arch/powerpc/kvm/book3s_hv.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
   4  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
   5  *
   6  * Authors:
   7  *    Paul Mackerras <paulus@au1.ibm.com>
   8  *    Alexander Graf <agraf@suse.de>
   9  *    Kevin Wolf <mail@kevin-wolf.de>
  10  *
  11  * Description: KVM functions specific to running on Book 3S
  12  * processors in hypervisor mode (specifically POWER7 and later).
  13  *
  14  * This file is derived from arch/powerpc/kvm/book3s.c,
  15  * by Alexander Graf <agraf@suse.de>.
  16  */
  17
  18 #include <linux/kvm_host.h>
  19 #include <linux/kernel.h>
  20 #include <linux/err.h>
  21 #include <linux/slab.h>
  22 #include <linux/preempt.h>
  23 #include <linux/sched/signal.h>
  24 #include <linux/sched/stat.h>
  25 #include <linux/delay.h>
  26 #include <linux/export.h>
  27 #include <linux/fs.h>
  28 #include <linux/anon_inodes.h>
  29 #include <linux/cpu.h>
  30 #include <linux/cpumask.h>
  31 #include <linux/spinlock.h>
  32 #include <linux/page-flags.h>
  33 #include <linux/srcu.h>
  34 #include <linux/miscdevice.h>
  35 #include <linux/debugfs.h>
  36 #include <linux/gfp.h>
  37 #include <linux/vmalloc.h>
  38 #include <linux/highmem.h>
  39 #include <linux/hugetlb.h>
  40 #include <linux/kvm_irqfd.h>
  41 #include <linux/irqbypass.h>
  42 #include <linux/module.h>
  43 #include <linux/compiler.h>
  44 #include <linux/of.h>
  45 #include <linux/irqdomain.h>
  46 #include <linux/smp.h>
  47
  48 #include <asm/ftrace.h>
  49 #include <asm/reg.h>
  50 #include <asm/ppc-opcode.h>
  51 #include <asm/asm-prototypes.h>
  52 #include <asm/archrandom.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/interrupt.h>
  59 #include <asm/io.h>
  60 #include <asm/kvm_ppc.h>
  61 #include <asm/kvm_book3s.h>
  62 #include <asm/mmu_context.h>
  63 #include <asm/lppaca.h>
  64 #include <asm/pmc.h>
  65 #include <asm/processor.h>
  66 #include <asm/cputhreads.h>
  67 #include <asm/page.h>
  68 #include <asm/hvcall.h>
  69 #include <asm/switch_to.h>
  70 #include <asm/smp.h>
  71 #include <asm/dbell.h>
  72 #include <asm/hmi.h>
  73 #include <asm/pnv-pci.h>
  74 #include <asm/mmu.h>
  75 #include <asm/opal.h>
  76 #include <asm/xics.h>
  77 #include <asm/xive.h>
  78 #include <asm/hw_breakpoint.h>
  79 #include <asm/kvm_book3s_uvmem.h>
  80 #include <asm/ultravisor.h>
  81 #include <asm/dtl.h>
  82 #include <asm/plpar_wrappers.h>
  83
  84 #include <trace/events/ipi.h>
  85
  86 #include "book3s.h"
  87 #include "book3s_hv.h"
  88
  89 #define CREATE_TRACE_POINTS
  90 #include "trace_hv.h"
  91
  92 /* #define EXIT_DEBUG */
  93 /* #define EXIT_DEBUG_SIMPLE */
  94 /* #define EXIT_DEBUG_INT */
  95
  96 /* Used to indicate that a guest page fault needs to be handled */
  97 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  98 /* Used to indicate that a guest passthrough interrupt needs to be handled */
  99 #define RESUME_PASSTHROUGH      (RESUME_GUEST | RESUME_FLAG_ARCH2)
 100
 101 /* Used as a "null" value for timebase values */
 102 #define TB_NIL  (~(u64)0)
 103
 104 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
 105
 106 static int dynamic_mt_modes = 6;
 107 module_param(dynamic_mt_modes, int, 0644);
 108 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
 109 static int target_smt_mode;
 110 module_param(target_smt_mode, int, 0644);
 111 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
 112
 113 static bool one_vm_per_core;
 114 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
 115 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires POWER8 or older)");
 116
 117 #ifdef CONFIG_KVM_XICS
 118 static const struct kernel_param_ops module_param_ops = {
 119         .set = param_set_int,
 120         .get = param_get_int,
 121 };
 122
 123 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
 124 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
 125
 126 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
 127 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
 128 #endif
 129
 130 /* If set, guests are allowed to create and control nested guests */
 131 static bool nested = true;
 132 module_param(nested, bool, S_IRUGO | S_IWUSR);
 133 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
 134
 135 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
 136
 137 /*
 138  * RWMR values for POWER8.  These control the rate at which PURR
 139  * and SPURR count and should be set according to the number of
 140  * online threads in the vcore being run.
 141  */
 142 #define RWMR_RPA_P8_1THREAD     0x164520C62609AECAUL
 143 #define RWMR_RPA_P8_2THREAD     0x7FFF2908450D8DA9UL
 144 #define RWMR_RPA_P8_3THREAD     0x164520C62609AECAUL
 145 #define RWMR_RPA_P8_4THREAD     0x199A421245058DA9UL
 146 #define RWMR_RPA_P8_5THREAD     0x164520C62609AECAUL
 147 #define RWMR_RPA_P8_6THREAD     0x164520C62609AECAUL
 148 #define RWMR_RPA_P8_7THREAD     0x164520C62609AECAUL
 149 #define RWMR_RPA_P8_8THREAD     0x164520C62609AECAUL
 150
 151 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
 152         RWMR_RPA_P8_1THREAD,
 153         RWMR_RPA_P8_1THREAD,
 154         RWMR_RPA_P8_2THREAD,
 155         RWMR_RPA_P8_3THREAD,
 156         RWMR_RPA_P8_4THREAD,
 157         RWMR_RPA_P8_5THREAD,
 158         RWMR_RPA_P8_6THREAD,
 159         RWMR_RPA_P8_7THREAD,
 160         RWMR_RPA_P8_8THREAD,
 161 };
 162
 163 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
 164                 int *ip)
 165 {
 166         int i = *ip;
 167         struct kvm_vcpu *vcpu;
 168
 169         while (++i < MAX_SMT_THREADS) {
 170                 vcpu = READ_ONCE(vc->runnable_threads[i]);
 171                 if (vcpu) {
 172                         *ip = i;
 173                         return vcpu;
 174                 }
 175         }
 176         return NULL;
 177 }
 178
 179 /* Used to traverse the list of runnable threads for a given vcore */
 180 #define for_each_runnable_thread(i, vcpu, vc) \
 181         for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
 182
 183 static bool kvmppc_ipi_thread(int cpu)
 184 {
 185         unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
 186
 187         /* If we're a nested hypervisor, fall back to ordinary IPIs for now */
 188         if (kvmhv_on_pseries())
 189                 return false;
 190
 191         /* On POWER9 we can use msgsnd to IPI any cpu */
 192         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
 193                 msg |= get_hard_smp_processor_id(cpu);
 194                 smp_mb();
 195                 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 196                 return true;
 197         }
 198
 199         /* On POWER8 for IPIs to threads in the same core, use msgsnd */
 200         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
 201                 preempt_disable();
 202                 if (cpu_first_thread_sibling(cpu) ==
 203                     cpu_first_thread_sibling(smp_processor_id())) {
 204                         msg |= cpu_thread_in_core(cpu);
 205                         smp_mb();
 206                         __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 207                         preempt_enable();
 208                         return true;
 209                 }
 210                 preempt_enable();
 211         }
 212
 213 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
 214         if (cpu >= 0 && cpu < nr_cpu_ids) {
 215                 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
 216                         xics_wake_cpu(cpu);
 217                         return true;
 218                 }
 219                 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
 220                 return true;
 221         }
 222 #endif
 223
 224         return false;
 225 }
 226
 227 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
 228 {
 229         int cpu;
 230         struct rcuwait *waitp;
 231
 232         /*
 233          * rcuwait_wake_up contains smp_mb() which orders prior stores that
 234          * create pending work vs below loads of cpu fields. The other side
 235          * is the barrier in vcpu run that orders setting the cpu fields vs
 236          * testing for pending work.
 237          */
 238
 239         waitp = kvm_arch_vcpu_get_wait(vcpu);
 240         if (rcuwait_wake_up(waitp))
 241                 ++vcpu->stat.generic.halt_wakeup;
 242
 243         cpu = READ_ONCE(vcpu->arch.thread_cpu);
 244         if (cpu >= 0 && kvmppc_ipi_thread(cpu))
 245                 return;
 246
 247         /* CPU points to the first thread of the core */
 248         cpu = vcpu->cpu;
 249         if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
 250                 smp_send_reschedule(cpu);
 251 }
 252
 253 /*
 254  * We use the vcpu_load/put functions to measure stolen time.
 255  *
 256  * Stolen time is counted as time when either the vcpu is able to
 257  * run as part of a virtual core, but the task running the vcore
 258  * is preempted or sleeping, or when the vcpu needs something done
 259  * in the kernel by the task running the vcpu, but that task is
 260  * preempted or sleeping.  Those two things have to be counted
 261  * separately, since one of the vcpu tasks will take on the job
 262  * of running the core, and the other vcpu tasks in the vcore will
 263  * sleep waiting for it to do that, but that sleep shouldn't count
 264  * as stolen time.
 265  *
 266  * Hence we accumulate stolen time when the vcpu can run as part of
 267  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 268  * needs its task to do other things in the kernel (for example,
 269  * service a page fault) in busy_stolen.  We don't accumulate
 270  * stolen time for a vcore when it is inactive, or for a vcpu
 271  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 272  * a misnomer; it means that the vcpu task is not executing in
 273  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 274  * the kernel.  We don't have any way of dividing up that time
 275  * between time that the vcpu is genuinely stopped, time that
 276  * the task is actively working on behalf of the vcpu, and time
 277  * that the task is preempted, so we don't count any of it as
 278  * stolen.
 279  *
 280  * Updates to busy_stolen are protected by arch.tbacct_lock;
 281  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
 282  * lock.  The stolen times are measured in units of timebase ticks.
 283  * (Note that the != TB_NIL checks below are purely defensive;
 284  * they should never fail.)
 285  *
 286  * The POWER9 path is simpler, one vcpu per virtual core so the
 287  * former case does not exist. If a vcpu is preempted when it is
 288  * BUSY_IN_HOST and not ceded or otherwise blocked, then accumulate
 289  * the stolen cycles in busy_stolen. RUNNING is not a preemptible
 290  * state in the P9 path.
 291  */
 292
 293 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc, u64 tb)
 294 {
 295         unsigned long flags;
 296
 297         WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
 298
 299         spin_lock_irqsave(&vc->stoltb_lock, flags);
 300         vc->preempt_tb = tb;
 301         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 302 }
 303
 304 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc, u64 tb)
 305 {
 306         unsigned long flags;
 307
 308         WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
 309
 310         spin_lock_irqsave(&vc->stoltb_lock, flags);
 311         if (vc->preempt_tb != TB_NIL) {
 312                 vc->stolen_tb += tb - vc->preempt_tb;
 313                 vc->preempt_tb = TB_NIL;
 314         }
 315         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 316 }
 317
 318 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 319 {
 320         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 321         unsigned long flags;
 322         u64 now;
 323
 324         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
 325                 if (vcpu->arch.busy_preempt != TB_NIL) {
 326                         WARN_ON_ONCE(vcpu->arch.state != KVMPPC_VCPU_BUSY_IN_HOST);
 327                         vc->stolen_tb += mftb() - vcpu->arch.busy_preempt;
 328                         vcpu->arch.busy_preempt = TB_NIL;
 329                 }
 330                 return;
 331         }
 332
 333         now = mftb();
 334
 335         /*
 336          * We can test vc->runner without taking the vcore lock,
 337          * because only this task ever sets vc->runner to this
 338          * vcpu, and once it is set to this vcpu, only this task
 339          * ever sets it to NULL.
 340          */
 341         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 342                 kvmppc_core_end_stolen(vc, now);
 343
 344         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 345         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 346             vcpu->arch.busy_preempt != TB_NIL) {
 347                 vcpu->arch.busy_stolen += now - vcpu->arch.busy_preempt;
 348                 vcpu->arch.busy_preempt = TB_NIL;
 349         }
 350         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 351 }
 352
 353 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 354 {
 355         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 356         unsigned long flags;
 357         u64 now;
 358
 359         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
 360                 /*
 361                  * In the P9 path, RUNNABLE is not preemptible
 362                  * (nor takes host interrupts)
 363                  */
 364                 WARN_ON_ONCE(vcpu->arch.state == KVMPPC_VCPU_RUNNABLE);
 365                 /*
 366                  * Account stolen time when preempted while the vcpu task is
 367                  * running in the kernel (but not in qemu, which is INACTIVE).
 368                  */
 369                 if (task_is_running(current) &&
 370                                 vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 371                         vcpu->arch.busy_preempt = mftb();
 372                 return;
 373         }
 374
 375         now = mftb();
 376
 377         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 378                 kvmppc_core_start_stolen(vc, now);
 379
 380         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 381         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 382                 vcpu->arch.busy_preempt = now;
 383         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 384 }
 385
 386 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 387 {
 388         vcpu->arch.pvr = pvr;
 389 }
 390
 391 /* Dummy value used in computing PCR value below */
 392 #define PCR_ARCH_31    (PCR_ARCH_300 << 1)
 393
 394 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 395 {
 396         unsigned long host_pcr_bit = 0, guest_pcr_bit = 0, cap = 0;
 397         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 398
 399         /* We can (emulate) our own architecture version and anything older */
 400         if (cpu_has_feature(CPU_FTR_ARCH_31))
 401                 host_pcr_bit = PCR_ARCH_31;
 402         else if (cpu_has_feature(CPU_FTR_ARCH_300))
 403                 host_pcr_bit = PCR_ARCH_300;
 404         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
 405                 host_pcr_bit = PCR_ARCH_207;
 406         else if (cpu_has_feature(CPU_FTR_ARCH_206))
 407                 host_pcr_bit = PCR_ARCH_206;
 408         else
 409                 host_pcr_bit = PCR_ARCH_205;
 410
 411         /* Determine lowest PCR bit needed to run guest in given PVR level */
 412         guest_pcr_bit = host_pcr_bit;
 413         if (arch_compat) {
 414                 switch (arch_compat) {
 415                 case PVR_ARCH_205:
 416                         guest_pcr_bit = PCR_ARCH_205;
 417                         break;
 418                 case PVR_ARCH_206:
 419                 case PVR_ARCH_206p:
 420                         guest_pcr_bit = PCR_ARCH_206;
 421                         break;
 422                 case PVR_ARCH_207:
 423                         guest_pcr_bit = PCR_ARCH_207;
 424                         break;
 425                 case PVR_ARCH_300:
 426                         guest_pcr_bit = PCR_ARCH_300;
 427                         cap = H_GUEST_CAP_POWER9;
 428                         break;
 429                 case PVR_ARCH_31:
 430                         guest_pcr_bit = PCR_ARCH_31;
 431                         cap = H_GUEST_CAP_POWER10;
 432                         break;
 433                 default:
 434                         return -EINVAL;
 435                 }
 436         }
 437
 438         /* Check requested PCR bits don't exceed our capabilities */
 439         if (guest_pcr_bit > host_pcr_bit)
 440                 return -EINVAL;
 441
 442         if (kvmhv_on_pseries() && kvmhv_is_nestedv2()) {
 443                 if (!(cap & nested_capabilities))
 444                         return -EINVAL;
 445         }
 446
 447         spin_lock(&vc->lock);
 448         vc->arch_compat = arch_compat;
 449         kvmhv_nestedv2_mark_dirty(vcpu, KVMPPC_GSID_LOGICAL_PVR);
 450         /*
 451          * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
 452          * Also set all reserved PCR bits
 453          */
 454         vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
 455         spin_unlock(&vc->lock);
 456
 457         return 0;
 458 }
 459
 460 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 461 {
 462         int r;
 463
 464         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 465         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 466                vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
 467         for (r = 0; r < 16; ++r)
 468                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 469                        r, kvmppc_get_gpr(vcpu, r),
 470                        r+16, kvmppc_get_gpr(vcpu, r+16));
 471         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 472                vcpu->arch.regs.ctr, vcpu->arch.regs.link);
 473         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 474                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 475         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 476                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 477         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 478                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 479         pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
 480                vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
 481         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 482         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 483                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 484         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 485         for (r = 0; r < vcpu->arch.slb_max; ++r)
 486                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 487                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 488         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.16lx\n",
 489                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 490                vcpu->arch.last_inst);
 491 }
 492
 493 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 494 {
 495         return kvm_get_vcpu_by_id(kvm, id);
 496 }
 497
 498 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 499 {
 500         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 501         vpa->yield_count = cpu_to_be32(1);
 502 }
 503
 504 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 505                    unsigned long addr, unsigned long len)
 506 {
 507         /* check address is cacheline aligned */
 508         if (addr & (L1_CACHE_BYTES - 1))
 509                 return -EINVAL;
 510         spin_lock(&vcpu->arch.vpa_update_lock);
 511         if (v->next_gpa != addr || v->len != len) {
 512                 v->next_gpa = addr;
 513                 v->len = addr ? len : 0;
 514                 v->update_pending = 1;
 515         }
 516         spin_unlock(&vcpu->arch.vpa_update_lock);
 517         return 0;
 518 }
 519
 520 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 521 struct reg_vpa {
 522         u32 dummy;
 523         union {
 524                 __be16 hword;
 525                 __be32 word;
 526         } length;
 527 };
 528
 529 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 530 {
 531         if (vpap->update_pending)
 532                 return vpap->next_gpa != 0;
 533         return vpap->pinned_addr != NULL;
 534 }
 535
 536 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 537                                        unsigned long flags,
 538                                        unsigned long vcpuid, unsigned long vpa)
 539 {
 540         struct kvm *kvm = vcpu->kvm;
 541         unsigned long len, nb;
 542         void *va;
 543         struct kvm_vcpu *tvcpu;
 544         int err;
 545         int subfunc;
 546         struct kvmppc_vpa *vpap;
 547
 548         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 549         if (!tvcpu)
 550                 return H_PARAMETER;
 551
 552         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 553         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 554             subfunc == H_VPA_REG_SLB) {
 555                 /* Registering new area - address must be cache-line aligned */
 556                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 557                         return H_PARAMETER;
 558
 559                 /* convert logical addr to kernel addr and read length */
 560                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 561                 if (va == NULL)
 562                         return H_PARAMETER;
 563                 if (subfunc == H_VPA_REG_VPA)
 564                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 565                 else
 566                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 567                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 568
 569                 /* Check length */
 570                 if (len > nb || len < sizeof(struct reg_vpa))
 571                         return H_PARAMETER;
 572         } else {
 573                 vpa = 0;
 574                 len = 0;
 575         }
 576
 577         err = H_PARAMETER;
 578         vpap = NULL;
 579         spin_lock(&tvcpu->arch.vpa_update_lock);
 580
 581         switch (subfunc) {
 582         case H_VPA_REG_VPA:             /* register VPA */
 583                 /*
 584                  * The size of our lppaca is 1kB because of the way we align
 585                  * it for the guest to avoid crossing a 4kB boundary. We only
 586                  * use 640 bytes of the structure though, so we should accept
 587                  * clients that set a size of 640.
 588                  */
 589                 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
 590                 if (len < sizeof(struct lppaca))
 591                         break;
 592                 vpap = &tvcpu->arch.vpa;
 593                 err = 0;
 594                 break;
 595
 596         case H_VPA_REG_DTL:             /* register DTL */
 597                 if (len < sizeof(struct dtl_entry))
 598                         break;
 599                 len -= len % sizeof(struct dtl_entry);
 600
 601                 /* Check that they have previously registered a VPA */
 602                 err = H_RESOURCE;
 603                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 604                         break;
 605
 606                 vpap = &tvcpu->arch.dtl;
 607                 err = 0;
 608                 break;
 609
 610         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 611                 /* Check that they have previously registered a VPA */
 612                 err = H_RESOURCE;
 613                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 614                         break;
 615
 616                 vpap = &tvcpu->arch.slb_shadow;
 617                 err = 0;
 618                 break;
 619
 620         case H_VPA_DEREG_VPA:           /* deregister VPA */
 621                 /* Check they don't still have a DTL or SLB buf registered */
 622                 err = H_RESOURCE;
 623                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 624                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 625                         break;
 626
 627                 vpap = &tvcpu->arch.vpa;
 628                 err = 0;
 629                 break;
 630
 631         case H_VPA_DEREG_DTL:           /* deregister DTL */
 632                 vpap = &tvcpu->arch.dtl;
 633                 err = 0;
 634                 break;
 635
 636         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 637                 vpap = &tvcpu->arch.slb_shadow;
 638                 err = 0;
 639                 break;
 640         }
 641
 642         if (vpap) {
 643                 vpap->next_gpa = vpa;
 644                 vpap->len = len;
 645                 vpap->update_pending = 1;
 646         }
 647
 648         spin_unlock(&tvcpu->arch.vpa_update_lock);
 649
 650         return err;
 651 }
 652
 653 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 654 {
 655         struct kvm *kvm = vcpu->kvm;
 656         void *va;
 657         unsigned long nb;
 658         unsigned long gpa;
 659
 660         /*
 661          * We need to pin the page pointed to by vpap->next_gpa,
 662          * but we can't call kvmppc_pin_guest_page under the lock
 663          * as it does get_user_pages() and down_read().  So we
 664          * have to drop the lock, pin the page, then get the lock
 665          * again and check that a new area didn't get registered
 666          * in the meantime.
 667          */
 668         for (;;) {
 669                 gpa = vpap->next_gpa;
 670                 spin_unlock(&vcpu->arch.vpa_update_lock);
 671                 va = NULL;
 672                 nb = 0;
 673                 if (gpa)
 674                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 675                 spin_lock(&vcpu->arch.vpa_update_lock);
 676                 if (gpa == vpap->next_gpa)
 677                         break;
 678                 /* sigh... unpin that one and try again */
 679                 if (va)
 680                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 681         }
 682
 683         vpap->update_pending = 0;
 684         if (va && nb < vpap->len) {
 685                 /*
 686                  * If it's now too short, it must be that userspace
 687                  * has changed the mappings underlying guest memory,
 688                  * so unregister the region.
 689                  */
 690                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 691                 va = NULL;
 692         }
 693         if (vpap->pinned_addr)
 694                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 695                                         vpap->dirty);
 696         vpap->gpa = gpa;
 697         vpap->pinned_addr = va;
 698         vpap->dirty = false;
 699         if (va)
 700                 vpap->pinned_end = va + vpap->len;
 701 }
 702
 703 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 704 {
 705         if (!(vcpu->arch.vpa.update_pending ||
 706               vcpu->arch.slb_shadow.update_pending ||
 707               vcpu->arch.dtl.update_pending))
 708                 return;
 709
 710         spin_lock(&vcpu->arch.vpa_update_lock);
 711         if (vcpu->arch.vpa.update_pending) {
 712                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 713                 if (vcpu->arch.vpa.pinned_addr)
 714                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 715         }
 716         if (vcpu->arch.dtl.update_pending) {
 717                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 718                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 719                 vcpu->arch.dtl_index = 0;
 720         }
 721         if (vcpu->arch.slb_shadow.update_pending)
 722                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 723         spin_unlock(&vcpu->arch.vpa_update_lock);
 724 }
 725
 726 /*
 727  * Return the accumulated stolen time for the vcore up until `now'.
 728  * The caller should hold the vcore lock.
 729  */
 730 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 731 {
 732         u64 p;
 733         unsigned long flags;
 734
 735         WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
 736
 737         spin_lock_irqsave(&vc->stoltb_lock, flags);
 738         p = vc->stolen_tb;
 739         if (vc->vcore_state != VCORE_INACTIVE &&
 740             vc->preempt_tb != TB_NIL)
 741                 p += now - vc->preempt_tb;
 742         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 743         return p;
 744 }
 745
 746 static void __kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 747                                         struct lppaca *vpa,
 748                                         unsigned int pcpu, u64 now,
 749                                         unsigned long stolen)
 750 {
 751         struct dtl_entry *dt;
 752
 753         dt = vcpu->arch.dtl_ptr;
 754
 755         if (!dt)
 756                 return;
 757
 758         dt->dispatch_reason = 7;
 759         dt->preempt_reason = 0;
 760         dt->processor_id = cpu_to_be16(pcpu + vcpu->arch.ptid);
 761         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 762         dt->ready_to_enqueue_time = 0;
 763         dt->waiting_to_ready_time = 0;
 764         dt->timebase = cpu_to_be64(now);
 765         dt->fault_addr = 0;
 766         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 767         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 768
 769         ++dt;
 770         if (dt == vcpu->arch.dtl.pinned_end)
 771                 dt = vcpu->arch.dtl.pinned_addr;
 772         vcpu->arch.dtl_ptr = dt;
 773         /* order writing *dt vs. writing vpa->dtl_idx */
 774         smp_wmb();
 775         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 776
 777         /* vcpu->arch.dtl.dirty is set by the caller */
 778 }
 779
 780 static void kvmppc_update_vpa_dispatch(struct kvm_vcpu *vcpu,
 781                                        struct kvmppc_vcore *vc)
 782 {
 783         struct lppaca *vpa;
 784         unsigned long stolen;
 785         unsigned long core_stolen;
 786         u64 now;
 787         unsigned long flags;
 788
 789         vpa = vcpu->arch.vpa.pinned_addr;
 790         if (!vpa)
 791                 return;
 792
 793         now = mftb();
 794
 795         core_stolen = vcore_stolen_time(vc, now);
 796         stolen = core_stolen - vcpu->arch.stolen_logged;
 797         vcpu->arch.stolen_logged = core_stolen;
 798         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 799         stolen += vcpu->arch.busy_stolen;
 800         vcpu->arch.busy_stolen = 0;
 801         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 802
 803         vpa->enqueue_dispatch_tb = cpu_to_be64(be64_to_cpu(vpa->enqueue_dispatch_tb) + stolen);
 804
 805         __kvmppc_create_dtl_entry(vcpu, vpa, vc->pcpu, now + kvmppc_get_tb_offset(vcpu), stolen);
 806
 807         vcpu->arch.vpa.dirty = true;
 808 }
 809
 810 static void kvmppc_update_vpa_dispatch_p9(struct kvm_vcpu *vcpu,
 811                                        struct kvmppc_vcore *vc,
 812                                        u64 now)
 813 {
 814         struct lppaca *vpa;
 815         unsigned long stolen;
 816         unsigned long stolen_delta;
 817
 818         vpa = vcpu->arch.vpa.pinned_addr;
 819         if (!vpa)
 820                 return;
 821
 822         stolen = vc->stolen_tb;
 823         stolen_delta = stolen - vcpu->arch.stolen_logged;
 824         vcpu->arch.stolen_logged = stolen;
 825
 826         vpa->enqueue_dispatch_tb = cpu_to_be64(stolen);
 827
 828         __kvmppc_create_dtl_entry(vcpu, vpa, vc->pcpu, now, stolen_delta);
 829
 830         vcpu->arch.vpa.dirty = true;
 831 }
 832
 833 /* See if there is a doorbell interrupt pending for a vcpu */
 834 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
 835 {
 836         int thr;
 837         struct kvmppc_vcore *vc;
 838
 839         if (vcpu->arch.doorbell_request)
 840                 return true;
 841         if (cpu_has_feature(CPU_FTR_ARCH_300))
 842                 return false;
 843         /*
 844          * Ensure that the read of vcore->dpdes comes after the read
 845          * of vcpu->doorbell_request.  This barrier matches the
 846          * smp_wmb() in kvmppc_guest_entry_inject().
