;
3208                 /* If we polled, count this as a successful poll */
3209                 if (vc->halt_poll_ns)
3210                         ++vc->runner->stat.halt_successful_poll;
3211                 goto out;
3212         }
3213
3214         start_wait = ktime_get();
3215
3216         vc->vcore_state = VCORE_SLEEPING;
3217         trace_kvmppc_vcore_blocked(vc, 0);
3218         spin_unlock(&vc->lock);
3219         schedule();
3220         finish_swait(&vc->wq, &wait);
3221         spin_lock(&vc->lock);
3222         vc->vcore_state = VCORE_INACTIVE;
3223         trace_kvmppc_vcore_blocked(vc, 1);
3224         ++vc->runner->stat.halt_successful_wait;
3225
3226         cur = ktime_get();
3227
3228 out:
3229         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3230
3231         /* Attribute wait time */
3232         if (do_sleep) {
3233                 vc->runner->stat.halt_wait_ns +=
3234                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
3235                 /* Attribute failed poll time */
3236                 if (vc->halt_poll_ns)
3237                         vc->runner->stat.halt_poll_fail_ns +=
3238                                 ktime_to_ns(start_wait) -
3239                                 ktime_to_ns(start_poll);
3240         } else {
3241                 /* Attribute successful poll time */
3242                 if (vc->halt_poll_ns)
3243                         vc->runner->stat.halt_poll_success_ns +=
3244                                 ktime_to_ns(cur) -
3245                                 ktime_to_ns(start_poll);
3246         }
3247
3248         /* Adjust poll time */
3249         if (halt_poll_ns) {
3250                 if (block_ns <= vc->halt_poll_ns)
3251                         ;
3252                 /* We slept and blocked for longer than the max halt time */
3253                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3254                         shrink_halt_poll_ns(vc);
3255                 /* We slept and our poll time is too small */
3256                 else if (vc->halt_poll_ns < halt_poll_ns &&
3257                                 block_ns < halt_poll_ns)
3258                         grow_halt_poll_ns(vc);
3259                 if (vc->halt_poll_ns > halt_poll_ns)
3260                         vc->halt_poll_ns = halt_poll_ns;
3261         } else
3262                 vc->halt_poll_ns = 0;
3263
3264         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3265 }
3266
3267 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3268 {
3269         int r = 0;
3270         struct kvm *kvm = vcpu->kvm;
3271
3272         mutex_lock(&kvm->lock);
3273         if (!kvm->arch.mmu_ready) {
3274                 if (!kvm_is_radix(kvm))
3275                         r = kvmppc_hv_setup_htab_rma(vcpu);
3276                 if (!r) {
3277                         if (cpu_has_feature(CPU_FTR_ARCH_300))
3278                                 kvmppc_setup_partition_table(kvm);
3279                         kvm->arch.mmu_ready = 1;
3280                 }
3281         }
3282         mutex_unlock(&kvm->lock);
3283         return r;
3284 }
3285
3286 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3287 {
3288         int n_ceded, i, r;
3289         struct kvmppc_vcore *vc;
3290         struct kvm_vcpu *v;
3291
3292         trace_kvmppc_run_vcpu_enter(vcpu);
3293
3294         kvm_run->exit_reason = 0;
3295         vcpu->arch.ret = RESUME_GUEST;
3296         vcpu->arch.trap = 0;
3297         kvmppc_update_vpas(vcpu);
3298
3299         /*
3300          * Synchronize with other threads in this virtual core
3301          */
3302         vc = vcpu->arch.vcore;
3303         spin_lock(&vc->lock);
3304         vcpu->arch.ceded = 0;
3305         vcpu->arch.run_task = current;
3306         vcpu->arch.kvm_run = kvm_run;
3307         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3308         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3309         vcpu->arch.busy_preempt = TB_NIL;
3310         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3311         ++vc->n_runnable;
3312
3313         /*
3314          * This happens the first time this is called for a vcpu.
3315          * If the vcore is already running, we may be able to start
3316          * this thread straight away and have it join in.
3317          */
3318         if (!signal_pending(current)) {
3319                 if ((vc->vcore_state == VCORE_PIGGYBACK ||
3320                      vc->vcore_state == VCORE_RUNNING) &&
3321                            !VCORE_IS_EXITING(vc)) {
3322                         kvmppc_create_dtl_entry(vcpu, vc);
3323                         kvmppc_start_thread(vcpu, vc);
3324                         trace_kvm_guest_enter(vcpu);
3325                 } else if (vc->vcore_state == VCORE_SLEEPING) {
3326                         swake_up_one(&vc->wq);
3327                 }
3328
3329         }
3330
3331         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3332                !signal_pending(current)) {
3333                 /* See if the MMU is ready to go */
3334                 if (!vcpu->kvm->arch.mmu_ready) {
3335                         spin_unlock(&vc->lock);
3336                         r = kvmhv_setup_mmu(vcpu);
3337                         spin_lock(&vc->lock);
3338                         if (r) {
3339                                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3340                                 kvm_run->fail_entry.
