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
  46 #include <asm/ftrace.h>
  47 #include <asm/reg.h>
  48 #include <asm/ppc-opcode.h>
  49 #include <asm/asm-prototypes.h>
  50 #include <asm/archrandom.h>
  51 #include <asm/debug.h>
  52 #include <asm/disassemble.h>
  53 #include <asm/cputable.h>
  54 #include <asm/cacheflush.h>
  55 #include <linux/uaccess.h>
  56 #include <asm/interrupt.h>
  57 #include <asm/io.h>
  58 #include <asm/kvm_ppc.h>
  59 #include <asm/kvm_book3s.h>
  60 #include <asm/mmu_context.h>
  61 #include <asm/lppaca.h>
  62 #include <asm/pmc.h>
  63 #include <asm/processor.h>
  64 #include <asm/cputhreads.h>
  65 #include <asm/page.h>
  66 #include <asm/hvcall.h>
  67 #include <asm/switch_to.h>
  68 #include <asm/smp.h>
  69 #include <asm/dbell.h>
  70 #include <asm/hmi.h>
  71 #include <asm/pnv-pci.h>
  72 #include <asm/mmu.h>
  73 #include <asm/opal.h>
  74 #include <asm/xics.h>
  75 #include <asm/xive.h>
  76 #include <asm/hw_breakpoint.h>
  77 #include <asm/kvm_book3s_uvmem.h>
  78 #include <asm/ultravisor.h>
  79 #include <asm/dtl.h>
  80 #include <asm/plpar_wrappers.h>
  81
  82 #include "book3s.h"
  83
  84 #define CREATE_TRACE_POINTS
  85 #include "trace_hv.h"
  86
  87 /* #define EXIT_DEBUG */
  88 /* #define EXIT_DEBUG_SIMPLE */
  89 /* #define EXIT_DEBUG_INT */
  90
  91 /* Used to indicate that a guest page fault needs to be handled */
  92 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  93 /* Used to indicate that a guest passthrough interrupt needs to be handled */
  94 #define RESUME_PASSTHROUGH      (RESUME_GUEST | RESUME_FLAG_ARCH2)
  95
  96 /* Used as a "null" value for timebase values */
  97 #define TB_NIL  (~(u64)0)
  98
  99 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
 100
 101 static int dynamic_mt_modes = 6;
 102 module_param(dynamic_mt_modes, int, 0644);
 103 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
 104 static int target_smt_mode;
 105 module_param(target_smt_mode, int, 0644);
 106 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
 107
 108 static bool one_vm_per_core;
 109 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
 110 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires POWER8 or older)");
 111
 112 #ifdef CONFIG_KVM_XICS
 113 static const struct kernel_param_ops module_param_ops = {
 114         .set = param_set_int,
 115         .get = param_get_int,
 116 };
 117
 118 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
 119 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
 120
 121 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
 122 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
 123 #endif
 124
 125 /* If set, guests are allowed to create and control nested guests */
 126 static bool nested = true;
 127 module_param(nested, bool, S_IRUGO | S_IWUSR);
 128 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
 129
 130 static inline bool nesting_enabled(struct kvm *kvm)
 131 {
 132         return kvm->arch.nested_enable && kvm_is_radix(kvm);
 133 }
 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         waitp = kvm_arch_vcpu_get_wait(vcpu);
 233         if (rcuwait_wake_up(waitp))
 234                 ++vcpu->stat.generic.halt_wakeup;
 235
 236         cpu = READ_ONCE(vcpu->arch.thread_cpu);
 237         if (cpu >= 0 && kvmppc_ipi_thread(cpu))
 238                 return;
 239
 240         /* CPU points to the first thread of the core */
 241         cpu = vcpu->cpu;
 242         if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
 243                 smp_send_reschedule(cpu);
 244 }
 245
 246 /*
 247  * We use the vcpu_load/put functions to measure stolen time.
 248  * Stolen time is counted as time when either the vcpu is able to
 249  * run as part of a virtual core, but the task running the vcore
 250  * is preempted or sleeping, or when the vcpu needs something done
 251  * in the kernel by the task running the vcpu, but that task is
 252  * preempted or sleeping.  Those two things have to be counted
 253  * separately, since one of the vcpu tasks will take on the job
 254  * of running the core, and the other vcpu tasks in the vcore will
 255  * sleep waiting for it to do that, but that sleep shouldn't count
 256  * as stolen time.
 257  *
 258  * Hence we accumulate stolen time when the vcpu can run as part of
 259  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 260  * needs its task to do other things in the kernel (for example,
 261  * service a page fault) in busy_stolen.  We don't accumulate
 262  * stolen time for a vcore when it is inactive, or for a vcpu
 263  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 264  * a misnomer; it means that the vcpu task is not executing in
 265  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 266  * the kernel.  We don't have any way of dividing up that time
 267  * between time that the vcpu is genuinely stopped, time that
 268  * the task is actively working on behalf of the vcpu, and time
 269  * that the task is preempted, so we don't count any of it as
 270  * stolen.
 271  *
 272  * Updates to busy_stolen are protected by arch.tbacct_lock;
 273  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
 274  * lock.  The stolen times are measured in units of timebase ticks.
 275  * (Note that the != TB_NIL checks below are purely defensive;
 276  * they should never fail.)
 277  */
 278
 279 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
 280 {
 281         unsigned long flags;
 282
 283         spin_lock_irqsave(&vc->stoltb_lock, flags);
 284         vc->preempt_tb = mftb();
 285         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 286 }
 287
 288 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
 289 {
 290         unsigned long flags;
 291
 292         spin_lock_irqsave(&vc->stoltb_lock, flags);
 293         if (vc->preempt_tb != TB_NIL) {
 294                 vc->stolen_tb += mftb() - vc->preempt_tb;
 295                 vc->preempt_tb = TB_NIL;
 296         }
 297         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 298 }
 299
 300 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 301 {
 302         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 303         unsigned long flags;
 304
 305         /*
 306          * We can test vc->runner without taking the vcore lock,
 307          * because only this task ever sets vc->runner to this
 308          * vcpu, and once it is set to this vcpu, only this task
 309          * ever sets it to NULL.
 310          */
 311         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 312                 kvmppc_core_end_stolen(vc);
 313
 314         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 315         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 316             vcpu->arch.busy_preempt != TB_NIL) {
 317                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 318                 vcpu->arch.busy_preempt = TB_NIL;
 319         }
 320         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 321 }
 322
 323 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 324 {
 325         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 326         unsigned long flags;
 327
 328         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 329                 kvmppc_core_start_stolen(vc);
 330
 331         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 332         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 333                 vcpu->arch.busy_preempt = mftb();
 334         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 335 }
 336
 337 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 338 {
 339         vcpu->arch.pvr = pvr;
 340 }
 341
 342 /* Dummy value used in computing PCR value below */
 343 #define PCR_ARCH_31    (PCR_ARCH_300 << 1)
 344
 345 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 346 {
 347         unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
 348         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 349
 350         /* We can (emulate) our own architecture version and anything older */
 351         if (cpu_has_feature(CPU_FTR_ARCH_31))
 352                 host_pcr_bit = PCR_ARCH_31;
 353         else if (cpu_has_feature(CPU_FTR_ARCH_300))
 354                 host_pcr_bit = PCR_ARCH_300;
 355         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
 356                 host_pcr_bit = PCR_ARCH_207;
 357         else if (cpu_has_feature(CPU_FTR_ARCH_206))
 358                 host_pcr_bit = PCR_ARCH_206;
 359         else
 360                 host_pcr_bit = PCR_ARCH_205;
 361
 362         /* Determine lowest PCR bit needed to run guest in given PVR level */
 363         guest_pcr_bit = host_pcr_bit;
 364         if (arch_compat) {
 365                 switch (arch_compat) {
 366                 case PVR_ARCH_205:
 367                         guest_pcr_bit = PCR_ARCH_205;
 368                         break;
 369                 case PVR_ARCH_206:
 370                 case PVR_ARCH_206p:
 371                         guest_pcr_bit = PCR_ARCH_206;
 372                         break;
 373                 case PVR_ARCH_207:
 374                         guest_pcr_bit = PCR_ARCH_207;
 375                         break;
 376                 case PVR_ARCH_300:
 377                         guest_pcr_bit = PCR_ARCH_300;
 378                         break;
 379                 case PVR_ARCH_31:
 380                         guest_pcr_bit = PCR_ARCH_31;
 381                         break;
 382                 default:
 383                         return -EINVAL;
 384                 }
 385         }
 386
 387         /* Check requested PCR bits don't exceed our capabilities */
 388         if (guest_pcr_bit > host_pcr_bit)
 389                 return -EINVAL;
 390
 391         spin_lock(&vc->lock);
 392         vc->arch_compat = arch_compat;
 393         /*
 394          * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
 395          * Also set all reserved PCR bits
 396          */
 397         vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
 398         spin_unlock(&vc->lock);
 399
 400         return 0;
 401 }
 402
 403 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 404 {
 405         int r;
 406
 407         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 408         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 409                vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
 410         for (r = 0; r < 16; ++r)
 411                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 412                        r, kvmppc_get_gpr(vcpu, r),
 413                        r+16, kvmppc_get_gpr(vcpu, r+16));
 414         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 415                vcpu->arch.regs.ctr, vcpu->arch.regs.link);
 416         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 417                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 418         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 419                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 420         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 421                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 422         pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
 423                vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
 424         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 425         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 426                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 427         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 428         for (r = 0; r < vcpu->arch.slb_max; ++r)
 429                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 430                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 431         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 432                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 433                vcpu->arch.last_inst);
 434 }
 435
 436 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 437 {
 438         return kvm_get_vcpu_by_id(kvm, id);
 439 }
 440
 441 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 442 {
 443         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 444         vpa->yield_count = cpu_to_be32(1);
 445 }
 446
 447 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 448                    unsigned long addr, unsigned long len)
 449 {
 450         /* check address is cacheline aligned */
 451         if (addr & (L1_CACHE_BYTES - 1))
 452                 return -EINVAL;
 453         spin_lock(&vcpu->arch.vpa_update_lock);
 454         if (v->next_gpa != addr || v->len != len) {
 455                 v->next_gpa = addr;
 456                 v->len = addr ? len : 0;
 457                 v->update_pending = 1;
 458         }
 459         spin_unlock(&vcpu->arch.vpa_update_lock);
 460         return 0;
 461 }
 462
 463 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 464 struct reg_vpa {
 465         u32 dummy;
 466         union {
 467                 __be16 hword;
 468                 __be32 word;
 469         } length;
 470 };
 471
 472 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 473 {
 474         if (vpap->update_pending)
 475                 return vpap->next_gpa != 0;
 476         return vpap->pinned_addr != NULL;
 477 }
 478
 479 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 480                                        unsigned long flags,
 481                                        unsigned long vcpuid, unsigned long vpa)
 482 {
 483         struct kvm *kvm = vcpu->kvm;
 484         unsigned long len, nb;
 485         void *va;
 486         struct kvm_vcpu *tvcpu;
 487         int err;
 488         int subfunc;
 489         struct kvmppc_vpa *vpap;
 490
 491         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 492         if (!tvcpu)
 493                 return H_PARAMETER;
 494
 495         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 496         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 497             subfunc == H_VPA_REG_SLB) {
 498                 /* Registering new area - address must be cache-line aligned */
 499                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 500                         return H_PARAMETER;
 501
 502                 /* convert logical addr to kernel addr and read length */
 503                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 504                 if (va == NULL)
 505                         return H_PARAMETER;
 506                 if (subfunc == H_VPA_REG_VPA)
 507                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 508                 else
 509                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 510                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 511
 512                 /* Check length */
 513                 if (len > nb || len < sizeof(struct reg_vpa))
 514                         return H_PARAMETER;
 515         } else {
 516                 vpa = 0;
 517                 len = 0;
 518         }
 519
 520         err = H_PARAMETER;
 521         vpap = NULL;
 522         spin_lock(&tvcpu->arch.vpa_update_lock);
 523
 524         switch (subfunc) {
 525         case H_VPA_REG_VPA:             /* register VPA */
 526                 /*
 527                  * The size of our lppaca is 1kB because of the way we align
 528                  * it for the guest to avoid crossing a 4kB boundary. We only
 529                  * use 640 bytes of the structure though, so we should accept
 530                  * clients that set a size of 640.
 531                  */
 532                 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
 533                 if (len < sizeof(struct lppaca))
 534                         break;
 535                 vpap = &tvcpu->arch.vpa;
 536                 err = 0;
 537                 break;
 538
 539         case H_VPA_REG_DTL:             /* register DTL */
 540                 if (len < sizeof(struct dtl_entry))
 541                         break;
 542                 len -= len % sizeof(struct dtl_entry);
 543
 544                 /* Check that they have previously registered a VPA */
 545                 err = H_RESOURCE;
 546                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 547                         break;
 548
 549                 vpap = &tvcpu->arch.dtl;
 550                 err = 0;
 551                 break;
 552
 553         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 554                 /* Check that they have previously registered a VPA */
 555                 err = H_RESOURCE;
 556                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 557                         break;
 558
 559                 vpap = &tvcpu->arch.slb_shadow;
 560                 err = 0;
 561                 break;
 562
 563         case H_VPA_DEREG_VPA:           /* deregister VPA */
 564                 /* Check they don't still have a DTL or SLB buf registered */
 565                 err = H_RESOURCE;
 566                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 567                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 568                         break;
 569
 570                 vpap = &tvcpu->arch.vpa;
 571                 err = 0;
 572                 break;
 573
 574         case H_VPA_DEREG_DTL:           /* deregister DTL */
 575                 vpap = &tvcpu->arch.dtl;
 576                 err = 0;
 577                 break;
 578
 579         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 580                 vpap = &tvcpu->arch.slb_shadow;
 581                 err = 0;
 582                 break;
 583         }
 584
 585         if (vpap) {
 586                 vpap->next_gpa = vpa;
 587                 vpap->len = len;
 588                 vpap->update_pending = 1;
 589         }
 590
 591         spin_unlock(&tvcpu->arch.vpa_update_lock);
 592
 593         return err;
 594 }
 595
 596 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 597 {
 598         struct kvm *kvm = vcpu->kvm;
 599         void *va;
 600         unsigned long nb;
 601         unsigned long gpa;
 602
 603         /*
 604          * We need to pin the page pointed to by vpap->next_gpa,
 605          * but we can't call kvmppc_pin_guest_page under the lock
 606          * as it does get_user_pages() and down_read().  So we
 607          * have to drop the lock, pin the page, then get the lock
 608          * again and check that a new area didn't get registered
 609          * in the meantime.
 610          */
 611         for (;;) {
 612                 gpa = vpap->next_gpa;
 613                 spin_unlock(&vcpu->arch.vpa_update_lock);
 614                 va = NULL;
 615                 nb = 0;
 616                 if (gpa)
 617                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 618                 spin_lock(&vcpu->arch.vpa_update_lock);
 619                 if (gpa == vpap->next_gpa)
 620                         break;
 621                 /* sigh... unpin that one and try again */
 622                 if (va)
 623                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 624         }
 625
 626         vpap->update_pending = 0;
 627         if (va && nb < vpap->len) {
 628                 /*
 629                  * If it's now too short, it must be that userspace
 630                  * has changed the mappings underlying guest memory,
 631                  * so unregister the region.
 632                  */
 633                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 634                 va = NULL;
 635         }
 636         if (vpap->pinned_addr)
 637                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 638                                         vpap->dirty);
 639         vpap->gpa = gpa;
 640         vpap->pinned_addr = va;
 641         vpap->dirty = false;
 642         if (va)
 643                 vpap->pinned_end = va + vpap->len;
 644 }
 645
 646 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 647 {
 648         if (!(vcpu->arch.vpa.update_pending ||
 649               vcpu->arch.slb_shadow.update_pending ||
 650               vcpu->arch.dtl.update_pending))
 651                 return;
 652
 653         spin_lock(&vcpu->arch.vpa_update_lock);
 654         if (vcpu->arch.vpa.update_pending) {
 655                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 656                 if (vcpu->arch.vpa.pinned_addr)
 657                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 658         }
 659         if (vcpu->arch.dtl.update_pending) {
 660                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 661                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 662                 vcpu->arch.dtl_index = 0;
 663         }
 664         if (vcpu->arch.slb_shadow.update_pending)
 665                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 666         spin_unlock(&vcpu->arch.vpa_update_lock);
 667 }
 668
 669 /*
 670  * Return the accumulated stolen time for the vcore up until `now'.
 671  * The caller should hold the vcore lock.
 672  */
 673 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 674 {
 675         u64 p;
 676         unsigned long flags;
 677
 678         spin_lock_irqsave(&vc->stoltb_lock, flags);
 679         p = vc->stolen_tb;
 680         if (vc->vcore_state != VCORE_INACTIVE &&
 681             vc->preempt_tb != TB_NIL)
 682                 p += now - vc->preempt_tb;
 683         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 684         return p;
 685 }
 686
 687 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 688                                     struct kvmppc_vcore *vc)
 689 {
 690         struct dtl_entry *dt;
 691         struct lppaca *vpa;
 692         unsigned long stolen;
 693         unsigned long core_stolen;
 694         u64 now;
 695         unsigned long flags;
 696
 697         dt = vcpu->arch.dtl_ptr;
 698         vpa = vcpu->arch.vpa.pinned_addr;
 699         now = mftb();
 700         core_stolen = vcore_stolen_time(vc, now);
 701         stolen = core_stolen - vcpu->arch.stolen_logged;
 702         vcpu->arch.stolen_logged = core_stolen;
 703         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 704         stolen += vcpu->arch.busy_stolen;
 705         vcpu->arch.busy_stolen = 0;
 706         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 707         if (!dt || !vpa)
 708                 return;
 709         memset(dt, 0, sizeof(struct dtl_entry));
 710         dt->dispatch_reason = 7;
 711         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 712         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 713         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 714         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 715         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 716         ++dt;
 717         if (dt == vcpu->arch.dtl.pinned_end)
 718                 dt = vcpu->arch.dtl.pinned_addr;
 719         vcpu->arch.dtl_ptr = dt;
 720         /* order writing *dt vs. writing vpa->dtl_idx */
 721         smp_wmb();
 722         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 723         vcpu->arch.dtl.dirty = true;
 724 }
 725
 726 /* See if there is a doorbell interrupt pending for a vcpu */
 727 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
 728 {
 729         int thr;
 730         struct kvmppc_vcore *vc;
 731
 732         if (vcpu->arch.doorbell_request)
 733                 return true;
 734         /*
 735          * Ensure that the read of vcore->dpdes comes after the read
 736          * of vcpu->doorbell_request.  This barrier matches the
 737          * smp_wmb() in kvmppc_guest_entry_inject().
 738          */
 739         smp_rmb();
 740         vc = vcpu->arch.vcore;
 741         thr = vcpu->vcpu_id - vc->first_vcpuid;
 742         return !!(vc->dpdes & (1 << thr));
 743 }
 744
 745 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 746 {
 747         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 748                 return true;
 749         if ((!vcpu->arch.vcore->arch_compat) &&
 750             cpu_has_feature(CPU_FTR_ARCH_207S))
 751                 return true;
 752         return false;
 753 }
 754
 755 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 756                              unsigned long resource, unsigned long value1,
 757                              unsigned long value2)
 758 {
 759         switch (resource) {
 760         case H_SET_MODE_RESOURCE_SET_CIABR:
 761                 if (!kvmppc_power8_compatible(vcpu))
 762                         return H_P2;
 763                 if (value2)
 764                         return H_P4;
 765                 if (mflags)
 766                         return H_UNSUPPORTED_FLAG_START;
 767                 /* Guests can't breakpoint the hypervisor */
 768                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 769                         return H_P3;
 770                 vcpu->arch.ciabr  = value1;
 771                 return H_SUCCESS;
 772         case H_SET_MODE_RESOURCE_SET_DAWR0:
 773                 if (!kvmppc_power8_compatible(vcpu))
 774                         return H_P2;
 775                 if (!ppc_breakpoint_available())
 776                         return H_P2;
 777                 if (mflags)
 778                         return H_UNSUPPORTED_FLAG_START;
 779                 if (value2 & DABRX_HYP)
 780                         return H_P4;
 781                 vcpu->arch.dawr0  = value1;
 782                 vcpu->arch.dawrx0 = value2;
 783                 return H_SUCCESS;
 784         case H_SET_MODE_RESOURCE_SET_DAWR1:
 785                 if (!kvmppc_power8_compatible(vcpu))
 786                         return H_P2;
 787                 if (!ppc_breakpoint_available())
 788                         return H_P2;
 789                 if (!cpu_has_feature(CPU_FTR_DAWR1))
 790                         return H_P2;
 791                 if (!vcpu->kvm->arch.dawr1_enabled)
 792                         return H_FUNCTION;
 793                 if (mflags)
 794                         return H_UNSUPPORTED_FLAG_START;
 795                 if (value2 & DABRX_HYP)
 796                         return H_P4;
 797                 vcpu->arch.dawr1  = value1;
 798                 vcpu->arch.dawrx1 = value2;
 799                 return H_SUCCESS;
 800         case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
 801                 /*
 802                  * KVM does not support mflags=2 (AIL=2) and AIL=1 is reserved.
 803                  * Keep this in synch with kvmppc_filter_guest_lpcr_hv.
 804                  */
 805                 if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
 806                                 kvmhv_vcpu_is_radix(vcpu) && mflags == 3)
 807                         return H_UNSUPPORTED_FLAG_START;
 808                 return H_TOO_HARD;
 809         default:
 810                 return H_TOO_HARD;
 811         }
 812 }
 813
 814 /* Copy guest memory in place - must reside within a single memslot */
 815 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
 816                                   unsigned long len)
 817 {
 818         struct kvm_memory_slot *to_memslot = NULL;
 819         struct kvm_memory_slot *from_memslot = NULL;
 820         unsigned long to_addr, from_addr;
 821         int r;
 822
 823         /* Get HPA for from address */
 824         from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
 825         if (!from_memslot)
 826                 return -EFAULT;
 827         if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
 828                              << PAGE_SHIFT))
 829                 return -EINVAL;
 830         from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
 831         if (kvm_is_error_hva(from_addr))
 832                 return -EFAULT;
 833         from_addr |= (from & (PAGE_SIZE - 1));
 834
 835         /* Get HPA for to address */
 836         to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
 837         if (!to_memslot)
 838                 return -EFAULT;
 839         if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
 840                            << PAGE_SHIFT))
 841                 return -EINVAL;
 842         to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
 843         if (kvm_is_error_hva(to_addr))
 844                 return -EFAULT;
 845         to_addr |= (to & (PAGE_SIZE - 1));
 846
 847         /* Perform copy */
 848         r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
 849                              len);
 850         if (r)
 851                 return -EFAULT;
 852         mark_page_dirty(kvm, to >> PAGE_SHIFT);
 853         return 0;
 854 }
 855
 856 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
 857                                unsigned long dest, unsigned long src)
 858 {
 859         u64 pg_sz = SZ_4K;              /* 4K page size */
 860         u64 pg_mask = SZ_4K - 1;
 861         int ret;
 862
 863         /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
 864         if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
 865                       H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
 866                 return H_PARAMETER;
 867
 868         /* dest (and src if copy_page flag set) must be page aligned */
 869         if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
 870                 return H_PARAMETER;
 871
 872         /* zero and/or copy the page as determined by the flags */
 873         if (flags & H_COPY_PAGE) {
 874                 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
 875                 if (ret < 0)
 876                         return H_PARAMETER;
 877         } else if (flags & H_ZERO_PAGE) {
 878                 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
 879                 if (ret < 0)
 880                         return H_PARAMETER;
 881         }
 882
 883         /* We can ignore the remaining flags */
 884
 885         return H_SUCCESS;
 886 }
 887
 888 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 889 {
 890         struct kvmppc_vcore *vcore = target->arch.vcore;
 891
 892         /*
 893          * We expect to have been called by the real mode handler
 894          * (kvmppc_rm_h_confer()) which would have directly returned
 895          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 896          * have useful work to do and should not confer) so we don't
 897          * recheck that here.
 898          *
 899          * In the case of the P9 single vcpu per vcore case, the real
 900          * mode handler is not called but no other threads are in the
 901          * source vcore.
 902          */
 903
 904         spin_lock(&vcore->lock);
 905         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 906             vcore->vcore_state != VCORE_INACTIVE &&
 907             vcore->runner)
 908                 target = vcore->runner;
 909         spin_unlock(&vcore->lock);
 910
 911         return kvm_vcpu_yield_to(target);
 912 }
 913
 914 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 915 {
 916         int yield_count = 0;
 917         struct lppaca *lppaca;
 918
 919         spin_lock(&vcpu->arch.vpa_update_lock);
 920         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 921         if (lppaca)
 922                 yield_count = be32_to_cpu(lppaca->yield_count);
 923         spin_unlock(&vcpu->arch.vpa_update_lock);
 924         return yield_count;
 925 }
 926
 927 /*
 928  * H_RPT_INVALIDATE hcall handler for nested guests.
 929  *
 930  * Handles only nested process-scoped invalidation requests in L0.
 931  */
 932 static int kvmppc_nested_h_rpt_invalidate(struct kvm_vcpu *vcpu)
 933 {
 934         unsigned long type = kvmppc_get_gpr(vcpu, 6);
 935         unsigned long pid, pg_sizes, start, end;
 936
 937         /*
 938          * The partition-scoped invalidations aren't handled here in L0.
