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/io.h>
  57 #include <asm/kvm_ppc.h>
  58 #include <asm/kvm_book3s.h>
  59 #include <asm/mmu_context.h>
  60 #include <asm/lppaca.h>
  61 #include <asm/pmc.h>
  62 #include <asm/processor.h>
  63 #include <asm/cputhreads.h>
  64 #include <asm/page.h>
  65 #include <asm/hvcall.h>
  66 #include <asm/switch_to.h>
  67 #include <asm/smp.h>
  68 #include <asm/dbell.h>
  69 #include <asm/hmi.h>
  70 #include <asm/pnv-pci.h>
  71 #include <asm/mmu.h>
  72 #include <asm/opal.h>
  73 #include <asm/xics.h>
  74 #include <asm/xive.h>
  75 #include <asm/hw_breakpoint.h>
  76 #include <asm/kvm_book3s_uvmem.h>
  77 #include <asm/ultravisor.h>
  78 #include <asm/dtl.h>
  79
  80 #include "book3s.h"
  81
  82 #define CREATE_TRACE_POINTS
  83 #include "trace_hv.h"
  84
  85 /* #define EXIT_DEBUG */
  86 /* #define EXIT_DEBUG_SIMPLE */
  87 /* #define EXIT_DEBUG_INT */
  88
  89 /* Used to indicate that a guest page fault needs to be handled */
  90 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  91 /* Used to indicate that a guest passthrough interrupt needs to be handled */
  92 #define RESUME_PASSTHROUGH      (RESUME_GUEST | RESUME_FLAG_ARCH2)
  93
  94 /* Used as a "null" value for timebase values */
  95 #define TB_NIL  (~(u64)0)
  96
  97 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
  98
  99 static int dynamic_mt_modes = 6;
 100 module_param(dynamic_mt_modes, int, 0644);
 101 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
 102 static int target_smt_mode;
 103 module_param(target_smt_mode, int, 0644);
 104 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
 105
 106 static bool indep_threads_mode = true;
 107 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
 108 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
 109
 110 static bool one_vm_per_core;
 111 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
 112 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)");
 113
 114 #ifdef CONFIG_KVM_XICS
 115 static const struct kernel_param_ops module_param_ops = {
 116         .set = param_set_int,
 117         .get = param_get_int,
 118 };
 119
 120 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
 121 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
 122
 123 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
 124 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
 125 #endif
 126
 127 /* If set, guests are allowed to create and control nested guests */
 128 static bool nested = true;
 129 module_param(nested, bool, S_IRUGO | S_IWUSR);
 130 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
 131
 132 static inline bool nesting_enabled(struct kvm *kvm)
 133 {
 134         return kvm->arch.nested_enable && kvm_is_radix(kvm);
 135 }
 136
 137 /* If set, the threads on each CPU core have to be in the same MMU mode */
 138 static bool no_mixing_hpt_and_radix;
 139
 140 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
 141
 142 /*
 143  * RWMR values for POWER8.  These control the rate at which PURR
 144  * and SPURR count and should be set according to the number of
 145  * online threads in the vcore being run.
 146  */
 147 #define RWMR_RPA_P8_1THREAD     0x164520C62609AECAUL
 148 #define RWMR_RPA_P8_2THREAD     0x7FFF2908450D8DA9UL
 149 #define RWMR_RPA_P8_3THREAD     0x164520C62609AECAUL
 150 #define RWMR_RPA_P8_4THREAD     0x199A421245058DA9UL
 151 #define RWMR_RPA_P8_5THREAD     0x164520C62609AECAUL
 152 #define RWMR_RPA_P8_6THREAD     0x164520C62609AECAUL
 153 #define RWMR_RPA_P8_7THREAD     0x164520C62609AECAUL
 154 #define RWMR_RPA_P8_8THREAD     0x164520C62609AECAUL
 155
 156 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
 157         RWMR_RPA_P8_1THREAD,
 158         RWMR_RPA_P8_1THREAD,
 159         RWMR_RPA_P8_2THREAD,
 160         RWMR_RPA_P8_3THREAD,
 161         RWMR_RPA_P8_4THREAD,
 162         RWMR_RPA_P8_5THREAD,
 163         RWMR_RPA_P8_6THREAD,
 164         RWMR_RPA_P8_7THREAD,
 165         RWMR_RPA_P8_8THREAD,
 166 };
 167
 168 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
 169                 int *ip)
 170 {
 171         int i = *ip;
 172         struct kvm_vcpu *vcpu;
 173
 174         while (++i < MAX_SMT_THREADS) {
 175                 vcpu = READ_ONCE(vc->runnable_threads[i]);
 176                 if (vcpu) {
 177                         *ip = i;
 178                         return vcpu;
 179                 }
 180         }
 181         return NULL;
 182 }
 183
 184 /* Used to traverse the list of runnable threads for a given vcore */
 185 #define for_each_runnable_thread(i, vcpu, vc) \
 186         for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
 187
 188 static bool kvmppc_ipi_thread(int cpu)
 189 {
 190         unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
 191
 192         /* If we're a nested hypervisor, fall back to ordinary IPIs for now */
 193         if (kvmhv_on_pseries())
 194                 return false;
 195
 196         /* On POWER9 we can use msgsnd to IPI any cpu */
 197         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
 198                 msg |= get_hard_smp_processor_id(cpu);
 199                 smp_mb();
 200                 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 201                 return true;
 202         }
 203
 204         /* On POWER8 for IPIs to threads in the same core, use msgsnd */