3341                                         hardware_entry_failure_reason = 0;
3342                                 vcpu->arch.ret = r;
3343                                 break;
3344                         }
3345                 }
3346
3347                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3348                         kvmppc_vcore_end_preempt(vc);
3349
3350                 if (vc->vcore_state != VCORE_INACTIVE) {
3351                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3352                         continue;
3353                 }
3354                 for_each_runnable_thread(i, v, vc) {
3355                         kvmppc_core_prepare_to_enter(v);
3356                         if (signal_pending(v->arch.run_task)) {
3357                                 kvmppc_remove_runnable(vc, v);
3358                                 v->stat.signal_exits++;
3359                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3360                                 v->arch.ret = -EINTR;
3361                                 wake_up(&v->arch.cpu_run);
3362                         }
3363                 }
3364                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3365                         break;
3366                 n_ceded = 0;
3367                 for_each_runnable_thread(i, v, vc) {
3368                         if (!kvmppc_vcpu_woken(v))
3369                                 n_ceded += v->arch.ceded;
3370                         else
3371                                 v->arch.ceded = 0;
3372                 }
3373                 vc->runner = vcpu;
3374                 if (n_ceded == vc->n_runnable) {
3375                         kvmppc_vcore_blocked(vc);
3376                 } else if (need_resched()) {
3377                         kvmppc_vcore_preempt(vc);
3378                         /* Let something else run */
3379                         cond_resched_lock(&vc->lock);
3380                         if (vc->vcore_state == VCORE_PREEMPT)
3381                                 kvmppc_vcore_end_preempt(vc);
3382                 } else {
3383                         kvmppc_run_core(vc);
3384                 }
3385                 vc->runner = NULL;
3386         }
3387
3388         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3389                (vc->vcore_state == VCORE_RUNNING ||
3390                 vc->vcore_state == VCORE_EXITING ||
3391                 vc->vcore_state == VCORE_PIGGYBACK))
3392                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3393
3394         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3395                 kvmppc_vcore_end_preempt(vc);
3396
3397         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3398                 kvmppc_remove_runnable(vc, vcpu);
3399                 vcpu->stat.signal_exits++;
3400                 kvm_run->exit_reason = KVM_EXIT_INTR;
3401                 vcpu->arch.ret = -EINTR;
3402         }
3403
3404         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3405                 /* Wake up some vcpu to run the core */
3406                 i = -1;
3407                 v = next_runnable_thread(vc, &i);
3408                 wake_up(&v->arch.cpu_run);
3409         }
3410
3411         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3412         spin_unlock(&vc->lock);
3413         return vcpu->arch.ret;
3414 }
3415
3416 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
3417 {
3418         int r;
3419         int srcu_idx;
3420         unsigned long ebb_regs[3] = {}; /* shut up GCC */
3421         unsigned long user_tar = 0;
3422         unsigned int user_vrsave;
3423         struct kvm *kvm;
3424
3425         if (!vcpu->arch.sane) {
3426                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3427                 return -EINVAL;
3428         }
3429
3430         /*
3431          * Don't allow entry with a suspended transaction, because
3432          * the guest entry/exit code will lose it.
3433          * If the guest has TM enabled, save away their TM-related SPRs
3434          * (they will get restored by the TM unavailable interrupt).
3435          */
3436 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3437         if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
3438             (current->thread.regs->msr & MSR_TM)) {
3439                 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
3440                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3441                         run->fail_entry.hardware_entry_failure_reason = 0;
3442                         return -EINVAL;
3443                 }
3444                 /* Enable TM so we can read the TM SPRs */
3445                 mtmsr(mfmsr() | MSR_TM);
3446                 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
3447                 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
3448                 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
3449                 current->thread.regs->msr &= ~MSR_TM;
3450         }
3451 #endif
3452
3453         /*
3454          * Force online to 1 for the sake of old userspace which doesn't
3455          * set it.
3456          */
3457         if (!vcpu->arch.online) {
3458                 atomic_inc(&vcpu->arch.vcore->online_count);
3459                 vcpu->arch.online = 1;
3460         }
3461
3462         kvmppc_core_prepare_to_enter(vcpu);
3463
3464         /* No need to go into the guest when all we'll do is come back out */
3465         if (signal_pending(current)) {
3466                 run->exit_reason = KVM_EXIT_INTR;
3467                 return -EINTR;
3468         }
3469
3470         kvm = vcpu->kvm;
3471         atomic_inc(&kvm->arch.vcpus_running);
3472         /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
3473         smp_mb();
3474
3475         flush_all_to_thread(current);
3476
3477         /* Save userspace EBB and other register values */
3478         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3479                 ebb_regs[0] = mfspr(SPRN_EBBHR);
3480                 ebb_regs[1] = mfspr(SPRN_EBBRR);
3481                 ebb_regs[2] = mfspr(SPRN_BESCR);
3482                 user_tar = mfspr(SPRN_TAR);
3483         }
3484         user_vrsave = mfspr(SPRN_VRSAVE);
3485
3486         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
3487         vcpu->arch.pgdir = current->mm->pgd;
3488         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3489
3490         do {
3491                 r = kvmppc_run_vcpu(run, vcpu);
3492
3493                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
3494                     !(vcpu->arch.shregs.msr & MSR_PR)) {
3495                         trace_kvm_hcall_enter(vcpu);
3496                         r = kvmppc_pseries_do_hcall(vcpu);
3497                         trace_kvm_hcall_exit(vcpu, r);
3498                         kvmppc_core_prepare_to_enter(vcpu);
3499                 } else if (r == RESUME_PAGE_FAULT) {
3500                         srcu_idx = srcu_read_lock(&kvm->srcu);
3501                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
3502                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
3503                         srcu_read_unlock(&kvm->srcu, srcu_idx);
3504                 } else if (r == RESUME_PASSTHROUGH) {
3505                         if (WARN_ON(xive_enabled()))
3506                                 r = H_SUCCESS;
3507                         else
3508                                 r = kvmppc_xics_rm_complete(vcpu, 0);
3509                 }
3510         } while (is_kvmppc_resume_guest(r));
3511
3512         /* Restore userspace EBB and other register values */
3513         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3514                 mtspr(SPRN_EBBHR, ebb_regs[0]);
3515                 mtspr(SPRN_EBBRR, ebb_regs[1]);
3516                 mtspr(SPRN_BESCR, ebb_regs[2]);
3517                 mtspr(SPRN_TAR, user_tar);
3518                 mtspr(SPRN_FSCR, current->thread.fscr);
3519         }
3520         mtspr(SPRN_VRSAVE, user_vrsave);
3521
3522         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
3523         atomic_dec(&kvm->arch.vcpus_running);
3524         return r;
3525 }
3526
3527 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
3528                                      int shift, int sllp)
3529 {
3530         (*sps)->page_shift = shift;
3531         (*sps)->slb_enc = sllp;
3532         (*sps)->enc[0].page_shift = shift;
3533         (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
3534         /*
3535          * Add 16MB MPSS support (may get filtered out by userspace)
3536          */
3537         if (shift != 24) {
3538                 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
3539                 if (penc != -1) {
3540                         (*sps)->enc[1].page_shift = 24;
3541                         (*sps)->enc[1].pte_enc = penc;
3542                 }
3543         }
3544         (*sps)++;
3545 }
3546
3547 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
3548                                          struct kvm_ppc_smmu_info *info)
3549 {
3550         struct kvm_ppc_one_seg_page_size *sps;
3551
3552         /*
3553          * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
3554          * POWER7 doesn't support keys for instruction accesses,
3555          * POWER8 and POWER9 do.