 847          */
 848         smp_rmb();
 849         vc = vcpu->arch.vcore;
 850         thr = vcpu->vcpu_id - vc->first_vcpuid;
 851         return !!(vc->dpdes & (1 << thr));
 852 }
 853
 854 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 855 {
 856         if (kvmppc_get_arch_compat(vcpu) >= PVR_ARCH_207)
 857                 return true;
 858         if ((!kvmppc_get_arch_compat(vcpu)) &&
 859             cpu_has_feature(CPU_FTR_ARCH_207S))
 860                 return true;
 861         return false;
 862 }
 863
 864 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 865                              unsigned long resource, unsigned long value1,
 866                              unsigned long value2)
 867 {
 868         switch (resource) {
 869         case H_SET_MODE_RESOURCE_SET_CIABR:
 870                 if (!kvmppc_power8_compatible(vcpu))
 871                         return H_P2;
 872                 if (value2)
 873                         return H_P4;
 874                 if (mflags)
 875                         return H_UNSUPPORTED_FLAG_START;
 876                 /* Guests can't breakpoint the hypervisor */
 877                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 878                         return H_P3;
 879                 kvmppc_set_ciabr_hv(vcpu, value1);
 880                 return H_SUCCESS;
 881         case H_SET_MODE_RESOURCE_SET_DAWR0:
 882                 if (!kvmppc_power8_compatible(vcpu))
 883                         return H_P2;
 884                 if (!ppc_breakpoint_available())
 885                         return H_P2;
 886                 if (mflags)
 887                         return H_UNSUPPORTED_FLAG_START;
 888                 if (value2 & DABRX_HYP)
 889                         return H_P4;
 890                 kvmppc_set_dawr0_hv(vcpu, value1);
 891                 kvmppc_set_dawrx0_hv(vcpu, value2);
 892                 return H_SUCCESS;
 893         case H_SET_MODE_RESOURCE_SET_DAWR1:
 894                 if (!kvmppc_power8_compatible(vcpu))
 895                         return H_P2;
 896                 if (!ppc_breakpoint_available())
 897                         return H_P2;
 898                 if (!cpu_has_feature(CPU_FTR_DAWR1))
 899                         return H_P2;
 900                 if (!vcpu->kvm->arch.dawr1_enabled)
 901                         return H_FUNCTION;
 902                 if (mflags)
 903                         return H_UNSUPPORTED_FLAG_START;
 904                 if (value2 & DABRX_HYP)
 905                         return H_P4;
 906                 kvmppc_set_dawr1_hv(vcpu, value1);
 907                 kvmppc_set_dawrx1_hv(vcpu, value2);
 908                 return H_SUCCESS;
 909         case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
 910                 /*
 911                  * KVM does not support mflags=2 (AIL=2) and AIL=1 is reserved.
 912                  * Keep this in synch with kvmppc_filter_guest_lpcr_hv.
 913                  */
 914                 if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
 915                                 kvmhv_vcpu_is_radix(vcpu) && mflags == 3)
 916                         return H_UNSUPPORTED_FLAG_START;
 917                 return H_TOO_HARD;
 918         default:
 919                 return H_TOO_HARD;
 920         }
 921 }
 922
 923 /* Copy guest memory in place - must reside within a single memslot */
 924 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
 925                                   unsigned long len)
 926 {
 927         struct kvm_memory_slot *to_memslot = NULL;
 928         struct kvm_memory_slot *from_memslot = NULL;
 929         unsigned long to_addr, from_addr;
 930         int r;
 931
 932         /* Get HPA for from address */
 933         from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
 934         if (!from_memslot)
 935                 return -EFAULT;
 936         if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
 937                              << PAGE_SHIFT))
 938                 return -EINVAL;
 939         from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
 940         if (kvm_is_error_hva(from_addr))
 941                 return -EFAULT;
 942         from_addr |= (from & (PAGE_SIZE - 1));
 943
 944         /* Get HPA for to address */
 945         to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
 946         if (!to_memslot)
 947                 return -EFAULT;
 948         if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
 949                            << PAGE_SHIFT))
 950                 return -EINVAL;
 951         to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
 952         if (kvm_is_error_hva(to_addr))
 953                 return -EFAULT;
 954         to_addr |= (to & (PAGE_SIZE - 1));
 955
 956         /* Perform copy */
 957         r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
 958                              len);
 959         if (r)
 960                 return -EFAULT;
 961         mark_page_dirty(kvm, to >> PAGE_SHIFT);
 962         return 0;
 963 }
 964
 965 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
 966                                unsigned long dest, unsigned long src)
 967 {
 968         u64 pg_sz = SZ_4K;              /* 4K page size */
 969         u64 pg_mask = SZ_4K - 1;
 970         int ret;
 971
 972         /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
 973         if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
 974                       H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
 975                 return H_PARAMETER;
 976
 977         /* dest (and src if copy_page flag set) must be page aligned */
 978         if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
 979                 return H_PARAMETER;
 980
 981         /* zero and/or copy the page as determined by the flags */
 982         if (flags & H_COPY_PAGE) {
 983                 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
 984                 if (ret < 0)
 985                         return H_PARAMETER;
 986         } else if (flags & H_ZERO_PAGE) {
 987                 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
 988                 if (ret < 0)
 989                         return H_PARAMETER;
 990         }
 991
 992         /* We can ignore the remaining flags */
 993
 994         return H_SUCCESS;
 995 }
 996
 997 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 998 {
 999         struct kvmppc_vcore *vcore = target->arch.vcore;
1000
1001         /*
1002          * We expect to have been called by the real mode handler
1003          * (kvmppc_rm_h_confer()) which would have directly returned
1004          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
1005          * have useful work to do and should not confer) so we don't
1006          * recheck that here.
1007          *
1008          * In the case of the P9 single vcpu per vcore case, the real
1009          * mode handler is not called but no other threads are in the
1010          * source vcore.
1011          */
1012         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
1013                 spin_lock(&vcore->lock);
1014                 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
1015                     vcore->vcore_state != VCORE_INACTIVE &&
1016                     vcore->runner)
1017                         target = vcore->runner;
1018                 spin_unlock(&vcore->lock);
1019         }
1020
1021         return kvm_vcpu_yield_to(target);
1022 }
1023
1024 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
1025 {
1026         int yield_count = 0;
1027         struct lppaca *lppaca;
1028
1029         spin_lock(&vcpu->arch.vpa_update_lock);
1030         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
1031         if (lppaca)
1032                 yield_count = be32_to_cpu(lppaca->yield_count);
1033         spin_unlock(&vcpu->arch.vpa_update_lock);
1034         return yield_count;
1035 }
1036
1037 /*
1038  * H_RPT_INVALIDATE hcall handler for nested guests.
1039  *
1040  * Handles only nested process-scoped invalidation requests in L0.
1041  */
1042 static int kvmppc_nested_h_rpt_invalidate(struct kvm_vcpu *vcpu)
1043 {
1044         unsigned long type = kvmppc_get_gpr(vcpu, 6);
1045         unsigned long pid, pg_sizes, start, end;
1046
1047         /*
1048          * The partition-scoped invalidations aren't handled here in L0.
1049          */
1050         if (type & H_RPTI_TYPE_NESTED)
1051                 return RESUME_HOST;
1052
1053         pid = kvmppc_get_gpr(vcpu, 4);
1054         pg_sizes = kvmppc_get_gpr(vcpu, 7);
1055         start = kvmppc_get_gpr(vcpu, 8);
1056         end = kvmppc_get_gpr(vcpu, 9);
1057
1058         do_h_rpt_invalidate_prt(pid, vcpu->arch.nested->shadow_lpid,
1059                                 type, pg_sizes, start, end);
1060
1061         kvmppc_set_gpr(vcpu, 3, H_SUCCESS);
1062         return RESUME_GUEST;
1063 }
1064
1065 static long kvmppc_h_rpt_invalidate(struct kvm_vcpu *vcpu,
1066                                     unsigned long id, unsigned long target,
1067                                     unsigned long type, unsigned long pg_sizes,
1068                                     unsigned long start, unsigned long end)
1069 {
1070         if (!kvm_is_radix(vcpu->kvm))
1071                 return H_UNSUPPORTED;
1072
1073         if (end < start)
1074                 return H_P5;
1075
1076         /*
1077          * Partition-scoped invalidation for nested guests.
1078          */
1079         if (type & H_RPTI_TYPE_NESTED) {
1080                 if (!nesting_enabled(vcpu->kvm))
1081                         return H_FUNCTION;
1082
1083                 /* Support only cores as target */
1084                 if (target != H_RPTI_TARGET_CMMU)
1085                         return H_P2;
1086
1087                 return do_h_rpt_invalidate_pat(vcpu, id, type, pg_sizes,
1088                                                start, end);
1089         }
1090
1091         /*
1092          * Process-scoped invalidation for L1 guests.
1093          */
1094         do_h_rpt_invalidate_prt(id, vcpu->kvm->arch.lpid,
1095                                 type, pg_sizes, start, end);
1096         return H_SUCCESS;
1097 }
1098
1099 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
1100 {
1101         struct kvm *kvm = vcpu->kvm;
1102         unsigned long req = kvmppc_get_gpr(vcpu, 3);
1103         unsigned long target, ret = H_SUCCESS;
1104         int yield_count;
1105         struct kvm_vcpu *tvcpu;
1106         int idx, rc;
1107
1108         if (req <= MAX_HCALL_OPCODE &&
1109             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
1110                 return RESUME_HOST;
1111
1112         switch (req) {
1113         case H_REMOVE:
1114                 ret = kvmppc_h_remove(vcpu, kvmppc_get_gpr(vcpu, 4),
1115                                         kvmppc_get_gpr(vcpu, 5),
1116                                         kvmppc_get_gpr(vcpu, 6));
1117                 if (ret == H_TOO_HARD)
1118                         return RESUME_HOST;
1119                 break;
1120         case H_ENTER:
1121                 ret = kvmppc_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
1122                                         kvmppc_get_gpr(vcpu, 5),
1123                                         kvmppc_get_gpr(vcpu, 6),
1124                                         kvmppc_get_gpr(vcpu, 7));
1125                 if (ret == H_TOO_HARD)
1126                         return RESUME_HOST;
1127                 break;
1128         case H_READ:
1129                 ret = kvmppc_h_read(vcpu, kvmppc_get_gpr(vcpu, 4),
1130                                         kvmppc_get_gpr(vcpu, 5));
1131                 if (ret == H_TOO_HARD)
1132                         return RESUME_HOST;
1133                 break;
1134         case H_CLEAR_MOD:
1135                 ret = kvmppc_h_clear_mod(vcpu, kvmppc_get_gpr(vcpu, 4),
1136                                         kvmppc_get_gpr(vcpu, 5));
1137                 if (ret == H_TOO_HARD)
1138                         return RESUME_HOST;
1139                 break;
1140         case H_CLEAR_REF:
1141                 ret = kvmppc_h_clear_ref(vcpu, kvmppc_get_gpr(vcpu, 4),
1142                                         kvmppc_get_gpr(vcpu, 5));
1143                 if (ret == H_TOO_HARD)
1144                         return RESUME_HOST;
1145                 break;
1146         case H_PROTECT:
1147                 ret = kvmppc_h_protect(vcpu, kvmppc_get_gpr(vcpu, 4),
1148                                         kvmppc_get_gpr(vcpu, 5),
1149                                         kvmppc_get_gpr(vcpu, 6));
1150                 if (ret == H_TOO_HARD)
1151                         return RESUME_HOST;
1152                 break;
1153         case H_BULK_REMOVE:
1154                 ret = kvmppc_h_bulk_remove(vcpu);
1155                 if (ret == H_TOO_HARD)
1156                         return RESUME_HOST;
1157                 break;
1158
1159         case H_CEDE:
1160                 break;
1161         case H_PROD:
1162                 target = kvmppc_get_gpr(vcpu, 4);
1163                 tvcpu = kvmppc_find_vcpu(kvm, target);
1164                 if (!tvcpu) {
1165                         ret = H_PARAMETER;
1166                         break;
1167                 }
1168                 tvcpu->arch.prodded = 1;
1169                 smp_mb(); /* This orders prodded store vs ceded load */
1170                 if (tvcpu->arch.ceded)
1171                         kvmppc_fast_vcpu_kick_hv(tvcpu);
1172                 break;
1173         case H_CONFER:
1174                 target = kvmppc_get_gpr(vcpu, 4);
1175                 if (target == -1)
1176                         break;
1177                 tvcpu = kvmppc_find_vcpu(kvm, target);
1178                 if (!tvcpu) {
1179                         ret = H_PARAMETER;
1180                         break;
1181                 }
1182                 yield_count = kvmppc_get_gpr(vcpu, 5);
1183                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
1184                         break;
1185                 kvm_arch_vcpu_yield_to(tvcpu);
1186                 break;
1187         case H_REGISTER_VPA:
1188                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
1189                                         kvmppc_get_gpr(vcpu, 5),
1190                                         kvmppc_get_gpr(vcpu, 6));
1191                 break;
1192         case H_RTAS:
1193                 if (list_empty(&kvm->arch.rtas_tokens))
1194                         return RESUME_HOST;
1195
1196                 idx = srcu_read_lock(&kvm->srcu);
1197                 rc = kvmppc_rtas_hcall(vcpu);
1198                 srcu_read_unlock(&kvm->srcu, idx);
1199
1200                 if (rc == -ENOENT)
1201                         return RESUME_HOST;
1202                 else if (rc == 0)
1203                         break;
1204
1205                 /* Send the error out to userspace via KVM_RUN */
1206                 return rc;
1207         case H_LOGICAL_CI_LOAD:
1208                 ret = kvmppc_h_logical_ci_load(vcpu);
1209                 if (ret == H_TOO_HARD)
1210                         return RESUME_HOST;
1211                 break;
1212         case H_LOGICAL_CI_STORE:
1213                 ret = kvmppc_h_logical_ci_store(vcpu);
1214                 if (ret == H_TOO_HARD)
1215                         return RESUME_HOST;
1216                 break;
1217         case H_SET_MODE:
1218                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
1219                                         kvmppc_get_gpr(vcpu, 5),
1220                                         kvmppc_get_gpr(vcpu, 6),
1221                                         kvmppc_get_gpr(vcpu, 7));
1222                 if (ret == H_TOO_HARD)
1223                         return RESUME_HOST;
1224                 break;
1225         case H_XIRR:
1226         case H_CPPR:
1227         case H_EOI:
1228         case H_IPI:
1229         case H_IPOLL:
1230         case H_XIRR_X:
1231                 if (kvmppc_xics_enabled(vcpu)) {
1232                         if (xics_on_xive()) {
1233                                 ret = H_NOT_AVAILABLE;
1234                                 return RESUME_GUEST;
1235                         }
1236                         ret = kvmppc_xics_hcall(vcpu, req);
1237                         break;
1238                 }
1239                 return RESUME_HOST;
1240         case H_SET_DABR:
1241                 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1242                 break;
1243         case H_SET_XDABR:
1244                 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1245                                                 kvmppc_get_gpr(vcpu, 5));
1246                 break;
1247 #ifdef CONFIG_SPAPR_TCE_IOMMU
1248         case H_GET_TCE:
1249                 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1250                                                 kvmppc_get_gpr(vcpu, 5));
1251                 if (ret == H_TOO_HARD)
1252                         return RESUME_HOST;
1253                 break;
1254         case H_PUT_TCE:
1255                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1256                                                 kvmppc_get_gpr(vcpu, 5),
1257                                                 kvmppc_get_gpr(vcpu, 6));
1258                 if (ret == H_TOO_HARD)
1259                         return RESUME_HOST;
1260                 break;
1261         case H_PUT_TCE_INDIRECT:
1262                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1263                                                 kvmppc_get_gpr(vcpu, 5),
1264                                                 kvmppc_get_gpr(vcpu, 6),
1265                                                 kvmppc_get_gpr(vcpu, 7));
1266                 if (ret == H_TOO_HARD)
1267                         return RESUME_HOST;
1268                 break;
1269         case H_STUFF_TCE:
1270                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1271                                                 kvmppc_get_gpr(vcpu, 5),
1272                                                 kvmppc_get_gpr(vcpu, 6),
1273                                                 kvmppc_get_gpr(vcpu, 7));
1274                 if (ret == H_TOO_HARD)
1275                         return RESUME_HOST;
1276                 break;
1277 #endif
1278         case H_RANDOM: {
1279                 unsigned long rand;
1280
1281                 if (!arch_get_random_seed_longs(&rand, 1))
1282                         ret = H_HARDWARE;
1283                 kvmppc_set_gpr(vcpu, 4, rand);
1284                 break;
1285         }
1286         case H_RPT_INVALIDATE:
1287                 ret = kvmppc_h_rpt_invalidate(vcpu, kvmppc_get_gpr(vcpu, 4),
1288                                               kvmppc_get_gpr(vcpu, 5),
1289                                               kvmppc_get_gpr(vcpu, 6),
1290                                               kvmppc_get_gpr(vcpu, 7),
1291                                               kvmppc_get_gpr(vcpu, 8),
1292                                               kvmppc_get_gpr(vcpu, 9));
1293                 break;
1294
1295         case H_SET_PARTITION_TABLE:
1296                 ret = H_FUNCTION;
1297                 if (nesting_enabled(kvm))
1298                         ret = kvmhv_set_partition_table(vcpu);
1299                 break;
1300         case H_ENTER_NESTED:
1301                 ret = H_FUNCTION;
1302                 if (!nesting_enabled(kvm))
1303                         break;
1304                 ret = kvmhv_enter_nested_guest(vcpu);
1305                 if (ret == H_INTERRUPT) {
1306                         kvmppc_set_gpr(vcpu, 3, 0);
1307                         vcpu->arch.hcall_needed = 0;
1308                         return -EINTR;
1309                 } else if (ret == H_TOO_HARD) {
1310                         kvmppc_set_gpr(vcpu, 3, 0);
1311                         vcpu->arch.hcall_needed = 0;
1312                         return RESUME_HOST;
1313                 }
1314                 break;
1315         case H_TLB_INVALIDATE:
1316                 ret = H_FUNCTION;
1317                 if (nesting_enabled(kvm))
1318                         ret = kvmhv_do_nested_tlbie(vcpu);
1319                 break;
1320         case H_COPY_TOFROM_GUEST:
1321                 ret = H_FUNCTION;
1322                 if (nesting_enabled(kvm))
1323                         ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1324                 break;
1325         case H_PAGE_INIT:
1326                 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1327                                          kvmppc_get_gpr(vcpu, 5),
1328                                          kvmppc_get_gpr(vcpu, 6));
1329                 break;
1330         case H_SVM_PAGE_IN:
1331                 ret = H_UNSUPPORTED;
1332                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1333                         ret = kvmppc_h_svm_page_in(kvm,
1334                                                    kvmppc_get_gpr(vcpu, 4),
1335                                                    kvmppc_get_gpr(vcpu, 5),
1336                                                    kvmppc_get_gpr(vcpu, 6));
1337                 break;
1338         case H_SVM_PAGE_OUT:
1339                 ret = H_UNSUPPORTED;
1340                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1341                         ret = kvmppc_h_svm_page_out(kvm,
1342                                                     kvmppc_get_gpr(vcpu, 4),
1343                                                     kvmppc_get_gpr(vcpu, 5),
1344                                                     kvmppc_get_gpr(vcpu, 6));
1345                 break;
1346         case H_SVM_INIT_START:
1347                 ret = H_UNSUPPORTED;
1348                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1349                         ret = kvmppc_h_svm_init_start(kvm);
1350                 break;
1351         case H_SVM_INIT_DONE:
1352                 ret = H_UNSUPPORTED;
1353                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1354                         ret = kvmppc_h_svm_init_done(kvm);
1355                 break;
1356         case H_SVM_INIT_ABORT:
1357                 /*
1358                  * Even if that call is made by the Ultravisor, the SSR1 value
1359                  * is the guest context one, with the secure bit clear as it has
1360                  * not yet been secured. So we can't check it here.
1361                  * Instead the kvm->arch.secure_guest flag is checked inside
1362                  * kvmppc_h_svm_init_abort().
1363                  */
1364                 ret = kvmppc_h_svm_init_abort(kvm);
1365                 break;
1366
1367         default:
1368                 return RESUME_HOST;
1369         }
1370         WARN_ON_ONCE(ret == H_TOO_HARD);
1371         kvmppc_set_gpr(vcpu, 3, ret);
1372         vcpu->arch.hcall_needed = 0;
1373         return RESUME_GUEST;
1374 }
1375
1376 /*
1377  * Handle H_CEDE in the P9 path where we don't call the real-mode hcall
1378  * handlers in book3s_hv_rmhandlers.S.
1379  *
1380  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1381  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1382  */
1383 static void kvmppc_cede(struct kvm_vcpu *vcpu)
1384 {
1385         __kvmppc_set_msr_hv(vcpu, __kvmppc_get_msr_hv(vcpu) | MSR_EE);
1386         vcpu->arch.ceded = 1;
1387         smp_mb();
1388         if (vcpu->arch.prodded) {
1389                 vcpu->arch.prodded = 0;
1390                 smp_mb();
1391                 vcpu->arch.ceded = 0;
1392         }
1393 }
1394
1395 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1396 {
1397         switch (cmd) {
1398         case H_CEDE:
1399         case H_PROD:
1400         case H_CONFER:
1401         case H_REGISTER_VPA:
1402         case H_SET_MODE:
1403 #ifdef CONFIG_SPAPR_TCE_IOMMU
1404         case H_GET_TCE:
1405         case H_PUT_TCE:
1406         case H_PUT_TCE_INDIRECT:
1407         case H_STUFF_TCE:
1408 #endif
1409         case H_LOGICAL_CI_LOAD:
1410         case H_LOGICAL_CI_STORE:
1411 #ifdef CONFIG_KVM_XICS
1412         case H_XIRR:
1413         case H_CPPR:
1414         case H_EOI:
1415         case H_IPI:
1416         case H_IPOLL:
1417         case H_XIRR_X:
1418 #endif
1419         case H_PAGE_INIT:
1420         case H_RPT_INVALIDATE:
1421                 return 1;
1422         }
1423
1424         /* See if it's in the real-mode table */
1425         return kvmppc_hcall_impl_hv_realmode(cmd);
1426 }
1427
1428 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1429 {
1430         ppc_inst_t last_inst;
1431
1432         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1433                                         EMULATE_DONE) {
1434                 /*
1435                  * Fetch failed, so return to guest and
1436                  * try executing it again.
1437                  */
1438                 return RESUME_GUEST;
1439         }
1440
1441         if (ppc_inst_val(last_inst) == KVMPPC_INST_SW_BREAKPOINT) {
1442                 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1443                 vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1444                 return RESUME_HOST;
1445         } else {
1446                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL |
1447                                 (kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1448                 return RESUME_GUEST;
1449         }
1450 }
1451
1452 static void do_nothing(void *x)
1453 {
1454 }
1455
1456 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1457 {
1458         int thr, cpu, pcpu, nthreads;
1459         struct kvm_vcpu *v;
1460         unsigned long dpdes;
1461
1462         nthreads = vcpu->kvm->arch.emul_smt_mode;
1463         dpdes = 0;
1464         cpu = vcpu->vcpu_id & ~(nthreads - 1);
1465         for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1466                 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1467                 if (!v)
1468                         continue;
1469                 /*
1470                  * If the vcpu is currently running on a physical cpu thread,
1471                  * interrupt it in order to pull it out of the guest briefly,
1472                  * which will update its vcore->dpdes value.
1473                  */
1474                 pcpu = READ_ONCE(v->cpu);
1475                 if (pcpu >= 0)
1476                         smp_call_function_single(pcpu, do_nothing, NULL, 1);
1477                 if (kvmppc_doorbell_pending(v))
1478                         dpdes |= 1 << thr;
1479         }
1480         return dpdes;
1481 }
1482
1483 /*
1484  * On POWER9, emulate doorbell-related instructions in order to
1485  * give the guest the illusion of running on a multi-threaded core.
1486  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1487  * and mfspr DPDES.
1488  */
1489 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1490 {
1491         u32 inst, rb, thr;
1492         unsigned long arg;
1493         struct kvm *kvm = vcpu->kvm;
1494         struct kvm_vcpu *tvcpu;
1495         ppc_inst_t pinst;
1496
1497         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &pinst) != EMULATE_DONE)
1498                 return RESUME_GUEST;
1499         inst = ppc_inst_val(pinst);
1500         if (get_op(inst) != 31)
1501                 return EMULATE_FAIL;
1502         rb = get_rb(inst);
1503         thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1504         switch (get_xop(inst)) {
1505         case OP_31_XOP_MSGSNDP:
1506                 arg = kvmppc_get_gpr(vcpu, rb);
1507                 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1508                         break;
1509                 arg &= 0x7f;
1510                 if (arg >= kvm->arch.emul_smt_mode)
1511                         break;
1512                 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1513                 if (!tvcpu)
1514                         break;
1515                 if (!tvcpu->arch.doorbell_request) {
1516                         tvcpu->arch.doorbell_request = 1;
1517                         kvmppc_fast_vcpu_kick_hv(tvcpu);
1518                 }
1519                 break;
1520         case OP_31_XOP_MSGCLRP:
1521                 arg = kvmppc_get_gpr(vcpu, rb);
1522                 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1523                         break;
1524                 vcpu->arch.vcore->dpdes = 0;
1525                 vcpu->arch.doorbell_request = 0;
1526                 break;
1527         case OP_31_XOP_MFSPR:
1528                 switch (get_sprn(inst)) {
1529                 case SPRN_TIR:
1530                         arg = thr;
1531                         break;
1532                 case SPRN_DPDES:
1533                         arg = kvmppc_read_dpdes(vcpu);
1534                         break;
1535                 default:
1536                         return EMULATE_FAIL;
1537                 }
1538                 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1539                 break;
1540         default:
1541                 return EMULATE_FAIL;
1542         }
1543         kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1544         return RESUME_GUEST;
1545 }
1546
1547 /*
1548  * If the lppaca had pmcregs_in_use clear when we exited the guest, then
1549  * HFSCR_PM is cleared for next entry. If the guest then tries to access
1550  * the PMU SPRs, we get this facility unavailable interrupt. Putting HFSCR_PM
1551  * back in the guest HFSCR will cause the next entry to load the PMU SPRs and
1552  * allow the guest access to continue.
1553  */
1554 static int kvmppc_pmu_unavailable(struct kvm_vcpu *vcpu)
1555 {
1556         if (!(vcpu->arch.hfscr_permitted & HFSCR_PM))
1557                 return EMULATE_FAIL;
1558
1559         kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_PM);
1560
1561         return RESUME_GUEST;
1562 }
1563
1564 static int kvmppc_ebb_unavailable(struct kvm_vcpu *vcpu)
1565 {
1566         if (!(vcpu->arch.hfscr_permitted & HFSCR_EBB))
1567                 return EMULATE_FAIL;
1568
1569         kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_EBB);
1570
1571         return RESUME_GUEST;
1572 }
1573
1574 static int kvmppc_tm_unavailable(struct kvm_vcpu *vcpu)
1575 {
1576         if (!(vcpu->arch.hfscr_permitted & HFSCR_TM))
1577                 return EMULATE_FAIL;
1578
1579         kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_TM);
1580
1581         return RESUME_GUEST;
1582 }
1583
1584 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1585                                  struct task_struct *tsk)
1586 {
1587         struct kvm_run *run = vcpu->run;
1588         int r = RESUME_HOST;
1589
1590         vcpu->stat.sum_exits++;
1591
1592         /*
1593          * This can happen if an interrupt occurs in the last stages
1594          * of guest entry or the first stages of guest exit (i.e. after
1595          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1596          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1597          * That can happen due to a bug, or due to a machine check
1598          * occurring at just the wrong time.
1599          */
1600         if (__kvmppc_get_msr_hv(vcpu) & MSR_HV) {
1601                 printk(KERN_EMERG "KVM trap in HV mode!\n");
1602                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1603                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1604                         vcpu->arch.shregs.msr);
1605                 kvmppc_dump_regs(vcpu);
1606                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1607                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1608                 return RESUME_HOST;
1609         }
1610         run->exit_reason = KVM_EXIT_UNKNOWN;
1611         run->ready_for_interrupt_injection = 1;
1612         switch (vcpu->arch.trap) {
1613         /* We're good on these - the host merely wanted to get our attention */
1614         case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER:
1615                 WARN_ON_ONCE(1); /* Should never happen */
1616                 vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
1617                 fallthrough;
1618         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1619                 vcpu->stat.dec_exits++;
1620                 r = RESUME_GUEST;
1621                 break;
1622         case BOOK3S_INTERRUPT_EXTERNAL:
1623         case BOOK3S_INTERRUPT_H_DOORBELL:
1624         case BOOK3S_INTERRUPT_H_VIRT:
1625                 vcpu->stat.ext_intr_exits++;
1626                 r = RESUME_GUEST;
1627                 break;
1628         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1629         case BOOK3S_INTERRUPT_HMI:
1630         case BOOK3S_INTERRUPT_PERFMON:
1631         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1632                 r = RESUME_GUEST;
1633                 break;
1634         case BOOK3S_INTERRUPT_MACHINE_CHECK: {
1635                 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1636                                               DEFAULT_RATELIMIT_BURST);
1637                 /*
1638                  * Print the MCE event to host console. Ratelimit so the guest
1639                  * can't flood the host log.