 939          */
 940         if (type & H_RPTI_TYPE_NESTED)
 941                 return RESUME_HOST;
 942
 943         pid = kvmppc_get_gpr(vcpu, 4);
 944         pg_sizes = kvmppc_get_gpr(vcpu, 7);
 945         start = kvmppc_get_gpr(vcpu, 8);
 946         end = kvmppc_get_gpr(vcpu, 9);
 947
 948         do_h_rpt_invalidate_prt(pid, vcpu->arch.nested->shadow_lpid,
 949                                 type, pg_sizes, start, end);
 950
 951         kvmppc_set_gpr(vcpu, 3, H_SUCCESS);
 952         return RESUME_GUEST;
 953 }
 954
 955 static long kvmppc_h_rpt_invalidate(struct kvm_vcpu *vcpu,
 956                                     unsigned long id, unsigned long target,
 957                                     unsigned long type, unsigned long pg_sizes,
 958                                     unsigned long start, unsigned long end)
 959 {
 960         if (!kvm_is_radix(vcpu->kvm))
 961                 return H_UNSUPPORTED;
 962
 963         if (end < start)
 964                 return H_P5;
 965
 966         /*
 967          * Partition-scoped invalidation for nested guests.
 968          */
 969         if (type & H_RPTI_TYPE_NESTED) {
 970                 if (!nesting_enabled(vcpu->kvm))
 971                         return H_FUNCTION;
 972
 973                 /* Support only cores as target */
 974                 if (target != H_RPTI_TARGET_CMMU)
 975                         return H_P2;
 976
 977                 return do_h_rpt_invalidate_pat(vcpu, id, type, pg_sizes,
 978                                                start, end);
 979         }
 980
 981         /*
 982          * Process-scoped invalidation for L1 guests.
 983          */
 984         do_h_rpt_invalidate_prt(id, vcpu->kvm->arch.lpid,
 985                                 type, pg_sizes, start, end);
 986         return H_SUCCESS;
 987 }
 988
 989 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 990 {
 991         struct kvm *kvm = vcpu->kvm;
 992         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 993         unsigned long target, ret = H_SUCCESS;
 994         int yield_count;
 995         struct kvm_vcpu *tvcpu;
 996         int idx, rc;
 997
 998         if (req <= MAX_HCALL_OPCODE &&
 999             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
1000                 return RESUME_HOST;
1001
1002         switch (req) {
1003         case H_REMOVE:
1004                 ret = kvmppc_h_remove(vcpu, kvmppc_get_gpr(vcpu, 4),
1005                                         kvmppc_get_gpr(vcpu, 5),
1006                                         kvmppc_get_gpr(vcpu, 6));
1007                 if (ret == H_TOO_HARD)
1008                         return RESUME_HOST;
1009                 break;
1010         case H_ENTER:
1011                 ret = kvmppc_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
1012                                         kvmppc_get_gpr(vcpu, 5),
1013                                         kvmppc_get_gpr(vcpu, 6),
1014                                         kvmppc_get_gpr(vcpu, 7));
1015                 if (ret == H_TOO_HARD)
1016                         return RESUME_HOST;
1017                 break;
1018         case H_READ:
1019                 ret = kvmppc_h_read(vcpu, kvmppc_get_gpr(vcpu, 4),
1020                                         kvmppc_get_gpr(vcpu, 5));
1021                 if (ret == H_TOO_HARD)
1022                         return RESUME_HOST;
1023                 break;
1024         case H_CLEAR_MOD:
1025                 ret = kvmppc_h_clear_mod(vcpu, kvmppc_get_gpr(vcpu, 4),
1026                                         kvmppc_get_gpr(vcpu, 5));
1027                 if (ret == H_TOO_HARD)
1028                         return RESUME_HOST;
1029                 break;
1030         case H_CLEAR_REF:
1031                 ret = kvmppc_h_clear_ref(vcpu, kvmppc_get_gpr(vcpu, 4),
1032                                         kvmppc_get_gpr(vcpu, 5));
1033                 if (ret == H_TOO_HARD)
1034                         return RESUME_HOST;
1035                 break;
1036         case H_PROTECT:
1037                 ret = kvmppc_h_protect(vcpu, kvmppc_get_gpr(vcpu, 4),
1038                                         kvmppc_get_gpr(vcpu, 5),
1039                                         kvmppc_get_gpr(vcpu, 6));
1040                 if (ret == H_TOO_HARD)
1041                         return RESUME_HOST;
1042                 break;
1043         case H_BULK_REMOVE:
1044                 ret = kvmppc_h_bulk_remove(vcpu);
1045                 if (ret == H_TOO_HARD)
1046                         return RESUME_HOST;
1047                 break;
1048
1049         case H_CEDE:
1050                 break;
1051         case H_PROD:
1052                 target = kvmppc_get_gpr(vcpu, 4);
1053                 tvcpu = kvmppc_find_vcpu(kvm, target);
1054                 if (!tvcpu) {
1055                         ret = H_PARAMETER;
1056                         break;
1057                 }
1058                 tvcpu->arch.prodded = 1;
1059                 smp_mb();
1060                 if (tvcpu->arch.ceded)
1061                         kvmppc_fast_vcpu_kick_hv(tvcpu);
1062                 break;
1063         case H_CONFER:
1064                 target = kvmppc_get_gpr(vcpu, 4);
1065                 if (target == -1)
1066                         break;
1067                 tvcpu = kvmppc_find_vcpu(kvm, target);
1068                 if (!tvcpu) {
1069                         ret = H_PARAMETER;
1070                         break;
1071                 }
1072                 yield_count = kvmppc_get_gpr(vcpu, 5);
1073                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
1074                         break;
1075                 kvm_arch_vcpu_yield_to(tvcpu);
1076                 break;
1077         case H_REGISTER_VPA:
1078                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
1079                                         kvmppc_get_gpr(vcpu, 5),
1080                                         kvmppc_get_gpr(vcpu, 6));
1081                 break;
1082         case H_RTAS:
1083                 if (list_empty(&kvm->arch.rtas_tokens))
1084                         return RESUME_HOST;
1085
1086                 idx = srcu_read_lock(&kvm->srcu);
1087                 rc = kvmppc_rtas_hcall(vcpu);
1088                 srcu_read_unlock(&kvm->srcu, idx);
1089
1090                 if (rc == -ENOENT)
1091                         return RESUME_HOST;
1092                 else if (rc == 0)
1093                         break;
1094
1095                 /* Send the error out to userspace via KVM_RUN */
1096                 return rc;
1097         case H_LOGICAL_CI_LOAD:
1098                 ret = kvmppc_h_logical_ci_load(vcpu);
1099                 if (ret == H_TOO_HARD)
1100                         return RESUME_HOST;
1101                 break;
1102         case H_LOGICAL_CI_STORE:
1103                 ret = kvmppc_h_logical_ci_store(vcpu);
1104                 if (ret == H_TOO_HARD)
1105                         return RESUME_HOST;
1106                 break;
1107         case H_SET_MODE:
1108                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
1109                                         kvmppc_get_gpr(vcpu, 5),
1110                                         kvmppc_get_gpr(vcpu, 6),
1111                                         kvmppc_get_gpr(vcpu, 7));
1112                 if (ret == H_TOO_HARD)
1113                         return RESUME_HOST;
1114                 break;
1115         case H_XIRR:
1116         case H_CPPR:
1117         case H_EOI:
1118         case H_IPI:
1119         case H_IPOLL:
1120         case H_XIRR_X:
1121                 if (kvmppc_xics_enabled(vcpu)) {
1122                         if (xics_on_xive()) {
1123                                 ret = H_NOT_AVAILABLE;
1124                                 return RESUME_GUEST;
1125                         }
1126                         ret = kvmppc_xics_hcall(vcpu, req);
1127                         break;
1128                 }
1129                 return RESUME_HOST;
1130         case H_SET_DABR:
1131                 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1132                 break;
1133         case H_SET_XDABR:
1134                 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1135                                                 kvmppc_get_gpr(vcpu, 5));
1136                 break;
1137 #ifdef CONFIG_SPAPR_TCE_IOMMU
1138         case H_GET_TCE:
1139                 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1140                                                 kvmppc_get_gpr(vcpu, 5));
1141                 if (ret == H_TOO_HARD)
1142                         return RESUME_HOST;
1143                 break;
1144         case H_PUT_TCE:
1145                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1146                                                 kvmppc_get_gpr(vcpu, 5),
1147                                                 kvmppc_get_gpr(vcpu, 6));
1148                 if (ret == H_TOO_HARD)
1149                         return RESUME_HOST;
1150                 break;
1151         case H_PUT_TCE_INDIRECT:
1152                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1153                                                 kvmppc_get_gpr(vcpu, 5),
1154                                                 kvmppc_get_gpr(vcpu, 6),
1155                                                 kvmppc_get_gpr(vcpu, 7));
1156                 if (ret == H_TOO_HARD)
1157                         return RESUME_HOST;
1158                 break;
1159         case H_STUFF_TCE:
1160                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1161                                                 kvmppc_get_gpr(vcpu, 5),
1162                                                 kvmppc_get_gpr(vcpu, 6),
1163                                                 kvmppc_get_gpr(vcpu, 7));
1164                 if (ret == H_TOO_HARD)
1165                         return RESUME_HOST;
1166                 break;
1167 #endif
1168         case H_RANDOM:
1169                 if (!arch_get_random_seed_long(&vcpu->arch.regs.gpr[4]))
1170                         ret = H_HARDWARE;
1171                 break;
1172         case H_RPT_INVALIDATE:
1173                 ret = kvmppc_h_rpt_invalidate(vcpu, kvmppc_get_gpr(vcpu, 4),
1174                                               kvmppc_get_gpr(vcpu, 5),
1175                                               kvmppc_get_gpr(vcpu, 6),
1176                                               kvmppc_get_gpr(vcpu, 7),
1177                                               kvmppc_get_gpr(vcpu, 8),
1178                                               kvmppc_get_gpr(vcpu, 9));
1179                 break;
1180
1181         case H_SET_PARTITION_TABLE:
1182                 ret = H_FUNCTION;
1183                 if (nesting_enabled(kvm))
1184                         ret = kvmhv_set_partition_table(vcpu);
1185                 break;
1186         case H_ENTER_NESTED:
1187                 ret = H_FUNCTION;
1188                 if (!nesting_enabled(kvm))
1189                         break;
1190                 ret = kvmhv_enter_nested_guest(vcpu);
1191                 if (ret == H_INTERRUPT) {
1192                         kvmppc_set_gpr(vcpu, 3, 0);
1193                         vcpu->arch.hcall_needed = 0;
1194                         return -EINTR;
1195                 } else if (ret == H_TOO_HARD) {
1196                         kvmppc_set_gpr(vcpu, 3, 0);
1197                         vcpu->arch.hcall_needed = 0;
1198                         return RESUME_HOST;
1199                 }
1200                 break;
1201         case H_TLB_INVALIDATE:
1202                 ret = H_FUNCTION;
1203                 if (nesting_enabled(kvm))
1204                         ret = kvmhv_do_nested_tlbie(vcpu);
1205                 break;
1206         case H_COPY_TOFROM_GUEST:
1207                 ret = H_FUNCTION;
1208                 if (nesting_enabled(kvm))
1209                         ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1210                 break;
1211         case H_PAGE_INIT:
1212                 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1213                                          kvmppc_get_gpr(vcpu, 5),
1214                                          kvmppc_get_gpr(vcpu, 6));
1215                 break;
1216         case H_SVM_PAGE_IN:
1217                 ret = H_UNSUPPORTED;
1218                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1219                         ret = kvmppc_h_svm_page_in(kvm,
1220                                                    kvmppc_get_gpr(vcpu, 4),
1221                                                    kvmppc_get_gpr(vcpu, 5),
1222                                                    kvmppc_get_gpr(vcpu, 6));
1223                 break;
1224         case H_SVM_PAGE_OUT:
1225                 ret = H_UNSUPPORTED;
1226                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1227                         ret = kvmppc_h_svm_page_out(kvm,
1228                                                     kvmppc_get_gpr(vcpu, 4),
1229                                                     kvmppc_get_gpr(vcpu, 5),
1230                                                     kvmppc_get_gpr(vcpu, 6));
1231                 break;
1232         case H_SVM_INIT_START:
1233                 ret = H_UNSUPPORTED;
1234                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1235                         ret = kvmppc_h_svm_init_start(kvm);
1236                 break;
1237         case H_SVM_INIT_DONE:
1238                 ret = H_UNSUPPORTED;
1239                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1240                         ret = kvmppc_h_svm_init_done(kvm);
1241                 break;
1242         case H_SVM_INIT_ABORT:
1243                 /*
1244                  * Even if that call is made by the Ultravisor, the SSR1 value
1245                  * is the guest context one, with the secure bit clear as it has
1246                  * not yet been secured. So we can't check it here.
1247                  * Instead the kvm->arch.secure_guest flag is checked inside
1248                  * kvmppc_h_svm_init_abort().
1249                  */
1250                 ret = kvmppc_h_svm_init_abort(kvm);
1251                 break;
1252
1253         default:
1254                 return RESUME_HOST;
1255         }
1256         WARN_ON_ONCE(ret == H_TOO_HARD);
1257         kvmppc_set_gpr(vcpu, 3, ret);
1258         vcpu->arch.hcall_needed = 0;
1259         return RESUME_GUEST;
1260 }
1261
1262 /*
1263  * Handle H_CEDE in the P9 path where we don't call the real-mode hcall
1264  * handlers in book3s_hv_rmhandlers.S.
1265  *
1266  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1267  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1268  */
1269 static void kvmppc_cede(struct kvm_vcpu *vcpu)
1270 {
1271         vcpu->arch.shregs.msr |= MSR_EE;
1272         vcpu->arch.ceded = 1;
1273         smp_mb();
1274         if (vcpu->arch.prodded) {
1275                 vcpu->arch.prodded = 0;
1276                 smp_mb();
1277                 vcpu->arch.ceded = 0;
1278         }
1279 }
1280
1281 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1282 {
1283         switch (cmd) {
1284         case H_CEDE:
1285         case H_PROD:
1286         case H_CONFER:
1287         case H_REGISTER_VPA:
1288         case H_SET_MODE:
1289         case H_LOGICAL_CI_LOAD:
1290         case H_LOGICAL_CI_STORE:
1291 #ifdef CONFIG_KVM_XICS
1292         case H_XIRR:
1293         case H_CPPR:
1294         case H_EOI:
1295         case H_IPI:
1296         case H_IPOLL:
1297         case H_XIRR_X:
1298 #endif
1299         case H_PAGE_INIT:
1300         case H_RPT_INVALIDATE:
1301                 return 1;
1302         }
1303
1304         /* See if it's in the real-mode table */
1305         return kvmppc_hcall_impl_hv_realmode(cmd);
1306 }
1307
1308 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1309 {
1310         u32 last_inst;
1311
1312         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1313                                         EMULATE_DONE) {
1314                 /*
1315                  * Fetch failed, so return to guest and
1316                  * try executing it again.
1317                  */
1318                 return RESUME_GUEST;
1319         }
1320
1321         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1322                 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1323                 vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1324                 return RESUME_HOST;
1325         } else {
1326                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1327                 return RESUME_GUEST;
1328         }
1329 }
1330
1331 static void do_nothing(void *x)
1332 {
1333 }
1334
1335 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1336 {
1337         int thr, cpu, pcpu, nthreads;
1338         struct kvm_vcpu *v;
1339         unsigned long dpdes;
1340
1341         nthreads = vcpu->kvm->arch.emul_smt_mode;
1342         dpdes = 0;
1343         cpu = vcpu->vcpu_id & ~(nthreads - 1);
1344         for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1345                 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1346                 if (!v)
1347                         continue;
1348                 /*
1349                  * If the vcpu is currently running on a physical cpu thread,
1350                  * interrupt it in order to pull it out of the guest briefly,
1351                  * which will update its vcore->dpdes value.
1352                  */
1353                 pcpu = READ_ONCE(v->cpu);
1354                 if (pcpu >= 0)
1355                         smp_call_function_single(pcpu, do_nothing, NULL, 1);
1356                 if (kvmppc_doorbell_pending(v))
1357                         dpdes |= 1 << thr;
1358         }
1359         return dpdes;
1360 }
1361
1362 /*
1363  * On POWER9, emulate doorbell-related instructions in order to
1364  * give the guest the illusion of running on a multi-threaded core.
1365  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1366  * and mfspr DPDES.
1367  */
1368 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1369 {
1370         u32 inst, rb, thr;
1371         unsigned long arg;
1372         struct kvm *kvm = vcpu->kvm;
1373         struct kvm_vcpu *tvcpu;
1374
1375         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1376                 return RESUME_GUEST;
1377         if (get_op(inst) != 31)
1378                 return EMULATE_FAIL;
1379         rb = get_rb(inst);
1380         thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1381         switch (get_xop(inst)) {
1382         case OP_31_XOP_MSGSNDP:
1383                 arg = kvmppc_get_gpr(vcpu, rb);
1384                 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1385                         break;
1386                 arg &= 0x7f;
1387                 if (arg >= kvm->arch.emul_smt_mode)
1388                         break;
1389                 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1390                 if (!tvcpu)
1391                         break;
1392                 if (!tvcpu->arch.doorbell_request) {
1393                         tvcpu->arch.doorbell_request = 1;
1394                         kvmppc_fast_vcpu_kick_hv(tvcpu);
1395                 }
1396                 break;
1397         case OP_31_XOP_MSGCLRP:
1398                 arg = kvmppc_get_gpr(vcpu, rb);
1399                 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1400                         break;
1401                 vcpu->arch.vcore->dpdes = 0;
1402                 vcpu->arch.doorbell_request = 0;
1403                 break;
1404         case OP_31_XOP_MFSPR:
1405                 switch (get_sprn(inst)) {
1406                 case SPRN_TIR:
1407                         arg = thr;
1408                         break;
1409                 case SPRN_DPDES:
1410                         arg = kvmppc_read_dpdes(vcpu);
1411                         break;
1412                 default:
1413                         return EMULATE_FAIL;
1414                 }
1415                 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1416                 break;
1417         default:
1418                 return EMULATE_FAIL;
1419         }
1420         kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1421         return RESUME_GUEST;
1422 }
1423
1424 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1425                                  struct task_struct *tsk)
1426 {
1427         struct kvm_run *run = vcpu->run;
1428         int r = RESUME_HOST;
1429
1430         vcpu->stat.sum_exits++;
1431
1432         /*
1433          * This can happen if an interrupt occurs in the last stages
1434          * of guest entry or the first stages of guest exit (i.e. after
1435          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1436          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1437          * That can happen due to a bug, or due to a machine check
1438          * occurring at just the wrong time.
1439          */
1440         if (vcpu->arch.shregs.msr & MSR_HV) {
1441                 printk(KERN_EMERG "KVM trap in HV mode!\n");
1442                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1443                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1444                         vcpu->arch.shregs.msr);
1445                 kvmppc_dump_regs(vcpu);
1446                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1447                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1448                 return RESUME_HOST;
1449         }
1450         run->exit_reason = KVM_EXIT_UNKNOWN;
1451         run->ready_for_interrupt_injection = 1;
1452         switch (vcpu->arch.trap) {
1453         /* We're good on these - the host merely wanted to get our attention */
1454         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1455                 vcpu->stat.dec_exits++;
1456                 r = RESUME_GUEST;
1457                 break;
1458         case BOOK3S_INTERRUPT_EXTERNAL:
1459         case BOOK3S_INTERRUPT_H_DOORBELL:
1460         case BOOK3S_INTERRUPT_H_VIRT:
1461                 vcpu->stat.ext_intr_exits++;
1462                 r = RESUME_GUEST;
1463                 break;
1464         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1465         case BOOK3S_INTERRUPT_HMI:
1466         case BOOK3S_INTERRUPT_PERFMON:
1467         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1468                 r = RESUME_GUEST;
1469                 break;
1470         case BOOK3S_INTERRUPT_MACHINE_CHECK: {
1471                 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1472                                               DEFAULT_RATELIMIT_BURST);
1473                 /*
1474                  * Print the MCE event to host console. Ratelimit so the guest
1475                  * can't flood the host log.
1476                  */
1477                 if (__ratelimit(&rs))
1478                         machine_check_print_event_info(&vcpu->arch.mce_evt,false, true);
1479
1480                 /*
1481                  * If the guest can do FWNMI, exit to userspace so it can
1482                  * deliver a FWNMI to the guest.
1483                  * Otherwise we synthesize a machine check for the guest
1484                  * so that it knows that the machine check occurred.
1485                  */
1486                 if (!vcpu->kvm->arch.fwnmi_enabled) {
1487                         ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1488                         kvmppc_core_queue_machine_check(vcpu, flags);
1489                         r = RESUME_GUEST;
1490                         break;
1491                 }
1492
1493                 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1494                 run->exit_reason = KVM_EXIT_NMI;
1495                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1496                 /* Clear out the old NMI status from run->flags */
1497                 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1498                 /* Now set the NMI status */
1499                 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1500                         run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1501                 else
1502                         run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1503
1504                 r = RESUME_HOST;
1505                 break;
1506         }
1507         case BOOK3S_INTERRUPT_PROGRAM:
1508         {
1509                 ulong flags;
1510                 /*
1511                  * Normally program interrupts are delivered directly
1512                  * to the guest by the hardware, but we can get here
1513                  * as a result of a hypervisor emulation interrupt
1514                  * (e40) getting turned into a 700 by BML RTAS.
1515                  */
1516                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1517                 kvmppc_core_queue_program(vcpu, flags);
1518                 r = RESUME_GUEST;
1519                 break;
1520         }
1521         case BOOK3S_INTERRUPT_SYSCALL:
1522         {
1523                 int i;
1524
1525                 if (unlikely(vcpu->arch.shregs.msr & MSR_PR)) {
1526                         /*
1527                          * Guest userspace executed sc 1. This can only be
1528                          * reached by the P9 path because the old path
1529                          * handles this case in realmode hcall handlers.
1530                          */
1531                         if (!kvmhv_vcpu_is_radix(vcpu)) {
1532                                 /*
1533                                  * A guest could be running PR KVM, so this
1534                                  * may be a PR KVM hcall. It must be reflected
1535                                  * to the guest kernel as a sc interrupt.
1536                                  */
1537                                 kvmppc_core_queue_syscall(vcpu);
1538                         } else {
1539                                 /*
1540                                  * Radix guests can not run PR KVM or nested HV
1541                                  * hash guests which might run PR KVM, so this
1542                                  * is always a privilege fault. Send a program
1543                                  * check to guest kernel.
1544                                  */
1545                                 kvmppc_core_queue_program(vcpu, SRR1_PROGPRIV);
1546                         }
1547                         r = RESUME_GUEST;
1548                         break;
1549                 }
1550
1551                 /*
1552                  * hcall - gather args and set exit_reason. This will next be
1553                  * handled by kvmppc_pseries_do_hcall which may be able to deal
1554                  * with it and resume guest, or may punt to userspace.
1555                  */
1556                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1557                 for (i = 0; i < 9; ++i)
1558                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1559                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1560                 vcpu->arch.hcall_needed = 1;
1561                 r = RESUME_HOST;
1562                 break;
1563         }
1564         /*
1565          * We get these next two if the guest accesses a page which it thinks
1566          * it has mapped but which is not actually present, either because
1567          * it is for an emulated I/O device or because the corresonding
1568          * host page has been paged out.
1569          *
1570          * Any other HDSI/HISI interrupts have been handled already for P7/8
1571          * guests. For POWER9 hash guests not using rmhandlers, basic hash
1572          * fault handling is done here.
1573          */
1574         case BOOK3S_INTERRUPT_H_DATA_STORAGE: {
1575                 unsigned long vsid;
1576                 long err;
1577
1578                 if (vcpu->arch.fault_dsisr == HDSISR_CANARY) {
1579                         r = RESUME_GUEST; /* Just retry if it's the canary */
1580                         break;
1581                 }
1582
1583                 if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1584                         /*
1585                          * Radix doesn't require anything, and pre-ISAv3.0 hash
1586                          * already attempted to handle this in rmhandlers. The
1587                          * hash fault handling below is v3 only (it uses ASDR
1588                          * via fault_gpa).
1589                          */
1590                         r = RESUME_PAGE_FAULT;
1591                         break;
1592                 }
1593
1594                 if (!(vcpu->arch.fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT))) {
1595                         kvmppc_core_queue_data_storage(vcpu,
1596                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1597                         r = RESUME_GUEST;
1598                         break;
1599                 }
1600
1601                 if (!(vcpu->arch.shregs.msr & MSR_DR))
1602                         vsid = vcpu->kvm->arch.vrma_slb_v;
1603                 else
1604                         vsid = vcpu->arch.fault_gpa;
1605
1606                 err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1607                                 vsid, vcpu->arch.fault_dsisr, true);
1608                 if (err == 0) {
1609                         r = RESUME_GUEST;
1610                 } else if (err == -1 || err == -2) {
1611                         r = RESUME_PAGE_FAULT;
1612                 } else {
1613                         kvmppc_core_queue_data_storage(vcpu,
1614                                 vcpu->arch.fault_dar, err);
1615                         r = RESUME_GUEST;
1616                 }
1617                 break;
1618         }
1619         case BOOK3S_INTERRUPT_H_INST_STORAGE: {
1620                 unsigned long vsid;
1621                 long err;
1622
1623                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1624                 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1625                         DSISR_SRR1_MATCH_64S;
1626                 if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1627                         /*
1628                          * Radix doesn't require anything, and pre-ISAv3.0 hash
1629                          * already attempted to handle this in rmhandlers. The
1630                          * hash fault handling below is v3 only (it uses ASDR
1631                          * via fault_gpa).