3556          */
3557         info->data_keys = 32;
3558         info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
3559
3560         /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
3561         info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
3562         info->slb_size = 32;
3563
3564         /* We only support these sizes for now, and no muti-size segments */
3565         sps = &info->sps[0];
3566         kvmppc_add_seg_page_size(&sps, 12, 0);
3567         kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
3568         kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
3569
3570         return 0;
3571 }
3572
3573 /*
3574  * Get (and clear) the dirty memory log for a memory slot.
3575  */
3576 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
3577                                          struct kvm_dirty_log *log)
3578 {
3579         struct kvm_memslots *slots;
3580         struct kvm_memory_slot *memslot;
3581         int i, r;
3582         unsigned long n;
3583         unsigned long *buf, *p;
3584         struct kvm_vcpu *vcpu;
3585
3586         mutex_lock(&kvm->slots_lock);
3587
3588         r = -EINVAL;
3589         if (log->slot >= KVM_USER_MEM_SLOTS)
3590                 goto out;
3591
3592         slots = kvm_memslots(kvm);
3593         memslot = id_to_memslot(slots, log->slot);
3594         r = -ENOENT;
3595         if (!memslot->dirty_bitmap)
3596                 goto out;
3597
3598         /*
3599          * Use second half of bitmap area because both HPT and radix
3600          * accumulate bits in the first half.
3601          */
3602         n = kvm_dirty_bitmap_bytes(memslot);
3603         buf = memslot->dirty_bitmap + n / sizeof(long);
3604         memset(buf, 0, n);
3605
3606         if (kvm_is_radix(kvm))
3607                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
3608         else
3609                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
3610         if (r)
3611                 goto out;
3612
3613         /*
3614          * We accumulate dirty bits in the first half of the
3615          * memslot's dirty_bitmap area, for when pages are paged
3616          * out or modified by the host directly.  Pick up these
3617          * bits and add them to the map.
3618          */
3619         p = memslot->dirty_bitmap;
3620         for (i = 0; i < n / sizeof(long); ++i)
3621                 buf[i] |= xchg(&p[i], 0);
3622
3623         /* Harvest dirty bits from VPA and DTL updates */
3624         /* Note: we never modify the SLB shadow buffer areas */
3625         kvm_for_each_vcpu(i, vcpu, kvm) {
3626                 spin_lock(&vcpu->arch.vpa_update_lock);
3627                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
3628                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
3629                 spin_unlock(&vcpu->arch.vpa_update_lock);
3630         }
3631
3632         r = -EFAULT;
3633         if (copy_to_user(log->dirty_bitmap, buf, n))
3634                 goto out;
3635
3636         r = 0;
3637 out:
3638         mutex_unlock(&kvm->slots_lock);
3639         return r;
3640 }
3641
3642 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
3643                                         struct kvm_memory_slot *dont)
3644 {
3645         if (!dont || free->arch.rmap != dont->arch.rmap) {
3646                 vfree(free->arch.rmap);
3647                 free->arch.rmap = NULL;
3648         }
3649 }
3650
3651 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3652                                          unsigned long npages)
3653 {
3654         slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
3655         if (!slot->arch.rmap)
3656                 return -ENOMEM;
3657
3658         return 0;
3659 }
3660
3661 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3662                                         struct kvm_memory_slot *memslot,
3663                                         const struct kvm_userspace_memory_region *mem)
3664 {
3665         return 0;
3666 }
3667
3668 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3669                                 const struct kvm_userspace_memory_region *mem,
3670                                 const struct kvm_memory_slot *old,
3671                                 const struct kvm_memory_slot *new)
3672 {
3673         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3674
3675         /*
3676          * If we are making a new memslot, it might make
3677          * some address that was previously cached as emulated
3678          * MMIO be no longer emulated MMIO, so invalidate
3679          * all the caches of emulated MMIO translations.
3680          */
3681         if (npages)
3682                 atomic64_inc(&kvm->arch.mmio_update);
3683 }
3684
3685 /*
3686  * Update LPCR values in kvm->arch and in vcores.
3687  * Caller must hold kvm->lock.
3688  */
3689 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3690 {
3691         long int i;
3692         u32 cores_done = 0;
3693
3694         if ((kvm->arch.lpcr & mask) == lpcr)
3695                 return;
3696
3697         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3698
3699         for (i = 0; i < KVM_MAX_VCORES; ++i) {
3700                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3701                 if (!vc)
3702                         continue;
3703                 spin_lock(&vc->lock);
3704                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3705                 spin_unlock(&vc->lock);
3706                 if (++cores_done >= kvm->arch.online_vcores)
3707                         break;
3708         }
3709 }
3710
3711 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3712 {
3713         return;
3714 }
3715
3716 void kvmppc_setup_partition_table(struct kvm *kvm)
3717 {
3718         unsigned long dw0, dw1;
3719
3720         if (!kvm_is_radix(kvm)) {
3721                 /* PS field - page size for VRMA */
3722                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
3723                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
3724                 /* HTABSIZE and HTABORG fields */
3725                 dw0 |= kvm->arch.sdr1;
3726
3727                 /* Second dword as set by userspace */
3728                 dw1 = kvm->arch.process_table;
3729         } else {
3730                 dw0 = PATB_HR | radix__get_tree_size() |
3731                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
3732                 dw1 = PATB_GR | kvm->arch.process_table;
3733         }
3734
3735         mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
3736 }
3737
3738 /*
3739  * Set up HPT (hashed page table) and RMA (real-mode area).