1640                  */
1641                 if (__ratelimit(&rs))
1642                         machine_check_print_event_info(&vcpu->arch.mce_evt,false, true);
1643
1644                 /*
1645                  * If the guest can do FWNMI, exit to userspace so it can
1646                  * deliver a FWNMI to the guest.
1647                  * Otherwise we synthesize a machine check for the guest
1648                  * so that it knows that the machine check occurred.
1649                  */
1650                 if (!vcpu->kvm->arch.fwnmi_enabled) {
1651                         ulong flags = (__kvmppc_get_msr_hv(vcpu) & 0x083c0000) |
1652                                         (kvmppc_get_msr(vcpu) & SRR1_PREFIXED);
1653                         kvmppc_core_queue_machine_check(vcpu, flags);
1654                         r = RESUME_GUEST;
1655                         break;
1656                 }
1657
1658                 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1659                 run->exit_reason = KVM_EXIT_NMI;
1660                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1661                 /* Clear out the old NMI status from run->flags */
1662                 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1663                 /* Now set the NMI status */
1664                 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1665                         run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1666                 else
1667                         run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1668
1669                 r = RESUME_HOST;
1670                 break;
1671         }
1672         case BOOK3S_INTERRUPT_PROGRAM:
1673         {
1674                 ulong flags;
1675                 /*
1676                  * Normally program interrupts are delivered directly
1677                  * to the guest by the hardware, but we can get here
1678                  * as a result of a hypervisor emulation interrupt
1679                  * (e40) getting turned into a 700 by BML RTAS.
1680                  */
1681                 flags = (__kvmppc_get_msr_hv(vcpu) & 0x1f0000ull) |
1682                         (kvmppc_get_msr(vcpu) & SRR1_PREFIXED);
1683                 kvmppc_core_queue_program(vcpu, flags);
1684                 r = RESUME_GUEST;
1685                 break;
1686         }
1687         case BOOK3S_INTERRUPT_SYSCALL:
1688         {
1689                 int i;
1690
1691                 if (unlikely(__kvmppc_get_msr_hv(vcpu) & MSR_PR)) {
1692                         /*
1693                          * Guest userspace executed sc 1. This can only be
1694                          * reached by the P9 path because the old path
1695                          * handles this case in realmode hcall handlers.
1696                          */
1697                         if (!kvmhv_vcpu_is_radix(vcpu)) {
1698                                 /*
1699                                  * A guest could be running PR KVM, so this
1700                                  * may be a PR KVM hcall. It must be reflected
1701                                  * to the guest kernel as a sc interrupt.
1702                                  */
1703                                 kvmppc_core_queue_syscall(vcpu);
1704                         } else {
1705                                 /*
1706                                  * Radix guests can not run PR KVM or nested HV
1707                                  * hash guests which might run PR KVM, so this
1708                                  * is always a privilege fault. Send a program
1709                                  * check to guest kernel.
1710                                  */
1711                                 kvmppc_core_queue_program(vcpu, SRR1_PROGPRIV);
1712                         }
1713                         r = RESUME_GUEST;
1714                         break;
1715                 }
1716
1717                 /*
1718                  * hcall - gather args and set exit_reason. This will next be
1719                  * handled by kvmppc_pseries_do_hcall which may be able to deal
1720                  * with it and resume guest, or may punt to userspace.
1721                  */
1722                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1723                 for (i = 0; i < 9; ++i)
1724                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1725                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1726                 vcpu->arch.hcall_needed = 1;
1727                 r = RESUME_HOST;
1728                 break;
1729         }
1730         /*
1731          * We get these next two if the guest accesses a page which it thinks
1732          * it has mapped but which is not actually present, either because
1733          * it is for an emulated I/O device or because the corresonding
1734          * host page has been paged out.
1735          *
1736          * Any other HDSI/HISI interrupts have been handled already for P7/8
1737          * guests. For POWER9 hash guests not using rmhandlers, basic hash
1738          * fault handling is done here.
1739          */
1740         case BOOK3S_INTERRUPT_H_DATA_STORAGE: {
1741                 unsigned long vsid;
1742                 long err;
1743
1744                 if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
1745                     unlikely(vcpu->arch.fault_dsisr == HDSISR_CANARY)) {
1746                         r = RESUME_GUEST; /* Just retry if it's the canary */
1747                         break;
1748                 }
1749
1750                 if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1751                         /*
1752                          * Radix doesn't require anything, and pre-ISAv3.0 hash
1753                          * already attempted to handle this in rmhandlers. The
1754                          * hash fault handling below is v3 only (it uses ASDR
1755                          * via fault_gpa).
1756                          */
1757                         r = RESUME_PAGE_FAULT;
1758                         break;
1759                 }
1760
1761                 if (!(vcpu->arch.fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT))) {
1762                         kvmppc_core_queue_data_storage(vcpu,
1763                                 kvmppc_get_msr(vcpu) & SRR1_PREFIXED,
1764                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1765                         r = RESUME_GUEST;
1766                         break;
1767                 }
1768
1769                 if (!(__kvmppc_get_msr_hv(vcpu) & MSR_DR))
1770                         vsid = vcpu->kvm->arch.vrma_slb_v;
1771                 else
1772                         vsid = vcpu->arch.fault_gpa;
1773
1774                 err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1775                                 vsid, vcpu->arch.fault_dsisr, true);
1776                 if (err == 0) {
1777                         r = RESUME_GUEST;
1778                 } else if (err == -1 || err == -2) {
1779                         r = RESUME_PAGE_FAULT;
1780                 } else {
1781                         kvmppc_core_queue_data_storage(vcpu,
1782                                 kvmppc_get_msr(vcpu) & SRR1_PREFIXED,
1783                                 vcpu->arch.fault_dar, err);
1784                         r = RESUME_GUEST;
1785                 }
1786                 break;
1787         }
1788         case BOOK3S_INTERRUPT_H_INST_STORAGE: {
1789                 unsigned long vsid;
1790                 long err;
1791
1792                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1793                 vcpu->arch.fault_dsisr = __kvmppc_get_msr_hv(vcpu) &
1794                         DSISR_SRR1_MATCH_64S;
1795                 if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1796                         /*
1797                          * Radix doesn't require anything, and pre-ISAv3.0 hash
1798                          * already attempted to handle this in rmhandlers. The
1799                          * hash fault handling below is v3 only (it uses ASDR
1800                          * via fault_gpa).
1801                          */
1802                         if (__kvmppc_get_msr_hv(vcpu) & HSRR1_HISI_WRITE)
1803                                 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1804                         r = RESUME_PAGE_FAULT;
1805                         break;
1806                 }
1807
1808                 if (!(vcpu->arch.fault_dsisr & SRR1_ISI_NOPT)) {
1809                         kvmppc_core_queue_inst_storage(vcpu,
1810                                 vcpu->arch.fault_dsisr |
1811                                 (kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1812                         r = RESUME_GUEST;
1813                         break;
1814                 }
1815
1816                 if (!(__kvmppc_get_msr_hv(vcpu) & MSR_IR))
1817                         vsid = vcpu->kvm->arch.vrma_slb_v;
1818                 else
1819                         vsid = vcpu->arch.fault_gpa;
1820
1821                 err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1822                                 vsid, vcpu->arch.fault_dsisr, false);
1823                 if (err == 0) {
1824                         r = RESUME_GUEST;
1825                 } else if (err == -1) {
1826                         r = RESUME_PAGE_FAULT;
1827                 } else {
1828                         kvmppc_core_queue_inst_storage(vcpu,
1829                                 err | (kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1830                         r = RESUME_GUEST;
1831                 }
1832                 break;
1833         }
1834
1835         /*
1836          * This occurs if the guest executes an illegal instruction.
1837          * If the guest debug is disabled, generate a program interrupt
1838          * to the guest. If guest debug is enabled, we need to check
1839          * whether the instruction is a software breakpoint instruction.
1840          * Accordingly return to Guest or Host.
1841          */
1842         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1843                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1844                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1845                                 swab32(vcpu->arch.emul_inst) :
1846                                 vcpu->arch.emul_inst;
1847                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1848                         r = kvmppc_emulate_debug_inst(vcpu);
1849                 } else {
1850                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL |
1851                                 (kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1852                         r = RESUME_GUEST;
1853                 }
1854                 break;
1855
1856 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1857         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1858                 /*
1859                  * This occurs for various TM-related instructions that
1860                  * we need to emulate on POWER9 DD2.2.  We have already
1861                  * handled the cases where the guest was in real-suspend
1862                  * mode and was transitioning to transactional state.
1863                  */
1864                 r = kvmhv_p9_tm_emulation(vcpu);
1865                 if (r != -1)
1866                         break;
1867                 fallthrough; /* go to facility unavailable handler */
1868 #endif
1869
1870         /*
1871          * This occurs if the guest (kernel or userspace), does something that
1872          * is prohibited by HFSCR.
1873          * On POWER9, this could be a doorbell instruction that we need
1874          * to emulate.
1875          * Otherwise, we just generate a program interrupt to the guest.
1876          */
1877         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: {
1878                 u64 cause = kvmppc_get_hfscr_hv(vcpu) >> 56;
1879
1880                 r = EMULATE_FAIL;
1881                 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1882                         if (cause == FSCR_MSGP_LG)
1883                                 r = kvmppc_emulate_doorbell_instr(vcpu);
1884                         if (cause == FSCR_PM_LG)
1885                                 r = kvmppc_pmu_unavailable(vcpu);
1886                         if (cause == FSCR_EBB_LG)
1887                                 r = kvmppc_ebb_unavailable(vcpu);
1888                         if (cause == FSCR_TM_LG)
1889                                 r = kvmppc_tm_unavailable(vcpu);
1890                 }
1891                 if (r == EMULATE_FAIL) {
1892                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL |
1893                                 (kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1894                         r = RESUME_GUEST;
1895                 }
1896                 break;
1897         }
1898
1899         case BOOK3S_INTERRUPT_HV_RM_HARD:
1900                 r = RESUME_PASSTHROUGH;
1901                 break;
1902         default:
1903                 kvmppc_dump_regs(vcpu);
1904                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1905                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1906                         __kvmppc_get_msr_hv(vcpu));
1907                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1908                 r = RESUME_HOST;
1909                 break;
1910         }
1911
1912         return r;
1913 }
1914
1915 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1916 {
1917         int r;
1918         int srcu_idx;
1919
1920         vcpu->stat.sum_exits++;
1921
1922         /*
1923          * This can happen if an interrupt occurs in the last stages
1924          * of guest entry or the first stages of guest exit (i.e. after
1925          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1926          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1927          * That can happen due to a bug, or due to a machine check
1928          * occurring at just the wrong time.
1929          */
1930         if (__kvmppc_get_msr_hv(vcpu) & MSR_HV) {
1931                 pr_emerg("KVM trap in HV mode while nested!\n");
1932                 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1933                          vcpu->arch.trap, kvmppc_get_pc(vcpu),
1934                          __kvmppc_get_msr_hv(vcpu));
1935                 kvmppc_dump_regs(vcpu);
1936                 return RESUME_HOST;
1937         }
1938         switch (vcpu->arch.trap) {
1939         /* We're good on these - the host merely wanted to get our attention */
1940         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1941                 vcpu->stat.dec_exits++;
1942                 r = RESUME_GUEST;
1943                 break;
1944         case BOOK3S_INTERRUPT_EXTERNAL:
1945                 vcpu->stat.ext_intr_exits++;
1946                 r = RESUME_HOST;
1947                 break;
1948         case BOOK3S_INTERRUPT_H_DOORBELL:
1949         case BOOK3S_INTERRUPT_H_VIRT:
1950                 vcpu->stat.ext_intr_exits++;
1951                 r = RESUME_GUEST;
1952                 break;
1953         /* These need to go to the nested HV */
1954         case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER:
1955                 vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
1956                 vcpu->stat.dec_exits++;
1957                 r = RESUME_HOST;
1958                 break;
1959         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1960         case BOOK3S_INTERRUPT_HMI:
1961         case BOOK3S_INTERRUPT_PERFMON:
1962         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1963                 r = RESUME_GUEST;
1964                 break;
1965         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1966         {
1967                 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1968                                               DEFAULT_RATELIMIT_BURST);
1969                 /* Pass the machine check to the L1 guest */
1970                 r = RESUME_HOST;
1971                 /* Print the MCE event to host console. */
1972                 if (__ratelimit(&rs))
1973                         machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1974                 break;
1975         }
1976         /*
1977          * We get these next two if the guest accesses a page which it thinks
1978          * it has mapped but which is not actually present, either because
1979          * it is for an emulated I/O device or because the corresonding
1980          * host page has been paged out.
1981          */
1982         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1983                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1984                 r = kvmhv_nested_page_fault(vcpu);
1985                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1986                 break;
1987         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1988                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1989                 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1990                                          DSISR_SRR1_MATCH_64S;
1991                 if (__kvmppc_get_msr_hv(vcpu) & HSRR1_HISI_WRITE)
1992                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1993                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1994                 r = kvmhv_nested_page_fault(vcpu);
1995                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1996                 break;
1997
1998 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1999         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
2000                 /*
2001                  * This occurs for various TM-related instructions that
2002                  * we need to emulate on POWER9 DD2.2.  We have already
2003                  * handled the cases where the guest was in real-suspend
2004                  * mode and was transitioning to transactional state.
2005                  */
2006                 r = kvmhv_p9_tm_emulation(vcpu);
2007                 if (r != -1)
2008                         break;
2009                 fallthrough; /* go to facility unavailable handler */
2010 #endif
2011
2012         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: {
2013                 u64 cause = vcpu->arch.hfscr >> 56;
2014
2015                 /*
2016                  * Only pass HFU interrupts to the L1 if the facility is
2017                  * permitted but disabled by the L1's HFSCR, otherwise
2018                  * the interrupt does not make sense to the L1 so turn
2019                  * it into a HEAI.
2020                  */
2021                 if (!(vcpu->arch.hfscr_permitted & (1UL << cause)) ||
2022                                 (vcpu->arch.nested_hfscr & (1UL << cause))) {
2023                         ppc_inst_t pinst;
2024                         vcpu->arch.trap = BOOK3S_INTERRUPT_H_EMUL_ASSIST;
2025
2026                         /*
2027                          * If the fetch failed, return to guest and
2028                          * try executing it again.
2029                          */
2030                         r = kvmppc_get_last_inst(vcpu, INST_GENERIC, &pinst);
2031                         vcpu->arch.emul_inst = ppc_inst_val(pinst);
2032                         if (r != EMULATE_DONE)
2033                                 r = RESUME_GUEST;
2034                         else
2035                                 r = RESUME_HOST;
2036                 } else {
2037                         r = RESUME_HOST;
2038                 }
2039
2040                 break;
2041         }
2042
2043         case BOOK3S_INTERRUPT_HV_RM_HARD:
2044                 vcpu->arch.trap = 0;
2045                 r = RESUME_GUEST;
2046                 if (!xics_on_xive())
2047                         kvmppc_xics_rm_complete(vcpu, 0);
2048                 break;
2049         case BOOK3S_INTERRUPT_SYSCALL:
2050         {
2051                 unsigned long req = kvmppc_get_gpr(vcpu, 3);
2052
2053                 /*
2054                  * The H_RPT_INVALIDATE hcalls issued by nested
2055                  * guests for process-scoped invalidations when
2056                  * GTSE=0, are handled here in L0.
2057                  */
2058                 if (req == H_RPT_INVALIDATE) {
2059                         r = kvmppc_nested_h_rpt_invalidate(vcpu);
2060                         break;
2061                 }
2062
2063                 r = RESUME_HOST;
2064                 break;
2065         }
2066         default:
2067                 r = RESUME_HOST;
2068                 break;
2069         }
2070
2071         return r;
2072 }
2073
2074 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
2075                                             struct kvm_sregs *sregs)
2076 {
2077         int i;
2078
2079         memset(sregs, 0, sizeof(struct kvm_sregs));
2080         sregs->pvr = vcpu->arch.pvr;
2081         for (i = 0; i < vcpu->arch.slb_max; i++) {
2082                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
2083                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
2084         }
2085
2086         return 0;
2087 }
2088
2089 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
2090                                             struct kvm_sregs *sregs)
2091 {
2092         int i, j;
2093
2094         /* Only accept the same PVR as the host's, since we can't spoof it */
2095         if (sregs->pvr != vcpu->arch.pvr)
2096                 return -EINVAL;
2097
2098         j = 0;
2099         for (i = 0; i < vcpu->arch.slb_nr; i++) {
2100                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
2101                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
2102                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
2103                         ++j;
2104                 }
2105         }
2106         vcpu->arch.slb_max = j;
2107
2108         return 0;
2109 }
2110
2111 /*
2112  * Enforce limits on guest LPCR values based on hardware availability,
2113  * guest configuration, and possibly hypervisor support and security
2114  * concerns.
2115  */
2116 unsigned long kvmppc_filter_lpcr_hv(struct kvm *kvm, unsigned long lpcr)
2117 {
2118         /* LPCR_TC only applies to HPT guests */
2119         if (kvm_is_radix(kvm))
2120                 lpcr &= ~LPCR_TC;
2121
2122         /* On POWER8 and above, userspace can modify AIL */
2123         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2124                 lpcr &= ~LPCR_AIL;
2125         if ((lpcr & LPCR_AIL) != LPCR_AIL_3)
2126                 lpcr &= ~LPCR_AIL; /* LPCR[AIL]=1/2 is disallowed */
2127         /*
2128          * On some POWER9s we force AIL off for radix guests to prevent
2129          * executing in MSR[HV]=1 mode with the MMU enabled and PIDR set to
2130          * guest, which can result in Q0 translations with LPID=0 PID=PIDR to
2131          * be cached, which the host TLB management does not expect.
2132          */
2133         if (kvm_is_radix(kvm) && cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
2134                 lpcr &= ~LPCR_AIL;
2135
2136         /*
2137          * On POWER9, allow userspace to enable large decrementer for the
2138          * guest, whether or not the host has it enabled.
2139          */
2140         if (!cpu_has_feature(CPU_FTR_ARCH_300))
2141                 lpcr &= ~LPCR_LD;
2142
2143         return lpcr;
2144 }
2145
2146 static void verify_lpcr(struct kvm *kvm, unsigned long lpcr)
2147 {
2148         if (lpcr != kvmppc_filter_lpcr_hv(kvm, lpcr)) {
2149                 WARN_ONCE(1, "lpcr 0x%lx differs from filtered 0x%lx\n",
2150                           lpcr, kvmppc_filter_lpcr_hv(kvm, lpcr));
2151         }
2152 }
2153
2154 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
2155                 bool preserve_top32)
2156 {
2157         struct kvm *kvm = vcpu->kvm;
2158         struct kvmppc_vcore *vc = vcpu->arch.vcore;
2159         u64 mask;
2160
2161         spin_lock(&vc->lock);
2162
2163         /*
2164          * Userspace can only modify
2165          * DPFD (default prefetch depth), ILE (interrupt little-endian),
2166          * TC (translation control), AIL (alternate interrupt location),
2167          * LD (large decrementer).
2168          * These are subject to restrictions from kvmppc_filter_lcpr_hv().
2169          */
2170         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD;
2171
2172         /* Broken 32-bit version of LPCR must not clear top bits */
2173         if (preserve_top32)
2174                 mask &= 0xFFFFFFFF;
2175
2176         new_lpcr = kvmppc_filter_lpcr_hv(kvm,
2177                         (vc->lpcr & ~mask) | (new_lpcr & mask));
2178
2179         /*
2180          * If ILE (interrupt little-endian) has changed, update the
2181          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
2182          */
2183         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
2184                 struct kvm_vcpu *vcpu;
2185                 unsigned long i;
2186
2187                 kvm_for_each_vcpu(i, vcpu, kvm) {
2188                         if (vcpu->arch.vcore != vc)
2189                                 continue;
2190                         if (new_lpcr & LPCR_ILE)
2191                                 vcpu->arch.intr_msr |= MSR_LE;
2192                         else
2193                                 vcpu->arch.intr_msr &= ~MSR_LE;
2194                 }
2195         }
2196
2197         vc->lpcr = new_lpcr;
2198         kvmhv_nestedv2_mark_dirty(vcpu, KVMPPC_GSID_LPCR);
2199
2200         spin_unlock(&vc->lock);
2201 }
2202
2203 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2204                                  union kvmppc_one_reg *val)
2205 {
2206         int r = 0;
2207         long int i;
2208
2209         switch (id) {
2210         case KVM_REG_PPC_DEBUG_INST:
2211                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
2212                 break;
2213         case KVM_REG_PPC_HIOR:
2214                 *val = get_reg_val(id, 0);
2215                 break;
2216         case KVM_REG_PPC_DABR:
2217                 *val = get_reg_val(id, vcpu->arch.dabr);
2218                 break;
2219         case KVM_REG_PPC_DABRX:
2220                 *val = get_reg_val(id, vcpu->arch.dabrx);
2221                 break;
2222         case KVM_REG_PPC_DSCR:
2223                 *val = get_reg_val(id, kvmppc_get_dscr_hv(vcpu));
2224                 break;
2225         case KVM_REG_PPC_PURR:
2226                 *val = get_reg_val(id, kvmppc_get_purr_hv(vcpu));
2227                 break;
2228         case KVM_REG_PPC_SPURR:
2229                 *val = get_reg_val(id, kvmppc_get_spurr_hv(vcpu));
2230                 break;
2231         case KVM_REG_PPC_AMR:
2232                 *val = get_reg_val(id, kvmppc_get_amr_hv(vcpu));
2233                 break;
2234         case KVM_REG_PPC_UAMOR:
2235                 *val = get_reg_val(id, kvmppc_get_uamor_hv(vcpu));
2236                 break;
2237         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2238                 i = id - KVM_REG_PPC_MMCR0;
2239                 *val = get_reg_val(id, kvmppc_get_mmcr_hv(vcpu, i));
2240                 break;
2241         case KVM_REG_PPC_MMCR2:
2242                 *val = get_reg_val(id, kvmppc_get_mmcr_hv(vcpu, 2));
2243                 break;
2244         case KVM_REG_PPC_MMCRA:
2245                 *val = get_reg_val(id, kvmppc_get_mmcra_hv(vcpu));
2246                 break;
2247         case KVM_REG_PPC_MMCRS:
2248                 *val = get_reg_val(id, vcpu->arch.mmcrs);
2249                 break;
2250         case KVM_REG_PPC_MMCR3:
2251                 *val = get_reg_val(id, kvmppc_get_mmcr_hv(vcpu, 3));
2252                 break;
2253         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2254                 i = id - KVM_REG_PPC_PMC1;
2255                 *val = get_reg_val(id, kvmppc_get_pmc_hv(vcpu, i));
2256                 break;
2257         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2258                 i = id - KVM_REG_PPC_SPMC1;
2259                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
2260                 break;
2261         case KVM_REG_PPC_SIAR:
2262                 *val = get_reg_val(id, kvmppc_get_siar_hv(vcpu));
2263                 break;
2264         case KVM_REG_PPC_SDAR:
2265                 *val = get_reg_val(id, kvmppc_get_siar_hv(vcpu));
2266                 break;
2267         case KVM_REG_PPC_SIER:
2268                 *val = get_reg_val(id, kvmppc_get_sier_hv(vcpu, 0));
2269                 break;
2270         case KVM_REG_PPC_SIER2:
2271                 *val = get_reg_val(id, kvmppc_get_sier_hv(vcpu, 1));
2272                 break;
2273         case KVM_REG_PPC_SIER3:
2274                 *val = get_reg_val(id, kvmppc_get_sier_hv(vcpu, 2));
2275                 break;
2276         case KVM_REG_PPC_IAMR:
2277                 *val = get_reg_val(id, kvmppc_get_iamr_hv(vcpu));
2278                 break;
2279         case KVM_REG_PPC_PSPB:
2280                 *val = get_reg_val(id, kvmppc_get_pspb_hv(vcpu));
2281                 break;
2282         case KVM_REG_PPC_DPDES:
2283                 /*
2284                  * On POWER9, where we are emulating msgsndp etc.,
2285                  * we return 1 bit for each vcpu, which can come from
2286                  * either vcore->dpdes or doorbell_request.
2287                  * On POWER8, doorbell_request is 0.