1632                          */
1633                         if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1634                                 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1635                         r = RESUME_PAGE_FAULT;
1636                         break;
1637                 }
1638
1639                 if (!(vcpu->arch.fault_dsisr & SRR1_ISI_NOPT)) {
1640                         kvmppc_core_queue_inst_storage(vcpu,
1641                                 vcpu->arch.fault_dsisr);
1642                         r = RESUME_GUEST;
1643                         break;
1644                 }
1645
1646                 if (!(vcpu->arch.shregs.msr & MSR_IR))
1647                         vsid = vcpu->kvm->arch.vrma_slb_v;
1648                 else
1649                         vsid = vcpu->arch.fault_gpa;
1650
1651                 err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1652                                 vsid, vcpu->arch.fault_dsisr, false);
1653                 if (err == 0) {
1654                         r = RESUME_GUEST;
1655                 } else if (err == -1) {
1656                         r = RESUME_PAGE_FAULT;
1657                 } else {
1658                         kvmppc_core_queue_inst_storage(vcpu, err);
1659                         r = RESUME_GUEST;
1660                 }
1661                 break;
1662         }
1663
1664         /*
1665          * This occurs if the guest executes an illegal instruction.
1666          * If the guest debug is disabled, generate a program interrupt
1667          * to the guest. If guest debug is enabled, we need to check
1668          * whether the instruction is a software breakpoint instruction.
1669          * Accordingly return to Guest or Host.
1670          */
1671         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1672                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1673                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1674                                 swab32(vcpu->arch.emul_inst) :
1675                                 vcpu->arch.emul_inst;
1676                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1677                         r = kvmppc_emulate_debug_inst(vcpu);
1678                 } else {
1679                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1680                         r = RESUME_GUEST;
1681                 }
1682                 break;
1683
1684 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1685         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1686                 /*
1687                  * This occurs for various TM-related instructions that
1688                  * we need to emulate on POWER9 DD2.2.  We have already
1689                  * handled the cases where the guest was in real-suspend
1690                  * mode and was transitioning to transactional state.
1691                  */
1692                 r = kvmhv_p9_tm_emulation(vcpu);
1693                 if (r != -1)
1694                         break;
1695                 fallthrough; /* go to facility unavailable handler */
1696 #endif
1697
1698         /*
1699          * This occurs if the guest (kernel or userspace), does something that
1700          * is prohibited by HFSCR.
1701          * On POWER9, this could be a doorbell instruction that we need
1702          * to emulate.
1703          * Otherwise, we just generate a program interrupt to the guest.
1704          */
1705         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1706                 r = EMULATE_FAIL;
1707                 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1708                     cpu_has_feature(CPU_FTR_ARCH_300))
1709                         r = kvmppc_emulate_doorbell_instr(vcpu);
1710                 if (r == EMULATE_FAIL) {
1711                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1712                         r = RESUME_GUEST;
1713                 }
1714                 break;
1715
1716         case BOOK3S_INTERRUPT_HV_RM_HARD:
1717                 r = RESUME_PASSTHROUGH;
1718                 break;
1719         default:
1720                 kvmppc_dump_regs(vcpu);
1721                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1722                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1723                         vcpu->arch.shregs.msr);
1724                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1725                 r = RESUME_HOST;
1726                 break;
1727         }
1728
1729         return r;
1730 }
1731
1732 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1733 {
1734         int r;
1735         int srcu_idx;
1736
1737         vcpu->stat.sum_exits++;
1738
1739         /*
1740          * This can happen if an interrupt occurs in the last stages
1741          * of guest entry or the first stages of guest exit (i.e. after
1742          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1743          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1744          * That can happen due to a bug, or due to a machine check
1745          * occurring at just the wrong time.
1746          */
1747         if (vcpu->arch.shregs.msr & MSR_HV) {
1748                 pr_emerg("KVM trap in HV mode while nested!\n");
1749                 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1750                          vcpu->arch.trap, kvmppc_get_pc(vcpu),
1751                          vcpu->arch.shregs.msr);
1752                 kvmppc_dump_regs(vcpu);
1753                 return RESUME_HOST;
1754         }
1755         switch (vcpu->arch.trap) {
1756         /* We're good on these - the host merely wanted to get our attention */
1757         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1758                 vcpu->stat.dec_exits++;
1759                 r = RESUME_GUEST;
1760                 break;
1761         case BOOK3S_INTERRUPT_EXTERNAL:
1762                 vcpu->stat.ext_intr_exits++;
1763                 r = RESUME_HOST;
1764                 break;
1765         case BOOK3S_INTERRUPT_H_DOORBELL:
1766         case BOOK3S_INTERRUPT_H_VIRT:
1767                 vcpu->stat.ext_intr_exits++;
1768                 r = RESUME_GUEST;
1769                 break;
1770         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1771         case BOOK3S_INTERRUPT_HMI:
1772         case BOOK3S_INTERRUPT_PERFMON:
1773         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1774                 r = RESUME_GUEST;
1775                 break;
1776         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1777         {
1778                 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1779                                               DEFAULT_RATELIMIT_BURST);
1780                 /* Pass the machine check to the L1 guest */
1781                 r = RESUME_HOST;
1782                 /* Print the MCE event to host console. */
1783                 if (__ratelimit(&rs))
1784                         machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1785                 break;
1786         }
1787         /*
1788          * We get these next two if the guest accesses a page which it thinks
1789          * it has mapped but which is not actually present, either because
1790          * it is for an emulated I/O device or because the corresonding
1791          * host page has been paged out.
1792          */
1793         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1794                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1795                 r = kvmhv_nested_page_fault(vcpu);
1796                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1797                 break;
1798         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1799                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1800                 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1801                                          DSISR_SRR1_MATCH_64S;
1802                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1803                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1804                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1805                 r = kvmhv_nested_page_fault(vcpu);
1806                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1807                 break;
1808
1809 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1810         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1811                 /*
1812                  * This occurs for various TM-related instructions that
1813                  * we need to emulate on POWER9 DD2.2.  We have already
1814                  * handled the cases where the guest was in real-suspend
1815                  * mode and was transitioning to transactional state.
1816                  */
1817                 r = kvmhv_p9_tm_emulation(vcpu);
1818                 if (r != -1)
1819                         break;
1820                 fallthrough; /* go to facility unavailable handler */
1821 #endif
1822
1823         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: {
1824                 u64 cause = vcpu->arch.hfscr >> 56;
1825
1826                 /*
1827                  * Only pass HFU interrupts to the L1 if the facility is
1828                  * permitted but disabled by the L1's HFSCR, otherwise
1829                  * the interrupt does not make sense to the L1 so turn
1830                  * it into a HEAI.
1831                  */
1832                 if (!(vcpu->arch.hfscr_permitted & (1UL << cause)) ||
1833                                 (vcpu->arch.nested_hfscr & (1UL << cause))) {
1834                         vcpu->arch.trap = BOOK3S_INTERRUPT_H_EMUL_ASSIST;
1835
1836                         /*
1837                          * If the fetch failed, return to guest and
1838                          * try executing it again.
1839                          */
1840                         r = kvmppc_get_last_inst(vcpu, INST_GENERIC,
1841                                                  &vcpu->arch.emul_inst);
1842                         if (r != EMULATE_DONE)
1843                                 r = RESUME_GUEST;
1844                         else
1845                                 r = RESUME_HOST;
1846                 } else {
1847                         r = RESUME_HOST;
1848                 }
1849
1850                 break;
1851         }
1852
1853         case BOOK3S_INTERRUPT_HV_RM_HARD:
1854                 vcpu->arch.trap = 0;
1855                 r = RESUME_GUEST;
1856                 if (!xics_on_xive())
1857                         kvmppc_xics_rm_complete(vcpu, 0);
1858                 break;
1859         case BOOK3S_INTERRUPT_SYSCALL:
1860         {
1861                 unsigned long req = kvmppc_get_gpr(vcpu, 3);
1862
1863                 /*
1864                  * The H_RPT_INVALIDATE hcalls issued by nested
1865                  * guests for process-scoped invalidations when
1866                  * GTSE=0, are handled here in L0.
1867                  */
1868                 if (req == H_RPT_INVALIDATE) {
1869                         r = kvmppc_nested_h_rpt_invalidate(vcpu);
1870                         break;
1871                 }
1872
1873                 r = RESUME_HOST;
1874                 break;
1875         }
1876         default:
1877                 r = RESUME_HOST;
1878                 break;
1879         }
1880
1881         return r;
1882 }
1883
1884 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1885                                             struct kvm_sregs *sregs)
1886 {
1887         int i;
1888
1889         memset(sregs, 0, sizeof(struct kvm_sregs));
1890         sregs->pvr = vcpu->arch.pvr;
1891         for (i = 0; i < vcpu->arch.slb_max; i++) {
1892                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1893                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1894         }
1895
1896         return 0;
1897 }
1898
1899 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1900                                             struct kvm_sregs *sregs)
1901 {
1902         int i, j;
1903
1904         /* Only accept the same PVR as the host's, since we can't spoof it */
1905         if (sregs->pvr != vcpu->arch.pvr)
1906                 return -EINVAL;
1907
1908         j = 0;
1909         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1910                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1911                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1912                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1913                         ++j;
1914                 }
1915         }
1916         vcpu->arch.slb_max = j;
1917
1918         return 0;
1919 }
1920
1921 /*
1922  * Enforce limits on guest LPCR values based on hardware availability,
1923  * guest configuration, and possibly hypervisor support and security
1924  * concerns.
1925  */
1926 unsigned long kvmppc_filter_lpcr_hv(struct kvm *kvm, unsigned long lpcr)
1927 {
1928         /* LPCR_TC only applies to HPT guests */
1929         if (kvm_is_radix(kvm))
1930                 lpcr &= ~LPCR_TC;
1931
1932         /* On POWER8 and above, userspace can modify AIL */
1933         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
1934                 lpcr &= ~LPCR_AIL;
1935         if ((lpcr & LPCR_AIL) != LPCR_AIL_3)
1936                 lpcr &= ~LPCR_AIL; /* LPCR[AIL]=1/2 is disallowed */
1937         /*
1938          * On some POWER9s we force AIL off for radix guests to prevent
1939          * executing in MSR[HV]=1 mode with the MMU enabled and PIDR set to
1940          * guest, which can result in Q0 translations with LPID=0 PID=PIDR to
1941          * be cached, which the host TLB management does not expect.
1942          */
1943         if (kvm_is_radix(kvm) && cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
1944                 lpcr &= ~LPCR_AIL;
1945
1946         /*
1947          * On POWER9, allow userspace to enable large decrementer for the
1948          * guest, whether or not the host has it enabled.
1949          */
1950         if (!cpu_has_feature(CPU_FTR_ARCH_300))
1951                 lpcr &= ~LPCR_LD;
1952
1953         return lpcr;
1954 }
1955
1956 static void verify_lpcr(struct kvm *kvm, unsigned long lpcr)
1957 {
1958         if (lpcr != kvmppc_filter_lpcr_hv(kvm, lpcr)) {
1959                 WARN_ONCE(1, "lpcr 0x%lx differs from filtered 0x%lx\n",
1960                           lpcr, kvmppc_filter_lpcr_hv(kvm, lpcr));
1961         }
1962 }
1963
1964 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1965                 bool preserve_top32)
1966 {
1967         struct kvm *kvm = vcpu->kvm;
1968         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1969         u64 mask;
1970
1971         spin_lock(&vc->lock);
1972
1973         /*
1974          * Userspace can only modify
1975          * DPFD (default prefetch depth), ILE (interrupt little-endian),
1976          * TC (translation control), AIL (alternate interrupt location),
1977          * LD (large decrementer).
1978          * These are subject to restrictions from kvmppc_filter_lcpr_hv().
1979          */
1980         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD;
1981
1982         /* Broken 32-bit version of LPCR must not clear top bits */
1983         if (preserve_top32)
1984                 mask &= 0xFFFFFFFF;
1985
1986         new_lpcr = kvmppc_filter_lpcr_hv(kvm,
1987                         (vc->lpcr & ~mask) | (new_lpcr & mask));
1988
1989         /*
1990          * If ILE (interrupt little-endian) has changed, update the
1991          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1992          */
1993         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1994                 struct kvm_vcpu *vcpu;
1995                 int i;
1996
1997                 kvm_for_each_vcpu(i, vcpu, kvm) {
1998                         if (vcpu->arch.vcore != vc)
1999                                 continue;
2000                         if (new_lpcr & LPCR_ILE)
2001                                 vcpu->arch.intr_msr |= MSR_LE;
2002                         else
2003                                 vcpu->arch.intr_msr &= ~MSR_LE;
2004                 }
2005         }
2006
2007         vc->lpcr = new_lpcr;
2008
2009         spin_unlock(&vc->lock);
2010 }
2011
2012 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2013                                  union kvmppc_one_reg *val)
2014 {
2015         int r = 0;
2016         long int i;
2017
2018         switch (id) {
2019         case KVM_REG_PPC_DEBUG_INST:
2020                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
2021                 break;
2022         case KVM_REG_PPC_HIOR:
2023                 *val = get_reg_val(id, 0);
2024                 break;
2025         case KVM_REG_PPC_DABR:
2026                 *val = get_reg_val(id, vcpu->arch.dabr);
2027                 break;
2028         case KVM_REG_PPC_DABRX:
2029                 *val = get_reg_val(id, vcpu->arch.dabrx);
2030                 break;
2031         case KVM_REG_PPC_DSCR:
2032                 *val = get_reg_val(id, vcpu->arch.dscr);
2033                 break;
2034         case KVM_REG_PPC_PURR:
2035                 *val = get_reg_val(id, vcpu->arch.purr);
2036                 break;
2037         case KVM_REG_PPC_SPURR:
2038                 *val = get_reg_val(id, vcpu->arch.spurr);
2039                 break;
2040         case KVM_REG_PPC_AMR:
2041                 *val = get_reg_val(id, vcpu->arch.amr);
2042                 break;
2043         case KVM_REG_PPC_UAMOR:
2044                 *val = get_reg_val(id, vcpu->arch.uamor);
2045                 break;
2046         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2047                 i = id - KVM_REG_PPC_MMCR0;
2048                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
2049                 break;
2050         case KVM_REG_PPC_MMCR2:
2051                 *val = get_reg_val(id, vcpu->arch.mmcr[2]);
2052                 break;
2053         case KVM_REG_PPC_MMCRA:
2054                 *val = get_reg_val(id, vcpu->arch.mmcra);
2055                 break;
2056         case KVM_REG_PPC_MMCRS:
2057                 *val = get_reg_val(id, vcpu->arch.mmcrs);
2058                 break;
2059         case KVM_REG_PPC_MMCR3:
2060                 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
2061                 break;
2062         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2063                 i = id - KVM_REG_PPC_PMC1;
2064                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
2065                 break;
2066         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2067                 i = id - KVM_REG_PPC_SPMC1;
2068                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
2069                 break;
2070         case KVM_REG_PPC_SIAR:
2071                 *val = get_reg_val(id, vcpu->arch.siar);
2072                 break;
2073         case KVM_REG_PPC_SDAR:
2074                 *val = get_reg_val(id, vcpu->arch.sdar);
2075                 break;
2076         case KVM_REG_PPC_SIER:
2077                 *val = get_reg_val(id, vcpu->arch.sier[0]);
2078                 break;
2079         case KVM_REG_PPC_SIER2:
2080                 *val = get_reg_val(id, vcpu->arch.sier[1]);
2081                 break;
2082         case KVM_REG_PPC_SIER3:
2083                 *val = get_reg_val(id, vcpu->arch.sier[2]);
2084                 break;
2085         case KVM_REG_PPC_IAMR:
2086                 *val = get_reg_val(id, vcpu->arch.iamr);
2087                 break;
2088         case KVM_REG_PPC_PSPB:
2089                 *val = get_reg_val(id, vcpu->arch.pspb);
2090                 break;
2091         case KVM_REG_PPC_DPDES:
2092                 /*
2093                  * On POWER9, where we are emulating msgsndp etc.,
2094                  * we return 1 bit for each vcpu, which can come from
2095                  * either vcore->dpdes or doorbell_request.
2096                  * On POWER8, doorbell_request is 0.
2097                  */
2098                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes |
2099                                    vcpu->arch.doorbell_request);
2100                 break;
2101         case KVM_REG_PPC_VTB:
2102                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
2103                 break;
2104         case KVM_REG_PPC_DAWR:
2105                 *val = get_reg_val(id, vcpu->arch.dawr0);
2106                 break;
2107         case KVM_REG_PPC_DAWRX:
2108                 *val = get_reg_val(id, vcpu->arch.dawrx0);
2109                 break;
2110         case KVM_REG_PPC_DAWR1:
2111                 *val = get_reg_val(id, vcpu->arch.dawr1);
2112                 break;
2113         case KVM_REG_PPC_DAWRX1:
2114                 *val = get_reg_val(id, vcpu->arch.dawrx1);
2115                 break;
2116         case KVM_REG_PPC_CIABR:
2117                 *val = get_reg_val(id, vcpu->arch.ciabr);
2118                 break;
2119         case KVM_REG_PPC_CSIGR:
2120                 *val = get_reg_val(id, vcpu->arch.csigr);
2121                 break;
2122         case KVM_REG_PPC_TACR:
2123                 *val = get_reg_val(id, vcpu->arch.tacr);
2124                 break;
2125         case KVM_REG_PPC_TCSCR:
2126                 *val = get_reg_val(id, vcpu->arch.tcscr);
2127                 break;
2128         case KVM_REG_PPC_PID:
2129                 *val = get_reg_val(id, vcpu->arch.pid);
2130                 break;
2131         case KVM_REG_PPC_ACOP:
2132                 *val = get_reg_val(id, vcpu->arch.acop);
2133                 break;
2134         case KVM_REG_PPC_WORT:
2135                 *val = get_reg_val(id, vcpu->arch.wort);
2136                 break;
2137         case KVM_REG_PPC_TIDR:
2138                 *val = get_reg_val(id, vcpu->arch.tid);
2139                 break;
2140         case KVM_REG_PPC_PSSCR:
2141                 *val = get_reg_val(id, vcpu->arch.psscr);
2142                 break;
2143         case KVM_REG_PPC_VPA_ADDR:
2144                 spin_lock(&vcpu->arch.vpa_update_lock);
2145                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
2146                 spin_unlock(&vcpu->arch.vpa_update_lock);
2147                 break;
2148         case KVM_REG_PPC_VPA_SLB:
2149                 spin_lock(&vcpu->arch.vpa_update_lock);
2150                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
2151                 val->vpaval.length = vcpu->arch.slb_shadow.len;
2152                 spin_unlock(&vcpu->arch.vpa_update_lock);
2153                 break;
2154         case KVM_REG_PPC_VPA_DTL:
2155                 spin_lock(&vcpu->arch.vpa_update_lock);
2156                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
2157                 val->vpaval.length = vcpu->arch.dtl.len;
2158                 spin_unlock(&vcpu->arch.vpa_update_lock);
2159                 break;
2160         case KVM_REG_PPC_TB_OFFSET:
2161                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
2162                 break;
2163         case KVM_REG_PPC_LPCR:
2164         case KVM_REG_PPC_LPCR_64:
2165                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
2166                 break;
2167         case KVM_REG_PPC_PPR:
2168                 *val = get_reg_val(id, vcpu->arch.ppr);
2169                 break;
2170 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2171         case KVM_REG_PPC_TFHAR:
2172                 *val = get_reg_val(id, vcpu->arch.tfhar);
2173                 break;
2174         case KVM_REG_PPC_TFIAR:
2175                 *val = get_reg_val(id, vcpu->arch.tfiar);
2176                 break;
2177         case KVM_REG_PPC_TEXASR:
2178                 *val = get_reg_val(id, vcpu->arch.texasr);
2179                 break;
2180         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2181                 i = id - KVM_REG_PPC_TM_GPR0;
2182                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
2183                 break;
2184         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2185         {
2186                 int j;
2187                 i = id - KVM_REG_PPC_TM_VSR0;
2188                 if (i < 32)
2189                         for (j = 0; j < TS_FPRWIDTH; j++)
2190                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
2191                 else {
2192                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
2193                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
2194                         else
2195                                 r = -ENXIO;
2196                 }
2197                 break;
2198         }
2199         case KVM_REG_PPC_TM_CR:
2200                 *val = get_reg_val(id, vcpu->arch.cr_tm);
2201                 break;
2202         case KVM_REG_PPC_TM_XER:
2203                 *val = get_reg_val(id, vcpu->arch.xer_tm);
2204                 break;
2205         case KVM_REG_PPC_TM_LR:
2206                 *val = get_reg_val(id, vcpu->arch.lr_tm);
2207                 break;
2208         case KVM_REG_PPC_TM_CTR:
2209                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
2210                 break;
2211         case KVM_REG_PPC_TM_FPSCR:
2212                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
2213                 break;
2214         case KVM_REG_PPC_TM_AMR:
2215                 *val = get_reg_val(id, vcpu->arch.amr_tm);
2216                 break;
2217         case KVM_REG_PPC_TM_PPR:
2218                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
2219                 break;
2220         case KVM_REG_PPC_TM_VRSAVE:
2221                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
2222                 break;
2223         case KVM_REG_PPC_TM_VSCR:
2224                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2225                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
2226                 else
2227                         r = -ENXIO;
2228                 break;
2229         case KVM_REG_PPC_TM_DSCR:
2230                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
2231                 break;
2232         case KVM_REG_PPC_TM_TAR:
2233                 *val = get_reg_val(id, vcpu->arch.tar_tm);
2234                 break;
2235 #endif
2236         case KVM_REG_PPC_ARCH_COMPAT:
2237                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
2238                 break;
2239         case KVM_REG_PPC_DEC_EXPIRY:
2240                 *val = get_reg_val(id, vcpu->arch.dec_expires +
2241                                    vcpu->arch.vcore->tb_offset);
2242                 break;
2243         case KVM_REG_PPC_ONLINE:
2244                 *val = get_reg_val(id, vcpu->arch.online);
2245                 break;
2246         case KVM_REG_PPC_PTCR:
2247                 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
2248                 break;
2249         default:
2250                 r = -EINVAL;
2251                 break;
2252         }
2253
2254         return r;
2255 }
2256
2257 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2258                                  union kvmppc_one_reg *val)
2259 {
2260         int r = 0;
2261         long int i;
2262         unsigned long addr, len;
2263
2264         switch (id) {
2265         case KVM_REG_PPC_HIOR:
2266                 /* Only allow this to be set to zero */
2267                 if (set_reg_val(id, *val))
2268                         r = -EINVAL;
2269                 break;
2270         case KVM_REG_PPC_DABR:
2271                 vcpu->arch.dabr = set_reg_val(id, *val);
2272                 break;
2273         case KVM_REG_PPC_DABRX:
2274                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
2275                 break;
2276         case KVM_REG_PPC_DSCR:
2277                 vcpu->arch.dscr = set_reg_val(id, *val);
2278                 break;
2279         case KVM_REG_PPC_PURR:
2280                 vcpu->arch.purr = set_reg_val(id, *val);
2281                 break;
2282         case KVM_REG_PPC_SPURR:
2283                 vcpu->arch.spurr = set_reg_val(id, *val);
2284                 break;
2285         case KVM_REG_PPC_AMR:
2286                 vcpu->arch.amr = set_reg_val(id, *val);
2287                 break;
2288         case KVM_REG_PPC_UAMOR:
2289                 vcpu->arch.uamor = set_reg_val(id, *val);
2290                 break;
2291         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2292                 i = id - KVM_REG_PPC_MMCR0;
2293                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
2294                 break;
2295         case KVM_REG_PPC_MMCR2:
2296                 vcpu->arch.mmcr[2] = set_reg_val(id, *val);
2297                 break;
2298         case KVM_REG_PPC_MMCRA:
2299                 vcpu->arch.mmcra = set_reg_val(id, *val);
2300                 break;
2301         case KVM_REG_PPC_MMCRS:
2302                 vcpu->arch.mmcrs = set_reg_val(id, *val);
2303                 break;
2304         case KVM_REG_PPC_MMCR3:
2305                 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
2306                 break;
2307         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2308                 i = id - KVM_REG_PPC_PMC1;
2309                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
2310                 break;
2311         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2312                 i = id - KVM_REG_PPC_SPMC1;
2313                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
2314                 break;
2315         case KVM_REG_PPC_SIAR:
2316                 vcpu->arch.siar = set_reg_val(id, *val);
2317                 break;
2318         case KVM_REG_PPC_SDAR:
2319                 vcpu->arch.sdar = set_reg_val(id, *val);
2320                 break;
2321         case KVM_REG_PPC_SIER:
2322                 vcpu->arch.sier[0] = set_reg_val(id, *val);
2323                 break;
2324         case KVM_REG_PPC_SIER2:
2325                 vcpu->arch.sier[1] = set_reg_val(id, *val);
2326                 break;
2327         case KVM_REG_PPC_SIER3:
2328                 vcpu->arch.sier[2] = set_reg_val(id, *val);
2329                 break;
2330         case KVM_REG_PPC_IAMR:
2331                 vcpu->arch.iamr = set_reg_val(id, *val);
2332                 break;
2333         case KVM_REG_PPC_PSPB:
2334                 vcpu->arch.pspb = set_reg_val(id, *val);
2335                 break;
2336         case KVM_REG_PPC_DPDES:
2337                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
2338                 break;
2339         case KVM_REG_PPC_VTB:
2340                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
2341                 break;
2342         case KVM_REG_PPC_DAWR:
2343                 vcpu->arch.dawr0 = set_reg_val(id, *val);
2344                 break;
2345         case KVM_REG_PPC_DAWRX:
2346                 vcpu->arch.dawrx0 = set_reg_val(id, *val) & ~DAWRX_HYP;
2347                 break;
2348         case KVM_REG_PPC_DAWR1:
2349                 vcpu->arch.dawr1 = set_reg_val(id, *val);
2350                 break;
2351         case KVM_REG_PPC_DAWRX1:
2352                 vcpu->arch.dawrx1 = set_reg_val(id, *val) & ~DAWRX_HYP;
2353                 break;
2354         case KVM_REG_PPC_CIABR:
2355                 vcpu->arch.ciabr = set_reg_val(id, *val);
2356                 /* Don't allow setting breakpoints in hypervisor code */
2357                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
2358                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
2359                 break;
2360         case KVM_REG_PPC_CSIGR:
2361                 vcpu->arch.csigr = set_reg_val(id, *val);
2362                 break;
2363         case KVM_REG_PPC_TACR:
2364                 vcpu->arch.tacr = set_reg_val(id, *val);
2365                 break;
2366         case KVM_REG_PPC_TCSCR:
2367                 vcpu->arch.tcscr = set_reg_val(id, *val);
2368                 break;
2369         case KVM_REG_PPC_PID:
2370                 vcpu->arch.pid = set_reg_val(id, *val);
2371                 break;
2372         case KVM_REG_PPC_ACOP:
2373                 vcpu->arch.acop = set_reg_val(id, *val);
2374                 break;
2375         case KVM_REG_PPC_WORT:
2376                 vcpu->arch.wort = set_reg_val(id, *val);
2377                 break;
2378         case KVM_REG_PPC_TIDR:
2379                 vcpu->arch.tid = set_reg_val(id, *val);
2380                 break;
2381         case KVM_REG_PPC_PSSCR:
2382                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2383                 break;
2384         case KVM_REG_PPC_VPA_ADDR:
2385                 addr = set_reg_val(id, *val);
2386                 r = -EINVAL;
2387                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2388                               vcpu->arch.dtl.next_gpa))
2389                         break;
2390                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2391                 break;
2392         case KVM_REG_PPC_VPA_SLB:
2393                 addr = val->vpaval.addr;
2394                 len = val->vpaval.length;
2395                 r = -EINVAL;
2396                 if (addr && !vcpu->arch.vpa.next_gpa)
2397                         break;
2398                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2399                 break;
2400         case KVM_REG_PPC_VPA_DTL:
2401                 addr = val->vpaval.addr;
2402                 len = val->vpaval.length;
2403                 r = -EINVAL;
2404                 if (addr && (len < sizeof(struct dtl_entry) ||
2405                              !vcpu->arch.vpa.next_gpa))
2406                         break;
2407                 len -= len % sizeof(struct dtl_entry);
2408                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2409                 break;
2410         case KVM_REG_PPC_TB_OFFSET:
2411                 /* round up to multiple of 2^24 */
2412                 vcpu->arch.vcore->tb_offset =
2413                         ALIGN(set_reg_val(id, *val), 1UL << 24);
2414                 break;
2415         case KVM_REG_PPC_LPCR:
2416                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2417                 break;
2418         case KVM_REG_PPC_LPCR_64:
2419                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2420                 break;
2421         case KVM_REG_PPC_PPR:
2422                 vcpu->arch.ppr = set_reg_val(id, *val);
2423                 break;
2424 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2425         case KVM_REG_PPC_TFHAR:
2426                 vcpu->arch.tfhar = set_reg_val(id, *val);
2427                 break;
2428         case KVM_REG_PPC_TFIAR:
2429                 vcpu->arch.tfiar = set_reg_val(id, *val);
2430                 break;
2431         case KVM_REG_PPC_TEXASR:
2432                 vcpu->arch.texasr = set_reg_val(id, *val);
2433                 break;
2434         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2435                 i = id - KVM_REG_PPC_TM_GPR0;
2436                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2437                 break;
2438         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2439         {
2440                 int j;
2441                 i = id - KVM_REG_PPC_TM_VSR0;
2442                 if (i < 32)
2443                         for (j = 0; j < TS_FPRWIDTH; j++)
2444                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2445                 else
2446                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
2447                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2448                         else
2449                                 r = -ENXIO;
2450                 break;
2451         }
2452         case KVM_REG_PPC_TM_CR:
2453                 vcpu->arch.cr_tm = set_reg_val(id, *val);
2454                 break;
2455         case KVM_REG_PPC_TM_XER:
2456                 vcpu->arch.xer_tm = set_reg_val(id, *val);
2457                 break;
2458         case KVM_REG_PPC_TM_LR:
2459                 vcpu->arch.lr_tm = set_reg_val(id, *val);
2460                 break;
2461         case KVM_REG_PPC_TM_CTR:
2462                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2463                 break;
2464         case KVM_REG_PPC_TM_FPSCR:
2465                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2466                 break;
2467         case KVM_REG_PPC_TM_AMR:
2468                 vcpu->arch.amr_tm = set_reg_val(id, *val);
2469                 break;
2470         case KVM_REG_PPC_TM_PPR:
2471                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2472                 break;
2473         case KVM_REG_PPC_TM_VRSAVE:
2474                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2475                 break;
2476         case KVM_REG_PPC_TM_VSCR:
2477                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2478                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2479                 else
2480                         r = - ENXIO;
2481                 break;
2482         case KVM_REG_PPC_TM_DSCR:
2483                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2484                 break;
2485         case KVM_REG_PPC_TM_TAR:
2486                 vcpu->arch.tar_tm = set_reg_val(id, *val);
2487                 break;
2488 #endif
2489         case KVM_REG_PPC_ARCH_COMPAT:
2490                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2491                 break;
2492         case KVM_REG_PPC_DEC_EXPIRY:
2493                 vcpu->arch.dec_expires = set_reg_val(id, *val) -
2494                         vcpu->arch.vcore->tb_offset;
2495                 break;
2496         case KVM_REG_PPC_ONLINE:
2497                 i = set_reg_val(id, *val);
2498                 if (i && !vcpu->arch.online)
2499                         atomic_inc(&vcpu->arch.vcore->online_count);
2500                 else if (!i && vcpu->arch.online)
2501                         atomic_dec(&vcpu->arch.vcore->online_count);
2502                 vcpu->arch.online = i;
2503                 break;
2504         case KVM_REG_PPC_PTCR:
2505                 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2506                 break;
2507         default:
2508                 r = -EINVAL;
2509                 break;
2510         }
2511
2512         return r;
2513 }
2514
2515 /*
2516  * On POWER9, threads are independent and can be in different partitions.