3740  * Must be called with kvm->lock held.
3741  */
3742 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3743 {
3744         int err = 0;
3745         struct kvm *kvm = vcpu->kvm;
3746         unsigned long hva;
3747         struct kvm_memory_slot *memslot;
3748         struct vm_area_struct *vma;
3749         unsigned long lpcr = 0, senc;
3750         unsigned long psize, porder;
3751         int srcu_idx;
3752
3753         /* Allocate hashed page table (if not done already) and reset it */
3754         if (!kvm->arch.hpt.virt) {
3755                 int order = KVM_DEFAULT_HPT_ORDER;
3756                 struct kvm_hpt_info info;
3757
3758                 err = kvmppc_allocate_hpt(&info, order);
3759                 /* If we get here, it means userspace didn't specify a
3760                  * size explicitly.  So, try successively smaller
3761                  * sizes if the default failed. */
3762                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
3763                         err  = kvmppc_allocate_hpt(&info, order);
3764
3765                 if (err < 0) {
3766                         pr_err("KVM: Couldn't alloc HPT\n");
3767                         goto out;
3768                 }
3769
3770                 kvmppc_set_hpt(kvm, &info);
3771         }
3772
3773         /* Look up the memslot for guest physical address 0 */
3774         srcu_idx = srcu_read_lock(&kvm->srcu);
3775         memslot = gfn_to_memslot(kvm, 0);
3776
3777         /* We must have some memory at 0 by now */
3778         err = -EINVAL;
3779         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3780                 goto out_srcu;
3781
3782         /* Look up the VMA for the start of this memory slot */
3783         hva = memslot->userspace_addr;
3784         down_read(&current->mm->mmap_sem);
3785         vma = find_vma(current->mm, hva);
3786         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3787                 goto up_out;
3788
3789         psize = vma_kernel_pagesize(vma);
3790
3791         up_read(&current->mm->mmap_sem);
3792
3793         /* We can handle 4k, 64k or 16M pages in the VRMA */
3794         if (psize >= 0x1000000)
3795                 psize = 0x1000000;
3796         else if (psize >= 0x10000)
3797                 psize = 0x10000;
3798         else
3799                 psize = 0x1000;
3800         porder = __ilog2(psize);
3801
3802         senc = slb_pgsize_encoding(psize);
3803         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3804                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3805         /* Create HPTEs in the hash page table for the VRMA */
3806         kvmppc_map_vrma(vcpu, memslot, porder);
3807
3808         /* Update VRMASD field in the LPCR */
3809         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3810                 /* the -4 is to account for senc values starting at 0x10 */
3811                 lpcr = senc << (LPCR_VRMASD_SH - 4);
3812                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3813         }
3814
3815         /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
3816         smp_wmb();
3817         err = 0;
3818  out_srcu:
3819         srcu_read_unlock(&kvm->srcu, srcu_idx);
3820  out:
3821         return err;
3822
3823  up_out:
3824         up_read(&current->mm->mmap_sem);
3825         goto out_srcu;
3826 }
3827
3828 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
3829 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
3830 {
3831         kvmppc_rmap_reset(kvm);
3832         kvm->arch.process_table = 0;
3833         /* Mutual exclusion with kvm_unmap_hva_range etc. */
3834         spin_lock(&kvm->mmu_lock);
3835         kvm->arch.radix = 0;
3836         spin_unlock(&kvm->mmu_lock);
3837         kvmppc_free_radix(kvm);
3838         kvmppc_update_lpcr(kvm, LPCR_VPM1,
3839                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
3840         return 0;
3841 }
3842
3843 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
3844 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
3845 {
3846         int err;
3847
3848         err = kvmppc_init_vm_radix(kvm);
3849         if (err)
3850                 return err;
3851
3852         kvmppc_rmap_reset(kvm);
3853         /* Mutual exclusion with kvm_unmap_hva_range etc. */
3854         spin_lock(&kvm->mmu_lock);
3855         kvm->arch.radix = 1;
3856         spin_unlock(&kvm->mmu_lock);
3857         kvmppc_free_hpt(&kvm->arch.hpt);
3858         kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
3859                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
3860         return 0;
3861 }
3862
3863 #ifdef CONFIG_KVM_XICS
3864 /*
3865  * Allocate a per-core structure for managing state about which cores are
3866  * running in the host versus the guest and for exchanging data between
3867  * real mode KVM and CPU running in the host.
3868  * This is only done for the first VM.
3869  * The allocated structure stays even if all VMs have stopped.
3870  * It is only freed when the kvm-hv module is unloaded.
3871  * It's OK for this routine to fail, we just don't support host
3872  * core operations like redirecting H_IPI wakeups.
3873  */
3874 void kvmppc_alloc_host_rm_ops(void)
3875 {
3876         struct kvmppc_host_rm_ops *ops;
3877         unsigned long l_ops;
3878         int cpu, core;
3879         int size;
3880
3881         /* Not the first time here ? */
3882         if (kvmppc_host_rm_ops_hv != NULL)
3883                 return;
3884
3885         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3886         if (!ops)
3887                 return;
3888
3889         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3890         ops->rm_core = kzalloc(size, GFP_KERNEL);
3891
3892         if (!ops->rm_core) {
3893                 kfree(ops);
3894                 return;
3895         }
3896
3897         cpus_read_lock();
3898
3899         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3900                 if (!cpu_online(cpu))
3901                         continue;
3902
3903                 core = cpu >> threads_shift;
3904                 ops->rm_core[core].rm_state.in_host = 1;
3905         }
3906
3907         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3908
3909         /*
3910          * Make the contents of the kvmppc_host_rm_ops structure visible
3911          * to other CPUs before we assign it to the global variable.