2288                  */
2289                 if (cpu_has_feature(CPU_FTR_ARCH_300))
2290                         *val = get_reg_val(id, vcpu->arch.doorbell_request);
2291                 else
2292                         *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
2293                 break;
2294         case KVM_REG_PPC_VTB:
2295                 *val = get_reg_val(id, kvmppc_get_vtb(vcpu));
2296                 break;
2297         case KVM_REG_PPC_DAWR:
2298                 *val = get_reg_val(id, kvmppc_get_dawr0_hv(vcpu));
2299                 break;
2300         case KVM_REG_PPC_DAWRX:
2301                 *val = get_reg_val(id, kvmppc_get_dawrx0_hv(vcpu));
2302                 break;
2303         case KVM_REG_PPC_DAWR1:
2304                 *val = get_reg_val(id, kvmppc_get_dawr1_hv(vcpu));
2305                 break;
2306         case KVM_REG_PPC_DAWRX1:
2307                 *val = get_reg_val(id, kvmppc_get_dawrx1_hv(vcpu));
2308                 break;
2309         case KVM_REG_PPC_CIABR:
2310                 *val = get_reg_val(id, kvmppc_get_ciabr_hv(vcpu));
2311                 break;
2312         case KVM_REG_PPC_CSIGR:
2313                 *val = get_reg_val(id, vcpu->arch.csigr);
2314                 break;
2315         case KVM_REG_PPC_TACR:
2316                 *val = get_reg_val(id, vcpu->arch.tacr);
2317                 break;
2318         case KVM_REG_PPC_TCSCR:
2319                 *val = get_reg_val(id, vcpu->arch.tcscr);
2320                 break;
2321         case KVM_REG_PPC_PID:
2322                 *val = get_reg_val(id, kvmppc_get_pid(vcpu));
2323                 break;
2324         case KVM_REG_PPC_ACOP:
2325                 *val = get_reg_val(id, vcpu->arch.acop);
2326                 break;
2327         case KVM_REG_PPC_WORT:
2328                 *val = get_reg_val(id, kvmppc_get_wort_hv(vcpu));
2329                 break;
2330         case KVM_REG_PPC_TIDR:
2331                 *val = get_reg_val(id, vcpu->arch.tid);
2332                 break;
2333         case KVM_REG_PPC_PSSCR:
2334                 *val = get_reg_val(id, vcpu->arch.psscr);
2335                 break;
2336         case KVM_REG_PPC_VPA_ADDR:
2337                 spin_lock(&vcpu->arch.vpa_update_lock);
2338                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
2339                 spin_unlock(&vcpu->arch.vpa_update_lock);
2340                 break;
2341         case KVM_REG_PPC_VPA_SLB:
2342                 spin_lock(&vcpu->arch.vpa_update_lock);
2343                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
2344                 val->vpaval.length = vcpu->arch.slb_shadow.len;
2345                 spin_unlock(&vcpu->arch.vpa_update_lock);
2346                 break;
2347         case KVM_REG_PPC_VPA_DTL:
2348                 spin_lock(&vcpu->arch.vpa_update_lock);
2349                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
2350                 val->vpaval.length = vcpu->arch.dtl.len;
2351                 spin_unlock(&vcpu->arch.vpa_update_lock);
2352                 break;
2353         case KVM_REG_PPC_TB_OFFSET:
2354                 *val = get_reg_val(id, kvmppc_get_tb_offset(vcpu));
2355                 break;
2356         case KVM_REG_PPC_LPCR:
2357         case KVM_REG_PPC_LPCR_64:
2358                 *val = get_reg_val(id, kvmppc_get_lpcr(vcpu));
2359                 break;
2360         case KVM_REG_PPC_PPR:
2361                 *val = get_reg_val(id, kvmppc_get_ppr_hv(vcpu));
2362                 break;
2363 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2364         case KVM_REG_PPC_TFHAR:
2365                 *val = get_reg_val(id, vcpu->arch.tfhar);
2366                 break;
2367         case KVM_REG_PPC_TFIAR:
2368                 *val = get_reg_val(id, vcpu->arch.tfiar);
2369                 break;
2370         case KVM_REG_PPC_TEXASR:
2371                 *val = get_reg_val(id, vcpu->arch.texasr);
2372                 break;
2373         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2374                 i = id - KVM_REG_PPC_TM_GPR0;
2375                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
2376                 break;
2377         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2378         {
2379                 int j;
2380                 i = id - KVM_REG_PPC_TM_VSR0;
2381                 if (i < 32)
2382                         for (j = 0; j < TS_FPRWIDTH; j++)
2383                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
2384                 else {
2385                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
2386                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
2387                         else
2388                                 r = -ENXIO;
2389                 }
2390                 break;
2391         }
2392         case KVM_REG_PPC_TM_CR:
2393                 *val = get_reg_val(id, vcpu->arch.cr_tm);
2394                 break;
2395         case KVM_REG_PPC_TM_XER:
2396                 *val = get_reg_val(id, vcpu->arch.xer_tm);
2397                 break;
2398         case KVM_REG_PPC_TM_LR:
2399                 *val = get_reg_val(id, vcpu->arch.lr_tm);
2400                 break;
2401         case KVM_REG_PPC_TM_CTR:
2402                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
2403                 break;
2404         case KVM_REG_PPC_TM_FPSCR:
2405                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
2406                 break;
2407         case KVM_REG_PPC_TM_AMR:
2408                 *val = get_reg_val(id, vcpu->arch.amr_tm);
2409                 break;
2410         case KVM_REG_PPC_TM_PPR:
2411                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
2412                 break;
2413         case KVM_REG_PPC_TM_VRSAVE:
2414                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
2415                 break;
2416         case KVM_REG_PPC_TM_VSCR:
2417                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2418                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
2419                 else
2420                         r = -ENXIO;
2421                 break;
2422         case KVM_REG_PPC_TM_DSCR:
2423                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
2424                 break;
2425         case KVM_REG_PPC_TM_TAR:
2426                 *val = get_reg_val(id, vcpu->arch.tar_tm);
2427                 break;
2428 #endif
2429         case KVM_REG_PPC_ARCH_COMPAT:
2430                 *val = get_reg_val(id, kvmppc_get_arch_compat(vcpu));
2431                 break;
2432         case KVM_REG_PPC_DEC_EXPIRY:
2433                 *val = get_reg_val(id, kvmppc_get_dec_expires(vcpu));
2434                 break;
2435         case KVM_REG_PPC_ONLINE:
2436                 *val = get_reg_val(id, vcpu->arch.online);
2437                 break;
2438         case KVM_REG_PPC_PTCR:
2439                 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
2440                 break;
2441         case KVM_REG_PPC_FSCR:
2442                 *val = get_reg_val(id, kvmppc_get_fscr_hv(vcpu));
2443                 break;
2444         default:
2445                 r = -EINVAL;
2446                 break;
2447         }
2448
2449         return r;
2450 }
2451
2452 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2453                                  union kvmppc_one_reg *val)
2454 {
2455         int r = 0;
2456         long int i;
2457         unsigned long addr, len;
2458
2459         switch (id) {
2460         case KVM_REG_PPC_HIOR:
2461                 /* Only allow this to be set to zero */
2462                 if (set_reg_val(id, *val))
2463                         r = -EINVAL;
2464                 break;
2465         case KVM_REG_PPC_DABR:
2466                 vcpu->arch.dabr = set_reg_val(id, *val);
2467                 break;
2468         case KVM_REG_PPC_DABRX:
2469                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
2470                 break;
2471         case KVM_REG_PPC_DSCR:
2472                 kvmppc_set_dscr_hv(vcpu, set_reg_val(id, *val));
2473                 break;
2474         case KVM_REG_PPC_PURR:
2475                 kvmppc_set_purr_hv(vcpu, set_reg_val(id, *val));
2476                 break;
2477         case KVM_REG_PPC_SPURR:
2478                 kvmppc_set_spurr_hv(vcpu, set_reg_val(id, *val));
2479                 break;
2480         case KVM_REG_PPC_AMR:
2481                 kvmppc_set_amr_hv(vcpu, set_reg_val(id, *val));
2482                 break;
2483         case KVM_REG_PPC_UAMOR:
2484                 kvmppc_set_uamor_hv(vcpu, set_reg_val(id, *val));
2485                 break;
2486         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2487                 i = id - KVM_REG_PPC_MMCR0;
2488                 kvmppc_set_mmcr_hv(vcpu, i, set_reg_val(id, *val));
2489                 break;
2490         case KVM_REG_PPC_MMCR2:
2491                 kvmppc_set_mmcr_hv(vcpu, 2, set_reg_val(id, *val));
2492                 break;
2493         case KVM_REG_PPC_MMCRA:
2494                 kvmppc_set_mmcra_hv(vcpu, set_reg_val(id, *val));
2495                 break;
2496         case KVM_REG_PPC_MMCRS:
2497                 vcpu->arch.mmcrs = set_reg_val(id, *val);
2498                 break;
2499         case KVM_REG_PPC_MMCR3:
2500                 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
2501                 break;
2502         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2503                 i = id - KVM_REG_PPC_PMC1;
2504                 kvmppc_set_pmc_hv(vcpu, i, set_reg_val(id, *val));
2505                 break;
2506         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2507                 i = id - KVM_REG_PPC_SPMC1;
2508                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
2509                 break;
2510         case KVM_REG_PPC_SIAR:
2511                 kvmppc_set_siar_hv(vcpu, set_reg_val(id, *val));
2512                 break;
2513         case KVM_REG_PPC_SDAR:
2514                 kvmppc_set_sdar_hv(vcpu, set_reg_val(id, *val));
2515                 break;
2516         case KVM_REG_PPC_SIER:
2517                 kvmppc_set_sier_hv(vcpu, 0, set_reg_val(id, *val));
2518                 break;
2519         case KVM_REG_PPC_SIER2:
2520                 kvmppc_set_sier_hv(vcpu, 1, set_reg_val(id, *val));
2521                 break;
2522         case KVM_REG_PPC_SIER3:
2523                 kvmppc_set_sier_hv(vcpu, 2, set_reg_val(id, *val));
2524                 break;
2525         case KVM_REG_PPC_IAMR:
2526                 kvmppc_set_iamr_hv(vcpu, set_reg_val(id, *val));
2527                 break;
2528         case KVM_REG_PPC_PSPB:
2529                 kvmppc_set_pspb_hv(vcpu, set_reg_val(id, *val));
2530                 break;
2531         case KVM_REG_PPC_DPDES:
2532                 if (cpu_has_feature(CPU_FTR_ARCH_300))
2533                         vcpu->arch.doorbell_request = set_reg_val(id, *val) & 1;
2534                 else
2535                         vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
2536                 break;
2537         case KVM_REG_PPC_VTB:
2538                 kvmppc_set_vtb(vcpu, set_reg_val(id, *val));
2539                 break;
2540         case KVM_REG_PPC_DAWR:
2541                 kvmppc_set_dawr0_hv(vcpu, set_reg_val(id, *val));
2542                 break;
2543         case KVM_REG_PPC_DAWRX:
2544                 kvmppc_set_dawrx0_hv(vcpu, set_reg_val(id, *val) & ~DAWRX_HYP);
2545                 break;
2546         case KVM_REG_PPC_DAWR1:
2547                 kvmppc_set_dawr1_hv(vcpu, set_reg_val(id, *val));
2548                 break;
2549         case KVM_REG_PPC_DAWRX1:
2550                 kvmppc_set_dawrx1_hv(vcpu, set_reg_val(id, *val) & ~DAWRX_HYP);
2551                 break;
2552         case KVM_REG_PPC_CIABR:
2553                 kvmppc_set_ciabr_hv(vcpu, set_reg_val(id, *val));
2554                 /* Don't allow setting breakpoints in hypervisor code */
2555                 if ((kvmppc_get_ciabr_hv(vcpu) & CIABR_PRIV) == CIABR_PRIV_HYPER)
2556                         kvmppc_set_ciabr_hv(vcpu, kvmppc_get_ciabr_hv(vcpu) & ~CIABR_PRIV);
2557                 break;
2558         case KVM_REG_PPC_CSIGR:
2559                 vcpu->arch.csigr = set_reg_val(id, *val);
2560                 break;
2561         case KVM_REG_PPC_TACR:
2562                 vcpu->arch.tacr = set_reg_val(id, *val);
2563                 break;
2564         case KVM_REG_PPC_TCSCR:
2565                 vcpu->arch.tcscr = set_reg_val(id, *val);
2566                 break;
2567         case KVM_REG_PPC_PID:
2568                 kvmppc_set_pid(vcpu, set_reg_val(id, *val));
2569                 break;
2570         case KVM_REG_PPC_ACOP:
2571                 vcpu->arch.acop = set_reg_val(id, *val);
2572                 break;
2573         case KVM_REG_PPC_WORT:
2574                 kvmppc_set_wort_hv(vcpu, set_reg_val(id, *val));
2575                 break;
2576         case KVM_REG_PPC_TIDR:
2577                 vcpu->arch.tid = set_reg_val(id, *val);
2578                 break;
2579         case KVM_REG_PPC_PSSCR:
2580                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2581                 break;
2582         case KVM_REG_PPC_VPA_ADDR:
2583                 addr = set_reg_val(id, *val);
2584                 r = -EINVAL;
2585                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2586                               vcpu->arch.dtl.next_gpa))
2587                         break;
2588                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2589                 break;
2590         case KVM_REG_PPC_VPA_SLB:
2591                 addr = val->vpaval.addr;
2592                 len = val->vpaval.length;
2593                 r = -EINVAL;
2594                 if (addr && !vcpu->arch.vpa.next_gpa)
2595                         break;
2596                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2597                 break;
2598         case KVM_REG_PPC_VPA_DTL:
2599                 addr = val->vpaval.addr;
2600                 len = val->vpaval.length;
2601                 r = -EINVAL;
2602                 if (addr && (len < sizeof(struct dtl_entry) ||
2603                              !vcpu->arch.vpa.next_gpa))
2604                         break;
2605                 len -= len % sizeof(struct dtl_entry);
2606                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2607                 break;
2608         case KVM_REG_PPC_TB_OFFSET:
2609         {
2610                 /* round up to multiple of 2^24 */
2611                 u64 tb_offset = ALIGN(set_reg_val(id, *val), 1UL << 24);
2612
2613                 /*
2614                  * Now that we know the timebase offset, update the
2615                  * decrementer expiry with a guest timebase value. If
2616                  * the userspace does not set DEC_EXPIRY, this ensures
2617                  * a migrated vcpu at least starts with an expired
2618                  * decrementer, which is better than a large one that
2619                  * causes a hang.
2620                  */
2621                 kvmppc_set_tb_offset(vcpu, tb_offset);
2622                 if (!kvmppc_get_dec_expires(vcpu) && tb_offset)
2623                         kvmppc_set_dec_expires(vcpu, get_tb() + tb_offset);
2624
2625                 kvmppc_set_tb_offset(vcpu, tb_offset);
2626                 break;
2627         }
2628         case KVM_REG_PPC_LPCR:
2629                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2630                 break;
2631         case KVM_REG_PPC_LPCR_64:
2632                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2633                 break;
2634         case KVM_REG_PPC_PPR:
2635                 kvmppc_set_ppr_hv(vcpu, set_reg_val(id, *val));
2636                 break;
2637 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2638         case KVM_REG_PPC_TFHAR:
2639                 vcpu->arch.tfhar = set_reg_val(id, *val);
2640                 break;
2641         case KVM_REG_PPC_TFIAR:
2642                 vcpu->arch.tfiar = set_reg_val(id, *val);
2643                 break;
2644         case KVM_REG_PPC_TEXASR:
2645                 vcpu->arch.texasr = set_reg_val(id, *val);
2646                 break;
2647         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2648                 i = id - KVM_REG_PPC_TM_GPR0;
2649                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2650                 break;
2651         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2652         {
2653                 int j;
2654                 i = id - KVM_REG_PPC_TM_VSR0;
2655                 if (i < 32)
2656                         for (j = 0; j < TS_FPRWIDTH; j++)
2657                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2658                 else
2659                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
2660                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2661                         else
2662                                 r = -ENXIO;
2663                 break;
2664         }
2665         case KVM_REG_PPC_TM_CR:
2666                 vcpu->arch.cr_tm = set_reg_val(id, *val);
2667                 break;
2668         case KVM_REG_PPC_TM_XER:
2669                 vcpu->arch.xer_tm = set_reg_val(id, *val);
2670                 break;
2671         case KVM_REG_PPC_TM_LR:
2672                 vcpu->arch.lr_tm = set_reg_val(id, *val);
2673                 break;
2674         case KVM_REG_PPC_TM_CTR:
2675                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2676                 break;
2677         case KVM_REG_PPC_TM_FPSCR:
2678                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2679                 break;
2680         case KVM_REG_PPC_TM_AMR:
2681                 vcpu->arch.amr_tm = set_reg_val(id, *val);
2682                 break;
2683         case KVM_REG_PPC_TM_PPR:
2684                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2685                 break;
2686         case KVM_REG_PPC_TM_VRSAVE:
2687                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2688                 break;
2689         case KVM_REG_PPC_TM_VSCR:
2690                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2691                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2692                 else
2693                         r = - ENXIO;
2694                 break;
2695         case KVM_REG_PPC_TM_DSCR:
2696                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2697                 break;
2698         case KVM_REG_PPC_TM_TAR:
2699                 vcpu->arch.tar_tm = set_reg_val(id, *val);
2700                 break;
2701 #endif
2702         case KVM_REG_PPC_ARCH_COMPAT:
2703                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2704                 break;
2705         case KVM_REG_PPC_DEC_EXPIRY:
2706                 kvmppc_set_dec_expires(vcpu, set_reg_val(id, *val));
2707                 break;
2708         case KVM_REG_PPC_ONLINE:
2709                 i = set_reg_val(id, *val);
2710                 if (i && !vcpu->arch.online)
2711                         atomic_inc(&vcpu->arch.vcore->online_count);
2712                 else if (!i && vcpu->arch.online)
2713                         atomic_dec(&vcpu->arch.vcore->online_count);
2714                 vcpu->arch.online = i;
2715                 break;
2716         case KVM_REG_PPC_PTCR:
2717                 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2718                 break;
2719         case KVM_REG_PPC_FSCR:
2720                 kvmppc_set_fscr_hv(vcpu, set_reg_val(id, *val));
2721                 break;
2722         default:
2723                 r = -EINVAL;
2724                 break;
2725         }
2726
2727         return r;
2728 }
2729
2730 /*
2731  * On POWER9, threads are independent and can be in different partitions.
2732  * Therefore we consider each thread to be a subcore.
2733  * There is a restriction that all threads have to be in the same
2734  * MMU mode (radix or HPT), unfortunately, but since we only support
2735  * HPT guests on a HPT host so far, that isn't an impediment yet.
2736  */
2737 static int threads_per_vcore(struct kvm *kvm)
2738 {
2739         if (cpu_has_feature(CPU_FTR_ARCH_300))
2740                 return 1;
2741         return threads_per_subcore;
2742 }
2743
2744 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2745 {
2746         struct kvmppc_vcore *vcore;
2747
2748         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2749
2750         if (vcore == NULL)
2751                 return NULL;
2752
2753         spin_lock_init(&vcore->lock);
2754         spin_lock_init(&vcore->stoltb_lock);
2755         rcuwait_init(&vcore->wait);
2756         vcore->preempt_tb = TB_NIL;
2757         vcore->lpcr = kvm->arch.lpcr;
2758         vcore->first_vcpuid = id;
2759         vcore->kvm = kvm;
2760         INIT_LIST_HEAD(&vcore->preempt_list);
2761
2762         return vcore;
2763 }
2764
2765 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2766 static struct debugfs_timings_element {
2767         const char *name;
2768         size_t offset;
2769 } timings[] = {
2770 #ifdef CONFIG_KVM_BOOK3S_HV_P9_TIMING
2771         {"vcpu_entry",  offsetof(struct kvm_vcpu, arch.vcpu_entry)},
2772         {"guest_entry", offsetof(struct kvm_vcpu, arch.guest_entry)},
2773         {"in_guest",    offsetof(struct kvm_vcpu, arch.in_guest)},
2774         {"guest_exit",  offsetof(struct kvm_vcpu, arch.guest_exit)},
2775         {"vcpu_exit",   offsetof(struct kvm_vcpu, arch.vcpu_exit)},
2776         {"hypercall",   offsetof(struct kvm_vcpu, arch.hcall)},
2777         {"page_fault",  offsetof(struct kvm_vcpu, arch.pg_fault)},
2778 #else
2779         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
2780         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
2781         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
2782         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
2783         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
2784 #endif
2785 };
2786
2787 #define N_TIMINGS       (ARRAY_SIZE(timings))
2788
2789 struct debugfs_timings_state {
2790         struct kvm_vcpu *vcpu;
2791         unsigned int    buflen;
2792         char            buf[N_TIMINGS * 100];
2793 };
2794
2795 static int debugfs_timings_open(struct inode *inode, struct file *file)
2796 {
2797         struct kvm_vcpu *vcpu = inode->i_private;
2798         struct debugfs_timings_state *p;
2799
2800         p = kzalloc(sizeof(*p), GFP_KERNEL);
2801         if (!p)
2802                 return -ENOMEM;
2803
2804         kvm_get_kvm(vcpu->kvm);
2805         p->vcpu = vcpu;
2806         file->private_data = p;
2807
2808         return nonseekable_open(inode, file);
2809 }
2810
2811 static int debugfs_timings_release(struct inode *inode, struct file *file)
2812 {
2813         struct debugfs_timings_state *p = file->private_data;
2814
2815         kvm_put_kvm(p->vcpu->kvm);
2816         kfree(p);
2817         return 0;
2818 }
2819
2820 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2821                                     size_t len, loff_t *ppos)
2822 {
2823         struct debugfs_timings_state *p = file->private_data;
2824         struct kvm_vcpu *vcpu = p->vcpu;
2825         char *s, *buf_end;
2826         struct kvmhv_tb_accumulator tb;
2827         u64 count;
2828         loff_t pos;
2829         ssize_t n;
2830         int i, loops;
2831         bool ok;
2832
2833         if (!p->buflen) {
2834                 s = p->buf;
2835                 buf_end = s + sizeof(p->buf);
2836                 for (i = 0; i < N_TIMINGS; ++i) {
2837                         struct kvmhv_tb_accumulator *acc;
2838
2839                         acc = (struct kvmhv_tb_accumulator *)
2840                                 ((unsigned long)vcpu + timings[i].offset);
2841                         ok = false;
2842                         for (loops = 0; loops < 1000; ++loops) {
2843                                 count = acc->seqcount;
2844                                 if (!(count & 1)) {
2845                                         smp_rmb();
2846                                         tb = *acc;
2847                                         smp_rmb();
2848                                         if (count == acc->seqcount) {
2849                                                 ok = true;
2850                                                 break;
2851                                         }
2852                                 }
2853                                 udelay(1);
2854                         }
2855                         if (!ok)
2856                                 snprintf(s, buf_end - s, "%s: stuck\n",
2857                                         timings[i].name);
2858                         else
2859                                 snprintf(s, buf_end - s,
2860                                         "%s: %llu %llu %llu %llu\n",
2861                                         timings[i].name, count / 2,
2862                                         tb_to_ns(tb.tb_total),
2863                                         tb_to_ns(tb.tb_min),
2864                                         tb_to_ns(tb.tb_max));
2865                         s += strlen(s);
2866                 }
2867                 p->buflen = s - p->buf;
2868         }
2869
2870         pos = *ppos;
2871         if (pos >= p->buflen)
2872                 return 0;
2873         if (len > p->buflen - pos)
2874                 len = p->buflen - pos;
2875         n = copy_to_user(buf, p->buf + pos, len);
2876         if (n) {
2877                 if (n == len)
2878                         return -EFAULT;
2879                 len -= n;
2880         }
2881         *ppos = pos + len;
2882         return len;
2883 }
2884
2885 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2886                                      size_t len, loff_t *ppos)
2887 {
2888         return -EACCES;
2889 }
2890
2891 static const struct file_operations debugfs_timings_ops = {
2892         .owner   = THIS_MODULE,
2893         .open    = debugfs_timings_open,
2894         .release = debugfs_timings_release,
2895         .read    = debugfs_timings_read,
2896         .write   = debugfs_timings_write,
2897         .llseek  = generic_file_llseek,
2898 };
2899
2900 /* Create a debugfs directory for the vcpu */
2901 static int kvmppc_arch_create_vcpu_debugfs_hv(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry)
2902 {
2903         if (cpu_has_feature(CPU_FTR_ARCH_300) == IS_ENABLED(CONFIG_KVM_BOOK3S_HV_P9_TIMING))
2904                 debugfs_create_file("timings", 0444, debugfs_dentry, vcpu,
2905                                     &debugfs_timings_ops);
2906         return 0;
2907 }
2908
2909 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2910 static int kvmppc_arch_create_vcpu_debugfs_hv(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry)
2911 {
2912         return 0;
2913 }
2914 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2915
2916 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2917 {
2918         int err;
2919         int core;
2920         struct kvmppc_vcore *vcore;
2921         struct kvm *kvm;
2922         unsigned int id;
2923
2924         kvm = vcpu->kvm;
2925         id = vcpu->vcpu_id;
2926
2927         vcpu->arch.shared = &vcpu->arch.shregs;
2928 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2929         /*
2930          * The shared struct is never shared on HV,
2931          * so we can always use host endianness
2932          */
2933 #ifdef __BIG_ENDIAN__
2934         vcpu->arch.shared_big_endian = true;
2935 #else
2936         vcpu->arch.shared_big_endian = false;
2937 #endif
2938 #endif
2939
2940         if (kvmhv_is_nestedv2()) {
2941                 err = kvmhv_nestedv2_vcpu_create(vcpu, &vcpu->arch.nestedv2_io);
2942                 if (err < 0)
2943                         return err;
2944         }
2945
2946         kvmppc_set_mmcr_hv(vcpu, 0, MMCR0_FC);
2947         if (cpu_has_feature(CPU_FTR_ARCH_31)) {
2948                 kvmppc_set_mmcr_hv(vcpu, 0, kvmppc_get_mmcr_hv(vcpu, 0) | MMCR0_PMCCEXT);
2949                 kvmppc_set_mmcra_hv(vcpu, MMCRA_BHRB_DISABLE);
2950         }
2951
2952         kvmppc_set_ctrl_hv(vcpu, CTRL_RUNLATCH);
2953         /* default to host PVR, since we can't spoof it */
2954         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2955         spin_lock_init(&vcpu->arch.vpa_update_lock);
2956         spin_lock_init(&vcpu->arch.tbacct_lock);
2957         vcpu->arch.busy_preempt = TB_NIL;
2958         __kvmppc_set_msr_hv(vcpu, MSR_ME);
2959         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2960
2961         /*
2962          * Set the default HFSCR for the guest from the host value.
2963          * This value is only used on POWER9 and later.
2964          * On >= POWER9, we want to virtualize the doorbell facility, so we
2965          * don't set the HFSCR_MSGP bit, and that causes those instructions
2966          * to trap and then we emulate them.
2967          */
2968         kvmppc_set_hfscr_hv(vcpu, HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2969                             HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP);
2970
2971         /* On POWER10 and later, allow prefixed instructions */
2972         if (cpu_has_feature(CPU_FTR_ARCH_31))
2973                 kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_PREFIX);
2974
2975         if (cpu_has_feature(CPU_FTR_HVMODE)) {
2976                 kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) & mfspr(SPRN_HFSCR));
2977
2978 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2979                 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2980                         kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_TM);
2981 #endif
2982         }
2983         if (cpu_has_feature(CPU_FTR_TM_COMP))
2984                 vcpu->arch.hfscr |= HFSCR_TM;
2985
2986         vcpu->arch.hfscr_permitted = kvmppc_get_hfscr_hv(vcpu);
2987
2988         /*
2989          * PM, EBB, TM are demand-faulted so start with it clear.
2990          */
2991         kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) & ~(HFSCR_PM | HFSCR_EBB | HFSCR_TM));
2992
2993         kvmppc_mmu_book3s_hv_init(vcpu);
2994
2995         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2996
2997         init_waitqueue_head(&vcpu->arch.cpu_run);
2998
2999         mutex_lock(&kvm->lock);
3000         vcore = NULL;
3001         err = -EINVAL;
3002         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3003                 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
3004                         pr_devel("KVM: VCPU ID too high\n");
3005                         core = KVM_MAX_VCORES;
3006                 } else {
3007                         BUG_ON(kvm->arch.smt_mode != 1);
3008                         core = kvmppc_pack_vcpu_id(kvm, id);
3009                 }
3010         } else {
3011                 core = id / kvm->arch.smt_mode;
3012         }
3013         if (core < KVM_MAX_VCORES) {
3014                 vcore = kvm->arch.vcores[core];
3015                 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
3016                         pr_devel("KVM: collision on id %u", id);
3017                         vcore = NULL;
3018                 } else if (!vcore) {
3019                         /*
3020                          * Take mmu_setup_lock for mutual exclusion
3021                          * with kvmppc_update_lpcr().
3022                          */
3023                         err = -ENOMEM;
3024                         vcore = kvmppc_vcore_create(kvm,
3025                                         id & ~(kvm->arch.smt_mode - 1));
3026                         mutex_lock(&kvm->arch.mmu_setup_lock);
3027                         kvm->arch.vcores[core] = vcore;
3028                         kvm->arch.online_vcores++;
3029                         mutex_unlock(&kvm->arch.mmu_setup_lock);
3030                 }
3031         }
3032         mutex_unlock(&kvm->lock);
3033
3034         if (!vcore)
3035                 return err;
3036
3037         spin_lock(&vcore->lock);
3038         ++vcore->num_threads;
3039         spin_unlock(&vcore->lock);
3040         vcpu->arch.vcore = vcore;
3041         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
3042         vcpu->arch.thread_cpu = -1;
3043         vcpu->arch.prev_cpu = -1;
3044
3045         vcpu->arch.cpu_type = KVM_CPU_3S_64;
3046         kvmppc_sanity_check(vcpu);
3047
3048         return 0;
3049 }
3050
3051 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
3052                               unsigned long flags)
3053 {
3054         int err;
3055         int esmt = 0;
3056
3057         if (flags)
3058                 return -EINVAL;
3059         if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
3060                 return -EINVAL;
3061         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3062                 /*
3063                  * On POWER8 (or POWER7), the threading mode is "strict",
3064                  * so we pack smt_mode vcpus per vcore.
3065                  */
3066                 if (smt_mode > threads_per_subcore)
3067                         return -EINVAL;
3068         } else {
3069                 /*
3070                  * On POWER9, the threading mode is "loose",
3071                  * so each vcpu gets its own vcore.
3072                  */
3073                 esmt = smt_mode;
3074                 smt_mode = 1;
3075         }
3076         mutex_lock(&kvm->lock);
3077         err = -EBUSY;
3078         if (!kvm->arch.online_vcores) {
3079                 kvm->arch.smt_mode = smt_mode;
3080                 kvm->arch.emul_smt_mode = esmt;
3081                 err = 0;
3082         }
3083         mutex_unlock(&kvm->lock);
3084
3085         return err;
3086 }
3087
3088 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
3089 {
3090         if (vpa->pinned_addr)
3091                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
3092                                         vpa->dirty);
3093 }
3094
3095 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
3096 {
3097         spin_lock(&vcpu->arch.vpa_update_lock);
3098         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
3099         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
3100         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
3101         spin_unlock(&vcpu->arch.vpa_update_lock);
3102         if (kvmhv_is_nestedv2())
3103                 kvmhv_nestedv2_vcpu_free(vcpu, &vcpu->arch.nestedv2_io);
3104 }
3105
3106 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
3107 {
3108         /* Indicate we want to get back into the guest */
3109         return 1;
3110 }
3111
3112 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
3113 {
3114         unsigned long dec_nsec, now;
3115
3116         now = get_tb();
3117         if (now > kvmppc_dec_expires_host_tb(vcpu)) {
3118                 /* decrementer has already gone negative */
3119                 kvmppc_core_queue_dec(vcpu);
3120                 kvmppc_core_prepare_to_enter(vcpu);
3121                 return;
3122         }
3123         dec_nsec = tb_to_ns(kvmppc_dec_expires_host_tb(vcpu) - now);
3124         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
3125         vcpu->arch.timer_running = 1;
3126 }
3127
3128 extern int __kvmppc_vcore_entry(void);
3129
3130 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
3131                                    struct kvm_vcpu *vcpu, u64 tb)
3132 {
3133         u64 now;
3134
3135         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3136                 return;
3137         spin_lock_irq(&vcpu->arch.tbacct_lock);
3138         now = tb;
3139         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
3140                 vcpu->arch.stolen_logged;
3141         vcpu->arch.busy_preempt = now;
3142         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3143         spin_unlock_irq(&vcpu->arch.tbacct_lock);
3144         --vc->n_runnable;
3145         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
3146 }
3147
3148 static int kvmppc_grab_hwthread(int cpu)
3149 {
3150         struct paca_struct *tpaca;
3151         long timeout = 10000;
3152
3153         tpaca = paca_ptrs[cpu];
3154
3155         /* Ensure the thread won't go into the kernel if it wakes */
3156         tpaca->kvm_hstate.kvm_vcpu = NULL;
3157         tpaca->kvm_hstate.kvm_vcore = NULL;
3158         tpaca->kvm_hstate.napping = 0;
3159         smp_wmb();
3160         tpaca->kvm_hstate.hwthread_req = 1;
3161
3162         /*
3163          * If the thread is already executing in the kernel (e.g. handling
3164          * a stray interrupt), wait for it to get back to nap mode.