2517  * Therefore we consider each thread to be a subcore.
2518  * There is a restriction that all threads have to be in the same
2519  * MMU mode (radix or HPT), unfortunately, but since we only support
2520  * HPT guests on a HPT host so far, that isn't an impediment yet.
2521  */
2522 static int threads_per_vcore(struct kvm *kvm)
2523 {
2524         if (cpu_has_feature(CPU_FTR_ARCH_300))
2525                 return 1;
2526         return threads_per_subcore;
2527 }
2528
2529 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2530 {
2531         struct kvmppc_vcore *vcore;
2532
2533         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2534
2535         if (vcore == NULL)
2536                 return NULL;
2537
2538         spin_lock_init(&vcore->lock);
2539         spin_lock_init(&vcore->stoltb_lock);
2540         rcuwait_init(&vcore->wait);
2541         vcore->preempt_tb = TB_NIL;
2542         vcore->lpcr = kvm->arch.lpcr;
2543         vcore->first_vcpuid = id;
2544         vcore->kvm = kvm;
2545         INIT_LIST_HEAD(&vcore->preempt_list);
2546
2547         return vcore;
2548 }
2549
2550 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2551 static struct debugfs_timings_element {
2552         const char *name;
2553         size_t offset;
2554 } timings[] = {
2555         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
2556         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
2557         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
2558         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
2559         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
2560 };
2561
2562 #define N_TIMINGS       (ARRAY_SIZE(timings))
2563
2564 struct debugfs_timings_state {
2565         struct kvm_vcpu *vcpu;
2566         unsigned int    buflen;
2567         char            buf[N_TIMINGS * 100];
2568 };
2569
2570 static int debugfs_timings_open(struct inode *inode, struct file *file)
2571 {
2572         struct kvm_vcpu *vcpu = inode->i_private;
2573         struct debugfs_timings_state *p;
2574
2575         p = kzalloc(sizeof(*p), GFP_KERNEL);
2576         if (!p)
2577                 return -ENOMEM;
2578
2579         kvm_get_kvm(vcpu->kvm);
2580         p->vcpu = vcpu;
2581         file->private_data = p;
2582
2583         return nonseekable_open(inode, file);
2584 }
2585
2586 static int debugfs_timings_release(struct inode *inode, struct file *file)
2587 {
2588         struct debugfs_timings_state *p = file->private_data;
2589
2590         kvm_put_kvm(p->vcpu->kvm);
2591         kfree(p);
2592         return 0;
2593 }
2594
2595 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2596                                     size_t len, loff_t *ppos)
2597 {
2598         struct debugfs_timings_state *p = file->private_data;
2599         struct kvm_vcpu *vcpu = p->vcpu;
2600         char *s, *buf_end;
2601         struct kvmhv_tb_accumulator tb;
2602         u64 count;
2603         loff_t pos;
2604         ssize_t n;
2605         int i, loops;
2606         bool ok;
2607
2608         if (!p->buflen) {
2609                 s = p->buf;
2610                 buf_end = s + sizeof(p->buf);
2611                 for (i = 0; i < N_TIMINGS; ++i) {
2612                         struct kvmhv_tb_accumulator *acc;
2613
2614                         acc = (struct kvmhv_tb_accumulator *)
2615                                 ((unsigned long)vcpu + timings[i].offset);
2616                         ok = false;
2617                         for (loops = 0; loops < 1000; ++loops) {
2618                                 count = acc->seqcount;
2619                                 if (!(count & 1)) {
2620                                         smp_rmb();
2621                                         tb = *acc;
2622                                         smp_rmb();
2623                                         if (count == acc->seqcount) {
2624                                                 ok = true;
2625                                                 break;
2626                                         }
2627                                 }
2628                                 udelay(1);
2629                         }
2630                         if (!ok)
2631                                 snprintf(s, buf_end - s, "%s: stuck\n",
2632                                         timings[i].name);
2633                         else
2634                                 snprintf(s, buf_end - s,
2635                                         "%s: %llu %llu %llu %llu\n",
2636                                         timings[i].name, count / 2,
2637                                         tb_to_ns(tb.tb_total),
2638                                         tb_to_ns(tb.tb_min),
2639                                         tb_to_ns(tb.tb_max));
2640                         s += strlen(s);
2641                 }
2642                 p->buflen = s - p->buf;
2643         }
2644
2645         pos = *ppos;
2646         if (pos >= p->buflen)
2647                 return 0;
2648         if (len > p->buflen - pos)
2649                 len = p->buflen - pos;
2650         n = copy_to_user(buf, p->buf + pos, len);
2651         if (n) {
2652                 if (n == len)
2653                         return -EFAULT;
2654                 len -= n;
2655         }
2656         *ppos = pos + len;
2657         return len;
2658 }
2659
2660 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2661                                      size_t len, loff_t *ppos)
2662 {
2663         return -EACCES;
2664 }
2665
2666 static const struct file_operations debugfs_timings_ops = {
2667         .owner   = THIS_MODULE,
2668         .open    = debugfs_timings_open,
2669         .release = debugfs_timings_release,
2670         .read    = debugfs_timings_read,
2671         .write   = debugfs_timings_write,
2672         .llseek  = generic_file_llseek,
2673 };
2674
2675 /* Create a debugfs directory for the vcpu */
2676 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2677 {
2678         char buf[16];
2679         struct kvm *kvm = vcpu->kvm;
2680
2681         snprintf(buf, sizeof(buf), "vcpu%u", id);
2682         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2683         debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir, vcpu,
2684                             &debugfs_timings_ops);
2685 }
2686
2687 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2688 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2689 {
2690 }
2691 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2692
2693 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2694 {
2695         int err;
2696         int core;
2697         struct kvmppc_vcore *vcore;
2698         struct kvm *kvm;
2699         unsigned int id;
2700
2701         kvm = vcpu->kvm;
2702         id = vcpu->vcpu_id;
2703
2704         vcpu->arch.shared = &vcpu->arch.shregs;
2705 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2706         /*
2707          * The shared struct is never shared on HV,
2708          * so we can always use host endianness
2709          */
2710 #ifdef __BIG_ENDIAN__
2711         vcpu->arch.shared_big_endian = true;
2712 #else
2713         vcpu->arch.shared_big_endian = false;
2714 #endif
2715 #endif
2716         vcpu->arch.mmcr[0] = MMCR0_FC;
2717         vcpu->arch.ctrl = CTRL_RUNLATCH;
2718         /* default to host PVR, since we can't spoof it */
2719         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2720         spin_lock_init(&vcpu->arch.vpa_update_lock);
2721         spin_lock_init(&vcpu->arch.tbacct_lock);
2722         vcpu->arch.busy_preempt = TB_NIL;
2723         vcpu->arch.shregs.msr = MSR_ME;
2724         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2725
2726         /*
2727          * Set the default HFSCR for the guest from the host value.
2728          * This value is only used on POWER9.
2729          * On POWER9, we want to virtualize the doorbell facility, so we
2730          * don't set the HFSCR_MSGP bit, and that causes those instructions
2731          * to trap and then we emulate them.
2732          */
2733         vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2734                 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP | HFSCR_PREFIX;
2735         if (cpu_has_feature(CPU_FTR_HVMODE)) {
2736                 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2737 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2738                 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2739                         vcpu->arch.hfscr |= HFSCR_TM;
2740 #endif
2741         }
2742         if (cpu_has_feature(CPU_FTR_TM_COMP))
2743                 vcpu->arch.hfscr |= HFSCR_TM;
2744
2745         vcpu->arch.hfscr_permitted = vcpu->arch.hfscr;
2746
2747         kvmppc_mmu_book3s_hv_init(vcpu);
2748
2749         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2750
2751         init_waitqueue_head(&vcpu->arch.cpu_run);
2752
2753         mutex_lock(&kvm->lock);
2754         vcore = NULL;
2755         err = -EINVAL;
2756         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2757                 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2758                         pr_devel("KVM: VCPU ID too high\n");
2759                         core = KVM_MAX_VCORES;
2760                 } else {
2761                         BUG_ON(kvm->arch.smt_mode != 1);
2762                         core = kvmppc_pack_vcpu_id(kvm, id);
2763                 }
2764         } else {
2765                 core = id / kvm->arch.smt_mode;
2766         }
2767         if (core < KVM_MAX_VCORES) {
2768                 vcore = kvm->arch.vcores[core];
2769                 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2770                         pr_devel("KVM: collision on id %u", id);
2771                         vcore = NULL;
2772                 } else if (!vcore) {
2773                         /*
2774                          * Take mmu_setup_lock for mutual exclusion
2775                          * with kvmppc_update_lpcr().
2776                          */
2777                         err = -ENOMEM;
2778                         vcore = kvmppc_vcore_create(kvm,
2779                                         id & ~(kvm->arch.smt_mode - 1));
2780                         mutex_lock(&kvm->arch.mmu_setup_lock);
2781                         kvm->arch.vcores[core] = vcore;
2782                         kvm->arch.online_vcores++;
2783                         mutex_unlock(&kvm->arch.mmu_setup_lock);
2784                 }
2785         }
2786         mutex_unlock(&kvm->lock);
2787
2788         if (!vcore)
2789                 return err;
2790
2791         spin_lock(&vcore->lock);
2792         ++vcore->num_threads;
2793         spin_unlock(&vcore->lock);
2794         vcpu->arch.vcore = vcore;
2795         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2796         vcpu->arch.thread_cpu = -1;
2797         vcpu->arch.prev_cpu = -1;
2798
2799         vcpu->arch.cpu_type = KVM_CPU_3S_64;
2800         kvmppc_sanity_check(vcpu);
2801
2802         debugfs_vcpu_init(vcpu, id);
2803
2804         return 0;
2805 }
2806
2807 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2808                               unsigned long flags)
2809 {
2810         int err;
2811         int esmt = 0;
2812
2813         if (flags)
2814                 return -EINVAL;
2815         if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2816                 return -EINVAL;
2817         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2818                 /*
2819                  * On POWER8 (or POWER7), the threading mode is "strict",
2820                  * so we pack smt_mode vcpus per vcore.
2821                  */
2822                 if (smt_mode > threads_per_subcore)
2823                         return -EINVAL;
2824         } else {
2825                 /*
2826                  * On POWER9, the threading mode is "loose",
2827                  * so each vcpu gets its own vcore.
2828                  */
2829                 esmt = smt_mode;
2830                 smt_mode = 1;
2831         }
2832         mutex_lock(&kvm->lock);
2833         err = -EBUSY;
2834         if (!kvm->arch.online_vcores) {
2835                 kvm->arch.smt_mode = smt_mode;
2836                 kvm->arch.emul_smt_mode = esmt;
2837                 err = 0;
2838         }
2839         mutex_unlock(&kvm->lock);
2840
2841         return err;
2842 }
2843
2844 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2845 {
2846         if (vpa->pinned_addr)
2847                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2848                                         vpa->dirty);
2849 }
2850
2851 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2852 {
2853         spin_lock(&vcpu->arch.vpa_update_lock);
2854         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2855         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2856         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2857         spin_unlock(&vcpu->arch.vpa_update_lock);
2858 }
2859
2860 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2861 {
2862         /* Indicate we want to get back into the guest */
2863         return 1;
2864 }
2865
2866 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2867 {
2868         unsigned long dec_nsec, now;
2869
2870         now = get_tb();
2871         if (now > vcpu->arch.dec_expires) {
2872                 /* decrementer has already gone negative */
2873                 kvmppc_core_queue_dec(vcpu);
2874                 kvmppc_core_prepare_to_enter(vcpu);
2875                 return;
2876         }
2877         dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2878         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2879         vcpu->arch.timer_running = 1;
2880 }
2881
2882 extern int __kvmppc_vcore_entry(void);
2883
2884 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2885                                    struct kvm_vcpu *vcpu)
2886 {
2887         u64 now;
2888
2889         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2890                 return;
2891         spin_lock_irq(&vcpu->arch.tbacct_lock);
2892         now = mftb();
2893         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2894                 vcpu->arch.stolen_logged;
2895         vcpu->arch.busy_preempt = now;
2896         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2897         spin_unlock_irq(&vcpu->arch.tbacct_lock);
2898         --vc->n_runnable;
2899         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2900 }
2901
2902 static int kvmppc_grab_hwthread(int cpu)
2903 {
2904         struct paca_struct *tpaca;
2905         long timeout = 10000;
2906
2907         tpaca = paca_ptrs[cpu];
2908
2909         /* Ensure the thread won't go into the kernel if it wakes */
2910         tpaca->kvm_hstate.kvm_vcpu = NULL;
2911         tpaca->kvm_hstate.kvm_vcore = NULL;
2912         tpaca->kvm_hstate.napping = 0;
2913         smp_wmb();
2914         tpaca->kvm_hstate.hwthread_req = 1;
2915
2916         /*
2917          * If the thread is already executing in the kernel (e.g. handling
2918          * a stray interrupt), wait for it to get back to nap mode.
2919          * The smp_mb() is to ensure that our setting of hwthread_req
2920          * is visible before we look at hwthread_state, so if this
2921          * races with the code at system_reset_pSeries and the thread
2922          * misses our setting of hwthread_req, we are sure to see its
2923          * setting of hwthread_state, and vice versa.
2924          */
2925         smp_mb();
2926         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2927                 if (--timeout <= 0) {
2928                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
2929                         return -EBUSY;
2930                 }
2931                 udelay(1);
2932         }
2933         return 0;
2934 }
2935
2936 static void kvmppc_release_hwthread(int cpu)
2937 {
2938         struct paca_struct *tpaca;
2939
2940         tpaca = paca_ptrs[cpu];
2941         tpaca->kvm_hstate.hwthread_req = 0;
2942         tpaca->kvm_hstate.kvm_vcpu = NULL;
2943         tpaca->kvm_hstate.kvm_vcore = NULL;
2944         tpaca->kvm_hstate.kvm_split_mode = NULL;
2945 }
2946
2947 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2948 {
2949         struct kvm_nested_guest *nested = vcpu->arch.nested;
2950         cpumask_t *cpu_in_guest;
2951         int i;
2952
2953         cpu = cpu_first_tlb_thread_sibling(cpu);
2954         if (nested) {
2955                 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2956                 cpu_in_guest = &nested->cpu_in_guest;
2957         } else {
2958                 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2959                 cpu_in_guest = &kvm->arch.cpu_in_guest;
2960         }
2961         /*
2962          * Make sure setting of bit in need_tlb_flush precedes
2963          * testing of cpu_in_guest bits.  The matching barrier on
2964          * the other side is the first smp_mb() in kvmppc_run_core().
2965          */
2966         smp_mb();
2967         for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
2968                                         i += cpu_tlb_thread_sibling_step())
2969                 if (cpumask_test_cpu(i, cpu_in_guest))
2970                         smp_call_function_single(i, do_nothing, NULL, 1);
2971 }
2972
2973 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2974 {
2975         struct kvm_nested_guest *nested = vcpu->arch.nested;
2976         struct kvm *kvm = vcpu->kvm;
2977         int prev_cpu;
2978
2979         if (!cpu_has_feature(CPU_FTR_HVMODE))
2980                 return;
2981
2982         if (nested)
2983                 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2984         else
2985                 prev_cpu = vcpu->arch.prev_cpu;
2986
2987         /*
2988          * With radix, the guest can do TLB invalidations itself,
2989          * and it could choose to use the local form (tlbiel) if
2990          * it is invalidating a translation that has only ever been
2991          * used on one vcpu.  However, that doesn't mean it has
2992          * only ever been used on one physical cpu, since vcpus
2993          * can move around between pcpus.  To cope with this, when
2994          * a vcpu moves from one pcpu to another, we need to tell
2995          * any vcpus running on the same core as this vcpu previously
2996          * ran to flush the TLB.  The TLB is shared between threads,
2997          * so we use a single bit in .need_tlb_flush for all 4 threads.
2998          */
2999         if (prev_cpu != pcpu) {
3000                 if (prev_cpu >= 0 &&
3001                     cpu_first_tlb_thread_sibling(prev_cpu) !=
3002                     cpu_first_tlb_thread_sibling(pcpu))
3003                         radix_flush_cpu(kvm, prev_cpu, vcpu);
3004                 if (nested)
3005                         nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
3006                 else
3007                         vcpu->arch.prev_cpu = pcpu;
3008         }
3009 }
3010
3011 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
3012 {
3013         int cpu;
3014         struct paca_struct *tpaca;
3015         struct kvm *kvm = vc->kvm;
3016
3017         cpu = vc->pcpu;
3018         if (vcpu) {
3019                 if (vcpu->arch.timer_running) {
3020                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
3021                         vcpu->arch.timer_running = 0;
3022                 }
3023                 cpu += vcpu->arch.ptid;
3024                 vcpu->cpu = vc->pcpu;
3025                 vcpu->arch.thread_cpu = cpu;
3026                 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
3027         }
3028         tpaca = paca_ptrs[cpu];
3029         tpaca->kvm_hstate.kvm_vcpu = vcpu;
3030         tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
3031         tpaca->kvm_hstate.fake_suspend = 0;
3032         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
3033         smp_wmb();
3034         tpaca->kvm_hstate.kvm_vcore = vc;
3035         if (cpu != smp_processor_id())
3036                 kvmppc_ipi_thread(cpu);
3037 }
3038
3039 static void kvmppc_wait_for_nap(int n_threads)
3040 {
3041         int cpu = smp_processor_id();
3042         int i, loops;
3043
3044         if (n_threads <= 1)
3045                 return;
3046         for (loops = 0; loops < 1000000; ++loops) {
3047                 /*
3048                  * Check if all threads are finished.
3049                  * We set the vcore pointer when starting a thread
3050                  * and the thread clears it when finished, so we look
3051                  * for any threads that still have a non-NULL vcore ptr.
3052                  */
3053                 for (i = 1; i < n_threads; ++i)
3054                         if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3055                                 break;
3056                 if (i == n_threads) {
3057                         HMT_medium();
3058                         return;
3059                 }
3060                 HMT_low();
3061         }
3062         HMT_medium();
3063         for (i = 1; i < n_threads; ++i)
3064                 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3065                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
3066 }
3067
3068 /*
3069  * Check that we are on thread 0 and that any other threads in
3070  * this core are off-line.  Then grab the threads so they can't
3071  * enter the kernel.