3912          * Do an atomic assignment (no locks used here), but if someone
3913          * beats us to it, just free our copy and return.
3914          */
3915         smp_wmb();
3916         l_ops = (unsigned long) ops;
3917
3918         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3919                 cpus_read_unlock();
3920                 kfree(ops->rm_core);
3921                 kfree(ops);
3922                 return;
3923         }
3924
3925         cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
3926                                              "ppc/kvm_book3s:prepare",
3927                                              kvmppc_set_host_core,
3928                                              kvmppc_clear_host_core);
3929         cpus_read_unlock();
3930 }
3931
3932 void kvmppc_free_host_rm_ops(void)
3933 {
3934         if (kvmppc_host_rm_ops_hv) {
3935                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
3936                 kfree(kvmppc_host_rm_ops_hv->rm_core);
3937                 kfree(kvmppc_host_rm_ops_hv);
3938                 kvmppc_host_rm_ops_hv = NULL;
3939         }
3940 }
3941 #endif
3942
3943 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3944 {
3945         unsigned long lpcr, lpid;
3946         char buf[32];
3947         int ret;
3948
3949         /* Allocate the guest's logical partition ID */
3950
3951         lpid = kvmppc_alloc_lpid();
3952         if ((long)lpid < 0)
3953                 return -ENOMEM;
3954         kvm->arch.lpid = lpid;
3955
3956         kvmppc_alloc_host_rm_ops();
3957
3958         /*
3959          * Since we don't flush the TLB when tearing down a VM,
3960          * and this lpid might have previously been used,
3961          * make sure we flush on each core before running the new VM.
3962          * On POWER9, the tlbie in mmu_partition_table_set_entry()
3963          * does this flush for us.
3964          */
3965         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3966                 cpumask_setall(&kvm->arch.need_tlb_flush);
3967
3968         /* Start out with the default set of hcalls enabled */
3969         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3970                sizeof(kvm->arch.enabled_hcalls));
3971
3972         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3973                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3974
3975         /* Init LPCR for virtual RMA mode */
3976         kvm->arch.host_lpid = mfspr(SPRN_LPID);
3977         kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3978         lpcr &= LPCR_PECE | LPCR_LPES;
3979         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3980                 LPCR_VPM0 | LPCR_VPM1;
3981         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3982                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3983         /* On POWER8 turn on online bit to enable PURR/SPURR */
3984         if (cpu_has_feature(CPU_FTR_ARCH_207S))
3985                 lpcr |= LPCR_ONL;
3986         /*
3987          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
3988          * Set HVICE bit to enable hypervisor virtualization interrupts.
3989          * Set HEIC to prevent OS interrupts to go to hypervisor (should
3990          * be unnecessary but better safe than sorry in case we re-enable
3991          * EE in HV mode with this LPCR still set)
3992          */
3993         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3994                 lpcr &= ~LPCR_VPM0;
3995                 lpcr |= LPCR_HVICE | LPCR_HEIC;
3996
3997                 /*
3998                  * If xive is enabled, we route 0x500 interrupts directly
3999                  * to the guest.
4000                  */
4001                 if (xive_enabled())
4002                         lpcr |= LPCR_LPES;
4003         }
4004
4005         /*
4006          * If the host uses radix, the guest starts out as radix.
4007          */
4008         if (radix_enabled()) {
4009                 kvm->arch.radix = 1;
4010                 kvm->arch.mmu_ready = 1;
4011                 lpcr &= ~LPCR_VPM1;
4012                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4013                 ret = kvmppc_init_vm_radix(kvm);
4014                 if (ret) {
4015                         kvmppc_free_lpid(kvm->arch.lpid);
4016                         return ret;
4017                 }
4018                 kvmppc_setup_partition_table(kvm);
4019         }
4020
4021         kvm->arch.lpcr = lpcr;
4022
4023         /* Initialization for future HPT resizes */
4024         kvm->arch.resize_hpt = NULL;
4025
4026         /*
4027          * Work out how many sets the TLB has, for the use of
4028          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4029          */
4030         if (radix_enabled())
4031                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
4032         else if (cpu_has_feature(CPU_FTR_ARCH_300))
4033                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
4034         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4035                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
4036         else
4037                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
4038
4039         /*
4040          * Track that we now have a HV mode VM active. This blocks secondary
4041          * CPU threads from coming online.
4042          * On POWER9, we only need to do this if the "indep_threads_mode"
4043          * module parameter has been set to N.
4044          */
4045         if (cpu_has_feature(CPU_FTR_ARCH_300))
4046                 kvm->arch.threads_indep = indep_threads_mode;
4047         if (!kvm->arch.threads_indep)
4048                 kvm_hv_vm_activated();
4049
4050         /*
4051          * Initialize smt_mode depending on processor.
4052          * POWER8 and earlier have to use "strict" threading, where
4053          * all vCPUs in a vcore have to run on the same (sub)core,
4054          * whereas on POWER9 the threads can each run a different
4055          * guest.