3165          * The smp_mb() is to ensure that our setting of hwthread_req
3166          * is visible before we look at hwthread_state, so if this
3167          * races with the code at system_reset_pSeries and the thread
3168          * misses our setting of hwthread_req, we are sure to see its
3169          * setting of hwthread_state, and vice versa.
3170          */
3171         smp_mb();
3172         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
3173                 if (--timeout <= 0) {
3174                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
3175                         return -EBUSY;
3176                 }
3177                 udelay(1);
3178         }
3179         return 0;
3180 }
3181
3182 static void kvmppc_release_hwthread(int cpu)
3183 {
3184         struct paca_struct *tpaca;
3185
3186         tpaca = paca_ptrs[cpu];
3187         tpaca->kvm_hstate.hwthread_req = 0;
3188         tpaca->kvm_hstate.kvm_vcpu = NULL;
3189         tpaca->kvm_hstate.kvm_vcore = NULL;
3190         tpaca->kvm_hstate.kvm_split_mode = NULL;
3191 }
3192
3193 static DEFINE_PER_CPU(struct kvm *, cpu_in_guest);
3194
3195 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
3196 {
3197         struct kvm_nested_guest *nested = vcpu->arch.nested;
3198         cpumask_t *need_tlb_flush;
3199         int i;
3200
3201         if (nested)
3202                 need_tlb_flush = &nested->need_tlb_flush;
3203         else
3204                 need_tlb_flush = &kvm->arch.need_tlb_flush;
3205
3206         cpu = cpu_first_tlb_thread_sibling(cpu);
3207         for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
3208                                         i += cpu_tlb_thread_sibling_step())
3209                 cpumask_set_cpu(i, need_tlb_flush);
3210
3211         /*
3212          * Make sure setting of bit in need_tlb_flush precedes testing of
3213          * cpu_in_guest. The matching barrier on the other side is hwsync
3214          * when switching to guest MMU mode, which happens between
3215          * cpu_in_guest being set to the guest kvm, and need_tlb_flush bit
3216          * being tested.
3217          */
3218         smp_mb();
3219
3220         for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
3221                                         i += cpu_tlb_thread_sibling_step()) {
3222                 struct kvm *running = *per_cpu_ptr(&cpu_in_guest, i);
3223
3224                 if (running == kvm)
3225                         smp_call_function_single(i, do_nothing, NULL, 1);
3226         }
3227 }
3228
3229 static void do_migrate_away_vcpu(void *arg)
3230 {
3231         struct kvm_vcpu *vcpu = arg;
3232         struct kvm *kvm = vcpu->kvm;
3233
3234         /*
3235          * If the guest has GTSE, it may execute tlbie, so do a eieio; tlbsync;
3236          * ptesync sequence on the old CPU before migrating to a new one, in
3237          * case we interrupted the guest between a tlbie ; eieio ;
3238          * tlbsync; ptesync sequence.
3239          *
3240          * Otherwise, ptesync is sufficient for ordering tlbiel sequences.
3241          */
3242         if (kvm->arch.lpcr & LPCR_GTSE)
3243                 asm volatile("eieio; tlbsync; ptesync");
3244         else
3245                 asm volatile("ptesync");
3246 }
3247
3248 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
3249 {
3250         struct kvm_nested_guest *nested = vcpu->arch.nested;
3251         struct kvm *kvm = vcpu->kvm;
3252         int prev_cpu;
3253
3254         if (!cpu_has_feature(CPU_FTR_HVMODE))
3255                 return;
3256
3257         if (nested)
3258                 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
3259         else
3260                 prev_cpu = vcpu->arch.prev_cpu;
3261
3262         /*
3263          * With radix, the guest can do TLB invalidations itself,
3264          * and it could choose to use the local form (tlbiel) if
3265          * it is invalidating a translation that has only ever been
3266          * used on one vcpu.  However, that doesn't mean it has
3267          * only ever been used on one physical cpu, since vcpus
3268          * can move around between pcpus.  To cope with this, when
3269          * a vcpu moves from one pcpu to another, we need to tell
3270          * any vcpus running on the same core as this vcpu previously
3271          * ran to flush the TLB.
3272          */
3273         if (prev_cpu != pcpu) {
3274                 if (prev_cpu >= 0) {
3275                         if (cpu_first_tlb_thread_sibling(prev_cpu) !=
3276                             cpu_first_tlb_thread_sibling(pcpu))
3277                                 radix_flush_cpu(kvm, prev_cpu, vcpu);
3278
3279                         smp_call_function_single(prev_cpu,
3280                                         do_migrate_away_vcpu, vcpu, 1);
3281                 }
3282                 if (nested)
3283                         nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
3284                 else
3285                         vcpu->arch.prev_cpu = pcpu;
3286         }
3287 }
3288
3289 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
3290 {
3291         int cpu;
3292         struct paca_struct *tpaca;
3293
3294         cpu = vc->pcpu;
3295         if (vcpu) {
3296                 if (vcpu->arch.timer_running) {
3297                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
3298                         vcpu->arch.timer_running = 0;
3299                 }
3300                 cpu += vcpu->arch.ptid;
3301                 vcpu->cpu = vc->pcpu;
3302                 vcpu->arch.thread_cpu = cpu;
3303         }
3304         tpaca = paca_ptrs[cpu];
3305         tpaca->kvm_hstate.kvm_vcpu = vcpu;
3306         tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
3307         tpaca->kvm_hstate.fake_suspend = 0;
3308         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
3309         smp_wmb();
3310         tpaca->kvm_hstate.kvm_vcore = vc;
3311         if (cpu != smp_processor_id())
3312                 kvmppc_ipi_thread(cpu);
3313 }
3314
3315 static void kvmppc_wait_for_nap(int n_threads)
3316 {
3317         int cpu = smp_processor_id();
3318         int i, loops;
3319
3320         if (n_threads <= 1)
3321                 return;
3322         for (loops = 0; loops < 1000000; ++loops) {
3323                 /*
3324                  * Check if all threads are finished.
3325                  * We set the vcore pointer when starting a thread
3326                  * and the thread clears it when finished, so we look
3327                  * for any threads that still have a non-NULL vcore ptr.
3328                  */
3329                 for (i = 1; i < n_threads; ++i)
3330                         if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3331                                 break;
3332                 if (i == n_threads) {
3333                         HMT_medium();
3334                         return;
3335                 }
3336                 HMT_low();
3337         }
3338         HMT_medium();
3339         for (i = 1; i < n_threads; ++i)
3340                 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3341                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
3342 }
3343
3344 /*
3345  * Check that we are on thread 0 and that any other threads in
3346  * this core are off-line.  Then grab the threads so they can't
3347  * enter the kernel.
3348  */
3349 static int on_primary_thread(void)
3350 {
3351         int cpu = smp_processor_id();
3352         int thr;
3353
3354         /* Are we on a primary subcore? */
3355         if (cpu_thread_in_subcore(cpu))
3356                 return 0;
3357
3358         thr = 0;
3359         while (++thr < threads_per_subcore)
3360                 if (cpu_online(cpu + thr))
3361                         return 0;
3362
3363         /* Grab all hw threads so they can't go into the kernel */
3364         for (thr = 1; thr < threads_per_subcore; ++thr) {
3365                 if (kvmppc_grab_hwthread(cpu + thr)) {
3366                         /* Couldn't grab one; let the others go */
3367                         do {
3368                                 kvmppc_release_hwthread(cpu + thr);
3369                         } while (--thr > 0);
3370                         return 0;
3371                 }
3372         }
3373         return 1;
3374 }
3375
3376 /*
3377  * A list of virtual cores for each physical CPU.
3378  * These are vcores that could run but their runner VCPU tasks are
3379  * (or may be) preempted.
3380  */
3381 struct preempted_vcore_list {
3382         struct list_head        list;
3383         spinlock_t              lock;
3384 };
3385
3386 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
3387
3388 static void init_vcore_lists(void)
3389 {
3390         int cpu;
3391
3392         for_each_possible_cpu(cpu) {
3393                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
3394                 spin_lock_init(&lp->lock);
3395                 INIT_LIST_HEAD(&lp->list);
3396         }
3397 }
3398
3399 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
3400 {
3401         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3402
3403         WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
3404
3405         vc->vcore_state = VCORE_PREEMPT;
3406         vc->pcpu = smp_processor_id();
3407         if (vc->num_threads < threads_per_vcore(vc->kvm)) {
3408                 spin_lock(&lp->lock);
3409                 list_add_tail(&vc->preempt_list, &lp->list);
3410                 spin_unlock(&lp->lock);
3411         }
3412
3413         /* Start accumulating stolen time */
3414         kvmppc_core_start_stolen(vc, mftb());
3415 }
3416
3417 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
3418 {
3419         struct preempted_vcore_list *lp;
3420
3421         WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
3422
3423         kvmppc_core_end_stolen(vc, mftb());
3424         if (!list_empty(&vc->preempt_list)) {
3425                 lp = &per_cpu(preempted_vcores, vc->pcpu);
3426                 spin_lock(&lp->lock);
3427                 list_del_init(&vc->preempt_list);
3428                 spin_unlock(&lp->lock);
3429         }
3430         vc->vcore_state = VCORE_INACTIVE;
3431 }
3432
3433 /*
3434  * This stores information about the virtual cores currently
3435  * assigned to a physical core.
3436  */
3437 struct core_info {
3438         int             n_subcores;
3439         int             max_subcore_threads;
3440         int             total_threads;
3441         int             subcore_threads[MAX_SUBCORES];
3442         struct kvmppc_vcore *vc[MAX_SUBCORES];
3443 };
3444
3445 /*
3446  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
3447  * respectively in 2-way micro-threading (split-core) mode on POWER8.
3448  */
3449 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
3450
3451 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
3452 {
3453         memset(cip, 0, sizeof(*cip));
3454         cip->n_subcores = 1;
3455         cip->max_subcore_threads = vc->num_threads;
3456         cip->total_threads = vc->num_threads;
3457         cip->subcore_threads[0] = vc->num_threads;
3458         cip->vc[0] = vc;
3459 }
3460
3461 static bool subcore_config_ok(int n_subcores, int n_threads)
3462 {
3463         /*
3464          * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
3465          * split-core mode, with one thread per subcore.
3466          */
3467         if (cpu_has_feature(CPU_FTR_ARCH_300))
3468                 return n_subcores <= 4 && n_threads == 1;
3469
3470         /* On POWER8, can only dynamically split if unsplit to begin with */
3471         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
3472                 return false;
3473         if (n_subcores > MAX_SUBCORES)
3474                 return false;
3475         if (n_subcores > 1) {
3476                 if (!(dynamic_mt_modes & 2))
3477                         n_subcores = 4;
3478                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
3479                         return false;
3480         }
3481
3482         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
3483 }
3484
3485 static void init_vcore_to_run(struct kvmppc_vcore *vc)
3486 {
3487         vc->entry_exit_map = 0;
3488         vc->in_guest = 0;
3489         vc->napping_threads = 0;
3490         vc->conferring_threads = 0;
3491         vc->tb_offset_applied = 0;
3492 }
3493
3494 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
3495 {
3496         int n_threads = vc->num_threads;
3497         int sub;
3498
3499         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
3500                 return false;
3501
3502         /* In one_vm_per_core mode, require all vcores to be from the same vm */
3503         if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
3504                 return false;
3505
3506         if (n_threads < cip->max_subcore_threads)
3507                 n_threads = cip->max_subcore_threads;
3508         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
3509                 return false;
3510         cip->max_subcore_threads = n_threads;
3511
3512         sub = cip->n_subcores;
3513         ++cip->n_subcores;
3514         cip->total_threads += vc->num_threads;
3515         cip->subcore_threads[sub] = vc->num_threads;
3516         cip->vc[sub] = vc;
3517         init_vcore_to_run(vc);
3518         list_del_init(&vc->preempt_list);
3519
3520         return true;
3521 }
3522
3523 /*
3524  * Work out whether it is possible to piggyback the execution of
3525  * vcore *pvc onto the execution of the other vcores described in *cip.
3526  */
3527 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
3528                           int target_threads)
3529 {
3530         if (cip->total_threads + pvc->num_threads > target_threads)
3531                 return false;
3532
3533         return can_dynamic_split(pvc, cip);
3534 }
3535
3536 static void prepare_threads(struct kvmppc_vcore *vc)
3537 {
3538         int i;
3539         struct kvm_vcpu *vcpu;
3540
3541         for_each_runnable_thread(i, vcpu, vc) {
3542                 if (signal_pending(vcpu->arch.run_task))
3543                         vcpu->arch.ret = -EINTR;
3544                 else if (vcpu->arch.vpa.update_pending ||
3545                          vcpu->arch.slb_shadow.update_pending ||
3546                          vcpu->arch.dtl.update_pending)
3547                         vcpu->arch.ret = RESUME_GUEST;
3548                 else
3549                         continue;
3550                 kvmppc_remove_runnable(vc, vcpu, mftb());
3551                 wake_up(&vcpu->arch.cpu_run);
3552         }
3553 }
3554
3555 static void collect_piggybacks(struct core_info *cip, int target_threads)
3556 {
3557         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3558         struct kvmppc_vcore *pvc, *vcnext;
3559
3560         spin_lock(&lp->lock);
3561         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
3562                 if (!spin_trylock(&pvc->lock))
3563                         continue;
3564                 prepare_threads(pvc);
3565                 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
3566                         list_del_init(&pvc->preempt_list);
3567                         if (pvc->runner == NULL) {
3568                                 pvc->vcore_state = VCORE_INACTIVE;
3569                                 kvmppc_core_end_stolen(pvc, mftb());
3570                         }
3571                         spin_unlock(&pvc->lock);
3572                         continue;
3573                 }
3574                 if (!can_piggyback(pvc, cip, target_threads)) {
3575                         spin_unlock(&pvc->lock);
3576                         continue;
3577                 }
3578                 kvmppc_core_end_stolen(pvc, mftb());
3579                 pvc->vcore_state = VCORE_PIGGYBACK;
3580                 if (cip->total_threads >= target_threads)
3581                         break;
3582         }
3583         spin_unlock(&lp->lock);
3584 }
3585
3586 static bool recheck_signals_and_mmu(struct core_info *cip)
3587 {
3588         int sub, i;
3589         struct kvm_vcpu *vcpu;
3590         struct kvmppc_vcore *vc;
3591
3592         for (sub = 0; sub < cip->n_subcores; ++sub) {
3593                 vc = cip->vc[sub];
3594                 if (!vc->kvm->arch.mmu_ready)
3595                         return true;
3596                 for_each_runnable_thread(i, vcpu, vc)
3597                         if (signal_pending(vcpu->arch.run_task))
3598                                 return true;
3599         }
3600         return false;
3601 }
3602
3603 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
3604 {
3605         int still_running = 0, i;
3606         u64 now;
3607         long ret;
3608         struct kvm_vcpu *vcpu;
3609
3610         spin_lock(&vc->lock);
3611         now = get_tb();
3612         for_each_runnable_thread(i, vcpu, vc) {
3613                 /*
3614                  * It's safe to unlock the vcore in the loop here, because
3615                  * for_each_runnable_thread() is safe against removal of
3616                  * the vcpu, and the vcore state is VCORE_EXITING here,
3617                  * so any vcpus becoming runnable will have their arch.trap
3618                  * set to zero and can't actually run in the guest.
3619                  */
3620                 spin_unlock(&vc->lock);
3621                 /* cancel pending dec exception if dec is positive */
3622                 if (now < kvmppc_dec_expires_host_tb(vcpu) &&
3623                     kvmppc_core_pending_dec(vcpu))
3624                         kvmppc_core_dequeue_dec(vcpu);
3625
3626                 trace_kvm_guest_exit(vcpu);
3627
3628                 ret = RESUME_GUEST;
3629                 if (vcpu->arch.trap)
3630                         ret = kvmppc_handle_exit_hv(vcpu,
3631                                                     vcpu->arch.run_task);
3632
3633                 vcpu->arch.ret = ret;
3634                 vcpu->arch.trap = 0;
3635
3636                 spin_lock(&vc->lock);
3637                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
3638                         if (vcpu->arch.pending_exceptions)
3639                                 kvmppc_core_prepare_to_enter(vcpu);
3640                         if (vcpu->arch.ceded)
3641                                 kvmppc_set_timer(vcpu);
3642                         else
3643                                 ++still_running;
3644                 } else {
3645                         kvmppc_remove_runnable(vc, vcpu, mftb());
3646                         wake_up(&vcpu->arch.cpu_run);
3647                 }
3648         }
3649         if (!is_master) {
3650                 if (still_running > 0) {
3651                         kvmppc_vcore_preempt(vc);
3652                 } else if (vc->runner) {
3653                         vc->vcore_state = VCORE_PREEMPT;
3654                         kvmppc_core_start_stolen(vc, mftb());
3655                 } else {
3656                         vc->vcore_state = VCORE_INACTIVE;
3657                 }
3658                 if (vc->n_runnable > 0 && vc->runner == NULL) {
3659                         /* make sure there's a candidate runner awake */
3660                         i = -1;
3661                         vcpu = next_runnable_thread(vc, &i);
3662                         wake_up(&vcpu->arch.cpu_run);
3663                 }
3664         }
3665         spin_unlock(&vc->lock);
3666 }
3667
3668 /*
3669  * Clear core from the list of active host cores as we are about to
3670  * enter the guest. Only do this if it is the primary thread of the
3671  * core (not if a subcore) that is entering the guest.
3672  */
3673 static inline int kvmppc_clear_host_core(unsigned int cpu)
3674 {
3675         int core;
3676
3677         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3678                 return 0;
3679         /*
3680          * Memory barrier can be omitted here as we will do a smp_wmb()
3681          * later in kvmppc_start_thread and we need ensure that state is
3682          * visible to other CPUs only after we enter guest.
3683          */
3684         core = cpu >> threads_shift;
3685         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3686         return 0;
3687 }
3688
3689 /*
3690  * Advertise this core as an active host core since we exited the guest
3691  * Only need to do this if it is the primary thread of the core that is
3692  * exiting.
3693  */
3694 static inline int kvmppc_set_host_core(unsigned int cpu)
3695 {
3696         int core;
3697
3698         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3699                 return 0;
3700
3701         /*
3702          * Memory barrier can be omitted here because we do a spin_unlock
3703          * immediately after this which provides the memory barrier.
3704          */
3705         core = cpu >> threads_shift;
3706         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3707         return 0;
3708 }
3709
3710 static void set_irq_happened(int trap)
3711 {
3712         switch (trap) {
3713         case BOOK3S_INTERRUPT_EXTERNAL:
3714                 local_paca->irq_happened |= PACA_IRQ_EE;
3715                 break;
3716         case BOOK3S_INTERRUPT_H_DOORBELL:
3717                 local_paca->irq_happened |= PACA_IRQ_DBELL;
3718                 break;
3719         case BOOK3S_INTERRUPT_HMI:
3720                 local_paca->irq_happened |= PACA_IRQ_HMI;
3721                 break;
3722         case BOOK3S_INTERRUPT_SYSTEM_RESET:
3723                 replay_system_reset();
3724                 break;
3725         }
3726 }
3727
3728 /*
3729  * Run a set of guest threads on a physical core.
3730  * Called with vc->lock held.
3731  */
3732 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3733 {
3734         struct kvm_vcpu *vcpu;
3735         int i;
3736         int srcu_idx;
3737         struct core_info core_info;
3738         struct kvmppc_vcore *pvc;
3739         struct kvm_split_mode split_info, *sip;
3740         int split, subcore_size, active;
3741         int sub;
3742         bool thr0_done;
3743         unsigned long cmd_bit, stat_bit;
3744         int pcpu, thr;
3745         int target_threads;
3746         int controlled_threads;
3747         int trap;
3748         bool is_power8;
3749
3750         if (WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)))
3751                 return;
3752
3753         /*
3754          * Remove from the list any threads that have a signal pending
3755          * or need a VPA update done
3756          */
3757         prepare_threads(vc);
3758
3759         /* if the runner is no longer runnable, let the caller pick a new one */
3760         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3761                 return;
3762
3763         /*
3764          * Initialize *vc.
3765          */
3766         init_vcore_to_run(vc);
3767         vc->preempt_tb = TB_NIL;
3768
3769         /*
3770          * Number of threads that we will be controlling: the same as
3771          * the number of threads per subcore, except on POWER9,
3772          * where it's 1 because the threads are (mostly) independent.
3773          */
3774         controlled_threads = threads_per_vcore(vc->kvm);
3775
3776         /*
3777          * Make sure we are running on primary threads, and that secondary
3778          * threads are offline.  Also check if the number of threads in this
3779          * guest are greater than the current system threads per guest.
3780          */
3781         if ((controlled_threads > 1) &&
3782             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
3783                 for_each_runnable_thread(i, vcpu, vc) {
3784                         vcpu->arch.ret = -EBUSY;
3785                         kvmppc_remove_runnable(vc, vcpu, mftb());
3786                         wake_up(&vcpu->arch.cpu_run);
3787                 }
3788                 goto out;
3789         }
3790
3791         /*
3792          * See if we could run any other vcores on the physical core
3793          * along with this one.
3794          */
3795         init_core_info(&core_info, vc);
3796         pcpu = smp_processor_id();
3797         target_threads = controlled_threads;
3798         if (target_smt_mode && target_smt_mode < target_threads)
3799                 target_threads = target_smt_mode;
3800         if (vc->num_threads < target_threads)
3801                 collect_piggybacks(&core_info, target_threads);
3802
3803         /*
3804          * Hard-disable interrupts, and check resched flag and signals.
3805          * If we need to reschedule or deliver a signal, clean up
3806          * and return without going into the guest(s).
3807          * If the mmu_ready flag has been cleared, don't go into the
3808          * guest because that means a HPT resize operation is in progress.
3809          */
3810         local_irq_disable();
3811         hard_irq_disable();
3812         if (lazy_irq_pending() || need_resched() ||
3813             recheck_signals_and_mmu(&core_info)) {
3814                 local_irq_enable();
3815                 vc->vcore_state = VCORE_INACTIVE;
3816                 /* Unlock all except the primary vcore */
3817                 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3818                         pvc = core_info.vc[sub];
3819                         /* Put back on to the preempted vcores list */
3820                         kvmppc_vcore_preempt(pvc);
3821                         spin_unlock(&pvc->lock);
3822                 }
3823                 for (i = 0; i < controlled_threads; ++i)
3824                         kvmppc_release_hwthread(pcpu + i);
3825                 return;
3826         }
3827
3828         kvmppc_clear_host_core(pcpu);
3829
3830         /* Decide on micro-threading (split-core) mode */
3831         subcore_size = threads_per_subcore;
3832         cmd_bit = stat_bit = 0;
3833         split = core_info.n_subcores;
3834         sip = NULL;
3835         is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S);
3836
3837         if (split > 1) {
3838                 sip = &split_info;
3839                 memset(&split_info, 0, sizeof(split_info));
3840                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3841                         split_info.vc[sub] = core_info.vc[sub];
3842
3843                 if (is_power8) {
3844                         if (split == 2 && (dynamic_mt_modes & 2)) {
3845                                 cmd_bit = HID0_POWER8_1TO2LPAR;
3846                                 stat_bit = HID0_POWER8_2LPARMODE;
3847                         } else {
3848                                 split = 4;
3849                                 cmd_bit = HID0_POWER8_1TO4LPAR;
3850                                 stat_bit = HID0_POWER8_4LPARMODE;
3851                         }
3852                         subcore_size = MAX_SMT_THREADS / split;
3853                         split_info.rpr = mfspr(SPRN_RPR);
3854                         split_info.pmmar = mfspr(SPRN_PMMAR);
3855                         split_info.ldbar = mfspr(SPRN_LDBAR);
3856                         split_info.subcore_size = subcore_size;
3857                 } else {
3858                         split_info.subcore_size = 1;
3859                 }
3860
3861                 /* order writes to split_info before kvm_split_mode pointer */
3862                 smp_wmb();
3863         }
3864
3865         for (thr = 0; thr < controlled_threads; ++thr) {
3866                 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3867
3868                 paca->kvm_hstate.napping = 0;
3869                 paca->kvm_hstate.kvm_split_mode = sip;
3870         }
3871
3872         /* Initiate micro-threading (split-core) on POWER8 if required */
3873         if (cmd_bit) {
3874                 unsigned long hid0 = mfspr(SPRN_HID0);
3875
3876                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3877                 mb();
3878                 mtspr(SPRN_HID0, hid0);
3879                 isync();
3880                 for (;;) {
3881                         hid0 = mfspr(SPRN_HID0);
3882                         if (hid0 & stat_bit)
3883                                 break;
3884                         cpu_relax();
3885                 }
3886         }
3887
3888         /*
3889          * On POWER8, set RWMR register.
3890          * Since it only affects PURR and SPURR, it doesn't affect
3891          * the host, so we don't save/restore the host value.
3892          */
3893         if (is_power8) {
3894                 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3895                 int n_online = atomic_read(&vc->online_count);
3896
3897                 /*
3898                  * Use the 8-thread value if we're doing split-core
3899                  * or if the vcore's online count looks bogus.
3900                  */
3901                 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3902                     n_online >= 1 && n_online <= MAX_SMT_THREADS)
3903                         rwmr_val = p8_rwmr_values[n_online];
3904                 mtspr(SPRN_RWMR, rwmr_val);
3905         }
3906
3907         /* Start all the threads */
3908         active = 0;
3909         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3910                 thr = is_power8 ? subcore_thread_map[sub] : sub;
3911                 thr0_done = false;
3912                 active |= 1 << thr;
3913                 pvc = core_info.vc[sub];
3914                 pvc->pcpu = pcpu + thr;
3915                 for_each_runnable_thread(i, vcpu, pvc) {
3916                         /*
3917                          * XXX: is kvmppc_start_thread called too late here?
3918                          * It updates vcpu->cpu and vcpu->arch.thread_cpu
3919                          * which are used by kvmppc_fast_vcpu_kick_hv(), but
3920                          * kick is called after new exceptions become available
3921                          * and exceptions are checked earlier than here, by
3922                          * kvmppc_core_prepare_to_enter.
3923                          */
3924                         kvmppc_start_thread(vcpu, pvc);
3925                         kvmppc_update_vpa_dispatch(vcpu, pvc);
3926                         trace_kvm_guest_enter(vcpu);
3927                         if (!vcpu->arch.ptid)
3928                                 thr0_done = true;
3929                         active |= 1 << (thr + vcpu->arch.ptid);
3930                 }
3931                 /*
3932                  * We need to start the first thread of each subcore
3933                  * even if it doesn't have a vcpu.
3934                  */
3935                 if (!thr0_done)
3936                         kvmppc_start_thread(NULL, pvc);
3937         }
3938
3939         /*
3940          * Ensure that split_info.do_nap is set after setting
3941          * the vcore pointer in the PACA of the secondaries.
3942          */
3943         smp_mb();
3944
3945         /*
3946          * When doing micro-threading, poke the inactive threads as well.
3947          * This gets them to the nap instruction after kvm_do_nap,
3948          * which reduces the time taken to unsplit later.