3072  */
3073 static int on_primary_thread(void)
3074 {
3075         int cpu = smp_processor_id();
3076         int thr;
3077
3078         /* Are we on a primary subcore? */
3079         if (cpu_thread_in_subcore(cpu))
3080                 return 0;
3081
3082         thr = 0;
3083         while (++thr < threads_per_subcore)
3084                 if (cpu_online(cpu + thr))
3085                         return 0;
3086
3087         /* Grab all hw threads so they can't go into the kernel */
3088         for (thr = 1; thr < threads_per_subcore; ++thr) {
3089                 if (kvmppc_grab_hwthread(cpu + thr)) {
3090                         /* Couldn't grab one; let the others go */
3091                         do {
3092                                 kvmppc_release_hwthread(cpu + thr);
3093                         } while (--thr > 0);
3094                         return 0;
3095                 }
3096         }
3097         return 1;
3098 }
3099
3100 /*
3101  * A list of virtual cores for each physical CPU.
3102  * These are vcores that could run but their runner VCPU tasks are
3103  * (or may be) preempted.
3104  */
3105 struct preempted_vcore_list {
3106         struct list_head        list;
3107         spinlock_t              lock;
3108 };
3109
3110 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
3111
3112 static void init_vcore_lists(void)
3113 {
3114         int cpu;
3115
3116         for_each_possible_cpu(cpu) {
3117                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
3118                 spin_lock_init(&lp->lock);
3119                 INIT_LIST_HEAD(&lp->list);
3120         }
3121 }
3122
3123 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
3124 {
3125         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3126
3127         vc->vcore_state = VCORE_PREEMPT;
3128         vc->pcpu = smp_processor_id();
3129         if (vc->num_threads < threads_per_vcore(vc->kvm)) {
3130                 spin_lock(&lp->lock);
3131                 list_add_tail(&vc->preempt_list, &lp->list);
3132                 spin_unlock(&lp->lock);
3133         }
3134
3135         /* Start accumulating stolen time */
3136         kvmppc_core_start_stolen(vc);
3137 }
3138
3139 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
3140 {
3141         struct preempted_vcore_list *lp;
3142
3143         kvmppc_core_end_stolen(vc);
3144         if (!list_empty(&vc->preempt_list)) {
3145                 lp = &per_cpu(preempted_vcores, vc->pcpu);
3146                 spin_lock(&lp->lock);
3147                 list_del_init(&vc->preempt_list);
3148                 spin_unlock(&lp->lock);
3149         }
3150         vc->vcore_state = VCORE_INACTIVE;
3151 }
3152
3153 /*
3154  * This stores information about the virtual cores currently
3155  * assigned to a physical core.
3156  */
3157 struct core_info {
3158         int             n_subcores;
3159         int             max_subcore_threads;
3160         int             total_threads;
3161         int             subcore_threads[MAX_SUBCORES];
3162         struct kvmppc_vcore *vc[MAX_SUBCORES];
3163 };
3164
3165 /*
3166  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
3167  * respectively in 2-way micro-threading (split-core) mode on POWER8.
3168  */
3169 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
3170
3171 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
3172 {
3173         memset(cip, 0, sizeof(*cip));
3174         cip->n_subcores = 1;
3175         cip->max_subcore_threads = vc->num_threads;
3176         cip->total_threads = vc->num_threads;
3177         cip->subcore_threads[0] = vc->num_threads;
3178         cip->vc[0] = vc;
3179 }
3180
3181 static bool subcore_config_ok(int n_subcores, int n_threads)
3182 {
3183         /*
3184          * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
3185          * split-core mode, with one thread per subcore.
3186          */
3187         if (cpu_has_feature(CPU_FTR_ARCH_300))
3188                 return n_subcores <= 4 && n_threads == 1;
3189
3190         /* On POWER8, can only dynamically split if unsplit to begin with */
3191         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
3192                 return false;
3193         if (n_subcores > MAX_SUBCORES)
3194                 return false;
3195         if (n_subcores > 1) {
3196                 if (!(dynamic_mt_modes & 2))
3197                         n_subcores = 4;
3198                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
3199                         return false;
3200         }
3201
3202         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
3203 }
3204
3205 static void init_vcore_to_run(struct kvmppc_vcore *vc)
3206 {
3207         vc->entry_exit_map = 0;
3208         vc->in_guest = 0;
3209         vc->napping_threads = 0;
3210         vc->conferring_threads = 0;
3211         vc->tb_offset_applied = 0;
3212 }
3213
3214 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
3215 {
3216         int n_threads = vc->num_threads;
3217         int sub;
3218
3219         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
3220                 return false;
3221
3222         /* In one_vm_per_core mode, require all vcores to be from the same vm */
3223         if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
3224                 return false;
3225
3226         if (n_threads < cip->max_subcore_threads)
3227                 n_threads = cip->max_subcore_threads;
3228         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
3229                 return false;
3230         cip->max_subcore_threads = n_threads;
3231
3232         sub = cip->n_subcores;
3233         ++cip->n_subcores;
3234         cip->total_threads += vc->num_threads;
3235         cip->subcore_threads[sub] = vc->num_threads;
3236         cip->vc[sub] = vc;
3237         init_vcore_to_run(vc);
3238         list_del_init(&vc->preempt_list);
3239
3240         return true;
3241 }
3242
3243 /*
3244  * Work out whether it is possible to piggyback the execution of
3245  * vcore *pvc onto the execution of the other vcores described in *cip.
3246  */
3247 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
3248                           int target_threads)
3249 {
3250         if (cip->total_threads + pvc->num_threads > target_threads)
3251                 return false;
3252
3253         return can_dynamic_split(pvc, cip);
3254 }
3255
3256 static void prepare_threads(struct kvmppc_vcore *vc)
3257 {
3258         int i;
3259         struct kvm_vcpu *vcpu;
3260
3261         for_each_runnable_thread(i, vcpu, vc) {
3262                 if (signal_pending(vcpu->arch.run_task))
3263                         vcpu->arch.ret = -EINTR;
3264                 else if (vcpu->arch.vpa.update_pending ||
3265                          vcpu->arch.slb_shadow.update_pending ||
3266                          vcpu->arch.dtl.update_pending)
3267                         vcpu->arch.ret = RESUME_GUEST;
3268                 else
3269                         continue;
3270                 kvmppc_remove_runnable(vc, vcpu);
3271                 wake_up(&vcpu->arch.cpu_run);
3272         }
3273 }
3274
3275 static void collect_piggybacks(struct core_info *cip, int target_threads)
3276 {
3277         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3278         struct kvmppc_vcore *pvc, *vcnext;
3279
3280         spin_lock(&lp->lock);
3281         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
3282                 if (!spin_trylock(&pvc->lock))
3283                         continue;
3284                 prepare_threads(pvc);
3285                 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
3286                         list_del_init(&pvc->preempt_list);
3287                         if (pvc->runner == NULL) {
3288                                 pvc->vcore_state = VCORE_INACTIVE;
3289                                 kvmppc_core_end_stolen(pvc);
3290                         }
3291                         spin_unlock(&pvc->lock);
3292                         continue;
3293                 }
3294                 if (!can_piggyback(pvc, cip, target_threads)) {
3295                         spin_unlock(&pvc->lock);
3296                         continue;
3297                 }
3298                 kvmppc_core_end_stolen(pvc);
3299                 pvc->vcore_state = VCORE_PIGGYBACK;
3300                 if (cip->total_threads >= target_threads)
3301                         break;
3302         }
3303         spin_unlock(&lp->lock);
3304 }
3305
3306 static bool recheck_signals_and_mmu(struct core_info *cip)
3307 {
3308         int sub, i;
3309         struct kvm_vcpu *vcpu;
3310         struct kvmppc_vcore *vc;
3311
3312         for (sub = 0; sub < cip->n_subcores; ++sub) {
3313                 vc = cip->vc[sub];
3314                 if (!vc->kvm->arch.mmu_ready)
3315                         return true;
3316                 for_each_runnable_thread(i, vcpu, vc)
3317                         if (signal_pending(vcpu->arch.run_task))
3318                                 return true;
3319         }
3320         return false;
3321 }
3322
3323 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
3324 {
3325         int still_running = 0, i;
3326         u64 now;
3327         long ret;
3328         struct kvm_vcpu *vcpu;
3329
3330         spin_lock(&vc->lock);
3331         now = get_tb();
3332         for_each_runnable_thread(i, vcpu, vc) {
3333                 /*
3334                  * It's safe to unlock the vcore in the loop here, because
3335                  * for_each_runnable_thread() is safe against removal of
3336                  * the vcpu, and the vcore state is VCORE_EXITING here,
3337                  * so any vcpus becoming runnable will have their arch.trap
3338                  * set to zero and can't actually run in the guest.
3339                  */
3340                 spin_unlock(&vc->lock);
3341                 /* cancel pending dec exception if dec is positive */
3342                 if (now < vcpu->arch.dec_expires &&
3343                     kvmppc_core_pending_dec(vcpu))
3344                         kvmppc_core_dequeue_dec(vcpu);
3345
3346                 trace_kvm_guest_exit(vcpu);
3347
3348                 ret = RESUME_GUEST;
3349                 if (vcpu->arch.trap)
3350                         ret = kvmppc_handle_exit_hv(vcpu,
3351                                                     vcpu->arch.run_task);
3352
3353                 vcpu->arch.ret = ret;
3354                 vcpu->arch.trap = 0;
3355
3356                 spin_lock(&vc->lock);
3357                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
3358                         if (vcpu->arch.pending_exceptions)
3359                                 kvmppc_core_prepare_to_enter(vcpu);
3360                         if (vcpu->arch.ceded)
3361                                 kvmppc_set_timer(vcpu);
3362                         else
3363                                 ++still_running;
3364                 } else {
3365                         kvmppc_remove_runnable(vc, vcpu);
3366                         wake_up(&vcpu->arch.cpu_run);
3367                 }
3368         }
3369         if (!is_master) {
3370                 if (still_running > 0) {
3371                         kvmppc_vcore_preempt(vc);
3372                 } else if (vc->runner) {
3373                         vc->vcore_state = VCORE_PREEMPT;
3374                         kvmppc_core_start_stolen(vc);
3375                 } else {
3376                         vc->vcore_state = VCORE_INACTIVE;
3377                 }
3378                 if (vc->n_runnable > 0 && vc->runner == NULL) {
3379                         /* make sure there's a candidate runner awake */
3380                         i = -1;
3381                         vcpu = next_runnable_thread(vc, &i);
3382                         wake_up(&vcpu->arch.cpu_run);
3383                 }
3384         }
3385         spin_unlock(&vc->lock);
3386 }
3387
3388 /*
3389  * Clear core from the list of active host cores as we are about to
3390  * enter the guest. Only do this if it is the primary thread of the
3391  * core (not if a subcore) that is entering the guest.
3392  */
3393 static inline int kvmppc_clear_host_core(unsigned int cpu)
3394 {
3395         int core;
3396
3397         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3398                 return 0;
3399         /*
3400          * Memory barrier can be omitted here as we will do a smp_wmb()
3401          * later in kvmppc_start_thread and we need ensure that state is
3402          * visible to other CPUs only after we enter guest.
3403          */
3404         core = cpu >> threads_shift;
3405         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3406         return 0;
3407 }
3408
3409 /*
3410  * Advertise this core as an active host core since we exited the guest
3411  * Only need to do this if it is the primary thread of the core that is
3412  * exiting.
3413  */
3414 static inline int kvmppc_set_host_core(unsigned int cpu)
3415 {
3416         int core;
3417
3418         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3419                 return 0;
3420
3421         /*
3422          * Memory barrier can be omitted here because we do a spin_unlock
3423          * immediately after this which provides the memory barrier.
3424          */
3425         core = cpu >> threads_shift;
3426         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3427         return 0;
3428 }
3429
3430 static void set_irq_happened(int trap)
3431 {
3432         switch (trap) {
3433         case BOOK3S_INTERRUPT_EXTERNAL:
3434                 local_paca->irq_happened |= PACA_IRQ_EE;
3435                 break;
3436         case BOOK3S_INTERRUPT_H_DOORBELL:
3437                 local_paca->irq_happened |= PACA_IRQ_DBELL;
3438                 break;
3439         case BOOK3S_INTERRUPT_HMI:
3440                 local_paca->irq_happened |= PACA_IRQ_HMI;
3441                 break;
3442         case BOOK3S_INTERRUPT_SYSTEM_RESET:
3443                 replay_system_reset();
3444                 break;
3445         }
3446 }
3447
3448 /*
3449  * Run a set of guest threads on a physical core.
3450  * Called with vc->lock held.
3451  */
3452 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3453 {
3454         struct kvm_vcpu *vcpu;
3455         int i;
3456         int srcu_idx;
3457         struct core_info core_info;
3458         struct kvmppc_vcore *pvc;
3459         struct kvm_split_mode split_info, *sip;
3460         int split, subcore_size, active;
3461         int sub;
3462         bool thr0_done;
3463         unsigned long cmd_bit, stat_bit;
3464         int pcpu, thr;
3465         int target_threads;
3466         int controlled_threads;
3467         int trap;
3468         bool is_power8;
3469
3470         if (WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)))
3471                 return;
3472
3473         /*
3474          * Remove from the list any threads that have a signal pending
3475          * or need a VPA update done
3476          */
3477         prepare_threads(vc);
3478
3479         /* if the runner is no longer runnable, let the caller pick a new one */
3480         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3481                 return;
3482
3483         /*
3484          * Initialize *vc.
3485          */
3486         init_vcore_to_run(vc);
3487         vc->preempt_tb = TB_NIL;
3488
3489         /*
3490          * Number of threads that we will be controlling: the same as
3491          * the number of threads per subcore, except on POWER9,
3492          * where it's 1 because the threads are (mostly) independent.
3493          */
3494         controlled_threads = threads_per_vcore(vc->kvm);
3495
3496         /*
3497          * Make sure we are running on primary threads, and that secondary
3498          * threads are offline.  Also check if the number of threads in this
3499          * guest are greater than the current system threads per guest.
3500          */
3501         if ((controlled_threads > 1) &&
3502             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
3503                 for_each_runnable_thread(i, vcpu, vc) {
3504                         vcpu->arch.ret = -EBUSY;
3505                         kvmppc_remove_runnable(vc, vcpu);
3506                         wake_up(&vcpu->arch.cpu_run);
3507                 }
3508                 goto out;
3509         }
3510
3511         /*
3512          * See if we could run any other vcores on the physical core
3513          * along with this one.
3514          */
3515         init_core_info(&core_info, vc);
3516         pcpu = smp_processor_id();
3517         target_threads = controlled_threads;
3518         if (target_smt_mode && target_smt_mode < target_threads)
3519                 target_threads = target_smt_mode;
3520         if (vc->num_threads < target_threads)
3521                 collect_piggybacks(&core_info, target_threads);
3522
3523         /*
3524          * Hard-disable interrupts, and check resched flag and signals.
3525          * If we need to reschedule or deliver a signal, clean up
3526          * and return without going into the guest(s).
3527          * If the mmu_ready flag has been cleared, don't go into the
3528          * guest because that means a HPT resize operation is in progress.
3529          */
3530         local_irq_disable();
3531         hard_irq_disable();
3532         if (lazy_irq_pending() || need_resched() ||
3533             recheck_signals_and_mmu(&core_info)) {
3534                 local_irq_enable();
3535                 vc->vcore_state = VCORE_INACTIVE;
3536                 /* Unlock all except the primary vcore */
3537                 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3538                         pvc = core_info.vc[sub];
3539                         /* Put back on to the preempted vcores list */
3540                         kvmppc_vcore_preempt(pvc);
3541                         spin_unlock(&pvc->lock);
3542                 }
3543                 for (i = 0; i < controlled_threads; ++i)
3544                         kvmppc_release_hwthread(pcpu + i);
3545                 return;
3546         }
3547
3548         kvmppc_clear_host_core(pcpu);
3549
3550         /* Decide on micro-threading (split-core) mode */
3551         subcore_size = threads_per_subcore;
3552         cmd_bit = stat_bit = 0;
3553         split = core_info.n_subcores;
3554         sip = NULL;
3555         is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S);
3556
3557         if (split > 1) {
3558                 sip = &split_info;
3559                 memset(&split_info, 0, sizeof(split_info));
3560                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3561                         split_info.vc[sub] = core_info.vc[sub];
3562
3563                 if (is_power8) {
3564                         if (split == 2 && (dynamic_mt_modes & 2)) {
3565                                 cmd_bit = HID0_POWER8_1TO2LPAR;
3566                                 stat_bit = HID0_POWER8_2LPARMODE;
3567                         } else {
3568                                 split = 4;
3569                                 cmd_bit = HID0_POWER8_1TO4LPAR;
3570                                 stat_bit = HID0_POWER8_4LPARMODE;
3571                         }
3572                         subcore_size = MAX_SMT_THREADS / split;
3573                         split_info.rpr = mfspr(SPRN_RPR);
3574                         split_info.pmmar = mfspr(SPRN_PMMAR);
3575                         split_info.ldbar = mfspr(SPRN_LDBAR);
3576                         split_info.subcore_size = subcore_size;
3577                 } else {
3578                         split_info.subcore_size = 1;
3579                 }
3580
3581                 /* order writes to split_info before kvm_split_mode pointer */
3582                 smp_wmb();
3583         }
3584
3585         for (thr = 0; thr < controlled_threads; ++thr) {
3586                 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3587
3588                 paca->kvm_hstate.napping = 0;
3589                 paca->kvm_hstate.kvm_split_mode = sip;
3590         }
3591
3592         /* Initiate micro-threading (split-core) on POWER8 if required */
3593         if (cmd_bit) {
3594                 unsigned long hid0 = mfspr(SPRN_HID0);
3595
3596                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3597                 mb();
3598                 mtspr(SPRN_HID0, hid0);
3599                 isync();
3600                 for (;;) {
3601                         hid0 = mfspr(SPRN_HID0);
3602                         if (hid0 & stat_bit)
3603                                 break;
3604                         cpu_relax();
3605                 }
3606         }
3607
3608         /*
3609          * On POWER8, set RWMR register.
3610          * Since it only affects PURR and SPURR, it doesn't affect
3611          * the host, so we don't save/restore the host value.
3612          */
3613         if (is_power8) {
3614                 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3615                 int n_online = atomic_read(&vc->online_count);
3616
3617                 /*
3618                  * Use the 8-thread value if we're doing split-core
3619                  * or if the vcore's online count looks bogus.
3620                  */
3621                 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3622                     n_online >= 1 && n_online <= MAX_SMT_THREADS)
3623                         rwmr_val = p8_rwmr_values[n_online];
3624                 mtspr(SPRN_RWMR, rwmr_val);
3625         }
3626
3627         /* Start all the threads */
3628         active = 0;
3629         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3630                 thr = is_power8 ? subcore_thread_map[sub] : sub;
3631                 thr0_done = false;
3632                 active |= 1 << thr;
3633                 pvc = core_info.vc[sub];
3634                 pvc->pcpu = pcpu + thr;
3635                 for_each_runnable_thread(i, vcpu, pvc) {
3636                         kvmppc_start_thread(vcpu, pvc);
3637                         kvmppc_create_dtl_entry(vcpu, pvc);
3638                         trace_kvm_guest_enter(vcpu);
3639                         if (!vcpu->arch.ptid)
3640                                 thr0_done = true;
3641                         active |= 1 << (thr + vcpu->arch.ptid);
3642                 }
3643                 /*
3644                  * We need to start the first thread of each subcore
3645                  * even if it doesn't have a vcpu.
3646                  */
3647                 if (!thr0_done)
3648                         kvmppc_start_thread(NULL, pvc);
3649         }
3650
3651         /*
3652          * Ensure that split_info.do_nap is set after setting
3653          * the vcore pointer in the PACA of the secondaries.
3654          */
3655         smp_mb();
3656
3657         /*
3658          * When doing micro-threading, poke the inactive threads as well.
3659          * This gets them to the nap instruction after kvm_do_nap,
3660          * which reduces the time taken to unsplit later.
3661          */
3662         if (cmd_bit) {
3663                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
3664                 for (thr = 1; thr < threads_per_subcore; ++thr)
3665                         if (!(active & (1 << thr)))
3666                                 kvmppc_ipi_thread(pcpu + thr);
3667         }
3668
3669         vc->vcore_state = VCORE_RUNNING;
3670         preempt_disable();
3671
3672         trace_kvmppc_run_core(vc, 0);
3673
3674         for (sub = 0; sub < core_info.n_subcores; ++sub)
3675                 spin_unlock(&core_info.vc[sub]->lock);
3676
3677         guest_enter_irqoff();
3678
3679         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3680
3681         this_cpu_disable_ftrace();
3682
3683         /*
3684          * Interrupts will be enabled once we get into the guest,
3685          * so tell lockdep that we're about to enable interrupts.
3686          */
3687         trace_hardirqs_on();
3688
3689         trap = __kvmppc_vcore_entry();
3690
3691         trace_hardirqs_off();
3692
3693         this_cpu_enable_ftrace();
3694
3695         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3696
3697         set_irq_happened(trap);
3698
3699         spin_lock(&vc->lock);
3700         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3701         vc->vcore_state = VCORE_EXITING;
3702
3703         /* wait for secondary threads to finish writing their state to memory */
3704         kvmppc_wait_for_nap(controlled_threads);
3705
3706         /* Return to whole-core mode if we split the core earlier */
3707         if (cmd_bit) {
3708                 unsigned long hid0 = mfspr(SPRN_HID0);
3709                 unsigned long loops = 0;
3710
3711                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3712                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3713                 mb();
3714                 mtspr(SPRN_HID0, hid0);
3715                 isync();
3716                 for (;;) {
3717                         hid0 = mfspr(SPRN_HID0);
3718                         if (!(hid0 & stat_bit))
3719                                 break;
3720                         cpu_relax();
3721                         ++loops;
3722                 }
3723                 split_info.do_nap = 0;
3724         }
3725
3726         kvmppc_set_host_core(pcpu);
3727
3728         context_tracking_guest_exit();
3729         if (!vtime_accounting_enabled_this_cpu()) {
3730                 local_irq_enable();
3731                 /*
3732                  * Service IRQs here before vtime_account_guest_exit() so any
3733                  * ticks that occurred while running the guest are accounted to
3734                  * the guest. If vtime accounting is enabled, accounting uses
3735                  * TB rather than ticks, so it can be done without enabling
3736                  * interrupts here, which has the problem that it accounts
3737                  * interrupt processing overhead to the host.
3738                  */
3739                 local_irq_disable();
3740         }
3741         vtime_account_guest_exit();
3742
3743         local_irq_enable();
3744
3745         /* Let secondaries go back to the offline loop */
3746         for (i = 0; i < controlled_threads; ++i) {
3747                 kvmppc_release_hwthread(pcpu + i);
3748                 if (sip && sip->napped[i])
3749                         kvmppc_ipi_thread(pcpu + i);
3750                 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3751         }
3752
3753         spin_unlock(&vc->lock);
3754
3755         /* make sure updates to secondary vcpu structs are visible now */
3756         smp_mb();
3757
3758         preempt_enable();
3759
3760         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3761                 pvc = core_info.vc[sub];
3762                 post_guest_process(pvc, pvc == vc);
3763         }
3764
3765         spin_lock(&vc->lock);
3766
3767  out:
3768         vc->vcore_state = VCORE_INACTIVE;
3769         trace_kvmppc_run_core(vc, 1);
3770 }
3771
3772 static void load_spr_state(struct kvm_vcpu *vcpu)
3773 {
3774         mtspr(SPRN_DSCR, vcpu->arch.dscr);
3775         mtspr(SPRN_IAMR, vcpu->arch.iamr);
3776         mtspr(SPRN_PSPB, vcpu->arch.pspb);
3777         mtspr(SPRN_FSCR, vcpu->arch.fscr);
3778         mtspr(SPRN_TAR, vcpu->arch.tar);
3779         mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3780         mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3781         mtspr(SPRN_BESCR, vcpu->arch.bescr);
3782         mtspr(SPRN_TIDR, vcpu->arch.tid);
3783         mtspr(SPRN_AMR, vcpu->arch.amr);
3784         mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3785
3786         /*
3787          * DAR, DSISR, and for nested HV, SPRGs must be set with MSR[RI]
3788          * clear (or hstate set appropriately to catch those registers
3789          * being clobbered if we take a MCE or SRESET), so those are done
3790          * later.
3791          */
3792
3793         if (!(vcpu->arch.ctrl & 1))
3794                 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3795 }
3796
3797 static void store_spr_state(struct kvm_vcpu *vcpu)
3798 {
3799         vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3800
3801         vcpu->arch.iamr = mfspr(SPRN_IAMR);
3802         vcpu->arch.pspb = mfspr(SPRN_PSPB);
3803         vcpu->arch.fscr = mfspr(SPRN_FSCR);
3804         vcpu->arch.tar = mfspr(SPRN_TAR);
3805         vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3806         vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3807         vcpu->arch.bescr = mfspr(SPRN_BESCR);
3808         vcpu->arch.tid = mfspr(SPRN_TIDR);
3809         vcpu->arch.amr = mfspr(SPRN_AMR);
3810         vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3811         vcpu->arch.dscr = mfspr(SPRN_DSCR);
3812 }
3813
3814 /*
3815  * Privileged (non-hypervisor) host registers to save.