4056          */
4057         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4058                 kvm->arch.smt_mode = threads_per_subcore;
4059         else
4060                 kvm->arch.smt_mode = 1;
4061         kvm->arch.emul_smt_mode = 1;
4062
4063         /*
4064          * Create a debugfs directory for the VM
4065          */
4066         snprintf(buf, sizeof(buf), "vm%d", current->pid);
4067         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4068         kvmppc_mmu_debugfs_init(kvm);
4069
4070         return 0;
4071 }
4072
4073 static void kvmppc_free_vcores(struct kvm *kvm)
4074 {
4075         long int i;
4076
4077         for (i = 0; i < KVM_MAX_VCORES; ++i)
4078                 kfree(kvm->arch.vcores[i]);
4079         kvm->arch.online_vcores = 0;
4080 }
4081
4082 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4083 {
4084         debugfs_remove_recursive(kvm->arch.debugfs_dir);
4085
4086         if (!kvm->arch.threads_indep)
4087                 kvm_hv_vm_deactivated();
4088
4089         kvmppc_free_vcores(kvm);
4090
4091         kvmppc_free_lpid(kvm->arch.lpid);
4092
4093         if (kvm_is_radix(kvm))
4094                 kvmppc_free_radix(kvm);
4095         else
4096                 kvmppc_free_hpt(&kvm->arch.hpt);
4097
4098         kvmppc_free_pimap(kvm);
4099 }
4100
4101 /* We don't need to emulate any privileged instructions or dcbz */
4102 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4103                                      unsigned int inst, int *advance)
4104 {
4105         return EMULATE_FAIL;
4106 }
4107
4108 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4109                                         ulong spr_val)
4110 {
4111         return EMULATE_FAIL;
4112 }
4113
4114 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4115                                         ulong *spr_val)
4116 {
4117         return EMULATE_FAIL;
4118 }
4119
4120 static int kvmppc_core_check_processor_compat_hv(void)
4121 {
4122         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
4123             !cpu_has_feature(CPU_FTR_ARCH_206))
4124                 return -EIO;
4125
4126         return 0;
4127 }
4128
4129 #ifdef CONFIG_KVM_XICS
4130
4131 void kvmppc_free_pimap(struct kvm *kvm)
4132 {
4133         kfree(kvm->arch.pimap);
4134 }
4135
4136 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
4137 {
4138         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
4139 }
4140
4141 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4142 {
4143         struct irq_desc *desc;
4144         struct kvmppc_irq_map *irq_map;
4145         struct kvmppc_passthru_irqmap *pimap;
4146         struct irq_chip *chip;
4147         int i, rc = 0;
4148
4149         if (!kvm_irq_bypass)
4150                 return 1;
4151
4152         desc = irq_to_desc(host_irq);
4153         if (!desc)
4154                 return -EIO;
4155
4156         mutex_lock(&kvm->lock);
4157
4158         pimap = kvm->arch.pimap;
4159         if (pimap == NULL) {
4160                 /* First call, allocate structure to hold IRQ map */
4161                 pimap = kvmppc_alloc_pimap();
4162                 if (pimap == NULL) {
4163                         mutex_unlock(&kvm->lock);
4164                         return -ENOMEM;
4165                 }
4166                 kvm->arch.pimap = pimap;
4167         }
4168
4169         /*
4170          * For now, we only support interrupts for which the EOI operation
4171          * is an OPAL call followed by a write to XIRR, since that's
4172          * what our real-mode EOI code does, or a XIVE interrupt
4173          */
4174         chip = irq_data_get_irq_chip(&desc->irq_data);
4175         if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
4176                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
4177                         host_irq, guest_gsi);
4178                 mutex_unlock(&kvm->lock);
4179                 return -ENOENT;
4180         }
4181
4182         /*
4183          * See if we already have an entry for this guest IRQ number.
4184          * If it's mapped to a hardware IRQ number, that's an error,
4185          * otherwise re-use this entry.
4186          */
4187         for (i = 0; i < pimap->n_mapped; i++) {
4188                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
4189                         if (pimap->mapped[i].r_hwirq) {
4190                                 mutex_unlock(&kvm->lock);
4191                                 return -EINVAL;
4192                         }
4193                         break;
4194                 }
4195         }
4196
4197         if (i == KVMPPC_PIRQ_MAPPED) {
4198                 mutex_unlock(&kvm->lock);
4199                 return -EAGAIN;         /* table is full */
4200         }
4201
4202         irq_map = &pimap->mapped[i];
4203
4204         irq_map->v_hwirq = guest_gsi;
4205         irq_map->desc = desc;
4206
4207         /*
4208          * Order the above two stores before the next to serialize with
4209          * the KVM real mode handler.
4210          */
4211         smp_wmb();
4212         irq_map->r_hwirq = desc->irq_data.hwirq;
4213
4214         if (i == pimap->n_mapped)
4215                 pimap->n_mapped++;
4216
4217         if (xive_enabled())
4218                 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
4219         else
4220                 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
4221         if (rc)
4222                 irq_map->r_hwirq = 0;
4223
4224         mutex_unlock(&kvm->lock);
4225
4226         return 0;
4227 }
4228
4229 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4230 {
4231         struct irq_desc *desc;
4232         struct kvmppc_passthru_irqmap *pimap;
4233         int i, rc = 0;
4234
4235         if (!kvm_irq_bypass)
4236                 return 0;
4237
4238         desc = irq_to_desc(host_irq);
4239         if (!desc)
4240                 return -EIO;
4241
4242         mutex_lock(&kvm->lock);
4243         if (!kvm->arch.pimap)
4244                 goto unlock;
4245
4246         pimap = kvm->arch.pimap;
4247
4248         for (i = 0; i < pimap->n_mapped; i++) {
4249                 if (guest_gsi == pimap->mapped[i].v_hwirq)
4250                         break;
4251         }
4252
4253         if (i == pimap->n_mapped) {
4254                 mutex_unlock(&kvm->lock);
4255                 return -ENODEV;
4256         }
4257
4258         if (xive_enabled())
4259                 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
4260         else
4261                 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
4262
4263         /* invalidate the entry (what do do on error from the above ?) */
4264         pimap->mapped[i].r_hwirq = 0;
4265
4266         /*
4267          * We don't free this structure even when the count goes to
4268          * zero. The structure is freed when we destroy the VM.