3949          */
3950         if (cmd_bit) {
3951                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
3952                 for (thr = 1; thr < threads_per_subcore; ++thr)
3953                         if (!(active & (1 << thr)))
3954                                 kvmppc_ipi_thread(pcpu + thr);
3955         }
3956
3957         vc->vcore_state = VCORE_RUNNING;
3958         preempt_disable();
3959
3960         trace_kvmppc_run_core(vc, 0);
3961
3962         for (sub = 0; sub < core_info.n_subcores; ++sub)
3963                 spin_unlock(&core_info.vc[sub]->lock);
3964
3965         guest_timing_enter_irqoff();
3966
3967         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3968
3969         guest_state_enter_irqoff();
3970         this_cpu_disable_ftrace();
3971
3972         trap = __kvmppc_vcore_entry();
3973
3974         this_cpu_enable_ftrace();
3975         guest_state_exit_irqoff();
3976
3977         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3978
3979         set_irq_happened(trap);
3980
3981         spin_lock(&vc->lock);
3982         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3983         vc->vcore_state = VCORE_EXITING;
3984
3985         /* wait for secondary threads to finish writing their state to memory */
3986         kvmppc_wait_for_nap(controlled_threads);
3987
3988         /* Return to whole-core mode if we split the core earlier */
3989         if (cmd_bit) {
3990                 unsigned long hid0 = mfspr(SPRN_HID0);
3991                 unsigned long loops = 0;
3992
3993                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3994                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3995                 mb();
3996                 mtspr(SPRN_HID0, hid0);
3997                 isync();
3998                 for (;;) {
3999                         hid0 = mfspr(SPRN_HID0);
4000                         if (!(hid0 & stat_bit))
4001                                 break;
4002                         cpu_relax();
4003                         ++loops;
4004                 }
4005                 split_info.do_nap = 0;
4006         }
4007
4008         kvmppc_set_host_core(pcpu);
4009
4010         if (!vtime_accounting_enabled_this_cpu()) {
4011                 local_irq_enable();
4012                 /*
4013                  * Service IRQs here before guest_timing_exit_irqoff() so any
4014                  * ticks that occurred while running the guest are accounted to
4015                  * the guest. If vtime accounting is enabled, accounting uses
4016                  * TB rather than ticks, so it can be done without enabling
4017                  * interrupts here, which has the problem that it accounts
4018                  * interrupt processing overhead to the host.
4019                  */
4020                 local_irq_disable();
4021         }
4022         guest_timing_exit_irqoff();
4023
4024         local_irq_enable();
4025
4026         /* Let secondaries go back to the offline loop */
4027         for (i = 0; i < controlled_threads; ++i) {
4028                 kvmppc_release_hwthread(pcpu + i);
4029                 if (sip && sip->napped[i])
4030                         kvmppc_ipi_thread(pcpu + i);
4031         }
4032
4033         spin_unlock(&vc->lock);
4034
4035         /* make sure updates to secondary vcpu structs are visible now */
4036         smp_mb();
4037
4038         preempt_enable();
4039
4040         for (sub = 0; sub < core_info.n_subcores; ++sub) {
4041                 pvc = core_info.vc[sub];
4042                 post_guest_process(pvc, pvc == vc);
4043         }
4044
4045         spin_lock(&vc->lock);
4046
4047  out:
4048         vc->vcore_state = VCORE_INACTIVE;
4049         trace_kvmppc_run_core(vc, 1);
4050 }
4051
4052 static inline bool hcall_is_xics(unsigned long req)
4053 {
4054         return req == H_EOI || req == H_CPPR || req == H_IPI ||
4055                 req == H_IPOLL || req == H_XIRR || req == H_XIRR_X;
4056 }
4057
4058 static void vcpu_vpa_increment_dispatch(struct kvm_vcpu *vcpu)
4059 {
4060         struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
4061         if (lp) {
4062                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
4063                 lp->yield_count = cpu_to_be32(yield_count);
4064                 vcpu->arch.vpa.dirty = 1;
4065         }
4066 }
4067
4068 static int kvmhv_vcpu_entry_nestedv2(struct kvm_vcpu *vcpu, u64 time_limit,
4069                                      unsigned long lpcr, u64 *tb)
4070 {
4071         struct kvmhv_nestedv2_io *io;
4072         unsigned long msr, i;
4073         int trap;
4074         long rc;
4075
4076         io = &vcpu->arch.nestedv2_io;
4077
4078         msr = mfmsr();
4079         kvmppc_msr_hard_disable_set_facilities(vcpu, msr);
4080         if (lazy_irq_pending())
4081                 return 0;
4082
4083         rc = kvmhv_nestedv2_flush_vcpu(vcpu, time_limit);
4084         if (rc < 0)
4085                 return -EINVAL;
4086
4087         accumulate_time(vcpu, &vcpu->arch.in_guest);
4088         rc = plpar_guest_run_vcpu(0, vcpu->kvm->arch.lpid, vcpu->vcpu_id,
4089                                   &trap, &i);
4090
4091         if (rc != H_SUCCESS) {
4092                 pr_err("KVM Guest Run VCPU hcall failed\n");
4093                 if (rc == H_INVALID_ELEMENT_ID)
4094                         pr_err("KVM: Guest Run VCPU invalid element id at %ld\n", i);
4095                 else if (rc == H_INVALID_ELEMENT_SIZE)
4096                         pr_err("KVM: Guest Run VCPU invalid element size at %ld\n", i);
4097                 else if (rc == H_INVALID_ELEMENT_VALUE)
4098                         pr_err("KVM: Guest Run VCPU invalid element value at %ld\n", i);
4099                 return -EINVAL;
4100         }
4101         accumulate_time(vcpu, &vcpu->arch.guest_exit);
4102
4103         *tb = mftb();
4104         kvmppc_gsm_reset(io->vcpu_message);
4105         kvmppc_gsm_reset(io->vcore_message);
4106         kvmppc_gsbm_zero(&io->valids);
4107
4108         rc = kvmhv_nestedv2_parse_output(vcpu);
4109         if (rc < 0)
4110                 return -EINVAL;
4111
4112         timer_rearm_host_dec(*tb);
4113
4114         return trap;
4115 }
4116
4117 /* call our hypervisor to load up HV regs and go */
4118 static int kvmhv_vcpu_entry_p9_nested(struct kvm_vcpu *vcpu, u64 time_limit, unsigned long lpcr, u64 *tb)
4119 {
4120         unsigned long host_psscr;
4121         unsigned long msr;
4122         struct hv_guest_state hvregs;
4123         struct p9_host_os_sprs host_os_sprs;
4124         s64 dec;
4125         int trap;
4126
4127         msr = mfmsr();
4128
4129         save_p9_host_os_sprs(&host_os_sprs);
4130
4131         /*
4132          * We need to save and restore the guest visible part of the
4133          * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
4134          * doesn't do this for us. Note only required if pseries since
4135          * this is done in kvmhv_vcpu_entry_p9() below otherwise.
4136          */
4137         host_psscr = mfspr(SPRN_PSSCR_PR);
4138
4139         kvmppc_msr_hard_disable_set_facilities(vcpu, msr);
4140         if (lazy_irq_pending())
4141                 return 0;
4142
4143         if (unlikely(load_vcpu_state(vcpu, &host_os_sprs)))
4144                 msr = mfmsr(); /* TM restore can update msr */
4145
4146         if (vcpu->arch.psscr != host_psscr)
4147                 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
4148
4149         kvmhv_save_hv_regs(vcpu, &hvregs);
4150         hvregs.lpcr = lpcr;
4151         hvregs.amor = ~0;
4152         vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
4153         hvregs.version = HV_GUEST_STATE_VERSION;
4154         if (vcpu->arch.nested) {
4155                 hvregs.lpid = vcpu->arch.nested->shadow_lpid;
4156                 hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
4157         } else {
4158                 hvregs.lpid = vcpu->kvm->arch.lpid;
4159                 hvregs.vcpu_token = vcpu->vcpu_id;
4160         }
4161         hvregs.hdec_expiry = time_limit;
4162
4163         /*
4164          * When setting DEC, we must always deal with irq_work_raise
4165          * via NMI vs setting DEC. The problem occurs right as we
4166          * switch into guest mode if a NMI hits and sets pending work
4167          * and sets DEC, then that will apply to the guest and not
4168          * bring us back to the host.
4169          *
4170          * irq_work_raise could check a flag (or possibly LPCR[HDICE]
4171          * for example) and set HDEC to 1? That wouldn't solve the
4172          * nested hv case which needs to abort the hcall or zero the
4173          * time limit.
4174          *
4175          * XXX: Another day's problem.
4176          */
4177         mtspr(SPRN_DEC, kvmppc_dec_expires_host_tb(vcpu) - *tb);
4178
4179         mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
4180         mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
4181         switch_pmu_to_guest(vcpu, &host_os_sprs);
4182         accumulate_time(vcpu, &vcpu->arch.in_guest);
4183         trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
4184                                   __pa(&vcpu->arch.regs));
4185         accumulate_time(vcpu, &vcpu->arch.guest_exit);
4186         kvmhv_restore_hv_return_state(vcpu, &hvregs);
4187         switch_pmu_to_host(vcpu, &host_os_sprs);
4188         vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
4189         vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
4190         vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
4191         vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
4192
4193         store_vcpu_state(vcpu);
4194
4195         dec = mfspr(SPRN_DEC);
4196         if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
4197                 dec = (s32) dec;
4198         *tb = mftb();
4199         vcpu->arch.dec_expires = dec + (*tb + kvmppc_get_tb_offset(vcpu));
4200
4201         timer_rearm_host_dec(*tb);
4202
4203         restore_p9_host_os_sprs(vcpu, &host_os_sprs);
4204         if (vcpu->arch.psscr != host_psscr)
4205                 mtspr(SPRN_PSSCR_PR, host_psscr);
4206
4207         return trap;
4208 }
4209
4210 /*
4211  * Guest entry for POWER9 and later CPUs.
4212  */
4213 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
4214                          unsigned long lpcr, u64 *tb)
4215 {
4216         struct kvm *kvm = vcpu->kvm;
4217         struct kvm_nested_guest *nested = vcpu->arch.nested;
4218         u64 next_timer;
4219         int trap;
4220
4221         next_timer = timer_get_next_tb();
4222         if (*tb >= next_timer)
4223                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
4224         if (next_timer < time_limit)
4225                 time_limit = next_timer;
4226         else if (*tb >= time_limit) /* nested time limit */
4227                 return BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER;
4228
4229         vcpu->arch.ceded = 0;
4230
4231         vcpu_vpa_increment_dispatch(vcpu);
4232
4233         if (kvmhv_on_pseries()) {
4234                 if (kvmhv_is_nestedv1())
4235                         trap = kvmhv_vcpu_entry_p9_nested(vcpu, time_limit, lpcr, tb);
4236                 else
4237                         trap = kvmhv_vcpu_entry_nestedv2(vcpu, time_limit, lpcr, tb);
4238
4239                 /* H_CEDE has to be handled now, not later */
4240                 if (trap == BOOK3S_INTERRUPT_SYSCALL && !nested &&
4241                     kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
4242                         kvmppc_cede(vcpu);
4243                         kvmppc_set_gpr(vcpu, 3, 0);
4244                         trap = 0;
4245                 }
4246
4247         } else if (nested) {
4248                 __this_cpu_write(cpu_in_guest, kvm);
4249                 trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb);
4250                 __this_cpu_write(cpu_in_guest, NULL);
4251
4252         } else {
4253                 kvmppc_xive_push_vcpu(vcpu);
4254
4255                 __this_cpu_write(cpu_in_guest, kvm);
4256                 trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb);
4257                 __this_cpu_write(cpu_in_guest, NULL);
4258
4259                 if (trap == BOOK3S_INTERRUPT_SYSCALL &&
4260                     !(__kvmppc_get_msr_hv(vcpu) & MSR_PR)) {
4261                         unsigned long req = kvmppc_get_gpr(vcpu, 3);
4262
4263                         /*
4264                          * XIVE rearm and XICS hcalls must be handled
4265                          * before xive context is pulled (is this
4266                          * true?)
4267                          */
4268                         if (req == H_CEDE) {
4269                                 /* H_CEDE has to be handled now */
4270                                 kvmppc_cede(vcpu);
4271                                 if (!kvmppc_xive_rearm_escalation(vcpu)) {
4272                                         /*
4273                                          * Pending escalation so abort
4274                                          * the cede.
4275                                          */
4276                                         vcpu->arch.ceded = 0;
4277                                 }
4278                                 kvmppc_set_gpr(vcpu, 3, 0);
4279                                 trap = 0;
4280
4281                         } else if (req == H_ENTER_NESTED) {
4282                                 /*
4283                                  * L2 should not run with the L1
4284                                  * context so rearm and pull it.
4285                                  */
4286                                 if (!kvmppc_xive_rearm_escalation(vcpu)) {
4287                                         /*
4288                                          * Pending escalation so abort
4289                                          * H_ENTER_NESTED.
4290                                          */
4291                                         kvmppc_set_gpr(vcpu, 3, 0);
4292                                         trap = 0;
4293                                 }
4294
4295                         } else if (hcall_is_xics(req)) {
4296                                 int ret;
4297
4298                                 ret = kvmppc_xive_xics_hcall(vcpu, req);
4299                                 if (ret != H_TOO_HARD) {
4300                                         kvmppc_set_gpr(vcpu, 3, ret);
4301                                         trap = 0;
4302                                 }
4303                         }
4304                 }
4305                 kvmppc_xive_pull_vcpu(vcpu);
4306
4307                 if (kvm_is_radix(kvm))
4308                         vcpu->arch.slb_max = 0;
4309         }
4310
4311         vcpu_vpa_increment_dispatch(vcpu);
4312
4313         return trap;
4314 }
4315
4316 /*
4317  * Wait for some other vcpu thread to execute us, and
4318  * wake us up when we need to handle something in the host.
4319  */
4320 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
4321                                  struct kvm_vcpu *vcpu, int wait_state)
4322 {
4323         DEFINE_WAIT(wait);
4324
4325         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
4326         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4327                 spin_unlock(&vc->lock);
4328                 schedule();
4329                 spin_lock(&vc->lock);
4330         }
4331         finish_wait(&vcpu->arch.cpu_run, &wait);
4332 }
4333
4334 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
4335 {
4336         if (!halt_poll_ns_grow)
4337                 return;
4338
4339         vc->halt_poll_ns *= halt_poll_ns_grow;
4340         if (vc->halt_poll_ns < halt_poll_ns_grow_start)
4341                 vc->halt_poll_ns = halt_poll_ns_grow_start;
4342 }
4343
4344 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
4345 {
4346         if (halt_poll_ns_shrink == 0)
4347                 vc->halt_poll_ns = 0;
4348         else
4349                 vc->halt_poll_ns /= halt_poll_ns_shrink;
4350 }
4351
4352 #ifdef CONFIG_KVM_XICS
4353 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4354 {
4355         if (!xics_on_xive())
4356                 return false;
4357         return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
4358                 vcpu->arch.xive_saved_state.cppr;
4359 }
4360 #else
4361 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4362 {
4363         return false;
4364 }
4365 #endif /* CONFIG_KVM_XICS */
4366
4367 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
4368 {
4369         if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
4370             kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
4371                 return true;
4372
4373         return false;
4374 }
4375
4376 static bool kvmppc_vcpu_check_block(struct kvm_vcpu *vcpu)
4377 {
4378         if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
4379                 return true;
4380         return false;
4381 }
4382
4383 /*
4384  * Check to see if any of the runnable vcpus on the vcore have pending
4385  * exceptions or are no longer ceded
4386  */
4387 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
4388 {
4389         struct kvm_vcpu *vcpu;
4390         int i;
4391
4392         for_each_runnable_thread(i, vcpu, vc) {
4393                 if (kvmppc_vcpu_check_block(vcpu))
4394                         return 1;
4395         }
4396
4397         return 0;
4398 }
4399
4400 /*
4401  * All the vcpus in this vcore are idle, so wait for a decrementer
4402  * or external interrupt to one of the vcpus.  vc->lock is held.
4403  */
4404 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
4405 {
4406         ktime_t cur, start_poll, start_wait;
4407         int do_sleep = 1;
4408         u64 block_ns;
4409
4410         WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
4411
4412         /* Poll for pending exceptions and ceded state */
4413         cur = start_poll = ktime_get();
4414         if (vc->halt_poll_ns) {
4415                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
4416                 ++vc->runner->stat.generic.halt_attempted_poll;
4417
4418                 vc->vcore_state = VCORE_POLLING;
4419                 spin_unlock(&vc->lock);
4420
4421                 do {
4422                         if (kvmppc_vcore_check_block(vc)) {
4423                                 do_sleep = 0;
4424                                 break;
4425                         }
4426                         cur = ktime_get();
4427                 } while (kvm_vcpu_can_poll(cur, stop));
4428
4429                 spin_lock(&vc->lock);
4430                 vc->vcore_state = VCORE_INACTIVE;
4431
4432                 if (!do_sleep) {
4433                         ++vc->runner->stat.generic.halt_successful_poll;
4434                         goto out;
4435                 }
4436         }
4437
4438         prepare_to_rcuwait(&vc->wait);
4439         set_current_state(TASK_INTERRUPTIBLE);
4440         if (kvmppc_vcore_check_block(vc)) {
4441                 finish_rcuwait(&vc->wait);
4442                 do_sleep = 0;
4443                 /* If we polled, count this as a successful poll */
4444                 if (vc->halt_poll_ns)
4445                         ++vc->runner->stat.generic.halt_successful_poll;
4446                 goto out;
4447         }
4448
4449         start_wait = ktime_get();
4450
4451         vc->vcore_state = VCORE_SLEEPING;
4452         trace_kvmppc_vcore_blocked(vc->runner, 0);
4453         spin_unlock(&vc->lock);
4454         schedule();
4455         finish_rcuwait(&vc->wait);
4456         spin_lock(&vc->lock);
4457         vc->vcore_state = VCORE_INACTIVE;
4458         trace_kvmppc_vcore_blocked(vc->runner, 1);
4459         ++vc->runner->stat.halt_successful_wait;
4460
4461         cur = ktime_get();
4462
4463 out:
4464         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
4465
4466         /* Attribute wait time */
4467         if (do_sleep) {
4468                 vc->runner->stat.generic.halt_wait_ns +=
4469                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
4470                 KVM_STATS_LOG_HIST_UPDATE(
4471                                 vc->runner->stat.generic.halt_wait_hist,
4472                                 ktime_to_ns(cur) - ktime_to_ns(start_wait));
4473                 /* Attribute failed poll time */
4474                 if (vc->halt_poll_ns) {
4475                         vc->runner->stat.generic.halt_poll_fail_ns +=
4476                                 ktime_to_ns(start_wait) -
4477                                 ktime_to_ns(start_poll);
4478                         KVM_STATS_LOG_HIST_UPDATE(
4479                                 vc->runner->stat.generic.halt_poll_fail_hist,
4480                                 ktime_to_ns(start_wait) -
4481                                 ktime_to_ns(start_poll));
4482                 }
4483         } else {
4484                 /* Attribute successful poll time */
4485                 if (vc->halt_poll_ns) {
4486                         vc->runner->stat.generic.halt_poll_success_ns +=
4487                                 ktime_to_ns(cur) -
4488                                 ktime_to_ns(start_poll);
4489                         KVM_STATS_LOG_HIST_UPDATE(
4490                                 vc->runner->stat.generic.halt_poll_success_hist,
4491                                 ktime_to_ns(cur) - ktime_to_ns(start_poll));
4492                 }
4493         }
4494
4495         /* Adjust poll time */
4496         if (halt_poll_ns) {
4497                 if (block_ns <= vc->halt_poll_ns)
4498                         ;
4499                 /* We slept and blocked for longer than the max halt time */
4500                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
4501                         shrink_halt_poll_ns(vc);
4502                 /* We slept and our poll time is too small */
4503                 else if (vc->halt_poll_ns < halt_poll_ns &&
4504                                 block_ns < halt_poll_ns)
4505                         grow_halt_poll_ns(vc);
4506                 if (vc->halt_poll_ns > halt_poll_ns)
4507                         vc->halt_poll_ns = halt_poll_ns;
4508         } else
4509                 vc->halt_poll_ns = 0;
4510
4511         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
4512 }
4513
4514 /*
4515  * This never fails for a radix guest, as none of the operations it does
4516  * for a radix guest can fail or have a way to report failure.
4517  */
4518 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
4519 {
4520         int r = 0;
4521         struct kvm *kvm = vcpu->kvm;
4522
4523         mutex_lock(&kvm->arch.mmu_setup_lock);
4524         if (!kvm->arch.mmu_ready) {
4525                 if (!kvm_is_radix(kvm))
4526                         r = kvmppc_hv_setup_htab_rma(vcpu);
4527                 if (!r) {
4528                         if (cpu_has_feature(CPU_FTR_ARCH_300))
4529                                 kvmppc_setup_partition_table(kvm);
4530                         kvm->arch.mmu_ready = 1;
4531                 }
4532         }
4533         mutex_unlock(&kvm->arch.mmu_setup_lock);
4534         return r;
4535 }
4536
4537 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
4538 {
4539         struct kvm_run *run = vcpu->run;
4540         int n_ceded, i, r;
4541         struct kvmppc_vcore *vc;
4542         struct kvm_vcpu *v;
4543
4544         trace_kvmppc_run_vcpu_enter(vcpu);
4545
4546         run->exit_reason = 0;
4547         vcpu->arch.ret = RESUME_GUEST;
4548         vcpu->arch.trap = 0;
4549         kvmppc_update_vpas(vcpu);
4550
4551         /*
4552          * Synchronize with other threads in this virtual core
4553          */
4554         vc = vcpu->arch.vcore;
4555         spin_lock(&vc->lock);
4556         vcpu->arch.ceded = 0;
4557         vcpu->arch.run_task = current;
4558         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4559         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4560         vcpu->arch.busy_preempt = TB_NIL;
4561         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4562         ++vc->n_runnable;
4563
4564         /*
4565          * This happens the first time this is called for a vcpu.
4566          * If the vcore is already running, we may be able to start
4567          * this thread straight away and have it join in.
4568          */
4569         if (!signal_pending(current)) {
4570                 if ((vc->vcore_state == VCORE_PIGGYBACK ||
4571                      vc->vcore_state == VCORE_RUNNING) &&
4572                            !VCORE_IS_EXITING(vc)) {
4573                         kvmppc_update_vpa_dispatch(vcpu, vc);
4574                         kvmppc_start_thread(vcpu, vc);
4575                         trace_kvm_guest_enter(vcpu);
4576                 } else if (vc->vcore_state == VCORE_SLEEPING) {
4577                         rcuwait_wake_up(&vc->wait);
4578                 }
4579
4580         }
4581
4582         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4583                !signal_pending(current)) {
4584                 /* See if the MMU is ready to go */
4585                 if (!vcpu->kvm->arch.mmu_ready) {
4586                         spin_unlock(&vc->lock);
4587                         r = kvmhv_setup_mmu(vcpu);
4588                         spin_lock(&vc->lock);
4589                         if (r) {
4590                                 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4591                                 run->fail_entry.
4592                                         hardware_entry_failure_reason = 0;
4593                                 vcpu->arch.ret = r;
4594                                 break;
4595                         }
4596                 }
4597
4598                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4599                         kvmppc_vcore_end_preempt(vc);
4600
4601                 if (vc->vcore_state != VCORE_INACTIVE) {
4602                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4603                         continue;
4604                 }
4605                 for_each_runnable_thread(i, v, vc) {
4606                         kvmppc_core_prepare_to_enter(v);
4607                         if (signal_pending(v->arch.run_task)) {
4608                                 kvmppc_remove_runnable(vc, v, mftb());
4609                                 v->stat.signal_exits++;
4610                                 v->run->exit_reason = KVM_EXIT_INTR;
4611                                 v->arch.ret = -EINTR;
4612                                 wake_up(&v->arch.cpu_run);
4613                         }
4614                 }
4615                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4616                         break;
4617                 n_ceded = 0;
4618                 for_each_runnable_thread(i, v, vc) {
4619                         if (!kvmppc_vcpu_woken(v))
4620                                 n_ceded += v->arch.ceded;
4621                         else
4622                                 v->arch.ceded = 0;
4623                 }
4624                 vc->runner = vcpu;
4625                 if (n_ceded == vc->n_runnable) {
4626                         kvmppc_vcore_blocked(vc);
4627                 } else if (need_resched()) {
4628                         kvmppc_vcore_preempt(vc);
4629                         /* Let something else run */
4630                         cond_resched_lock(&vc->lock);
4631                         if (vc->vcore_state == VCORE_PREEMPT)
4632                                 kvmppc_vcore_end_preempt(vc);
4633                 } else {
4634                         kvmppc_run_core(vc);
4635                 }
4636                 vc->runner = NULL;
4637         }
4638
4639         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4640                (vc->vcore_state == VCORE_RUNNING ||
4641                 vc->vcore_state == VCORE_EXITING ||
4642                 vc->vcore_state == VCORE_PIGGYBACK))
4643                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4644
4645         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4646                 kvmppc_vcore_end_preempt(vc);
4647
4648         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4649                 kvmppc_remove_runnable(vc, vcpu, mftb());
4650                 vcpu->stat.signal_exits++;
4651                 run->exit_reason = KVM_EXIT_INTR;
4652                 vcpu->arch.ret = -EINTR;
4653         }
4654
4655         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4656                 /* Wake up some vcpu to run the core */
4657                 i = -1;
4658                 v = next_runnable_thread(vc, &i);
4659                 wake_up(&v->arch.cpu_run);
4660         }
4661
4662         trace_kvmppc_run_vcpu_exit(vcpu);
4663         spin_unlock(&vc->lock);
4664         return vcpu->arch.ret;
4665 }
4666
4667 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4668                           unsigned long lpcr)
4669 {
4670         struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
4671         struct kvm_run *run = vcpu->run;
4672         int trap, r, pcpu;
4673         int srcu_idx;
4674         struct kvmppc_vcore *vc;
4675         struct kvm *kvm = vcpu->kvm;
4676         struct kvm_nested_guest *nested = vcpu->arch.nested;
4677         unsigned long flags;
4678         u64 tb;
4679
4680         trace_kvmppc_run_vcpu_enter(vcpu);
4681
4682         run->exit_reason = 0;
4683         vcpu->arch.ret = RESUME_GUEST;
4684         vcpu->arch.trap = 0;
4685
4686         vc = vcpu->arch.vcore;
4687         vcpu->arch.ceded = 0;
4688         vcpu->arch.run_task = current;
4689         vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4690
4691         /* See if the MMU is ready to go */
4692         if (unlikely(!kvm->arch.mmu_ready)) {
4693                 r = kvmhv_setup_mmu(vcpu);
4694                 if (r) {
4695                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4696                         run->fail_entry.hardware_entry_failure_reason = 0;
4697                         vcpu->arch.ret = r;
4698                         return r;
4699                 }
4700         }
4701
4702         if (need_resched())
4703                 cond_resched();
4704
4705         kvmppc_update_vpas(vcpu);
4706
4707         preempt_disable();
4708         pcpu = smp_processor_id();
4709         if (kvm_is_radix(kvm))
4710                 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4711
4712         /* flags save not required, but irq_pmu has no disable/enable API */
4713         powerpc_local_irq_pmu_save(flags);
4714
4715         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4716
4717         if (signal_pending(current))
4718                 goto sigpend;
4719         if (need_resched() || !kvm->arch.mmu_ready)
4720                 goto out;
4721
4722         vcpu->cpu = pcpu;
4723         vcpu->arch.thread_cpu = pcpu;
4724         vc->pcpu = pcpu;
4725         local_paca->kvm_hstate.kvm_vcpu = vcpu;
4726         local_paca->kvm_hstate.ptid = 0;
4727         local_paca->kvm_hstate.fake_suspend = 0;
4728
4729         /*
4730          * Orders set cpu/thread_cpu vs testing for pending interrupts and
4731          * doorbells below. The other side is when these fields are set vs
4732          * kvmppc_fast_vcpu_kick_hv reading the cpu/thread_cpu fields to
4733          * kick a vCPU to notice the pending interrupt.