3816  */
3817 struct p9_host_os_sprs {
3818         unsigned long dscr;
3819         unsigned long tidr;
3820         unsigned long iamr;
3821         unsigned long amr;
3822         unsigned long fscr;
3823 };
3824
3825 static void save_p9_host_os_sprs(struct p9_host_os_sprs *host_os_sprs)
3826 {
3827         host_os_sprs->dscr = mfspr(SPRN_DSCR);
3828         host_os_sprs->tidr = mfspr(SPRN_TIDR);
3829         host_os_sprs->iamr = mfspr(SPRN_IAMR);
3830         host_os_sprs->amr = mfspr(SPRN_AMR);
3831         host_os_sprs->fscr = mfspr(SPRN_FSCR);
3832 }
3833
3834 /* vcpu guest regs must already be saved */
3835 static void restore_p9_host_os_sprs(struct kvm_vcpu *vcpu,
3836                                     struct p9_host_os_sprs *host_os_sprs)
3837 {
3838         mtspr(SPRN_PSPB, 0);
3839         mtspr(SPRN_UAMOR, 0);
3840
3841         mtspr(SPRN_DSCR, host_os_sprs->dscr);
3842         mtspr(SPRN_TIDR, host_os_sprs->tidr);
3843         mtspr(SPRN_IAMR, host_os_sprs->iamr);
3844
3845         if (host_os_sprs->amr != vcpu->arch.amr)
3846                 mtspr(SPRN_AMR, host_os_sprs->amr);
3847
3848         if (host_os_sprs->fscr != vcpu->arch.fscr)
3849                 mtspr(SPRN_FSCR, host_os_sprs->fscr);
3850
3851         /* Save guest CTRL register, set runlatch to 1 */
3852         if (!(vcpu->arch.ctrl & 1))
3853                 mtspr(SPRN_CTRLT, 1);
3854 }
3855
3856 static inline bool hcall_is_xics(unsigned long req)
3857 {
3858         return req == H_EOI || req == H_CPPR || req == H_IPI ||
3859                 req == H_IPOLL || req == H_XIRR || req == H_XIRR_X;
3860 }
3861
3862 /*
3863  * Guest entry for POWER9 and later CPUs.
3864  */
3865 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3866                          unsigned long lpcr)
3867 {
3868         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3869         struct p9_host_os_sprs host_os_sprs;
3870         s64 dec;
3871         u64 tb;
3872         int trap, save_pmu;
3873
3874         WARN_ON_ONCE(vcpu->arch.ceded);
3875
3876         dec = mfspr(SPRN_DEC);
3877         tb = mftb();
3878         if (dec < 0)
3879                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3880         local_paca->kvm_hstate.dec_expires = dec + tb;
3881         if (local_paca->kvm_hstate.dec_expires < time_limit)
3882                 time_limit = local_paca->kvm_hstate.dec_expires;
3883
3884         save_p9_host_os_sprs(&host_os_sprs);
3885
3886         kvmhv_save_host_pmu();          /* saves it to PACA kvm_hstate */
3887
3888         kvmppc_subcore_enter_guest();
3889
3890         vc->entry_exit_map = 1;
3891         vc->in_guest = 1;
3892
3893         if (vcpu->arch.vpa.pinned_addr) {
3894                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3895                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3896                 lp->yield_count = cpu_to_be32(yield_count);
3897                 vcpu->arch.vpa.dirty = 1;
3898         }
3899
3900         if (cpu_has_feature(CPU_FTR_TM) ||
3901             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3902                 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3903
3904 #ifdef CONFIG_PPC_PSERIES
3905         if (kvmhv_on_pseries()) {
3906                 barrier();
3907                 if (vcpu->arch.vpa.pinned_addr) {
3908                         struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3909                         get_lppaca()->pmcregs_in_use = lp->pmcregs_in_use;
3910                 } else {
3911                         get_lppaca()->pmcregs_in_use = 1;
3912                 }
3913                 barrier();
3914         }
3915 #endif
3916         kvmhv_load_guest_pmu(vcpu);
3917
3918         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3919         load_fp_state(&vcpu->arch.fp);
3920 #ifdef CONFIG_ALTIVEC
3921         load_vr_state(&vcpu->arch.vr);
3922 #endif
3923         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3924
3925         load_spr_state(vcpu);
3926
3927         /*
3928          * When setting DEC, we must always deal with irq_work_raise via NMI vs
3929          * setting DEC. The problem occurs right as we switch into guest mode
3930          * if a NMI hits and sets pending work and sets DEC, then that will
3931          * apply to the guest and not bring us back to the host.
3932          *
3933          * irq_work_raise could check a flag (or possibly LPCR[HDICE] for
3934          * example) and set HDEC to 1? That wouldn't solve the nested hv
3935          * case which needs to abort the hcall or zero the time limit.
3936          *
3937          * XXX: Another day's problem.
3938          */
3939         mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3940
3941         if (kvmhv_on_pseries()) {
3942                 /*
3943                  * We need to save and restore the guest visible part of the
3944                  * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3945                  * doesn't do this for us. Note only required if pseries since
3946                  * this is done in kvmhv_vcpu_entry_p9() below otherwise.
3947                  */
3948                 unsigned long host_psscr;
3949                 /* call our hypervisor to load up HV regs and go */
3950                 struct hv_guest_state hvregs;
3951
3952                 host_psscr = mfspr(SPRN_PSSCR_PR);
3953                 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3954                 kvmhv_save_hv_regs(vcpu, &hvregs);
3955                 hvregs.lpcr = lpcr;
3956                 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3957                 hvregs.version = HV_GUEST_STATE_VERSION;
3958                 if (vcpu->arch.nested) {
3959                         hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3960                         hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3961                 } else {
3962                         hvregs.lpid = vcpu->kvm->arch.lpid;
3963                         hvregs.vcpu_token = vcpu->vcpu_id;
3964                 }
3965                 hvregs.hdec_expiry = time_limit;
3966                 mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3967                 mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3968                 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3969                                           __pa(&vcpu->arch.regs));
3970                 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3971                 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3972                 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3973                 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3974                 vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3975                 mtspr(SPRN_PSSCR_PR, host_psscr);
3976
3977                 /* H_CEDE has to be handled now, not later */
3978                 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3979                     kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3980                         kvmppc_cede(vcpu);
3981                         kvmppc_set_gpr(vcpu, 3, 0);
3982                         trap = 0;
3983                 }
3984         } else {
3985                 kvmppc_xive_push_vcpu(vcpu);
3986                 trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr);
3987                 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3988                     !(vcpu->arch.shregs.msr & MSR_PR)) {
3989                         unsigned long req = kvmppc_get_gpr(vcpu, 3);
3990
3991                         /* H_CEDE has to be handled now, not later */
3992                         if (req == H_CEDE) {
3993                                 kvmppc_cede(vcpu);
3994                                 kvmppc_xive_rearm_escalation(vcpu); /* may un-cede */
3995                                 kvmppc_set_gpr(vcpu, 3, 0);
3996                                 trap = 0;
3997
3998                         /* XICS hcalls must be handled before xive is pulled */
3999                         } else if (hcall_is_xics(req)) {
4000                                 int ret;
4001
4002                                 ret = kvmppc_xive_xics_hcall(vcpu, req);
4003                                 if (ret != H_TOO_HARD) {
4004                                         kvmppc_set_gpr(vcpu, 3, ret);
4005                                         trap = 0;
4006                                 }
4007                         }
4008                 }
4009                 kvmppc_xive_pull_vcpu(vcpu);
4010
4011                 if (kvm_is_radix(vcpu->kvm))
4012                         vcpu->arch.slb_max = 0;
4013         }
4014
4015         dec = mfspr(SPRN_DEC);
4016         if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
4017                 dec = (s32) dec;
4018         tb = mftb();
4019         vcpu->arch.dec_expires = dec + tb;
4020         vcpu->cpu = -1;
4021         vcpu->arch.thread_cpu = -1;
4022
4023         store_spr_state(vcpu);
4024
4025         restore_p9_host_os_sprs(vcpu, &host_os_sprs);
4026
4027         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
4028         store_fp_state(&vcpu->arch.fp);
4029 #ifdef CONFIG_ALTIVEC
4030         store_vr_state(&vcpu->arch.vr);
4031 #endif
4032         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
4033
4034         if (cpu_has_feature(CPU_FTR_TM) ||
4035             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
4036                 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
4037
4038         save_pmu = 1;
4039         if (vcpu->arch.vpa.pinned_addr) {
4040                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
4041                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
4042                 lp->yield_count = cpu_to_be32(yield_count);
4043                 vcpu->arch.vpa.dirty = 1;
4044                 save_pmu = lp->pmcregs_in_use;
4045         }
4046         /* Must save pmu if this guest is capable of running nested guests */
4047         save_pmu |= nesting_enabled(vcpu->kvm);
4048
4049         kvmhv_save_guest_pmu(vcpu, save_pmu);
4050 #ifdef CONFIG_PPC_PSERIES
4051         if (kvmhv_on_pseries()) {
4052                 barrier();
4053                 get_lppaca()->pmcregs_in_use = ppc_get_pmu_inuse();
4054                 barrier();
4055         }
4056 #endif
4057
4058         vc->entry_exit_map = 0x101;
4059         vc->in_guest = 0;
4060
4061         mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
4062         /* We may have raced with new irq work */
4063         if (test_irq_work_pending())
4064                 set_dec(1);
4065         mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
4066
4067         kvmhv_load_host_pmu();
4068
4069         kvmppc_subcore_exit_guest();
4070
4071         return trap;
4072 }
4073
4074 /*
4075  * Wait for some other vcpu thread to execute us, and
4076  * wake us up when we need to handle something in the host.
4077  */
4078 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
4079                                  struct kvm_vcpu *vcpu, int wait_state)
4080 {
4081         DEFINE_WAIT(wait);
4082
4083         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
4084         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4085                 spin_unlock(&vc->lock);
4086                 schedule();
4087                 spin_lock(&vc->lock);
4088         }
4089         finish_wait(&vcpu->arch.cpu_run, &wait);
4090 }
4091
4092 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
4093 {
4094         if (!halt_poll_ns_grow)
4095                 return;
4096
4097         vc->halt_poll_ns *= halt_poll_ns_grow;
4098         if (vc->halt_poll_ns < halt_poll_ns_grow_start)
4099                 vc->halt_poll_ns = halt_poll_ns_grow_start;
4100 }
4101
4102 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
4103 {
4104         if (halt_poll_ns_shrink == 0)
4105                 vc->halt_poll_ns = 0;
4106         else
4107                 vc->halt_poll_ns /= halt_poll_ns_shrink;
4108 }
4109
4110 #ifdef CONFIG_KVM_XICS
4111 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4112 {
4113         if (!xics_on_xive())
4114                 return false;
4115         return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
4116                 vcpu->arch.xive_saved_state.cppr;
4117 }
4118 #else
4119 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4120 {
4121         return false;
4122 }
4123 #endif /* CONFIG_KVM_XICS */
4124
4125 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
4126 {
4127         if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
4128             kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
4129                 return true;
4130
4131         return false;
4132 }
4133
4134 /*
4135  * Check to see if any of the runnable vcpus on the vcore have pending
4136  * exceptions or are no longer ceded
4137  */
4138 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
4139 {
4140         struct kvm_vcpu *vcpu;
4141         int i;
4142
4143         for_each_runnable_thread(i, vcpu, vc) {
4144                 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
4145                         return 1;
4146         }
4147
4148         return 0;
4149 }
4150
4151 /*
4152  * All the vcpus in this vcore are idle, so wait for a decrementer
4153  * or external interrupt to one of the vcpus.  vc->lock is held.
4154  */
4155 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
4156 {
4157         ktime_t cur, start_poll, start_wait;
4158         int do_sleep = 1;
4159         u64 block_ns;
4160
4161         /* Poll for pending exceptions and ceded state */
4162         cur = start_poll = ktime_get();
4163         if (vc->halt_poll_ns) {
4164                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
4165                 ++vc->runner->stat.generic.halt_attempted_poll;
4166
4167                 vc->vcore_state = VCORE_POLLING;
4168                 spin_unlock(&vc->lock);
4169
4170                 do {
4171                         if (kvmppc_vcore_check_block(vc)) {
4172                                 do_sleep = 0;
4173                                 break;
4174                         }
4175                         cur = ktime_get();
4176                 } while (kvm_vcpu_can_poll(cur, stop));
4177
4178                 spin_lock(&vc->lock);
4179                 vc->vcore_state = VCORE_INACTIVE;
4180
4181                 if (!do_sleep) {
4182                         ++vc->runner->stat.generic.halt_successful_poll;
4183                         goto out;
4184                 }
4185         }
4186
4187         prepare_to_rcuwait(&vc->wait);
4188         set_current_state(TASK_INTERRUPTIBLE);
4189         if (kvmppc_vcore_check_block(vc)) {
4190                 finish_rcuwait(&vc->wait);
4191                 do_sleep = 0;
4192                 /* If we polled, count this as a successful poll */
4193                 if (vc->halt_poll_ns)
4194                         ++vc->runner->stat.generic.halt_successful_poll;
4195                 goto out;
4196         }
4197
4198         start_wait = ktime_get();
4199
4200         vc->vcore_state = VCORE_SLEEPING;
4201         trace_kvmppc_vcore_blocked(vc, 0);
4202         spin_unlock(&vc->lock);
4203         schedule();
4204         finish_rcuwait(&vc->wait);
4205         spin_lock(&vc->lock);
4206         vc->vcore_state = VCORE_INACTIVE;
4207         trace_kvmppc_vcore_blocked(vc, 1);
4208         ++vc->runner->stat.halt_successful_wait;
4209
4210         cur = ktime_get();
4211
4212 out:
4213         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
4214
4215         /* Attribute wait time */
4216         if (do_sleep) {
4217                 vc->runner->stat.generic.halt_wait_ns +=
4218                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
4219                 KVM_STATS_LOG_HIST_UPDATE(
4220                                 vc->runner->stat.generic.halt_wait_hist,
4221                                 ktime_to_ns(cur) - ktime_to_ns(start_wait));
4222                 /* Attribute failed poll time */
4223                 if (vc->halt_poll_ns) {
4224                         vc->runner->stat.generic.halt_poll_fail_ns +=
4225                                 ktime_to_ns(start_wait) -
4226                                 ktime_to_ns(start_poll);
4227                         KVM_STATS_LOG_HIST_UPDATE(
4228                                 vc->runner->stat.generic.halt_poll_fail_hist,
4229                                 ktime_to_ns(start_wait) -
4230                                 ktime_to_ns(start_poll));
4231                 }
4232         } else {
4233                 /* Attribute successful poll time */
4234                 if (vc->halt_poll_ns) {
4235                         vc->runner->stat.generic.halt_poll_success_ns +=
4236                                 ktime_to_ns(cur) -
4237                                 ktime_to_ns(start_poll);
4238                         KVM_STATS_LOG_HIST_UPDATE(
4239                                 vc->runner->stat.generic.halt_poll_success_hist,
4240                                 ktime_to_ns(cur) - ktime_to_ns(start_poll));
4241                 }
4242         }
4243
4244         /* Adjust poll time */
4245         if (halt_poll_ns) {
4246                 if (block_ns <= vc->halt_poll_ns)
4247                         ;
4248                 /* We slept and blocked for longer than the max halt time */
4249                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
4250                         shrink_halt_poll_ns(vc);
4251                 /* We slept and our poll time is too small */
4252                 else if (vc->halt_poll_ns < halt_poll_ns &&
4253                                 block_ns < halt_poll_ns)
4254                         grow_halt_poll_ns(vc);
4255                 if (vc->halt_poll_ns > halt_poll_ns)
4256                         vc->halt_poll_ns = halt_poll_ns;
4257         } else
4258                 vc->halt_poll_ns = 0;
4259
4260         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
4261 }
4262
4263 /*
4264  * This never fails for a radix guest, as none of the operations it does
4265  * for a radix guest can fail or have a way to report failure.
4266  */
4267 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
4268 {
4269         int r = 0;
4270         struct kvm *kvm = vcpu->kvm;
4271
4272         mutex_lock(&kvm->arch.mmu_setup_lock);
4273         if (!kvm->arch.mmu_ready) {
4274                 if (!kvm_is_radix(kvm))
4275                         r = kvmppc_hv_setup_htab_rma(vcpu);
4276                 if (!r) {
4277                         if (cpu_has_feature(CPU_FTR_ARCH_300))
4278                                 kvmppc_setup_partition_table(kvm);
4279                         kvm->arch.mmu_ready = 1;
4280                 }
4281         }
4282         mutex_unlock(&kvm->arch.mmu_setup_lock);
4283         return r;
4284 }
4285
4286 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
4287 {
4288         struct kvm_run *run = vcpu->run;
4289         int n_ceded, i, r;
4290         struct kvmppc_vcore *vc;
4291         struct kvm_vcpu *v;
4292
4293         trace_kvmppc_run_vcpu_enter(vcpu);
4294
4295         run->exit_reason = 0;
4296         vcpu->arch.ret = RESUME_GUEST;
4297         vcpu->arch.trap = 0;
4298         kvmppc_update_vpas(vcpu);
4299
4300         /*
4301          * Synchronize with other threads in this virtual core
4302          */
4303         vc = vcpu->arch.vcore;
4304         spin_lock(&vc->lock);
4305         vcpu->arch.ceded = 0;
4306         vcpu->arch.run_task = current;
4307         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4308         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4309         vcpu->arch.busy_preempt = TB_NIL;
4310         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4311         ++vc->n_runnable;
4312
4313         /*
4314          * This happens the first time this is called for a vcpu.
4315          * If the vcore is already running, we may be able to start
4316          * this thread straight away and have it join in.
4317          */
4318         if (!signal_pending(current)) {
4319                 if ((vc->vcore_state == VCORE_PIGGYBACK ||
4320                      vc->vcore_state == VCORE_RUNNING) &&
4321                            !VCORE_IS_EXITING(vc)) {
4322                         kvmppc_create_dtl_entry(vcpu, vc);
4323                         kvmppc_start_thread(vcpu, vc);
4324                         trace_kvm_guest_enter(vcpu);
4325                 } else if (vc->vcore_state == VCORE_SLEEPING) {
4326                         rcuwait_wake_up(&vc->wait);
4327                 }
4328
4329         }
4330
4331         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4332                !signal_pending(current)) {
4333                 /* See if the MMU is ready to go */
4334                 if (!vcpu->kvm->arch.mmu_ready) {
4335                         spin_unlock(&vc->lock);
4336                         r = kvmhv_setup_mmu(vcpu);
4337                         spin_lock(&vc->lock);
4338                         if (r) {
4339                                 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4340                                 run->fail_entry.
4341                                         hardware_entry_failure_reason = 0;
4342                                 vcpu->arch.ret = r;
4343                                 break;
4344                         }
4345                 }
4346
4347                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4348                         kvmppc_vcore_end_preempt(vc);
4349
4350                 if (vc->vcore_state != VCORE_INACTIVE) {
4351                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4352                         continue;
4353                 }
4354                 for_each_runnable_thread(i, v, vc) {
4355                         kvmppc_core_prepare_to_enter(v);
4356                         if (signal_pending(v->arch.run_task)) {
4357                                 kvmppc_remove_runnable(vc, v);
4358                                 v->stat.signal_exits++;
4359                                 v->run->exit_reason = KVM_EXIT_INTR;
4360                                 v->arch.ret = -EINTR;
4361                                 wake_up(&v->arch.cpu_run);
4362                         }
4363                 }
4364                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4365                         break;
4366                 n_ceded = 0;
4367                 for_each_runnable_thread(i, v, vc) {
4368                         if (!kvmppc_vcpu_woken(v))
4369                                 n_ceded += v->arch.ceded;
4370                         else
4371                                 v->arch.ceded = 0;
4372                 }
4373                 vc->runner = vcpu;
4374                 if (n_ceded == vc->n_runnable) {
4375                         kvmppc_vcore_blocked(vc);
4376                 } else if (need_resched()) {
4377                         kvmppc_vcore_preempt(vc);
4378                         /* Let something else run */
4379                         cond_resched_lock(&vc->lock);
4380                         if (vc->vcore_state == VCORE_PREEMPT)
4381                                 kvmppc_vcore_end_preempt(vc);
4382                 } else {
4383                         kvmppc_run_core(vc);
4384                 }
4385                 vc->runner = NULL;
4386         }
4387
4388         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4389                (vc->vcore_state == VCORE_RUNNING ||
4390                 vc->vcore_state == VCORE_EXITING ||
4391                 vc->vcore_state == VCORE_PIGGYBACK))
4392                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4393
4394         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4395                 kvmppc_vcore_end_preempt(vc);
4396
4397         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4398                 kvmppc_remove_runnable(vc, vcpu);
4399                 vcpu->stat.signal_exits++;
4400                 run->exit_reason = KVM_EXIT_INTR;
4401                 vcpu->arch.ret = -EINTR;
4402         }
4403
4404         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4405                 /* Wake up some vcpu to run the core */
4406                 i = -1;
4407                 v = next_runnable_thread(vc, &i);
4408                 wake_up(&v->arch.cpu_run);
4409         }
4410
4411         trace_kvmppc_run_vcpu_exit(vcpu);
4412         spin_unlock(&vc->lock);
4413         return vcpu->arch.ret;
4414 }
4415
4416 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4417                           unsigned long lpcr)
4418 {
4419         struct kvm_run *run = vcpu->run;
4420         int trap, r, pcpu;
4421         int srcu_idx;
4422         struct kvmppc_vcore *vc;
4423         struct kvm *kvm = vcpu->kvm;
4424         struct kvm_nested_guest *nested = vcpu->arch.nested;
4425
4426         trace_kvmppc_run_vcpu_enter(vcpu);
4427
4428         run->exit_reason = 0;
4429         vcpu->arch.ret = RESUME_GUEST;
4430         vcpu->arch.trap = 0;
4431
4432         vc = vcpu->arch.vcore;
4433         vcpu->arch.ceded = 0;
4434         vcpu->arch.run_task = current;
4435         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4436         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4437         vcpu->arch.busy_preempt = TB_NIL;
4438         vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4439         vc->runnable_threads[0] = vcpu;
4440         vc->n_runnable = 1;
4441         vc->runner = vcpu;
4442
4443         /* See if the MMU is ready to go */
4444         if (!kvm->arch.mmu_ready) {
4445                 r = kvmhv_setup_mmu(vcpu);
4446                 if (r) {
4447                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4448                         run->fail_entry.hardware_entry_failure_reason = 0;
4449                         vcpu->arch.ret = r;
4450                         return r;
4451                 }
4452         }
4453
4454         if (need_resched())
4455                 cond_resched();
4456
4457         kvmppc_update_vpas(vcpu);
4458
4459         init_vcore_to_run(vc);
4460         vc->preempt_tb = TB_NIL;
4461
4462         preempt_disable();
4463         pcpu = smp_processor_id();
4464         vc->pcpu = pcpu;
4465         if (kvm_is_radix(kvm))
4466                 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4467
4468         local_irq_disable();
4469         hard_irq_disable();
4470         if (signal_pending(current))
4471                 goto sigpend;
4472         if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4473                 goto out;
4474
4475         if (!nested) {
4476                 kvmppc_core_prepare_to_enter(vcpu);
4477                 if (vcpu->arch.doorbell_request) {
4478                         vc->dpdes = 1;
4479                         smp_wmb();
4480                         vcpu->arch.doorbell_request = 0;
4481                 }
4482                 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4483                              &vcpu->arch.pending_exceptions))
4484                         lpcr |= LPCR_MER;
4485         } else if (vcpu->arch.pending_exceptions ||
4486                    vcpu->arch.doorbell_request ||
4487                    xive_interrupt_pending(vcpu)) {
4488                 vcpu->arch.ret = RESUME_HOST;
4489                 goto out;
4490         }
4491
4492         kvmppc_clear_host_core(pcpu);
4493
4494         local_paca->kvm_hstate.napping = 0;
4495         local_paca->kvm_hstate.kvm_split_mode = NULL;
4496         kvmppc_start_thread(vcpu, vc);
4497         kvmppc_create_dtl_entry(vcpu, vc);
4498         trace_kvm_guest_enter(vcpu);
4499
4500         vc->vcore_state = VCORE_RUNNING;
4501         trace_kvmppc_run_core(vc, 0);
4502
4503         guest_enter_irqoff();
4504
4505         srcu_idx = srcu_read_lock(&kvm->srcu);
4506
4507         this_cpu_disable_ftrace();
4508
4509         /* Tell lockdep that we're about to enable interrupts */
4510         trace_hardirqs_on();
4511
4512         trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4513         vcpu->arch.trap = trap;
4514
4515         trace_hardirqs_off();
4516
4517         this_cpu_enable_ftrace();
4518
4519         srcu_read_unlock(&kvm->srcu, srcu_idx);
4520
4521         set_irq_happened(trap);
4522
4523         kvmppc_set_host_core(pcpu);
4524
4525         context_tracking_guest_exit();
4526         if (!vtime_accounting_enabled_this_cpu()) {
4527                 local_irq_enable();
4528                 /*
4529                  * Service IRQs here before vtime_account_guest_exit() so any
4530                  * ticks that occurred while running the guest are accounted to
4531                  * the guest. If vtime accounting is enabled, accounting uses
4532                  * TB rather than ticks, so it can be done without enabling
4533                  * interrupts here, which has the problem that it accounts
4534                  * interrupt processing overhead to the host.