4269          */
4270  unlock:
4271         mutex_unlock(&kvm->lock);
4272         return rc;
4273 }
4274
4275 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
4276                                              struct irq_bypass_producer *prod)
4277 {
4278         int ret = 0;
4279         struct kvm_kernel_irqfd *irqfd =
4280                 container_of(cons, struct kvm_kernel_irqfd, consumer);
4281
4282         irqfd->producer = prod;
4283
4284         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4285         if (ret)
4286                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
4287                         prod->irq, irqfd->gsi, ret);
4288
4289         return ret;
4290 }
4291
4292 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
4293                                               struct irq_bypass_producer *prod)
4294 {
4295         int ret;
4296         struct kvm_kernel_irqfd *irqfd =
4297                 container_of(cons, struct kvm_kernel_irqfd, consumer);
4298
4299         irqfd->producer = NULL;
4300
4301         /*
4302          * When producer of consumer is unregistered, we change back to
4303          * default external interrupt handling mode - KVM real mode
4304          * will switch back to host.
4305          */
4306         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4307         if (ret)
4308                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
4309                         prod->irq, irqfd->gsi, ret);
4310 }
4311 #endif
4312
4313 static long kvm_arch_vm_ioctl_hv(struct file *filp,
4314                                  unsigned int ioctl, unsigned long arg)
4315 {
4316         struct kvm *kvm __maybe_unused = filp->private_data;
4317         void __user *argp = (void __user *)arg;
4318         long r;
4319
4320         switch (ioctl) {
4321
4322         case KVM_PPC_ALLOCATE_HTAB: {
4323                 u32 htab_order;
4324
4325                 r = -EFAULT;
4326                 if (get_user(htab_order, (u32 __user *)argp))
4327                         break;
4328                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
4329                 if (r)
4330                         break;
4331                 r = 0;
4332                 break;
4333         }
4334
4335         case KVM_PPC_GET_HTAB_FD: {
4336                 struct kvm_get_htab_fd ghf;
4337
4338                 r = -EFAULT;
4339                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
4340                         break;
4341                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
4342                 break;
4343         }
4344
4345         case KVM_PPC_RESIZE_HPT_PREPARE: {
4346                 struct kvm_ppc_resize_hpt rhpt;
4347
4348                 r = -EFAULT;
4349                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4350                         break;
4351
4352                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
4353                 break;
4354         }
4355
4356         case KVM_PPC_RESIZE_HPT_COMMIT: {
4357                 struct kvm_ppc_resize_hpt rhpt;
4358
4359                 r = -EFAULT;
4360                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4361                         break;
4362
4363                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
4364                 break;
4365         }
4366
4367         default:
4368                 r = -ENOTTY;
4369         }
4370
4371         return r;
4372 }
4373
4374 /*
4375  * List of hcall numbers to enable by default.
4376  * For compatibility with old userspace, we enable by default
4377  * all hcalls that were implemented before the hcall-enabling
4378  * facility was added.  Note this list should not include H_RTAS.
4379  */
4380 static unsigned int default_hcall_list[] = {
4381         H_REMOVE,
4382         H_ENTER,
4383         H_READ,
4384         H_PROTECT,
4385         H_BULK_REMOVE,
4386         H_GET_TCE,
4387         H_PUT_TCE,
4388         H_SET_DABR,
4389         H_SET_XDABR,
4390         H_CEDE,
4391         H_PROD,
4392         H_CONFER,
4393         H_REGISTER_VPA,
4394 #ifdef CONFIG_KVM_XICS
4395         H_EOI,
4396         H_CPPR,
4397         H_IPI,
4398         H_IPOLL,
4399         H_XIRR,
4400         H_XIRR_X,
4401 #endif
4402         0
4403 };
4404
4405 static void init_default_hcalls(void)
4406 {
4407         int i;
4408         unsigned int hcall;
4409
4410         for (i = 0; default_hcall_list[i]; ++i) {
4411                 hcall = default_hcall_list[i];
4412                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
4413                 __set_bit(hcall / 4, default_enabled_hcalls);
4414         }
4415 }
4416
4417 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
4418 {
4419         unsigned long lpcr;
4420         int radix;
4421         int err;
4422
4423         /* If not on a POWER9, reject it */
4424         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4425                 return -ENODEV;
4426
4427         /* If any unknown flags set, reject it */
4428         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
4429                 return -EINVAL;
4430
4431         /* GR (guest radix) bit in process_table field must match */
4432         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
4433         if (!!(cfg->process_table & PATB_GR) != radix)
4434                 return -EINVAL;
4435
4436         /* Process table size field must be reasonable, i.e. <= 24 */
4437         if ((cfg->process_table & PRTS_MASK) > 24)
4438                 return -EINVAL;
4439
4440         /* We can change a guest to/from radix now, if the host is radix */
4441         if (radix && !radix_enabled())
4442                 return -EINVAL;
4443
4444         mutex_lock(&kvm->lock);
4445         if (radix != kvm_is_radix(kvm)) {
4446                 if (kvm->arch.mmu_ready) {
4447                         kvm->arch.mmu_ready = 0;
4448                         /* order mmu_ready vs. vcpus_running */
4449                         smp_mb();
4450                         if (atomic_read(&kvm->arch.vcpus_running)) {
4451                                 kvm->arch.mmu_ready = 1;
4452                                 err = -EBUSY;
4453                                 goto out_unlock;
4454                         }
4455                 }
4456                 if (radix)
4457                         err = kvmppc_switch_mmu_to_radix(kvm);
4458                 else
4459                         err = kvmppc_switch_mmu_to_hpt(kvm);
4460                 if (err)