4734          */
4735         smp_mb();
4736
4737         if (!nested) {
4738                 kvmppc_core_prepare_to_enter(vcpu);
4739                 if (__kvmppc_get_msr_hv(vcpu) & MSR_EE) {
4740                         if (xive_interrupt_pending(vcpu))
4741                                 kvmppc_inject_interrupt_hv(vcpu,
4742                                                 BOOK3S_INTERRUPT_EXTERNAL, 0);
4743                 } else if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4744                              &vcpu->arch.pending_exceptions)) {
4745                         lpcr |= LPCR_MER;
4746                 }
4747         } else if (vcpu->arch.pending_exceptions ||
4748                    vcpu->arch.doorbell_request ||
4749                    xive_interrupt_pending(vcpu)) {
4750                 vcpu->arch.ret = RESUME_HOST;
4751                 goto out;
4752         }
4753
4754         if (vcpu->arch.timer_running) {
4755                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
4756                 vcpu->arch.timer_running = 0;
4757         }
4758
4759         tb = mftb();
4760
4761         kvmppc_update_vpa_dispatch_p9(vcpu, vc, tb + kvmppc_get_tb_offset(vcpu));
4762
4763         trace_kvm_guest_enter(vcpu);
4764
4765         guest_timing_enter_irqoff();
4766
4767         srcu_idx = srcu_read_lock(&kvm->srcu);
4768
4769         guest_state_enter_irqoff();
4770         this_cpu_disable_ftrace();
4771
4772         trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr, &tb);
4773         vcpu->arch.trap = trap;
4774
4775         this_cpu_enable_ftrace();
4776         guest_state_exit_irqoff();
4777
4778         srcu_read_unlock(&kvm->srcu, srcu_idx);
4779
4780         set_irq_happened(trap);
4781
4782         vcpu->cpu = -1;
4783         vcpu->arch.thread_cpu = -1;
4784         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4785
4786         if (!vtime_accounting_enabled_this_cpu()) {
4787                 powerpc_local_irq_pmu_restore(flags);
4788                 /*
4789                  * Service IRQs here before guest_timing_exit_irqoff() so any
4790                  * ticks that occurred while running the guest are accounted to
4791                  * the guest. If vtime accounting is enabled, accounting uses
4792                  * TB rather than ticks, so it can be done without enabling
4793                  * interrupts here, which has the problem that it accounts
4794                  * interrupt processing overhead to the host.
4795                  */
4796                 powerpc_local_irq_pmu_save(flags);
4797         }
4798         guest_timing_exit_irqoff();
4799
4800         powerpc_local_irq_pmu_restore(flags);
4801
4802         preempt_enable();
4803
4804         /*
4805          * cancel pending decrementer exception if DEC is now positive, or if
4806          * entering a nested guest in which case the decrementer is now owned
4807          * by L2 and the L1 decrementer is provided in hdec_expires
4808          */
4809         if (kvmppc_core_pending_dec(vcpu) &&
4810                         ((tb < kvmppc_dec_expires_host_tb(vcpu)) ||
4811                          (trap == BOOK3S_INTERRUPT_SYSCALL &&
4812                           kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4813                 kvmppc_core_dequeue_dec(vcpu);
4814
4815         trace_kvm_guest_exit(vcpu);
4816         r = RESUME_GUEST;
4817         if (trap) {
4818                 if (!nested)
4819                         r = kvmppc_handle_exit_hv(vcpu, current);
4820                 else
4821                         r = kvmppc_handle_nested_exit(vcpu);
4822         }
4823         vcpu->arch.ret = r;
4824
4825         if (is_kvmppc_resume_guest(r) && !kvmppc_vcpu_check_block(vcpu)) {
4826                 kvmppc_set_timer(vcpu);
4827
4828                 prepare_to_rcuwait(wait);
4829                 for (;;) {
4830                         set_current_state(TASK_INTERRUPTIBLE);
4831                         if (signal_pending(current)) {
4832                                 vcpu->stat.signal_exits++;
4833                                 run->exit_reason = KVM_EXIT_INTR;
4834                                 vcpu->arch.ret = -EINTR;
4835                                 break;
4836                         }
4837
4838                         if (kvmppc_vcpu_check_block(vcpu))
4839                                 break;
4840
4841                         trace_kvmppc_vcore_blocked(vcpu, 0);
4842                         schedule();
4843                         trace_kvmppc_vcore_blocked(vcpu, 1);
4844                 }
4845                 finish_rcuwait(wait);
4846         }
4847         vcpu->arch.ceded = 0;
4848
4849  done:
4850         trace_kvmppc_run_vcpu_exit(vcpu);
4851
4852         return vcpu->arch.ret;
4853
4854  sigpend:
4855         vcpu->stat.signal_exits++;
4856         run->exit_reason = KVM_EXIT_INTR;
4857         vcpu->arch.ret = -EINTR;
4858  out:
4859         vcpu->cpu = -1;
4860         vcpu->arch.thread_cpu = -1;
4861         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4862         powerpc_local_irq_pmu_restore(flags);
4863         preempt_enable();
4864         goto done;
4865 }
4866
4867 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
4868 {
4869         struct kvm_run *run = vcpu->run;
4870         int r;
4871         int srcu_idx;
4872         struct kvm *kvm;
4873         unsigned long msr;
4874
4875         start_timing(vcpu, &vcpu->arch.vcpu_entry);
4876
4877         if (!vcpu->arch.sane) {
4878                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4879                 return -EINVAL;
4880         }
4881
4882         /* No need to go into the guest when all we'll do is come back out */
4883         if (signal_pending(current)) {
4884                 run->exit_reason = KVM_EXIT_INTR;
4885                 return -EINTR;
4886         }
4887
4888 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4889         /*
4890          * Don't allow entry with a suspended transaction, because
4891          * the guest entry/exit code will lose it.
4892          */
4893         if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4894             (current->thread.regs->msr & MSR_TM)) {
4895                 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4896                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4897                         run->fail_entry.hardware_entry_failure_reason = 0;
4898                         return -EINVAL;
4899                 }
4900         }
4901 #endif
4902
4903         /*
4904          * Force online to 1 for the sake of old userspace which doesn't
4905          * set it.
4906          */
4907         if (!vcpu->arch.online) {
4908                 atomic_inc(&vcpu->arch.vcore->online_count);
4909                 vcpu->arch.online = 1;
4910         }
4911
4912         kvmppc_core_prepare_to_enter(vcpu);
4913
4914         kvm = vcpu->kvm;
4915         atomic_inc(&kvm->arch.vcpus_running);
4916         /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4917         smp_mb();
4918
4919         msr = 0;
4920         if (IS_ENABLED(CONFIG_PPC_FPU))
4921                 msr |= MSR_FP;
4922         if (cpu_has_feature(CPU_FTR_ALTIVEC))
4923                 msr |= MSR_VEC;
4924         if (cpu_has_feature(CPU_FTR_VSX))
4925                 msr |= MSR_VSX;
4926         if ((cpu_has_feature(CPU_FTR_TM) ||
4927             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) &&
4928                         (kvmppc_get_hfscr_hv(vcpu) & HFSCR_TM))
4929                 msr |= MSR_TM;
4930         msr = msr_check_and_set(msr);
4931
4932         kvmppc_save_user_regs();
4933
4934         kvmppc_save_current_sprs();
4935
4936         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4937                 vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4938         vcpu->arch.pgdir = kvm->mm->pgd;
4939         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4940
4941         do {
4942                 accumulate_time(vcpu, &vcpu->arch.guest_entry);
4943                 if (cpu_has_feature(CPU_FTR_ARCH_300))
4944                         r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
4945                                                   vcpu->arch.vcore->lpcr);
4946                 else
4947                         r = kvmppc_run_vcpu(vcpu);
4948
4949                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL) {
4950                         accumulate_time(vcpu, &vcpu->arch.hcall);
4951
4952                         if (WARN_ON_ONCE(__kvmppc_get_msr_hv(vcpu) & MSR_PR)) {
4953                                 /*
4954                                  * These should have been caught reflected
4955                                  * into the guest by now. Final sanity check:
4956                                  * don't allow userspace to execute hcalls in
4957                                  * the hypervisor.
4958                                  */
4959                                 r = RESUME_GUEST;
4960                                 continue;
4961                         }
4962                         trace_kvm_hcall_enter(vcpu);
4963                         r = kvmppc_pseries_do_hcall(vcpu);
4964                         trace_kvm_hcall_exit(vcpu, r);
4965                         kvmppc_core_prepare_to_enter(vcpu);
4966                 } else if (r == RESUME_PAGE_FAULT) {
4967                         accumulate_time(vcpu, &vcpu->arch.pg_fault);
4968                         srcu_idx = srcu_read_lock(&kvm->srcu);
4969                         r = kvmppc_book3s_hv_page_fault(vcpu,
4970                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4971                         srcu_read_unlock(&kvm->srcu, srcu_idx);
4972                 } else if (r == RESUME_PASSTHROUGH) {
4973                         if (WARN_ON(xics_on_xive()))
4974                                 r = H_SUCCESS;
4975                         else
4976                                 r = kvmppc_xics_rm_complete(vcpu, 0);
4977                 }
4978         } while (is_kvmppc_resume_guest(r));
4979         accumulate_time(vcpu, &vcpu->arch.vcpu_exit);
4980
4981         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4982         atomic_dec(&kvm->arch.vcpus_running);
4983
4984         srr_regs_clobbered();
4985
4986         end_timing(vcpu);
4987
4988         return r;
4989 }
4990
4991 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4992                                      int shift, int sllp)
4993 {
4994         (*sps)->page_shift = shift;
4995         (*sps)->slb_enc = sllp;
4996         (*sps)->enc[0].page_shift = shift;
4997         (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4998         /*
4999          * Add 16MB MPSS support (may get filtered out by userspace)
5000          */
5001         if (shift != 24) {
5002                 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
5003                 if (penc != -1) {
5004                         (*sps)->enc[1].page_shift = 24;
5005                         (*sps)->enc[1].pte_enc = penc;
5006                 }
5007         }
5008         (*sps)++;
5009 }
5010
5011 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
5012                                          struct kvm_ppc_smmu_info *info)
5013 {
5014         struct kvm_ppc_one_seg_page_size *sps;
5015
5016         /*
5017          * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
5018          * POWER7 doesn't support keys for instruction accesses,
5019          * POWER8 and POWER9 do.
5020          */
5021         info->data_keys = 32;
5022         info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
5023
5024         /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
5025         info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
5026         info->slb_size = 32;
5027
5028         /* We only support these sizes for now, and no muti-size segments */
5029         sps = &info->sps[0];
5030         kvmppc_add_seg_page_size(&sps, 12, 0);
5031         kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
5032         kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
5033
5034         /* If running as a nested hypervisor, we don't support HPT guests */
5035         if (kvmhv_on_pseries())
5036                 info->flags |= KVM_PPC_NO_HASH;
5037
5038         return 0;
5039 }
5040
5041 /*
5042  * Get (and clear) the dirty memory log for a memory slot.
5043  */
5044 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
5045                                          struct kvm_dirty_log *log)
5046 {
5047         struct kvm_memslots *slots;
5048         struct kvm_memory_slot *memslot;
5049         int r;
5050         unsigned long n, i;
5051         unsigned long *buf, *p;
5052         struct kvm_vcpu *vcpu;
5053
5054         mutex_lock(&kvm->slots_lock);
5055
5056         r = -EINVAL;
5057         if (log->slot >= KVM_USER_MEM_SLOTS)
5058                 goto out;
5059
5060         slots = kvm_memslots(kvm);
5061         memslot = id_to_memslot(slots, log->slot);
5062         r = -ENOENT;
5063         if (!memslot || !memslot->dirty_bitmap)
5064                 goto out;
5065
5066         /*
5067          * Use second half of bitmap area because both HPT and radix
5068          * accumulate bits in the first half.
5069          */
5070         n = kvm_dirty_bitmap_bytes(memslot);
5071         buf = memslot->dirty_bitmap + n / sizeof(long);
5072         memset(buf, 0, n);
5073
5074         if (kvm_is_radix(kvm))
5075                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
5076         else
5077                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
5078         if (r)
5079                 goto out;
5080
5081         /*
5082          * We accumulate dirty bits in the first half of the
5083          * memslot's dirty_bitmap area, for when pages are paged
5084          * out or modified by the host directly.  Pick up these
5085          * bits and add them to the map.
5086          */
5087         p = memslot->dirty_bitmap;
5088         for (i = 0; i < n / sizeof(long); ++i)
5089                 buf[i] |= xchg(&p[i], 0);
5090
5091         /* Harvest dirty bits from VPA and DTL updates */
5092         /* Note: we never modify the SLB shadow buffer areas */
5093         kvm_for_each_vcpu(i, vcpu, kvm) {
5094                 spin_lock(&vcpu->arch.vpa_update_lock);
5095                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
5096                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
5097                 spin_unlock(&vcpu->arch.vpa_update_lock);
5098         }
5099
5100         r = -EFAULT;
5101         if (copy_to_user(log->dirty_bitmap, buf, n))
5102                 goto out;
5103
5104         r = 0;
5105 out:
5106         mutex_unlock(&kvm->slots_lock);
5107         return r;
5108 }
5109
5110 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
5111 {
5112         vfree(slot->arch.rmap);
5113         slot->arch.rmap = NULL;
5114 }
5115
5116 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
5117                                 const struct kvm_memory_slot *old,
5118                                 struct kvm_memory_slot *new,
5119                                 enum kvm_mr_change change)
5120 {
5121         if (change == KVM_MR_CREATE) {
5122                 unsigned long size = array_size(new->npages, sizeof(*new->arch.rmap));
5123
5124                 if ((size >> PAGE_SHIFT) > totalram_pages())
5125                         return -ENOMEM;
5126
5127                 new->arch.rmap = vzalloc(size);
5128                 if (!new->arch.rmap)
5129                         return -ENOMEM;
5130         } else if (change != KVM_MR_DELETE) {
5131                 new->arch.rmap = old->arch.rmap;
5132         }
5133
5134         return 0;
5135 }
5136
5137 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
5138                                 struct kvm_memory_slot *old,
5139                                 const struct kvm_memory_slot *new,
5140                                 enum kvm_mr_change change)
5141 {
5142         /*
5143          * If we are creating or modifying a memslot, it might make
5144          * some address that was previously cached as emulated
5145          * MMIO be no longer emulated MMIO, so invalidate
5146          * all the caches of emulated MMIO translations.
5147          */
5148         if (change != KVM_MR_DELETE)
5149                 atomic64_inc(&kvm->arch.mmio_update);
5150
5151         /*
5152          * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
5153          * have already called kvm_arch_flush_shadow_memslot() to
5154          * flush shadow mappings.  For KVM_MR_CREATE we have no
5155          * previous mappings.  So the only case to handle is
5156          * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
5157          * has been changed.
5158          * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
5159          * to get rid of any THP PTEs in the partition-scoped page tables
5160          * so we can track dirtiness at the page level; we flush when
5161          * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
5162          * using THP PTEs.
5163          */
5164         if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
5165             ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
5166                 kvmppc_radix_flush_memslot(kvm, old);
5167         /*
5168          * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
5169          */
5170         if (!kvm->arch.secure_guest)
5171                 return;
5172
5173         switch (change) {
5174         case KVM_MR_CREATE:
5175                 /*
5176                  * @TODO kvmppc_uvmem_memslot_create() can fail and
5177                  * return error. Fix this.
5178                  */
5179                 kvmppc_uvmem_memslot_create(kvm, new);
5180                 break;
5181         case KVM_MR_DELETE:
5182                 kvmppc_uvmem_memslot_delete(kvm, old);
5183                 break;
5184         default:
5185                 /* TODO: Handle KVM_MR_MOVE */
5186                 break;
5187         }
5188 }
5189
5190 /*
5191  * Update LPCR values in kvm->arch and in vcores.
5192  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
5193  * of kvm->arch.lpcr update).
5194  */
5195 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
5196 {
5197         long int i;
5198         u32 cores_done = 0;
5199
5200         if ((kvm->arch.lpcr & mask) == lpcr)
5201                 return;
5202
5203         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
5204
5205         for (i = 0; i < KVM_MAX_VCORES; ++i) {
5206                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
5207                 if (!vc)
5208                         continue;
5209
5210                 spin_lock(&vc->lock);
5211                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
5212                 verify_lpcr(kvm, vc->lpcr);
5213                 spin_unlock(&vc->lock);
5214                 if (++cores_done >= kvm->arch.online_vcores)
5215                         break;
5216         }
5217
5218         if (kvmhv_is_nestedv2()) {
5219                 struct kvm_vcpu *vcpu;
5220
5221                 kvm_for_each_vcpu(i, vcpu, kvm) {
5222                         kvmhv_nestedv2_mark_dirty(vcpu, KVMPPC_GSID_LPCR);
5223                 }
5224         }
5225 }
5226
5227 void kvmppc_setup_partition_table(struct kvm *kvm)
5228 {
5229         unsigned long dw0, dw1;
5230
5231         if (!kvm_is_radix(kvm)) {
5232                 /* PS field - page size for VRMA */
5233                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
5234                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
5235                 /* HTABSIZE and HTABORG fields */
5236                 dw0 |= kvm->arch.sdr1;
5237
5238                 /* Second dword as set by userspace */
5239                 dw1 = kvm->arch.process_table;
5240         } else {
5241                 dw0 = PATB_HR | radix__get_tree_size() |
5242                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
5243                 dw1 = PATB_GR | kvm->arch.process_table;
5244         }
5245         kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
5246 }
5247
5248 /*
5249  * Set up HPT (hashed page table) and RMA (real-mode area).
5250  * Must be called with kvm->arch.mmu_setup_lock held.
5251  */
5252 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
5253 {
5254         int err = 0;
5255         struct kvm *kvm = vcpu->kvm;
5256         unsigned long hva;
5257         struct kvm_memory_slot *memslot;
5258         struct vm_area_struct *vma;
5259         unsigned long lpcr = 0, senc;
5260         unsigned long psize, porder;
5261         int srcu_idx;
5262
5263         /* Allocate hashed page table (if not done already) and reset it */
5264         if (!kvm->arch.hpt.virt) {
5265                 int order = KVM_DEFAULT_HPT_ORDER;
5266                 struct kvm_hpt_info info;
5267
5268                 err = kvmppc_allocate_hpt(&info, order);
5269                 /* If we get here, it means userspace didn't specify a
5270                  * size explicitly.  So, try successively smaller
5271                  * sizes if the default failed. */
5272                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
5273                         err  = kvmppc_allocate_hpt(&info, order);
5274
5275                 if (err < 0) {
5276                         pr_err("KVM: Couldn't alloc HPT\n");
5277                         goto out;
5278                 }
5279
5280                 kvmppc_set_hpt(kvm, &info);
5281         }
5282
5283         /* Look up the memslot for guest physical address 0 */
5284         srcu_idx = srcu_read_lock(&kvm->srcu);
5285         memslot = gfn_to_memslot(kvm, 0);
5286
5287         /* We must have some memory at 0 by now */
5288         err = -EINVAL;
5289         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
5290                 goto out_srcu;
5291
5292         /* Look up the VMA for the start of this memory slot */
5293         hva = memslot->userspace_addr;
5294         mmap_read_lock(kvm->mm);
5295         vma = vma_lookup(kvm->mm, hva);
5296         if (!vma || (vma->vm_flags & VM_IO))
5297                 goto up_out;
5298
5299         psize = vma_kernel_pagesize(vma);
5300
5301         mmap_read_unlock(kvm->mm);
5302
5303         /* We can handle 4k, 64k or 16M pages in the VRMA */
5304         if (psize >= 0x1000000)
5305                 psize = 0x1000000;
5306         else if (psize >= 0x10000)
5307                 psize = 0x10000;
5308         else
5309                 psize = 0x1000;
5310         porder = __ilog2(psize);
5311
5312         senc = slb_pgsize_encoding(psize);
5313         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
5314                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
5315         /* Create HPTEs in the hash page table for the VRMA */
5316         kvmppc_map_vrma(vcpu, memslot, porder);
5317
5318         /* Update VRMASD field in the LPCR */
5319         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
5320                 /* the -4 is to account for senc values starting at 0x10 */
5321                 lpcr = senc << (LPCR_VRMASD_SH - 4);
5322                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
5323         }
5324
5325         /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
5326         smp_wmb();
5327         err = 0;
5328  out_srcu:
5329         srcu_read_unlock(&kvm->srcu, srcu_idx);
5330  out:
5331         return err;
5332
5333  up_out:
5334         mmap_read_unlock(kvm->mm);
5335         goto out_srcu;
5336 }
5337
5338 /*
5339  * Must be called with kvm->arch.mmu_setup_lock held and
5340  * mmu_ready = 0 and no vcpus running.
5341  */
5342 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
5343 {
5344         unsigned long lpcr, lpcr_mask;
5345
5346         if (nesting_enabled(kvm))
5347                 kvmhv_release_all_nested(kvm);
5348         kvmppc_rmap_reset(kvm);
5349         kvm->arch.process_table = 0;
5350         /* Mutual exclusion with kvm_unmap_gfn_range etc. */
5351         spin_lock(&kvm->mmu_lock);
5352         kvm->arch.radix = 0;
5353         spin_unlock(&kvm->mmu_lock);
5354         kvmppc_free_radix(kvm);
5355
5356         lpcr = LPCR_VPM1;
5357         lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5358         if (cpu_has_feature(CPU_FTR_ARCH_31))
5359                 lpcr_mask |= LPCR_HAIL;
5360         kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
5361
5362         return 0;
5363 }
5364
5365 /*
5366  * Must be called with kvm->arch.mmu_setup_lock held and
5367  * mmu_ready = 0 and no vcpus running.
5368  */
5369 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
5370 {
5371         unsigned long lpcr, lpcr_mask;
5372         int err;
5373
5374         err = kvmppc_init_vm_radix(kvm);
5375         if (err)
5376                 return err;
5377         kvmppc_rmap_reset(kvm);
5378         /* Mutual exclusion with kvm_unmap_gfn_range etc. */
5379         spin_lock(&kvm->mmu_lock);
5380         kvm->arch.radix = 1;
5381         spin_unlock(&kvm->mmu_lock);
5382         kvmppc_free_hpt(&kvm->arch.hpt);
5383
5384         lpcr = LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5385         lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5386         if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5387                 lpcr_mask |= LPCR_HAIL;
5388                 if (cpu_has_feature(CPU_FTR_HVMODE) &&
5389                                 (kvm->arch.host_lpcr & LPCR_HAIL))
5390                         lpcr |= LPCR_HAIL;
5391         }
5392         kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
5393
5394         return 0;
5395 }
5396
5397 #ifdef CONFIG_KVM_XICS
5398 /*
5399  * Allocate a per-core structure for managing state about which cores are
5400  * running in the host versus the guest and for exchanging data between
5401  * real mode KVM and CPU running in the host.
5402  * This is only done for the first VM.
5403  * The allocated structure stays even if all VMs have stopped.
5404  * It is only freed when the kvm-hv module is unloaded.
5405  * It's OK for this routine to fail, we just don't support host
5406  * core operations like redirecting H_IPI wakeups.
5407  */
5408 void kvmppc_alloc_host_rm_ops(void)
5409 {
5410         struct kvmppc_host_rm_ops *ops;
5411         unsigned long l_ops;
5412         int cpu, core;
5413         int size;
5414
5415         if (cpu_has_feature(CPU_FTR_ARCH_300))
5416                 return;
5417
5418         /* Not the first time here ? */
5419         if (kvmppc_host_rm_ops_hv != NULL)
5420                 return;
5421
5422         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
5423         if (!ops)
5424                 return;
5425
5426         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
5427         ops->rm_core = kzalloc(size, GFP_KERNEL);
5428
5429         if (!ops->rm_core) {
5430                 kfree(ops);
5431                 return;
5432         }
5433
5434         cpus_read_lock();
5435
5436         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
5437                 if (!cpu_online(cpu))
5438                         continue;
5439
5440                 core = cpu >> threads_shift;
5441                 ops->rm_core[core].rm_state.in_host = 1;
5442         }
5443
5444         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
5445
5446         /*
5447          * Make the contents of the kvmppc_host_rm_ops structure visible
5448          * to other CPUs before we assign it to the global variable.
5449          * Do an atomic assignment (no locks used here), but if someone
5450          * beats us to it, just free our copy and return.
5451          */
5452         smp_wmb();
5453         l_ops = (unsigned long) ops;
5454
5455         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
5456                 cpus_read_unlock();
5457                 kfree(ops->rm_core);
5458                 kfree(ops);
5459                 return;
5460         }
5461
5462         cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
5463                                              "ppc/kvm_book3s:prepare",
5464                                              kvmppc_set_host_core,
5465                                              kvmppc_clear_host_core);
5466         cpus_read_unlock();
5467 }
5468
5469 void kvmppc_free_host_rm_ops(void)
5470 {
5471         if (kvmppc_host_rm_ops_hv) {
5472                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
5473                 kfree(kvmppc_host_rm_ops_hv->rm_core);
5474                 kfree(kvmppc_host_rm_ops_hv);
5475                 kvmppc_host_rm_ops_hv = NULL;
5476         }
5477 }
5478 #endif
5479
5480 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
5481 {
5482         unsigned long lpcr, lpid;
5483         int ret;
5484
5485         mutex_init(&kvm->arch.uvmem_lock);
5486         INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
5487         mutex_init(&kvm->arch.mmu_setup_lock);
5488
5489         /* Allocate the guest's logical partition ID */
5490
5491         if (!kvmhv_is_nestedv2()) {
5492                 lpid = kvmppc_alloc_lpid();
5493                 if ((long)lpid < 0)
5494                         return -ENOMEM;
5495                 kvm->arch.lpid = lpid;
5496         }
5497
5498         kvmppc_alloc_host_rm_ops();
5499
5500         kvmhv_vm_nested_init(kvm);
5501
5502         if (kvmhv_is_nestedv2()) {
5503                 long rc;
5504                 unsigned long guest_id;
5505
5506                 rc = plpar_guest_create(0, &guest_id);
5507
5508                 if (rc != H_SUCCESS)
5509                         pr_err("KVM: Create Guest hcall failed, rc=%ld\n", rc);
5510
5511                 switch (rc) {
5512                 case H_PARAMETER:
5513                 case H_FUNCTION:
5514                 case H_STATE:
5515                         return -EINVAL;
5516                 case H_NOT_ENOUGH_RESOURCES:
5517                 case H_ABORTED:
5518                         return -ENOMEM;
5519                 case H_AUTHORITY:
5520                         return -EPERM;
5521                 case H_NOT_AVAILABLE:
5522                         return -EBUSY;
5523                 }
5524                 kvm->arch.lpid = guest_id;
5525         }
5526
5527
5528         /*
5529          * Since we don't flush the TLB when tearing down a VM,
5530          * and this lpid might have previously been used,
5531          * make sure we flush on each core before running the new VM.
5532          * On POWER9, the tlbie in mmu_partition_table_set_entry()
5533          * does this flush for us.
5534          */
5535         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5536                 cpumask_setall(&kvm->arch.need_tlb_flush);
5537
5538         /* Start out with the default set of hcalls enabled */
5539         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
5540                sizeof(kvm->arch.enabled_hcalls));
5541
5542         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5543                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
5544
5545         /* Init LPCR for virtual RMA mode */
5546         if (cpu_has_feature(CPU_FTR_HVMODE)) {
5547                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
5548                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
5549                 lpcr &= LPCR_PECE | LPCR_LPES;
5550         } else {
5551                 /*
5552                  * The L2 LPES mode will be set by the L0 according to whether
5553                  * or not it needs to take external interrupts in HV mode.
5554                  */
5555                 lpcr = 0;
5556         }
5557         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
5558                 LPCR_VPM0 | LPCR_VPM1;
5559         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
5560                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
5561         /* On POWER8 turn on online bit to enable PURR/SPURR */
5562         if (cpu_has_feature(CPU_FTR_ARCH_207S))
5563                 lpcr |= LPCR_ONL;
5564         /*
5565          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
5566          * Set HVICE bit to enable hypervisor virtualization interrupts.
5567          * Set HEIC to prevent OS interrupts to go to hypervisor (should
5568          * be unnecessary but better safe than sorry in case we re-enable
5569          * EE in HV mode with this LPCR still set)
5570          */
5571         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5572                 lpcr &= ~LPCR_VPM0;
5573                 lpcr |= LPCR_HVICE | LPCR_HEIC;
5574
5575                 /*
5576                  * If xive is enabled, we route 0x500 interrupts directly
5577                  * to the guest.