4535                  */
4536                 local_irq_disable();
4537         }
4538         vtime_account_guest_exit();
4539
4540         local_irq_enable();
4541
4542         cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4543
4544         preempt_enable();
4545
4546         /*
4547          * cancel pending decrementer exception if DEC is now positive, or if
4548          * entering a nested guest in which case the decrementer is now owned
4549          * by L2 and the L1 decrementer is provided in hdec_expires
4550          */
4551         if (kvmppc_core_pending_dec(vcpu) &&
4552                         ((get_tb() < vcpu->arch.dec_expires) ||
4553                          (trap == BOOK3S_INTERRUPT_SYSCALL &&
4554                           kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4555                 kvmppc_core_dequeue_dec(vcpu);
4556
4557         trace_kvm_guest_exit(vcpu);
4558         r = RESUME_GUEST;
4559         if (trap) {
4560                 if (!nested)
4561                         r = kvmppc_handle_exit_hv(vcpu, current);
4562                 else
4563                         r = kvmppc_handle_nested_exit(vcpu);
4564         }
4565         vcpu->arch.ret = r;
4566
4567         if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4568             !kvmppc_vcpu_woken(vcpu)) {
4569                 kvmppc_set_timer(vcpu);
4570                 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4571                         if (signal_pending(current)) {
4572                                 vcpu->stat.signal_exits++;
4573                                 run->exit_reason = KVM_EXIT_INTR;
4574                                 vcpu->arch.ret = -EINTR;
4575                                 break;
4576                         }
4577                         spin_lock(&vc->lock);
4578                         kvmppc_vcore_blocked(vc);
4579                         spin_unlock(&vc->lock);
4580                 }
4581         }
4582         vcpu->arch.ceded = 0;
4583
4584         vc->vcore_state = VCORE_INACTIVE;
4585         trace_kvmppc_run_core(vc, 1);
4586
4587  done:
4588         kvmppc_remove_runnable(vc, vcpu);
4589         trace_kvmppc_run_vcpu_exit(vcpu);
4590
4591         return vcpu->arch.ret;
4592
4593  sigpend:
4594         vcpu->stat.signal_exits++;
4595         run->exit_reason = KVM_EXIT_INTR;
4596         vcpu->arch.ret = -EINTR;
4597  out:
4598         local_irq_enable();
4599         preempt_enable();
4600         goto done;
4601 }
4602
4603 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
4604 {
4605         struct kvm_run *run = vcpu->run;
4606         int r;
4607         int srcu_idx;
4608         unsigned long ebb_regs[3] = {}; /* shut up GCC */
4609         unsigned long user_tar = 0;
4610         unsigned int user_vrsave;
4611         struct kvm *kvm;
4612
4613         if (!vcpu->arch.sane) {
4614                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4615                 return -EINVAL;
4616         }
4617
4618         /*
4619          * Don't allow entry with a suspended transaction, because
4620          * the guest entry/exit code will lose it.
4621          * If the guest has TM enabled, save away their TM-related SPRs
4622          * (they will get restored by the TM unavailable interrupt).
4623          */
4624 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4625         if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4626             (current->thread.regs->msr & MSR_TM)) {
4627                 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4628                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4629                         run->fail_entry.hardware_entry_failure_reason = 0;
4630                         return -EINVAL;
4631                 }
4632                 /* Enable TM so we can read the TM SPRs */
4633                 mtmsr(mfmsr() | MSR_TM);
4634                 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4635                 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4636                 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4637                 current->thread.regs->msr &= ~MSR_TM;
4638         }
4639 #endif
4640
4641         /*
4642          * Force online to 1 for the sake of old userspace which doesn't
4643          * set it.
4644          */
4645         if (!vcpu->arch.online) {
4646                 atomic_inc(&vcpu->arch.vcore->online_count);
4647                 vcpu->arch.online = 1;
4648         }
4649
4650         kvmppc_core_prepare_to_enter(vcpu);
4651
4652         /* No need to go into the guest when all we'll do is come back out */
4653         if (signal_pending(current)) {
4654                 run->exit_reason = KVM_EXIT_INTR;
4655                 return -EINTR;
4656         }
4657
4658         kvm = vcpu->kvm;
4659         atomic_inc(&kvm->arch.vcpus_running);
4660         /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4661         smp_mb();
4662
4663         flush_all_to_thread(current);
4664
4665         /* Save userspace EBB and other register values */
4666         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4667                 ebb_regs[0] = mfspr(SPRN_EBBHR);
4668                 ebb_regs[1] = mfspr(SPRN_EBBRR);
4669                 ebb_regs[2] = mfspr(SPRN_BESCR);
4670                 user_tar = mfspr(SPRN_TAR);
4671         }
4672         user_vrsave = mfspr(SPRN_VRSAVE);
4673
4674         vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4675         vcpu->arch.pgdir = kvm->mm->pgd;
4676         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4677
4678         do {
4679                 if (cpu_has_feature(CPU_FTR_ARCH_300))
4680                         r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
4681                                                   vcpu->arch.vcore->lpcr);
4682                 else
4683                         r = kvmppc_run_vcpu(vcpu);
4684
4685                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL) {
4686                         if (WARN_ON_ONCE(vcpu->arch.shregs.msr & MSR_PR)) {
4687                                 /*
4688                                  * These should have been caught reflected
4689                                  * into the guest by now. Final sanity check:
4690                                  * don't allow userspace to execute hcalls in
4691                                  * the hypervisor.
4692                                  */
4693                                 r = RESUME_GUEST;
4694                                 continue;
4695                         }
4696                         trace_kvm_hcall_enter(vcpu);
4697                         r = kvmppc_pseries_do_hcall(vcpu);
4698                         trace_kvm_hcall_exit(vcpu, r);
4699                         kvmppc_core_prepare_to_enter(vcpu);
4700                 } else if (r == RESUME_PAGE_FAULT) {
4701                         srcu_idx = srcu_read_lock(&kvm->srcu);
4702                         r = kvmppc_book3s_hv_page_fault(vcpu,
4703                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4704                         srcu_read_unlock(&kvm->srcu, srcu_idx);
4705                 } else if (r == RESUME_PASSTHROUGH) {
4706                         if (WARN_ON(xics_on_xive()))
4707                                 r = H_SUCCESS;
4708                         else
4709                                 r = kvmppc_xics_rm_complete(vcpu, 0);
4710                 }
4711         } while (is_kvmppc_resume_guest(r));
4712
4713         /* Restore userspace EBB and other register values */
4714         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4715                 mtspr(SPRN_EBBHR, ebb_regs[0]);
4716                 mtspr(SPRN_EBBRR, ebb_regs[1]);
4717                 mtspr(SPRN_BESCR, ebb_regs[2]);
4718                 mtspr(SPRN_TAR, user_tar);
4719         }
4720         mtspr(SPRN_VRSAVE, user_vrsave);
4721
4722         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4723         atomic_dec(&kvm->arch.vcpus_running);
4724
4725         srr_regs_clobbered();
4726
4727         return r;
4728 }
4729
4730 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4731                                      int shift, int sllp)
4732 {
4733         (*sps)->page_shift = shift;
4734         (*sps)->slb_enc = sllp;
4735         (*sps)->enc[0].page_shift = shift;
4736         (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4737         /*
4738          * Add 16MB MPSS support (may get filtered out by userspace)
4739          */
4740         if (shift != 24) {
4741                 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4742                 if (penc != -1) {
4743                         (*sps)->enc[1].page_shift = 24;
4744                         (*sps)->enc[1].pte_enc = penc;
4745                 }
4746         }
4747         (*sps)++;
4748 }
4749
4750 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4751                                          struct kvm_ppc_smmu_info *info)
4752 {
4753         struct kvm_ppc_one_seg_page_size *sps;
4754
4755         /*
4756          * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4757          * POWER7 doesn't support keys for instruction accesses,
4758          * POWER8 and POWER9 do.
4759          */
4760         info->data_keys = 32;
4761         info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4762
4763         /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4764         info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4765         info->slb_size = 32;
4766
4767         /* We only support these sizes for now, and no muti-size segments */
4768         sps = &info->sps[0];
4769         kvmppc_add_seg_page_size(&sps, 12, 0);
4770         kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4771         kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4772
4773         /* If running as a nested hypervisor, we don't support HPT guests */
4774         if (kvmhv_on_pseries())
4775                 info->flags |= KVM_PPC_NO_HASH;
4776
4777         return 0;
4778 }
4779
4780 /*
4781  * Get (and clear) the dirty memory log for a memory slot.
4782  */
4783 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4784                                          struct kvm_dirty_log *log)
4785 {
4786         struct kvm_memslots *slots;
4787         struct kvm_memory_slot *memslot;
4788         int i, r;
4789         unsigned long n;
4790         unsigned long *buf, *p;
4791         struct kvm_vcpu *vcpu;
4792
4793         mutex_lock(&kvm->slots_lock);
4794
4795         r = -EINVAL;
4796         if (log->slot >= KVM_USER_MEM_SLOTS)
4797                 goto out;
4798
4799         slots = kvm_memslots(kvm);
4800         memslot = id_to_memslot(slots, log->slot);
4801         r = -ENOENT;
4802         if (!memslot || !memslot->dirty_bitmap)
4803                 goto out;
4804
4805         /*
4806          * Use second half of bitmap area because both HPT and radix
4807          * accumulate bits in the first half.
4808          */
4809         n = kvm_dirty_bitmap_bytes(memslot);
4810         buf = memslot->dirty_bitmap + n / sizeof(long);
4811         memset(buf, 0, n);
4812
4813         if (kvm_is_radix(kvm))
4814                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4815         else
4816                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4817         if (r)
4818                 goto out;
4819
4820         /*
4821          * We accumulate dirty bits in the first half of the
4822          * memslot's dirty_bitmap area, for when pages are paged
4823          * out or modified by the host directly.  Pick up these
4824          * bits and add them to the map.
4825          */
4826         p = memslot->dirty_bitmap;
4827         for (i = 0; i < n / sizeof(long); ++i)
4828                 buf[i] |= xchg(&p[i], 0);
4829
4830         /* Harvest dirty bits from VPA and DTL updates */
4831         /* Note: we never modify the SLB shadow buffer areas */
4832         kvm_for_each_vcpu(i, vcpu, kvm) {
4833                 spin_lock(&vcpu->arch.vpa_update_lock);
4834                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4835                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4836                 spin_unlock(&vcpu->arch.vpa_update_lock);
4837         }
4838
4839         r = -EFAULT;
4840         if (copy_to_user(log->dirty_bitmap, buf, n))
4841                 goto out;
4842
4843         r = 0;
4844 out:
4845         mutex_unlock(&kvm->slots_lock);
4846         return r;
4847 }
4848
4849 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
4850 {
4851         vfree(slot->arch.rmap);
4852         slot->arch.rmap = NULL;
4853 }
4854
4855 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4856                                         struct kvm_memory_slot *slot,
4857                                         const struct kvm_userspace_memory_region *mem,
4858                                         enum kvm_mr_change change)
4859 {
4860         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4861
4862         if (change == KVM_MR_CREATE) {
4863                 unsigned long size = array_size(npages, sizeof(*slot->arch.rmap));
4864
4865                 if ((size >> PAGE_SHIFT) > totalram_pages())
4866                         return -ENOMEM;
4867
4868                 slot->arch.rmap = vzalloc(size);
4869                 if (!slot->arch.rmap)
4870                         return -ENOMEM;
4871         }
4872
4873         return 0;
4874 }
4875
4876 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4877                                 const struct kvm_userspace_memory_region *mem,
4878                                 const struct kvm_memory_slot *old,
4879                                 const struct kvm_memory_slot *new,
4880                                 enum kvm_mr_change change)
4881 {
4882         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4883
4884         /*
4885          * If we are making a new memslot, it might make
4886          * some address that was previously cached as emulated
4887          * MMIO be no longer emulated MMIO, so invalidate
4888          * all the caches of emulated MMIO translations.
4889          */
4890         if (npages)
4891                 atomic64_inc(&kvm->arch.mmio_update);
4892
4893         /*
4894          * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4895          * have already called kvm_arch_flush_shadow_memslot() to
4896          * flush shadow mappings.  For KVM_MR_CREATE we have no
4897          * previous mappings.  So the only case to handle is
4898          * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4899          * has been changed.
4900          * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4901          * to get rid of any THP PTEs in the partition-scoped page tables
4902          * so we can track dirtiness at the page level; we flush when
4903          * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4904          * using THP PTEs.
4905          */
4906         if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4907             ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4908                 kvmppc_radix_flush_memslot(kvm, old);
4909         /*
4910          * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4911          */
4912         if (!kvm->arch.secure_guest)
4913                 return;
4914
4915         switch (change) {
4916         case KVM_MR_CREATE:
4917                 /*
4918                  * @TODO kvmppc_uvmem_memslot_create() can fail and
4919                  * return error. Fix this.
4920                  */
4921                 kvmppc_uvmem_memslot_create(kvm, new);
4922                 break;
4923         case KVM_MR_DELETE:
4924                 kvmppc_uvmem_memslot_delete(kvm, old);
4925                 break;
4926         default:
4927                 /* TODO: Handle KVM_MR_MOVE */
4928                 break;
4929         }
4930 }
4931
4932 /*
4933  * Update LPCR values in kvm->arch and in vcores.
4934  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4935  * of kvm->arch.lpcr update).
4936  */
4937 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4938 {
4939         long int i;
4940         u32 cores_done = 0;
4941
4942         if ((kvm->arch.lpcr & mask) == lpcr)
4943                 return;
4944
4945         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4946
4947         for (i = 0; i < KVM_MAX_VCORES; ++i) {
4948                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4949                 if (!vc)
4950                         continue;
4951
4952                 spin_lock(&vc->lock);
4953                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4954                 verify_lpcr(kvm, vc->lpcr);
4955                 spin_unlock(&vc->lock);
4956                 if (++cores_done >= kvm->arch.online_vcores)
4957                         break;
4958         }
4959 }
4960
4961 void kvmppc_setup_partition_table(struct kvm *kvm)
4962 {
4963         unsigned long dw0, dw1;
4964
4965         if (!kvm_is_radix(kvm)) {
4966                 /* PS field - page size for VRMA */
4967                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4968                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4969                 /* HTABSIZE and HTABORG fields */
4970                 dw0 |= kvm->arch.sdr1;
4971
4972                 /* Second dword as set by userspace */
4973                 dw1 = kvm->arch.process_table;
4974         } else {
4975                 dw0 = PATB_HR | radix__get_tree_size() |
4976                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4977                 dw1 = PATB_GR | kvm->arch.process_table;
4978         }
4979         kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4980 }
4981
4982 /*
4983  * Set up HPT (hashed page table) and RMA (real-mode area).
4984  * Must be called with kvm->arch.mmu_setup_lock held.
4985  */
4986 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4987 {
4988         int err = 0;
4989         struct kvm *kvm = vcpu->kvm;
4990         unsigned long hva;
4991         struct kvm_memory_slot *memslot;
4992         struct vm_area_struct *vma;
4993         unsigned long lpcr = 0, senc;
4994         unsigned long psize, porder;
4995         int srcu_idx;
4996
4997         /* Allocate hashed page table (if not done already) and reset it */
4998         if (!kvm->arch.hpt.virt) {
4999                 int order = KVM_DEFAULT_HPT_ORDER;
5000                 struct kvm_hpt_info info;
5001
5002                 err = kvmppc_allocate_hpt(&info, order);
5003                 /* If we get here, it means userspace didn't specify a
5004                  * size explicitly.  So, try successively smaller
5005                  * sizes if the default failed. */
5006                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
5007                         err  = kvmppc_allocate_hpt(&info, order);
5008
5009                 if (err < 0) {
5010                         pr_err("KVM: Couldn't alloc HPT\n");
5011                         goto out;
5012                 }
5013
5014                 kvmppc_set_hpt(kvm, &info);
5015         }
5016
5017         /* Look up the memslot for guest physical address 0 */
5018         srcu_idx = srcu_read_lock(&kvm->srcu);
5019         memslot = gfn_to_memslot(kvm, 0);
5020
5021         /* We must have some memory at 0 by now */
5022         err = -EINVAL;
5023         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
5024                 goto out_srcu;
5025
5026         /* Look up the VMA for the start of this memory slot */
5027         hva = memslot->userspace_addr;
5028         mmap_read_lock(kvm->mm);
5029         vma = vma_lookup(kvm->mm, hva);
5030         if (!vma || (vma->vm_flags & VM_IO))
5031                 goto up_out;
5032
5033         psize = vma_kernel_pagesize(vma);
5034
5035         mmap_read_unlock(kvm->mm);
5036
5037         /* We can handle 4k, 64k or 16M pages in the VRMA */
5038         if (psize >= 0x1000000)
5039                 psize = 0x1000000;
5040         else if (psize >= 0x10000)
5041                 psize = 0x10000;
5042         else
5043                 psize = 0x1000;
5044         porder = __ilog2(psize);
5045
5046         senc = slb_pgsize_encoding(psize);
5047         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
5048                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
5049         /* Create HPTEs in the hash page table for the VRMA */
5050         kvmppc_map_vrma(vcpu, memslot, porder);
5051
5052         /* Update VRMASD field in the LPCR */
5053         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
5054                 /* the -4 is to account for senc values starting at 0x10 */
5055                 lpcr = senc << (LPCR_VRMASD_SH - 4);
5056                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
5057         }
5058
5059         /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
5060         smp_wmb();
5061         err = 0;
5062  out_srcu:
5063         srcu_read_unlock(&kvm->srcu, srcu_idx);
5064  out:
5065         return err;
5066
5067  up_out:
5068         mmap_read_unlock(kvm->mm);
5069         goto out_srcu;
5070 }
5071
5072 /*
5073  * Must be called with kvm->arch.mmu_setup_lock held and
5074  * mmu_ready = 0 and no vcpus running.
5075  */
5076 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
5077 {
5078         if (nesting_enabled(kvm))
5079                 kvmhv_release_all_nested(kvm);
5080         kvmppc_rmap_reset(kvm);
5081         kvm->arch.process_table = 0;
5082         /* Mutual exclusion with kvm_unmap_gfn_range etc. */
5083         spin_lock(&kvm->mmu_lock);
5084         kvm->arch.radix = 0;
5085         spin_unlock(&kvm->mmu_lock);
5086         kvmppc_free_radix(kvm);
5087         kvmppc_update_lpcr(kvm, LPCR_VPM1,
5088                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
5089         return 0;
5090 }
5091
5092 /*
5093  * Must be called with kvm->arch.mmu_setup_lock held and
5094  * mmu_ready = 0 and no vcpus running.
5095  */
5096 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
5097 {
5098         int err;
5099
5100         err = kvmppc_init_vm_radix(kvm);
5101         if (err)
5102                 return err;
5103         kvmppc_rmap_reset(kvm);
5104         /* Mutual exclusion with kvm_unmap_gfn_range etc. */
5105         spin_lock(&kvm->mmu_lock);
5106         kvm->arch.radix = 1;
5107         spin_unlock(&kvm->mmu_lock);
5108         kvmppc_free_hpt(&kvm->arch.hpt);
5109         kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
5110                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
5111         return 0;
5112 }
5113
5114 #ifdef CONFIG_KVM_XICS
5115 /*
5116  * Allocate a per-core structure for managing state about which cores are
5117  * running in the host versus the guest and for exchanging data between
5118  * real mode KVM and CPU running in the host.
5119  * This is only done for the first VM.
5120  * The allocated structure stays even if all VMs have stopped.
5121  * It is only freed when the kvm-hv module is unloaded.
5122  * It's OK for this routine to fail, we just don't support host
5123  * core operations like redirecting H_IPI wakeups.
5124  */
5125 void kvmppc_alloc_host_rm_ops(void)
5126 {
5127         struct kvmppc_host_rm_ops *ops;
5128         unsigned long l_ops;
5129         int cpu, core;
5130         int size;
5131
5132         /* Not the first time here ? */
5133         if (kvmppc_host_rm_ops_hv != NULL)
5134                 return;
5135
5136         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
5137         if (!ops)
5138                 return;
5139
5140         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
5141         ops->rm_core = kzalloc(size, GFP_KERNEL);
5142
5143         if (!ops->rm_core) {
5144                 kfree(ops);
5145                 return;
5146         }
5147
5148         cpus_read_lock();
5149
5150         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
5151                 if (!cpu_online(cpu))
5152                         continue;
5153
5154                 core = cpu >> threads_shift;
5155                 ops->rm_core[core].rm_state.in_host = 1;
5156         }
5157
5158         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
5159
5160         /*
5161          * Make the contents of the kvmppc_host_rm_ops structure visible
5162          * to other CPUs before we assign it to the global variable.
5163          * Do an atomic assignment (no locks used here), but if someone
5164          * beats us to it, just free our copy and return.
5165          */
5166         smp_wmb();
5167         l_ops = (unsigned long) ops;
5168
5169         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
5170                 cpus_read_unlock();
5171                 kfree(ops->rm_core);
5172                 kfree(ops);
5173                 return;
5174         }
5175
5176         cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
5177                                              "ppc/kvm_book3s:prepare",
5178                                              kvmppc_set_host_core,
5179                                              kvmppc_clear_host_core);
5180         cpus_read_unlock();
5181 }
5182
5183 void kvmppc_free_host_rm_ops(void)
5184 {
5185         if (kvmppc_host_rm_ops_hv) {
5186                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
5187                 kfree(kvmppc_host_rm_ops_hv->rm_core);
5188                 kfree(kvmppc_host_rm_ops_hv);
5189                 kvmppc_host_rm_ops_hv = NULL;
5190         }
5191 }
5192 #endif
5193
5194 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
5195 {
5196         unsigned long lpcr, lpid;
5197         char buf[32];
5198         int ret;
5199
5200         mutex_init(&kvm->arch.uvmem_lock);
5201         INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
5202         mutex_init(&kvm->arch.mmu_setup_lock);
5203
5204         /* Allocate the guest's logical partition ID */
5205
5206         lpid = kvmppc_alloc_lpid();
5207         if ((long)lpid < 0)
5208                 return -ENOMEM;
5209         kvm->arch.lpid = lpid;
5210
5211         kvmppc_alloc_host_rm_ops();
5212
5213         kvmhv_vm_nested_init(kvm);
5214
5215         /*
5216          * Since we don't flush the TLB when tearing down a VM,
5217          * and this lpid might have previously been used,
5218          * make sure we flush on each core before running the new VM.
5219          * On POWER9, the tlbie in mmu_partition_table_set_entry()
5220          * does this flush for us.
5221          */
5222         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5223                 cpumask_setall(&kvm->arch.need_tlb_flush);
5224
5225         /* Start out with the default set of hcalls enabled */
5226         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
5227                sizeof(kvm->arch.enabled_hcalls));
5228
5229         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5230                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
5231
5232         /* Init LPCR for virtual RMA mode */
5233         if (cpu_has_feature(CPU_FTR_HVMODE)) {
5234                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
5235                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
5236                 lpcr &= LPCR_PECE | LPCR_LPES;
5237         } else {
5238                 /*
5239                  * The L2 LPES mode will be set by the L0 according to whether
5240                  * or not it needs to take external interrupts in HV mode.
5241                  */
5242                 lpcr = 0;
5243         }
5244         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
5245                 LPCR_VPM0 | LPCR_VPM1;
5246         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
5247                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
5248         /* On POWER8 turn on online bit to enable PURR/SPURR */
5249         if (cpu_has_feature(CPU_FTR_ARCH_207S))
5250                 lpcr |= LPCR_ONL;
5251         /*
5252          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
5253          * Set HVICE bit to enable hypervisor virtualization interrupts.
5254          * Set HEIC to prevent OS interrupts to go to hypervisor (should
5255          * be unnecessary but better safe than sorry in case we re-enable
5256          * EE in HV mode with this LPCR still set)
5257          */
5258         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5259                 lpcr &= ~LPCR_VPM0;
5260                 lpcr |= LPCR_HVICE | LPCR_HEIC;
5261
5262                 /*
5263                  * If xive is enabled, we route 0x500 interrupts directly
5264                  * to the guest.
5265                  */
5266                 if (xics_on_xive())
5267                         lpcr |= LPCR_LPES;
5268         }
5269
5270         /*
5271          * If the host uses radix, the guest starts out as radix.
5272          */
5273         if (radix_enabled()) {
5274                 kvm->arch.radix = 1;
5275                 kvm->arch.mmu_ready = 1;
5276                 lpcr &= ~LPCR_VPM1;
5277                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5278                 ret = kvmppc_init_vm_radix(kvm);
5279                 if (ret) {
5280                         kvmppc_free_lpid(kvm->arch.lpid);
5281                         return ret;
5282                 }
5283                 kvmppc_setup_partition_table(kvm);
5284         }
5285
5286         verify_lpcr(kvm, lpcr);
5287         kvm->arch.lpcr = lpcr;
5288
5289         /* Initialization for future HPT resizes */
5290         kvm->arch.resize_hpt = NULL;
5291
5292         /*
5293          * Work out how many sets the TLB has, for the use of
5294          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
5295          */
5296         if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5297                 /*
5298                  * P10 will flush all the congruence class with a single tlbiel
5299                  */
5300                 kvm->arch.tlb_sets = 1;
5301         } else if (radix_enabled())
5302                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
5303         else if (cpu_has_feature(CPU_FTR_ARCH_300))
5304                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
5305         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
5306                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
5307         else
5308                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
5309
5310         /*
5311          * Track that we now have a HV mode VM active. This blocks secondary
5312          * CPU threads from coming online.
5313          */
5314         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5315                 kvm_hv_vm_activated();
5316
5317         /*
5318          * Initialize smt_mode depending on processor.