4461                         goto out_unlock;
4462         }
4463
4464         kvm->arch.process_table = cfg->process_table;
4465         kvmppc_setup_partition_table(kvm);
4466
4467         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
4468         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
4469         err = 0;
4470
4471  out_unlock:
4472         mutex_unlock(&kvm->lock);
4473         return err;
4474 }
4475
4476 static struct kvmppc_ops kvm_ops_hv = {
4477         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
4478         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
4479         .get_one_reg = kvmppc_get_one_reg_hv,
4480         .set_one_reg = kvmppc_set_one_reg_hv,
4481         .vcpu_load   = kvmppc_core_vcpu_load_hv,
4482         .vcpu_put    = kvmppc_core_vcpu_put_hv,
4483         .set_msr     = kvmppc_set_msr_hv,
4484         .vcpu_run    = kvmppc_vcpu_run_hv,
4485         .vcpu_create = kvmppc_core_vcpu_create_hv,
4486         .vcpu_free   = kvmppc_core_vcpu_free_hv,
4487         .check_requests = kvmppc_core_check_requests_hv,
4488         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
4489         .flush_memslot  = kvmppc_core_flush_memslot_hv,
4490         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
4491         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
4492         .unmap_hva_range = kvm_unmap_hva_range_hv,
4493         .age_hva  = kvm_age_hva_hv,
4494         .test_age_hva = kvm_test_age_hva_hv,
4495         .set_spte_hva = kvm_set_spte_hva_hv,
4496         .mmu_destroy  = kvmppc_mmu_destroy_hv,
4497         .free_memslot = kvmppc_core_free_memslot_hv,
4498         .create_memslot = kvmppc_core_create_memslot_hv,
4499         .init_vm =  kvmppc_core_init_vm_hv,
4500         .destroy_vm = kvmppc_core_destroy_vm_hv,
4501         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
4502         .emulate_op = kvmppc_core_emulate_op_hv,
4503         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
4504         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
4505         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
4506         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
4507         .hcall_implemented = kvmppc_hcall_impl_hv,
4508 #ifdef CONFIG_KVM_XICS
4509         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
4510         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
4511 #endif
4512         .configure_mmu = kvmhv_configure_mmu,
4513         .get_rmmu_info = kvmhv_get_rmmu_info,
4514         .set_smt_mode = kvmhv_set_smt_mode,
4515 };
4516
4517 static int kvm_init_subcore_bitmap(void)
4518 {
4519         int i, j;
4520         int nr_cores = cpu_nr_cores();
4521         struct sibling_subcore_state *sibling_subcore_state;
4522
4523         for (i = 0; i < nr_cores; i++) {
4524                 int first_cpu = i * threads_per_core;
4525                 int node = cpu_to_node(first_cpu);
4526
4527                 /* Ignore if it is already allocated. */
4528                 if (paca_ptrs[first_cpu]->sibling_subcore_state)
4529                         continue;
4530
4531                 sibling_subcore_state =
4532                         kmalloc_node(sizeof(struct sibling_subcore_state),
4533                                                         GFP_KERNEL, node);
4534                 if (!sibling_subcore_state)
4535                         return -ENOMEM;
4536
4537                 memset(sibling_subcore_state, 0,
4538                                 sizeof(struct sibling_subcore_state));
4539
4540                 for (j = 0; j < threads_per_core; j++) {
4541                         int cpu = first_cpu + j;
4542
4543                         paca_ptrs[cpu]->sibling_subcore_state =
4544                                                 sibling_subcore_state;
4545                 }
4546         }
4547         return 0;
4548 }
4549
4550 static int kvmppc_radix_possible(void)
4551 {
4552         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
4553 }
4554
4555 static int kvmppc_book3s_init_hv(void)
4556 {
4557         int r;
4558         /*
4559          * FIXME!! Do we need to check on all cpus ?
4560          */
4561         r = kvmppc_core_check_processor_compat_hv();
4562         if (r < 0)
4563                 return -ENODEV;
4564
4565         r = kvm_init_subcore_bitmap();
4566         if (r)
4567                 return r;
4568
4569         /*
4570          * We need a way of accessing the XICS interrupt controller,
4571          * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
4572          * indirectly, via OPAL.
4573          */
4574 #ifdef CONFIG_SMP
4575         if (!xive_enabled() && !local_paca->kvm_hstate.xics_phys) {
4576                 struct device_node *np;
4577
4578                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
4579                 if (!np) {
4580                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
4581                         return -ENODEV;
4582                 }
4583                 /* presence of intc confirmed - node can be dropped again */
4584                 of_node_put(np);
4585         }
4586 #endif
4587
4588         kvm_ops_hv.owner = THIS_MODULE;
4589         kvmppc_hv_ops = &kvm_ops_hv;
4590
4591         init_default_hcalls();
4592
4593         init_vcore_lists();
4594
4595         r = kvmppc_mmu_hv_init();
4596         if (r)
4597                 return r;
4598
4599         if (kvmppc_radix_possible())
4600                 r = kvmppc_radix_init();
4601
4602         /*
4603          * POWER9 chips before version 2.02 can't have some threads in
4604          * HPT mode and some in radix mode on the same core.
4605          */
4606         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4607                 unsigned int pvr = mfspr(SPRN_PVR);
4608                 if ((pvr >> 16) == PVR_POWER9 &&
4609                     (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
4610                      ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
4611                         no_mixing_hpt_and_radix = true;
4612         }
4613
4614         return r;
4615 }
4616
4617 static void kvmppc_book3s_exit_hv(void)
4618 {
4619         kvmppc_free_host_rm_ops();
4620         if (kvmppc_radix_possible())
4621                 kvmppc_radix_exit();
4622         kvmppc_hv_ops = NULL;
4623 }
4624
4625 module_init(kvmppc_book3s_init_hv);
4626 module_exit(kvmppc_book3s_exit_hv);
4627 MODULE_LICENSE("GPL");
4628 MODULE_ALIAS_MISCDEV(KVM_MINOR);
4629 MODULE_ALIAS("devname:kvm");