5578                  */
5579                 if (xics_on_xive())
5580                         lpcr |= LPCR_LPES;
5581         }
5582
5583         /*
5584          * If the host uses radix, the guest starts out as radix.
5585          */
5586         if (radix_enabled()) {
5587                 kvm->arch.radix = 1;
5588                 kvm->arch.mmu_ready = 1;
5589                 lpcr &= ~LPCR_VPM1;
5590                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5591                 if (cpu_has_feature(CPU_FTR_HVMODE) &&
5592                     cpu_has_feature(CPU_FTR_ARCH_31) &&
5593                     (kvm->arch.host_lpcr & LPCR_HAIL))
5594                         lpcr |= LPCR_HAIL;
5595                 ret = kvmppc_init_vm_radix(kvm);
5596                 if (ret) {
5597                         if (kvmhv_is_nestedv2())
5598                                 plpar_guest_delete(0, kvm->arch.lpid);
5599                         else
5600                                 kvmppc_free_lpid(kvm->arch.lpid);
5601                         return ret;
5602                 }
5603                 kvmppc_setup_partition_table(kvm);
5604         }
5605
5606         verify_lpcr(kvm, lpcr);
5607         kvm->arch.lpcr = lpcr;
5608
5609         /* Initialization for future HPT resizes */
5610         kvm->arch.resize_hpt = NULL;
5611
5612         /*
5613          * Work out how many sets the TLB has, for the use of
5614          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
5615          */
5616         if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5617                 /*
5618                  * P10 will flush all the congruence class with a single tlbiel
5619                  */
5620                 kvm->arch.tlb_sets = 1;
5621         } else if (radix_enabled())
5622                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
5623         else if (cpu_has_feature(CPU_FTR_ARCH_300))
5624                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
5625         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
5626                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
5627         else
5628                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
5629
5630         /*
5631          * Track that we now have a HV mode VM active. This blocks secondary
5632          * CPU threads from coming online.
5633          */
5634         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5635                 kvm_hv_vm_activated();
5636
5637         /*
5638          * Initialize smt_mode depending on processor.
5639          * POWER8 and earlier have to use "strict" threading, where
5640          * all vCPUs in a vcore have to run on the same (sub)core,
5641          * whereas on POWER9 the threads can each run a different
5642          * guest.
5643          */
5644         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5645                 kvm->arch.smt_mode = threads_per_subcore;
5646         else
5647                 kvm->arch.smt_mode = 1;
5648         kvm->arch.emul_smt_mode = 1;
5649
5650         return 0;
5651 }
5652
5653 static int kvmppc_arch_create_vm_debugfs_hv(struct kvm *kvm)
5654 {
5655         kvmppc_mmu_debugfs_init(kvm);
5656         if (radix_enabled())
5657                 kvmhv_radix_debugfs_init(kvm);
5658         return 0;
5659 }
5660
5661 static void kvmppc_free_vcores(struct kvm *kvm)
5662 {
5663         long int i;
5664
5665         for (i = 0; i < KVM_MAX_VCORES; ++i)
5666                 kfree(kvm->arch.vcores[i]);
5667         kvm->arch.online_vcores = 0;
5668 }
5669
5670 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5671 {
5672         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5673                 kvm_hv_vm_deactivated();
5674
5675         kvmppc_free_vcores(kvm);
5676
5677
5678         if (kvm_is_radix(kvm))
5679                 kvmppc_free_radix(kvm);
5680         else
5681                 kvmppc_free_hpt(&kvm->arch.hpt);
5682
5683         /* Perform global invalidation and return lpid to the pool */
5684         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5685                 if (nesting_enabled(kvm))
5686                         kvmhv_release_all_nested(kvm);
5687                 kvm->arch.process_table = 0;
5688                 if (kvm->arch.secure_guest)
5689                         uv_svm_terminate(kvm->arch.lpid);
5690                 if (!kvmhv_is_nestedv2())
5691                         kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5692         }
5693
5694         if (kvmhv_is_nestedv2())
5695                 plpar_guest_delete(0, kvm->arch.lpid);
5696         else
5697                 kvmppc_free_lpid(kvm->arch.lpid);
5698
5699         kvmppc_free_pimap(kvm);
5700 }
5701
5702 /* We don't need to emulate any privileged instructions or dcbz */
5703 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5704                                      unsigned int inst, int *advance)
5705 {
5706         return EMULATE_FAIL;
5707 }
5708
5709 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5710                                         ulong spr_val)
5711 {
5712         return EMULATE_FAIL;
5713 }
5714
5715 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5716                                         ulong *spr_val)
5717 {
5718         return EMULATE_FAIL;
5719 }
5720
5721 static int kvmppc_core_check_processor_compat_hv(void)
5722 {
5723         if (cpu_has_feature(CPU_FTR_HVMODE) &&
5724             cpu_has_feature(CPU_FTR_ARCH_206))
5725                 return 0;
5726
5727         /* POWER9 in radix mode is capable of being a nested hypervisor. */
5728         if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5729                 return 0;
5730
5731         return -EIO;
5732 }
5733
5734 #ifdef CONFIG_KVM_XICS
5735
5736 void kvmppc_free_pimap(struct kvm *kvm)
5737 {
5738         kfree(kvm->arch.pimap);
5739 }
5740
5741 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5742 {
5743         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5744 }
5745
5746 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5747 {
5748         struct irq_desc *desc;
5749         struct kvmppc_irq_map *irq_map;
5750         struct kvmppc_passthru_irqmap *pimap;
5751         struct irq_chip *chip;
5752         int i, rc = 0;
5753         struct irq_data *host_data;
5754
5755         if (!kvm_irq_bypass)
5756                 return 1;
5757
5758         desc = irq_to_desc(host_irq);
5759         if (!desc)
5760                 return -EIO;
5761
5762         mutex_lock(&kvm->lock);
5763
5764         pimap = kvm->arch.pimap;
5765         if (pimap == NULL) {
5766                 /* First call, allocate structure to hold IRQ map */
5767                 pimap = kvmppc_alloc_pimap();
5768                 if (pimap == NULL) {
5769                         mutex_unlock(&kvm->lock);
5770                         return -ENOMEM;
5771                 }
5772                 kvm->arch.pimap = pimap;
5773         }
5774
5775         /*
5776          * For now, we only support interrupts for which the EOI operation
5777          * is an OPAL call followed by a write to XIRR, since that's
5778          * what our real-mode EOI code does, or a XIVE interrupt
5779          */
5780         chip = irq_data_get_irq_chip(&desc->irq_data);
5781         if (!chip || !is_pnv_opal_msi(chip)) {
5782                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5783                         host_irq, guest_gsi);
5784                 mutex_unlock(&kvm->lock);
5785                 return -ENOENT;
5786         }
5787
5788         /*
5789          * See if we already have an entry for this guest IRQ number.
5790          * If it's mapped to a hardware IRQ number, that's an error,
5791          * otherwise re-use this entry.
5792          */
5793         for (i = 0; i < pimap->n_mapped; i++) {
5794                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5795                         if (pimap->mapped[i].r_hwirq) {
5796                                 mutex_unlock(&kvm->lock);
5797                                 return -EINVAL;
5798                         }
5799                         break;
5800                 }
5801         }
5802
5803         if (i == KVMPPC_PIRQ_MAPPED) {
5804                 mutex_unlock(&kvm->lock);
5805                 return -EAGAIN;         /* table is full */
5806         }
5807
5808         irq_map = &pimap->mapped[i];
5809
5810         irq_map->v_hwirq = guest_gsi;
5811         irq_map->desc = desc;
5812
5813         /*
5814          * Order the above two stores before the next to serialize with
5815          * the KVM real mode handler.
5816          */
5817         smp_wmb();
5818
5819         /*
5820          * The 'host_irq' number is mapped in the PCI-MSI domain but
5821          * the underlying calls, which will EOI the interrupt in real
5822          * mode, need an HW IRQ number mapped in the XICS IRQ domain.
5823          */
5824         host_data = irq_domain_get_irq_data(irq_get_default_host(), host_irq);
5825         irq_map->r_hwirq = (unsigned int)irqd_to_hwirq(host_data);
5826
5827         if (i == pimap->n_mapped)
5828                 pimap->n_mapped++;
5829
5830         if (xics_on_xive())
5831                 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, host_irq);
5832         else
5833                 kvmppc_xics_set_mapped(kvm, guest_gsi, irq_map->r_hwirq);
5834         if (rc)
5835                 irq_map->r_hwirq = 0;
5836
5837         mutex_unlock(&kvm->lock);
5838
5839         return 0;
5840 }
5841
5842 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5843 {
5844         struct irq_desc *desc;
5845         struct kvmppc_passthru_irqmap *pimap;
5846         int i, rc = 0;
5847
5848         if (!kvm_irq_bypass)
5849                 return 0;
5850
5851         desc = irq_to_desc(host_irq);
5852         if (!desc)
5853                 return -EIO;
5854
5855         mutex_lock(&kvm->lock);
5856         if (!kvm->arch.pimap)
5857                 goto unlock;
5858
5859         pimap = kvm->arch.pimap;
5860
5861         for (i = 0; i < pimap->n_mapped; i++) {
5862                 if (guest_gsi == pimap->mapped[i].v_hwirq)
5863                         break;
5864         }
5865
5866         if (i == pimap->n_mapped) {
5867                 mutex_unlock(&kvm->lock);
5868                 return -ENODEV;
5869         }
5870
5871         if (xics_on_xive())
5872                 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, host_irq);
5873         else
5874                 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5875
5876         /* invalidate the entry (what to do on error from the above ?) */
5877         pimap->mapped[i].r_hwirq = 0;
5878
5879         /*
5880          * We don't free this structure even when the count goes to
5881          * zero. The structure is freed when we destroy the VM.
5882          */
5883  unlock:
5884         mutex_unlock(&kvm->lock);
5885         return rc;
5886 }
5887
5888 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5889                                              struct irq_bypass_producer *prod)
5890 {
5891         int ret = 0;
5892         struct kvm_kernel_irqfd *irqfd =
5893                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5894
5895         irqfd->producer = prod;
5896
5897         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5898         if (ret)
5899                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5900                         prod->irq, irqfd->gsi, ret);
5901
5902         return ret;
5903 }
5904
5905 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5906                                               struct irq_bypass_producer *prod)
5907 {
5908         int ret;
5909         struct kvm_kernel_irqfd *irqfd =
5910                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5911
5912         irqfd->producer = NULL;
5913
5914         /*
5915          * When producer of consumer is unregistered, we change back to
5916          * default external interrupt handling mode - KVM real mode
5917          * will switch back to host.
5918          */
5919         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5920         if (ret)
5921                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5922                         prod->irq, irqfd->gsi, ret);
5923 }
5924 #endif
5925
5926 static int kvm_arch_vm_ioctl_hv(struct file *filp,
5927                                 unsigned int ioctl, unsigned long arg)
5928 {
5929         struct kvm *kvm __maybe_unused = filp->private_data;
5930         void __user *argp = (void __user *)arg;
5931         int r;
5932
5933         switch (ioctl) {
5934
5935         case KVM_PPC_ALLOCATE_HTAB: {
5936                 u32 htab_order;
5937
5938                 /* If we're a nested hypervisor, we currently only support radix */
5939                 if (kvmhv_on_pseries()) {
5940                         r = -EOPNOTSUPP;
5941                         break;
5942                 }
5943
5944                 r = -EFAULT;
5945                 if (get_user(htab_order, (u32 __user *)argp))
5946                         break;
5947                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5948                 if (r)
5949                         break;
5950                 r = 0;
5951                 break;
5952         }
5953
5954         case KVM_PPC_GET_HTAB_FD: {
5955                 struct kvm_get_htab_fd ghf;
5956
5957                 r = -EFAULT;
5958                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5959                         break;
5960                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5961                 break;
5962         }
5963
5964         case KVM_PPC_RESIZE_HPT_PREPARE: {
5965                 struct kvm_ppc_resize_hpt rhpt;
5966
5967                 r = -EFAULT;
5968                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5969                         break;
5970
5971                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5972                 break;
5973         }
5974
5975         case KVM_PPC_RESIZE_HPT_COMMIT: {
5976                 struct kvm_ppc_resize_hpt rhpt;
5977
5978                 r = -EFAULT;
5979                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5980                         break;
5981
5982                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5983                 break;
5984         }
5985
5986         default:
5987                 r = -ENOTTY;
5988         }
5989
5990         return r;
5991 }
5992
5993 /*
5994  * List of hcall numbers to enable by default.
5995  * For compatibility with old userspace, we enable by default
5996  * all hcalls that were implemented before the hcall-enabling
5997  * facility was added.  Note this list should not include H_RTAS.
5998  */
5999 static unsigned int default_hcall_list[] = {
6000         H_REMOVE,
6001         H_ENTER,
6002         H_READ,
6003         H_PROTECT,
6004         H_BULK_REMOVE,
6005 #ifdef CONFIG_SPAPR_TCE_IOMMU
6006         H_GET_TCE,
6007         H_PUT_TCE,
6008 #endif
6009         H_SET_DABR,
6010         H_SET_XDABR,
6011         H_CEDE,
6012         H_PROD,
6013         H_CONFER,
6014         H_REGISTER_VPA,
6015 #ifdef CONFIG_KVM_XICS
6016         H_EOI,
6017         H_CPPR,
6018         H_IPI,
6019         H_IPOLL,
6020         H_XIRR,
6021         H_XIRR_X,
6022 #endif
6023         0
6024 };
6025
6026 static void init_default_hcalls(void)
6027 {
6028         int i;
6029         unsigned int hcall;
6030
6031         for (i = 0; default_hcall_list[i]; ++i) {
6032                 hcall = default_hcall_list[i];
6033                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
6034                 __set_bit(hcall / 4, default_enabled_hcalls);
6035         }
6036 }
6037
6038 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
6039 {
6040         unsigned long lpcr;
6041         int radix;
6042         int err;
6043
6044         /* If not on a POWER9, reject it */
6045         if (!cpu_has_feature(CPU_FTR_ARCH_300))
6046                 return -ENODEV;
6047
6048         /* If any unknown flags set, reject it */
6049         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
6050                 return -EINVAL;
6051
6052         /* GR (guest radix) bit in process_table field must match */
6053         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
6054         if (!!(cfg->process_table & PATB_GR) != radix)
6055                 return -EINVAL;
6056
6057         /* Process table size field must be reasonable, i.e. <= 24 */
6058         if ((cfg->process_table & PRTS_MASK) > 24)
6059                 return -EINVAL;
6060
6061         /* We can change a guest to/from radix now, if the host is radix */
6062         if (radix && !radix_enabled())
6063                 return -EINVAL;
6064
6065         /* If we're a nested hypervisor, we currently only support radix */
6066         if (kvmhv_on_pseries() && !radix)
6067                 return -EINVAL;
6068
6069         mutex_lock(&kvm->arch.mmu_setup_lock);
6070         if (radix != kvm_is_radix(kvm)) {
6071                 if (kvm->arch.mmu_ready) {
6072                         kvm->arch.mmu_ready = 0;
6073                         /* order mmu_ready vs. vcpus_running */
6074                         smp_mb();
6075                         if (atomic_read(&kvm->arch.vcpus_running)) {
6076                                 kvm->arch.mmu_ready = 1;
6077                                 err = -EBUSY;
6078                                 goto out_unlock;
6079                         }
6080                 }
6081                 if (radix)
6082                         err = kvmppc_switch_mmu_to_radix(kvm);
6083                 else
6084                         err = kvmppc_switch_mmu_to_hpt(kvm);
6085                 if (err)
6086                         goto out_unlock;
6087         }
6088
6089         kvm->arch.process_table = cfg->process_table;
6090         kvmppc_setup_partition_table(kvm);
6091
6092         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
6093         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
6094         err = 0;
6095
6096  out_unlock:
6097         mutex_unlock(&kvm->arch.mmu_setup_lock);
6098         return err;
6099 }
6100
6101 static int kvmhv_enable_nested(struct kvm *kvm)
6102 {
6103         if (!nested)
6104                 return -EPERM;
6105         if (!cpu_has_feature(CPU_FTR_ARCH_300))
6106                 return -ENODEV;
6107         if (!radix_enabled())
6108                 return -ENODEV;
6109         if (kvmhv_is_nestedv2())
6110                 return -ENODEV;
6111
6112         /* kvm == NULL means the caller is testing if the capability exists */
6113         if (kvm)
6114                 kvm->arch.nested_enable = true;
6115         return 0;
6116 }
6117
6118 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
6119                                  int size)
6120 {
6121         int rc = -EINVAL;
6122
6123         if (kvmhv_vcpu_is_radix(vcpu)) {
6124                 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
6125
6126                 if (rc > 0)
6127                         rc = -EINVAL;
6128         }
6129
6130         /* For now quadrants are the only way to access nested guest memory */
6131         if (rc && vcpu->arch.nested)
6132                 rc = -EAGAIN;
6133
6134         return rc;
6135 }
6136
6137 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
6138                                 int size)
6139 {
6140         int rc = -EINVAL;
6141
6142         if (kvmhv_vcpu_is_radix(vcpu)) {
6143                 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
6144
6145                 if (rc > 0)
6146                         rc = -EINVAL;
6147         }
6148
6149         /* For now quadrants are the only way to access nested guest memory */
6150         if (rc && vcpu->arch.nested)
6151                 rc = -EAGAIN;
6152
6153         return rc;
6154 }
6155
6156 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
6157 {
6158         unpin_vpa(kvm, vpa);
6159         vpa->gpa = 0;
6160         vpa->pinned_addr = NULL;
6161         vpa->dirty = false;
6162         vpa->update_pending = 0;
6163 }
6164
6165 /*
6166  * Enable a guest to become a secure VM, or test whether
6167  * that could be enabled.
6168  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
6169  * tested (kvm == NULL) or enabled (kvm != NULL).
6170  */
6171 static int kvmhv_enable_svm(struct kvm *kvm)
6172 {
6173         if (!kvmppc_uvmem_available())
6174                 return -EINVAL;
6175         if (kvm)
6176                 kvm->arch.svm_enabled = 1;
6177         return 0;
6178 }
6179
6180 /*
6181  *  IOCTL handler to turn off secure mode of guest
6182  *
6183  * - Release all device pages
6184  * - Issue ucall to terminate the guest on the UV side
6185  * - Unpin the VPA pages.
6186  * - Reinit the partition scoped page tables
6187  */
6188 static int kvmhv_svm_off(struct kvm *kvm)
6189 {
6190         struct kvm_vcpu *vcpu;
6191         int mmu_was_ready;
6192         int srcu_idx;
6193         int ret = 0;
6194         unsigned long i;
6195
6196         if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
6197                 return ret;
6198
6199         mutex_lock(&kvm->arch.mmu_setup_lock);
6200         mmu_was_ready = kvm->arch.mmu_ready;
6201         if (kvm->arch.mmu_ready) {
6202                 kvm->arch.mmu_ready = 0;
6203                 /* order mmu_ready vs. vcpus_running */
6204                 smp_mb();
6205                 if (atomic_read(&kvm->arch.vcpus_running)) {
6206                         kvm->arch.mmu_ready = 1;
6207                         ret = -EBUSY;
6208                         goto out;
6209                 }
6210         }
6211
6212         srcu_idx = srcu_read_lock(&kvm->srcu);
6213         for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) {
6214                 struct kvm_memory_slot *memslot;
6215                 struct kvm_memslots *slots = __kvm_memslots(kvm, i);
6216                 int bkt;
6217
6218                 if (!slots)
6219                         continue;
6220
6221                 kvm_for_each_memslot(memslot, bkt, slots) {
6222                         kvmppc_uvmem_drop_pages(memslot, kvm, true);
6223                         uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
6224                 }
6225         }
6226         srcu_read_unlock(&kvm->srcu, srcu_idx);
6227
6228         ret = uv_svm_terminate(kvm->arch.lpid);
6229         if (ret != U_SUCCESS) {
6230                 ret = -EINVAL;
6231                 goto out;
6232         }
6233
6234         /*
6235          * When secure guest is reset, all the guest pages are sent
6236          * to UV via UV_PAGE_IN before the non-boot vcpus get a
6237          * chance to run and unpin their VPA pages. Unpinning of all
6238          * VPA pages is done here explicitly so that VPA pages
6239          * can be migrated to the secure side.
6240          *
6241          * This is required to for the secure SMP guest to reboot
6242          * correctly.
6243          */
6244         kvm_for_each_vcpu(i, vcpu, kvm) {
6245                 spin_lock(&vcpu->arch.vpa_update_lock);
6246                 unpin_vpa_reset(kvm, &vcpu->arch.dtl);
6247                 unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
6248                 unpin_vpa_reset(kvm, &vcpu->arch.vpa);
6249                 spin_unlock(&vcpu->arch.vpa_update_lock);
6250         }
6251
6252         kvmppc_setup_partition_table(kvm);
6253         kvm->arch.secure_guest = 0;
6254         kvm->arch.mmu_ready = mmu_was_ready;
6255 out:
6256         mutex_unlock(&kvm->arch.mmu_setup_lock);
6257         return ret;
6258 }
6259
6260 static int kvmhv_enable_dawr1(struct kvm *kvm)
6261 {
6262         if (!cpu_has_feature(CPU_FTR_DAWR1))
6263                 return -ENODEV;
6264
6265         /* kvm == NULL means the caller is testing if the capability exists */
6266         if (kvm)
6267                 kvm->arch.dawr1_enabled = true;
6268         return 0;
6269 }
6270
6271 static bool kvmppc_hash_v3_possible(void)
6272 {
6273         if (!cpu_has_feature(CPU_FTR_ARCH_300))
6274                 return false;
6275
6276         if (!cpu_has_feature(CPU_FTR_HVMODE))
6277                 return false;
6278
6279         /*
6280          * POWER9 chips before version 2.02 can't have some threads in
6281          * HPT mode and some in radix mode on the same core.
6282          */
6283         if (radix_enabled()) {
6284                 unsigned int pvr = mfspr(SPRN_PVR);
6285                 if ((pvr >> 16) == PVR_POWER9 &&
6286                     (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
6287                      ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
6288                         return false;
6289         }
6290
6291         return true;
6292 }
6293
6294 static struct kvmppc_ops kvm_ops_hv = {
6295         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
6296         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
6297         .get_one_reg = kvmppc_get_one_reg_hv,
6298         .set_one_reg = kvmppc_set_one_reg_hv,
6299         .vcpu_load   = kvmppc_core_vcpu_load_hv,
6300         .vcpu_put    = kvmppc_core_vcpu_put_hv,
6301         .inject_interrupt = kvmppc_inject_interrupt_hv,
6302         .set_msr     = kvmppc_set_msr_hv,
6303         .vcpu_run    = kvmppc_vcpu_run_hv,
6304         .vcpu_create = kvmppc_core_vcpu_create_hv,
6305         .vcpu_free   = kvmppc_core_vcpu_free_hv,
6306         .check_requests = kvmppc_core_check_requests_hv,
6307         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
6308         .flush_memslot  = kvmppc_core_flush_memslot_hv,
6309         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
6310         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
6311         .unmap_gfn_range = kvm_unmap_gfn_range_hv,
6312         .age_gfn = kvm_age_gfn_hv,
6313         .test_age_gfn = kvm_test_age_gfn_hv,
6314         .set_spte_gfn = kvm_set_spte_gfn_hv,
6315         .free_memslot = kvmppc_core_free_memslot_hv,
6316         .init_vm =  kvmppc_core_init_vm_hv,
6317         .destroy_vm = kvmppc_core_destroy_vm_hv,
6318         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
6319         .emulate_op = kvmppc_core_emulate_op_hv,
6320         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
6321         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
6322         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
6323         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
6324         .hcall_implemented = kvmppc_hcall_impl_hv,
6325 #ifdef CONFIG_KVM_XICS
6326         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
6327         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
6328 #endif
6329         .configure_mmu = kvmhv_configure_mmu,
6330         .get_rmmu_info = kvmhv_get_rmmu_info,
6331         .set_smt_mode = kvmhv_set_smt_mode,
6332         .enable_nested = kvmhv_enable_nested,
6333         .load_from_eaddr = kvmhv_load_from_eaddr,
6334         .store_to_eaddr = kvmhv_store_to_eaddr,
6335         .enable_svm = kvmhv_enable_svm,
6336         .svm_off = kvmhv_svm_off,
6337         .enable_dawr1 = kvmhv_enable_dawr1,
6338         .hash_v3_possible = kvmppc_hash_v3_possible,
6339         .create_vcpu_debugfs = kvmppc_arch_create_vcpu_debugfs_hv,
6340         .create_vm_debugfs = kvmppc_arch_create_vm_debugfs_hv,
6341 };
6342
6343 static int kvm_init_subcore_bitmap(void)
6344 {
6345         int i, j;
6346         int nr_cores = cpu_nr_cores();
6347         struct sibling_subcore_state *sibling_subcore_state;
6348
6349         for (i = 0; i < nr_cores; i++) {
6350                 int first_cpu = i * threads_per_core;
6351                 int node = cpu_to_node(first_cpu);
6352
6353                 /* Ignore if it is already allocated. */
6354                 if (paca_ptrs[first_cpu]->sibling_subcore_state)
6355                         continue;
6356
6357                 sibling_subcore_state =
6358                         kzalloc_node(sizeof(struct sibling_subcore_state),
6359                                                         GFP_KERNEL, node);
6360                 if (!sibling_subcore_state)
6361                         return -ENOMEM;
6362
6363
6364                 for (j = 0; j < threads_per_core; j++) {
6365                         int cpu = first_cpu + j;
6366
6367                         paca_ptrs[cpu]->sibling_subcore_state =
6368                                                 sibling_subcore_state;
6369                 }
6370         }
6371         return 0;
6372 }
6373
6374 static int kvmppc_radix_possible(void)
6375 {
6376         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
6377 }
6378
6379 static int kvmppc_book3s_init_hv(void)
6380 {
6381         int r;
6382
6383         if (!tlbie_capable) {
6384                 pr_err("KVM-HV: Host does not support TLBIE\n");
6385                 return -ENODEV;
6386         }
6387
6388         /*
6389          * FIXME!! Do we need to check on all cpus ?
6390          */
6391         r = kvmppc_core_check_processor_compat_hv();
6392         if (r < 0)
6393                 return -ENODEV;
6394
6395         r = kvmhv_nested_init();
6396         if (r)
6397                 return r;
6398
6399         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
6400                 r = kvm_init_subcore_bitmap();
6401                 if (r)
6402                         goto err;
6403         }
6404
6405         /*
6406          * We need a way of accessing the XICS interrupt controller,
6407          * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
6408          * indirectly, via OPAL.
6409          */
6410 #ifdef CONFIG_SMP
6411         if (!xics_on_xive() && !kvmhv_on_pseries() &&
6412             !local_paca->kvm_hstate.xics_phys) {
6413                 struct device_node *np;
6414
6415                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
6416                 if (!np) {
6417                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
6418                         r = -ENODEV;
6419                         goto err;
6420                 }
6421                 /* presence of intc confirmed - node can be dropped again */
6422                 of_node_put(np);
6423         }
6424 #endif
6425
6426         init_default_hcalls();
6427
6428         init_vcore_lists();
6429
6430         r = kvmppc_mmu_hv_init();
6431         if (r)
6432                 goto err;
6433
6434         if (kvmppc_radix_possible()) {
6435                 r = kvmppc_radix_init();
6436                 if (r)
6437                         goto err;
6438         }
6439
6440         r = kvmppc_uvmem_init();
6441         if (r < 0) {
6442                 pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
6443                 return r;
6444         }
6445
6446         kvm_ops_hv.owner = THIS_MODULE;
6447         kvmppc_hv_ops = &kvm_ops_hv;
6448
6449         return 0;
6450
6451 err:
6452         kvmhv_nested_exit();
6453         kvmppc_radix_exit();
6454
6455         return r;
6456 }
6457
6458 static void kvmppc_book3s_exit_hv(void)
6459 {
6460         kvmppc_uvmem_free();
6461         kvmppc_free_host_rm_ops();
6462         if (kvmppc_radix_possible())
6463                 kvmppc_radix_exit();
6464         kvmppc_hv_ops = NULL;
6465         kvmhv_nested_exit();
6466 }
6467
6468 module_init(kvmppc_book3s_init_hv);
6469 module_exit(kvmppc_book3s_exit_hv);
6470 MODULE_LICENSE("GPL");
6471 MODULE_ALIAS_MISCDEV(KVM_MINOR);
6472 MODULE_ALIAS("devname:kvm");