5319          * POWER8 and earlier have to use "strict" threading, where
5320          * all vCPUs in a vcore have to run on the same (sub)core,
5321          * whereas on POWER9 the threads can each run a different
5322          * guest.
5323          */
5324         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5325                 kvm->arch.smt_mode = threads_per_subcore;
5326         else
5327                 kvm->arch.smt_mode = 1;
5328         kvm->arch.emul_smt_mode = 1;
5329
5330         /*
5331          * Create a debugfs directory for the VM
5332          */
5333         snprintf(buf, sizeof(buf), "vm%d", current->pid);
5334         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
5335         kvmppc_mmu_debugfs_init(kvm);
5336         if (radix_enabled())
5337                 kvmhv_radix_debugfs_init(kvm);
5338
5339         return 0;
5340 }
5341
5342 static void kvmppc_free_vcores(struct kvm *kvm)
5343 {
5344         long int i;
5345
5346         for (i = 0; i < KVM_MAX_VCORES; ++i)
5347                 kfree(kvm->arch.vcores[i]);
5348         kvm->arch.online_vcores = 0;
5349 }
5350
5351 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5352 {
5353         debugfs_remove_recursive(kvm->arch.debugfs_dir);
5354
5355         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5356                 kvm_hv_vm_deactivated();
5357
5358         kvmppc_free_vcores(kvm);
5359
5360
5361         if (kvm_is_radix(kvm))
5362                 kvmppc_free_radix(kvm);
5363         else
5364                 kvmppc_free_hpt(&kvm->arch.hpt);
5365
5366         /* Perform global invalidation and return lpid to the pool */
5367         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5368                 if (nesting_enabled(kvm))
5369                         kvmhv_release_all_nested(kvm);
5370                 kvm->arch.process_table = 0;
5371                 if (kvm->arch.secure_guest)
5372                         uv_svm_terminate(kvm->arch.lpid);
5373                 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5374         }
5375
5376         kvmppc_free_lpid(kvm->arch.lpid);
5377
5378         kvmppc_free_pimap(kvm);
5379 }
5380
5381 /* We don't need to emulate any privileged instructions or dcbz */
5382 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5383                                      unsigned int inst, int *advance)
5384 {
5385         return EMULATE_FAIL;
5386 }
5387
5388 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5389                                         ulong spr_val)
5390 {
5391         return EMULATE_FAIL;
5392 }
5393
5394 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5395                                         ulong *spr_val)
5396 {
5397         return EMULATE_FAIL;
5398 }
5399
5400 static int kvmppc_core_check_processor_compat_hv(void)
5401 {
5402         if (cpu_has_feature(CPU_FTR_HVMODE) &&
5403             cpu_has_feature(CPU_FTR_ARCH_206))
5404                 return 0;
5405
5406         /* POWER9 in radix mode is capable of being a nested hypervisor. */
5407         if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5408                 return 0;
5409
5410         return -EIO;
5411 }
5412
5413 #ifdef CONFIG_KVM_XICS
5414
5415 void kvmppc_free_pimap(struct kvm *kvm)
5416 {
5417         kfree(kvm->arch.pimap);
5418 }
5419
5420 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5421 {
5422         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5423 }
5424
5425 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5426 {
5427         struct irq_desc *desc;
5428         struct kvmppc_irq_map *irq_map;
5429         struct kvmppc_passthru_irqmap *pimap;
5430         struct irq_chip *chip;
5431         int i, rc = 0;
5432         struct irq_data *host_data;
5433
5434         if (!kvm_irq_bypass)
5435                 return 1;
5436
5437         desc = irq_to_desc(host_irq);
5438         if (!desc)
5439                 return -EIO;
5440
5441         mutex_lock(&kvm->lock);
5442
5443         pimap = kvm->arch.pimap;
5444         if (pimap == NULL) {
5445                 /* First call, allocate structure to hold IRQ map */
5446                 pimap = kvmppc_alloc_pimap();
5447                 if (pimap == NULL) {
5448                         mutex_unlock(&kvm->lock);
5449                         return -ENOMEM;
5450                 }
5451                 kvm->arch.pimap = pimap;
5452         }
5453
5454         /*
5455          * For now, we only support interrupts for which the EOI operation
5456          * is an OPAL call followed by a write to XIRR, since that's
5457          * what our real-mode EOI code does, or a XIVE interrupt
5458          */
5459         chip = irq_data_get_irq_chip(&desc->irq_data);
5460         if (!chip || !is_pnv_opal_msi(chip)) {
5461                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5462                         host_irq, guest_gsi);
5463                 mutex_unlock(&kvm->lock);
5464                 return -ENOENT;
5465         }
5466
5467         /*
5468          * See if we already have an entry for this guest IRQ number.
5469          * If it's mapped to a hardware IRQ number, that's an error,
5470          * otherwise re-use this entry.
5471          */
5472         for (i = 0; i < pimap->n_mapped; i++) {
5473                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5474                         if (pimap->mapped[i].r_hwirq) {
5475                                 mutex_unlock(&kvm->lock);
5476                                 return -EINVAL;
5477                         }
5478                         break;
5479                 }
5480         }
5481
5482         if (i == KVMPPC_PIRQ_MAPPED) {
5483                 mutex_unlock(&kvm->lock);
5484                 return -EAGAIN;         /* table is full */
5485         }
5486
5487         irq_map = &pimap->mapped[i];
5488
5489         irq_map->v_hwirq = guest_gsi;
5490         irq_map->desc = desc;
5491
5492         /*
5493          * Order the above two stores before the next to serialize with
5494          * the KVM real mode handler.
5495          */
5496         smp_wmb();
5497
5498         /*
5499          * The 'host_irq' number is mapped in the PCI-MSI domain but
5500          * the underlying calls, which will EOI the interrupt in real
5501          * mode, need an HW IRQ number mapped in the XICS IRQ domain.
5502          */
5503         host_data = irq_domain_get_irq_data(irq_get_default_host(), host_irq);
5504         irq_map->r_hwirq = (unsigned int)irqd_to_hwirq(host_data);
5505
5506         if (i == pimap->n_mapped)
5507                 pimap->n_mapped++;
5508
5509         if (xics_on_xive())
5510                 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, host_irq);
5511         else
5512                 kvmppc_xics_set_mapped(kvm, guest_gsi, irq_map->r_hwirq);
5513         if (rc)
5514                 irq_map->r_hwirq = 0;
5515
5516         mutex_unlock(&kvm->lock);
5517
5518         return 0;
5519 }
5520
5521 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5522 {
5523         struct irq_desc *desc;
5524         struct kvmppc_passthru_irqmap *pimap;
5525         int i, rc = 0;
5526
5527         if (!kvm_irq_bypass)
5528                 return 0;
5529
5530         desc = irq_to_desc(host_irq);
5531         if (!desc)
5532                 return -EIO;
5533
5534         mutex_lock(&kvm->lock);
5535         if (!kvm->arch.pimap)
5536                 goto unlock;
5537
5538         pimap = kvm->arch.pimap;
5539
5540         for (i = 0; i < pimap->n_mapped; i++) {
5541                 if (guest_gsi == pimap->mapped[i].v_hwirq)
5542                         break;
5543         }
5544
5545         if (i == pimap->n_mapped) {
5546                 mutex_unlock(&kvm->lock);
5547                 return -ENODEV;
5548         }
5549
5550         if (xics_on_xive())
5551                 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, host_irq);
5552         else
5553                 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5554
5555         /* invalidate the entry (what do do on error from the above ?) */
5556         pimap->mapped[i].r_hwirq = 0;
5557
5558         /*
5559          * We don't free this structure even when the count goes to
5560          * zero. The structure is freed when we destroy the VM.
5561          */
5562  unlock:
5563         mutex_unlock(&kvm->lock);
5564         return rc;
5565 }
5566
5567 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5568                                              struct irq_bypass_producer *prod)
5569 {
5570         int ret = 0;
5571         struct kvm_kernel_irqfd *irqfd =
5572                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5573
5574         irqfd->producer = prod;
5575
5576         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5577         if (ret)
5578                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5579                         prod->irq, irqfd->gsi, ret);
5580
5581         return ret;
5582 }
5583
5584 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5585                                               struct irq_bypass_producer *prod)
5586 {
5587         int ret;
5588         struct kvm_kernel_irqfd *irqfd =
5589                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5590
5591         irqfd->producer = NULL;
5592
5593         /*
5594          * When producer of consumer is unregistered, we change back to
5595          * default external interrupt handling mode - KVM real mode
5596          * will switch back to host.
5597          */
5598         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5599         if (ret)
5600                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5601                         prod->irq, irqfd->gsi, ret);
5602 }
5603 #endif
5604
5605 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5606                                  unsigned int ioctl, unsigned long arg)
5607 {
5608         struct kvm *kvm __maybe_unused = filp->private_data;
5609         void __user *argp = (void __user *)arg;
5610         long r;
5611
5612         switch (ioctl) {
5613
5614         case KVM_PPC_ALLOCATE_HTAB: {
5615                 u32 htab_order;
5616
5617                 /* If we're a nested hypervisor, we currently only support radix */
5618                 if (kvmhv_on_pseries()) {
5619                         r = -EOPNOTSUPP;
5620                         break;
5621                 }
5622
5623                 r = -EFAULT;
5624                 if (get_user(htab_order, (u32 __user *)argp))
5625                         break;
5626                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5627                 if (r)
5628                         break;
5629                 r = 0;
5630                 break;
5631         }
5632
5633         case KVM_PPC_GET_HTAB_FD: {
5634                 struct kvm_get_htab_fd ghf;
5635
5636                 r = -EFAULT;
5637                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5638                         break;
5639                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5640                 break;
5641         }
5642
5643         case KVM_PPC_RESIZE_HPT_PREPARE: {
5644                 struct kvm_ppc_resize_hpt rhpt;
5645
5646                 r = -EFAULT;
5647                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5648                         break;
5649
5650                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5651                 break;
5652         }
5653
5654         case KVM_PPC_RESIZE_HPT_COMMIT: {
5655                 struct kvm_ppc_resize_hpt rhpt;
5656
5657                 r = -EFAULT;
5658                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5659                         break;
5660
5661                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5662                 break;
5663         }
5664
5665         default:
5666                 r = -ENOTTY;
5667         }
5668
5669         return r;
5670 }
5671
5672 /*
5673  * List of hcall numbers to enable by default.
5674  * For compatibility with old userspace, we enable by default
5675  * all hcalls that were implemented before the hcall-enabling
5676  * facility was added.  Note this list should not include H_RTAS.
5677  */
5678 static unsigned int default_hcall_list[] = {
5679         H_REMOVE,
5680         H_ENTER,
5681         H_READ,
5682         H_PROTECT,
5683         H_BULK_REMOVE,
5684 #ifdef CONFIG_SPAPR_TCE_IOMMU
5685         H_GET_TCE,
5686         H_PUT_TCE,
5687 #endif
5688         H_SET_DABR,
5689         H_SET_XDABR,
5690         H_CEDE,
5691         H_PROD,
5692         H_CONFER,
5693         H_REGISTER_VPA,
5694 #ifdef CONFIG_KVM_XICS
5695         H_EOI,
5696         H_CPPR,
5697         H_IPI,
5698         H_IPOLL,
5699         H_XIRR,
5700         H_XIRR_X,
5701 #endif
5702         0
5703 };
5704
5705 static void init_default_hcalls(void)
5706 {
5707         int i;
5708         unsigned int hcall;
5709
5710         for (i = 0; default_hcall_list[i]; ++i) {
5711                 hcall = default_hcall_list[i];
5712                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5713                 __set_bit(hcall / 4, default_enabled_hcalls);
5714         }
5715 }
5716
5717 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5718 {
5719         unsigned long lpcr;
5720         int radix;
5721         int err;
5722
5723         /* If not on a POWER9, reject it */
5724         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5725                 return -ENODEV;
5726
5727         /* If any unknown flags set, reject it */
5728         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5729                 return -EINVAL;
5730
5731         /* GR (guest radix) bit in process_table field must match */
5732         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5733         if (!!(cfg->process_table & PATB_GR) != radix)
5734                 return -EINVAL;
5735
5736         /* Process table size field must be reasonable, i.e. <= 24 */
5737         if ((cfg->process_table & PRTS_MASK) > 24)
5738                 return -EINVAL;
5739
5740         /* We can change a guest to/from radix now, if the host is radix */
5741         if (radix && !radix_enabled())
5742                 return -EINVAL;
5743
5744         /* If we're a nested hypervisor, we currently only support radix */
5745         if (kvmhv_on_pseries() && !radix)
5746                 return -EINVAL;
5747
5748         mutex_lock(&kvm->arch.mmu_setup_lock);
5749         if (radix != kvm_is_radix(kvm)) {
5750                 if (kvm->arch.mmu_ready) {
5751                         kvm->arch.mmu_ready = 0;
5752                         /* order mmu_ready vs. vcpus_running */
5753                         smp_mb();
5754                         if (atomic_read(&kvm->arch.vcpus_running)) {
5755                                 kvm->arch.mmu_ready = 1;
5756                                 err = -EBUSY;
5757                                 goto out_unlock;
5758                         }
5759                 }
5760                 if (radix)
5761                         err = kvmppc_switch_mmu_to_radix(kvm);
5762                 else
5763                         err = kvmppc_switch_mmu_to_hpt(kvm);
5764                 if (err)
5765                         goto out_unlock;
5766         }
5767
5768         kvm->arch.process_table = cfg->process_table;
5769         kvmppc_setup_partition_table(kvm);
5770
5771         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5772         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5773         err = 0;
5774
5775  out_unlock:
5776         mutex_unlock(&kvm->arch.mmu_setup_lock);
5777         return err;
5778 }
5779
5780 static int kvmhv_enable_nested(struct kvm *kvm)
5781 {
5782         if (!nested)
5783                 return -EPERM;
5784         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5785                 return -ENODEV;
5786         if (!radix_enabled())
5787                 return -ENODEV;
5788
5789         /* kvm == NULL means the caller is testing if the capability exists */
5790         if (kvm)
5791                 kvm->arch.nested_enable = true;
5792         return 0;
5793 }
5794
5795 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5796                                  int size)
5797 {
5798         int rc = -EINVAL;
5799
5800         if (kvmhv_vcpu_is_radix(vcpu)) {
5801                 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5802
5803                 if (rc > 0)
5804                         rc = -EINVAL;
5805         }
5806
5807         /* For now quadrants are the only way to access nested guest memory */
5808         if (rc && vcpu->arch.nested)
5809                 rc = -EAGAIN;
5810
5811         return rc;
5812 }
5813
5814 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5815                                 int size)
5816 {
5817         int rc = -EINVAL;
5818
5819         if (kvmhv_vcpu_is_radix(vcpu)) {
5820                 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5821
5822                 if (rc > 0)
5823                         rc = -EINVAL;
5824         }
5825
5826         /* For now quadrants are the only way to access nested guest memory */
5827         if (rc && vcpu->arch.nested)
5828                 rc = -EAGAIN;
5829
5830         return rc;
5831 }
5832
5833 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5834 {
5835         unpin_vpa(kvm, vpa);
5836         vpa->gpa = 0;
5837         vpa->pinned_addr = NULL;
5838         vpa->dirty = false;
5839         vpa->update_pending = 0;
5840 }
5841
5842 /*
5843  * Enable a guest to become a secure VM, or test whether
5844  * that could be enabled.
5845  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
5846  * tested (kvm == NULL) or enabled (kvm != NULL).
5847  */
5848 static int kvmhv_enable_svm(struct kvm *kvm)
5849 {
5850         if (!kvmppc_uvmem_available())
5851                 return -EINVAL;
5852         if (kvm)
5853                 kvm->arch.svm_enabled = 1;
5854         return 0;
5855 }
5856
5857 /*
5858  *  IOCTL handler to turn off secure mode of guest
5859  *
5860  * - Release all device pages
5861  * - Issue ucall to terminate the guest on the UV side
5862  * - Unpin the VPA pages.
5863  * - Reinit the partition scoped page tables
5864  */
5865 static int kvmhv_svm_off(struct kvm *kvm)
5866 {
5867         struct kvm_vcpu *vcpu;
5868         int mmu_was_ready;
5869         int srcu_idx;
5870         int ret = 0;
5871         int i;
5872
5873         if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5874                 return ret;
5875
5876         mutex_lock(&kvm->arch.mmu_setup_lock);
5877         mmu_was_ready = kvm->arch.mmu_ready;
5878         if (kvm->arch.mmu_ready) {
5879                 kvm->arch.mmu_ready = 0;
5880                 /* order mmu_ready vs. vcpus_running */
5881                 smp_mb();
5882                 if (atomic_read(&kvm->arch.vcpus_running)) {
5883                         kvm->arch.mmu_ready = 1;
5884                         ret = -EBUSY;
5885                         goto out;
5886                 }
5887         }
5888
5889         srcu_idx = srcu_read_lock(&kvm->srcu);
5890         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5891                 struct kvm_memory_slot *memslot;
5892                 struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5893
5894                 if (!slots)
5895                         continue;
5896
5897                 kvm_for_each_memslot(memslot, slots) {
5898                         kvmppc_uvmem_drop_pages(memslot, kvm, true);
5899                         uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5900                 }
5901         }
5902         srcu_read_unlock(&kvm->srcu, srcu_idx);
5903
5904         ret = uv_svm_terminate(kvm->arch.lpid);
5905         if (ret != U_SUCCESS) {
5906                 ret = -EINVAL;
5907                 goto out;
5908         }
5909
5910         /*
5911          * When secure guest is reset, all the guest pages are sent
5912          * to UV via UV_PAGE_IN before the non-boot vcpus get a
5913          * chance to run and unpin their VPA pages. Unpinning of all
5914          * VPA pages is done here explicitly so that VPA pages
5915          * can be migrated to the secure side.
5916          *
5917          * This is required to for the secure SMP guest to reboot
5918          * correctly.
5919          */
5920         kvm_for_each_vcpu(i, vcpu, kvm) {
5921                 spin_lock(&vcpu->arch.vpa_update_lock);
5922                 unpin_vpa_reset(kvm, &vcpu->arch.dtl);
5923                 unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
5924                 unpin_vpa_reset(kvm, &vcpu->arch.vpa);
5925                 spin_unlock(&vcpu->arch.vpa_update_lock);
5926         }
5927
5928         kvmppc_setup_partition_table(kvm);
5929         kvm->arch.secure_guest = 0;
5930         kvm->arch.mmu_ready = mmu_was_ready;
5931 out:
5932         mutex_unlock(&kvm->arch.mmu_setup_lock);
5933         return ret;
5934 }
5935
5936 static int kvmhv_enable_dawr1(struct kvm *kvm)
5937 {
5938         if (!cpu_has_feature(CPU_FTR_DAWR1))
5939                 return -ENODEV;
5940
5941         /* kvm == NULL means the caller is testing if the capability exists */
5942         if (kvm)
5943                 kvm->arch.dawr1_enabled = true;
5944         return 0;
5945 }
5946
5947 static bool kvmppc_hash_v3_possible(void)
5948 {
5949         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5950                 return false;
5951
5952         if (!cpu_has_feature(CPU_FTR_HVMODE))
5953                 return false;
5954
5955         /*
5956          * POWER9 chips before version 2.02 can't have some threads in
5957          * HPT mode and some in radix mode on the same core.
5958          */
5959         if (radix_enabled()) {
5960                 unsigned int pvr = mfspr(SPRN_PVR);
5961                 if ((pvr >> 16) == PVR_POWER9 &&
5962                     (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5963                      ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5964                         return false;
5965         }
5966
5967         return true;
5968 }
5969
5970 static struct kvmppc_ops kvm_ops_hv = {
5971         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5972         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5973         .get_one_reg = kvmppc_get_one_reg_hv,
5974         .set_one_reg = kvmppc_set_one_reg_hv,
5975         .vcpu_load   = kvmppc_core_vcpu_load_hv,
5976         .vcpu_put    = kvmppc_core_vcpu_put_hv,
5977         .inject_interrupt = kvmppc_inject_interrupt_hv,
5978         .set_msr     = kvmppc_set_msr_hv,
5979         .vcpu_run    = kvmppc_vcpu_run_hv,
5980         .vcpu_create = kvmppc_core_vcpu_create_hv,
5981         .vcpu_free   = kvmppc_core_vcpu_free_hv,
5982         .check_requests = kvmppc_core_check_requests_hv,
5983         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5984         .flush_memslot  = kvmppc_core_flush_memslot_hv,
5985         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5986         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5987         .unmap_gfn_range = kvm_unmap_gfn_range_hv,
5988         .age_gfn = kvm_age_gfn_hv,
5989         .test_age_gfn = kvm_test_age_gfn_hv,
5990         .set_spte_gfn = kvm_set_spte_gfn_hv,
5991         .free_memslot = kvmppc_core_free_memslot_hv,
5992         .init_vm =  kvmppc_core_init_vm_hv,
5993         .destroy_vm = kvmppc_core_destroy_vm_hv,
5994         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5995         .emulate_op = kvmppc_core_emulate_op_hv,
5996         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5997         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5998         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5999         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
6000         .hcall_implemented = kvmppc_hcall_impl_hv,
6001 #ifdef CONFIG_KVM_XICS
6002         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
6003         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
6004 #endif
6005         .configure_mmu = kvmhv_configure_mmu,
6006         .get_rmmu_info = kvmhv_get_rmmu_info,
6007         .set_smt_mode = kvmhv_set_smt_mode,
6008         .enable_nested = kvmhv_enable_nested,
6009         .load_from_eaddr = kvmhv_load_from_eaddr,
6010         .store_to_eaddr = kvmhv_store_to_eaddr,
6011         .enable_svm = kvmhv_enable_svm,
6012         .svm_off = kvmhv_svm_off,
6013         .enable_dawr1 = kvmhv_enable_dawr1,
6014         .hash_v3_possible = kvmppc_hash_v3_possible,
6015 };
6016
6017 static int kvm_init_subcore_bitmap(void)
6018 {
6019         int i, j;
6020         int nr_cores = cpu_nr_cores();
6021         struct sibling_subcore_state *sibling_subcore_state;
6022
6023         for (i = 0; i < nr_cores; i++) {
6024                 int first_cpu = i * threads_per_core;
6025                 int node = cpu_to_node(first_cpu);
6026
6027                 /* Ignore if it is already allocated. */
6028                 if (paca_ptrs[first_cpu]->sibling_subcore_state)
6029                         continue;
6030
6031                 sibling_subcore_state =
6032                         kzalloc_node(sizeof(struct sibling_subcore_state),
6033                                                         GFP_KERNEL, node);
6034                 if (!sibling_subcore_state)
6035                         return -ENOMEM;
6036
6037
6038                 for (j = 0; j < threads_per_core; j++) {
6039                         int cpu = first_cpu + j;
6040
6041                         paca_ptrs[cpu]->sibling_subcore_state =
6042                                                 sibling_subcore_state;
6043                 }
6044         }
6045         return 0;
6046 }
6047
6048 static int kvmppc_radix_possible(void)
6049 {
6050         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
6051 }
6052
6053 static int kvmppc_book3s_init_hv(void)
6054 {
6055         int r;
6056
6057         if (!tlbie_capable) {
6058                 pr_err("KVM-HV: Host does not support TLBIE\n");
6059                 return -ENODEV;
6060         }
6061
6062         /*
6063          * FIXME!! Do we need to check on all cpus ?
6064          */
6065         r = kvmppc_core_check_processor_compat_hv();
6066         if (r < 0)
6067                 return -ENODEV;
6068
6069         r = kvmhv_nested_init();
6070         if (r)
6071                 return r;
6072
6073         r = kvm_init_subcore_bitmap();
6074         if (r)
6075                 return r;
6076
6077         /*
6078          * We need a way of accessing the XICS interrupt controller,
6079          * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
6080          * indirectly, via OPAL.
6081          */
6082 #ifdef CONFIG_SMP
6083         if (!xics_on_xive() && !kvmhv_on_pseries() &&
6084             !local_paca->kvm_hstate.xics_phys) {
6085                 struct device_node *np;
6086
6087                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
6088                 if (!np) {
6089                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
6090                         return -ENODEV;
6091                 }
6092                 /* presence of intc confirmed - node can be dropped again */
6093                 of_node_put(np);
6094         }
6095 #endif
6096
6097         kvm_ops_hv.owner = THIS_MODULE;
6098         kvmppc_hv_ops = &kvm_ops_hv;
6099
6100         init_default_hcalls();
6101
6102         init_vcore_lists();
6103
6104         r = kvmppc_mmu_hv_init();
6105         if (r)
6106                 return r;
6107
6108         if (kvmppc_radix_possible()) {
6109                 r = kvmppc_radix_init();
6110                 if (r)
6111                         return r;
6112         }
6113
6114         r = kvmppc_uvmem_init();
6115         if (r < 0)
6116                 pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
6117
6118         return r;
6119 }
6120
6121 static void kvmppc_book3s_exit_hv(void)
6122 {
6123         kvmppc_uvmem_free();
6124         kvmppc_free_host_rm_ops();
6125         if (kvmppc_radix_possible())
6126                 kvmppc_radix_exit();
6127         kvmppc_hv_ops = NULL;
6128         kvmhv_nested_exit();
6129 }
6130
6131 module_init(kvmppc_book3s_init_hv);
6132 module_exit(kvmppc_book3s_exit_hv);
6133 MODULE_LICENSE("GPL");
6134 MODULE_ALIAS_MISCDEV(KVM_MINOR);
6135 MODULE_ALIAS("devname:kvm");