 205         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
 206                 preempt_disable();
 207                 if (cpu_first_thread_sibling(cpu) ==
 208                     cpu_first_thread_sibling(smp_processor_id())) {
 209                         msg |= cpu_thread_in_core(cpu);
 210                         smp_mb();
 211                         __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 212                         preempt_enable();
 213                         return true;
 214                 }
 215                 preempt_enable();
 216         }
 217
 218 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
 219         if (cpu >= 0 && cpu < nr_cpu_ids) {
 220                 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
 221                         xics_wake_cpu(cpu);
 222                         return true;
 223                 }
 224                 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
 225                 return true;
 226         }
 227 #endif
 228
 229         return false;
 230 }
 231
 232 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
 233 {
 234         int cpu;
 235         struct rcuwait *waitp;
 236
 237         waitp = kvm_arch_vcpu_get_wait(vcpu);
 238         if (rcuwait_wake_up(waitp))
 239                 ++vcpu->stat.halt_wakeup;
 240
 241         cpu = READ_ONCE(vcpu->arch.thread_cpu);
 242         if (cpu >= 0 && kvmppc_ipi_thread(cpu))
 243                 return;
 244
 245         /* CPU points to the first thread of the core */
 246         cpu = vcpu->cpu;
 247         if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
 248                 smp_send_reschedule(cpu);
 249 }
 250
 251 /*
 252  * We use the vcpu_load/put functions to measure stolen time.
 253  * Stolen time is counted as time when either the vcpu is able to
 254  * run as part of a virtual core, but the task running the vcore
 255  * is preempted or sleeping, or when the vcpu needs something done
 256  * in the kernel by the task running the vcpu, but that task is
 257  * preempted or sleeping.  Those two things have to be counted
 258  * separately, since one of the vcpu tasks will take on the job
 259  * of running the core, and the other vcpu tasks in the vcore will
 260  * sleep waiting for it to do that, but that sleep shouldn't count
 261  * as stolen time.
 262  *
 263  * Hence we accumulate stolen time when the vcpu can run as part of
 264  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 265  * needs its task to do other things in the kernel (for example,
 266  * service a page fault) in busy_stolen.  We don't accumulate
 267  * stolen time for a vcore when it is inactive, or for a vcpu
 268  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 269  * a misnomer; it means that the vcpu task is not executing in
 270  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 271  * the kernel.  We don't have any way of dividing up that time
 272  * between time that the vcpu is genuinely stopped, time that
 273  * the task is actively working on behalf of the vcpu, and time
 274  * that the task is preempted, so we don't count any of it as
 275  * stolen.
 276  *
 277  * Updates to busy_stolen are protected by arch.tbacct_lock;
 278  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
 279  * lock.  The stolen times are measured in units of timebase ticks.
 280  * (Note that the != TB_NIL checks below are purely defensive;
 281  * they should never fail.)
 282  */
 283
 284 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
 285 {
 286         unsigned long flags;
 287
 288         spin_lock_irqsave(&vc->stoltb_lock, flags);
 289         vc->preempt_tb = mftb();
 290         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 291 }
 292
 293 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
 294 {
 295         unsigned long flags;
 296
 297         spin_lock_irqsave(&vc->stoltb_lock, flags);
 298         if (vc->preempt_tb != TB_NIL) {
 299                 vc->stolen_tb += mftb() - vc->preempt_tb;
 300                 vc->preempt_tb = TB_NIL;
 301         }
 302         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 303 }
 304
 305 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 306 {
 307         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 308         unsigned long flags;
 309
 310         /*
 311          * We can test vc->runner without taking the vcore lock,
 312          * because only this task ever sets vc->runner to this
 313          * vcpu, and once it is set to this vcpu, only this task
 314          * ever sets it to NULL.
 315          */
 316         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 317                 kvmppc_core_end_stolen(vc);
 318
 319         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 320         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 321             vcpu->arch.busy_preempt != TB_NIL) {
 322                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 323                 vcpu->arch.busy_preempt = TB_NIL;
 324         }
 325         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 326 }
 327
 328 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 329 {
 330         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 331         unsigned long flags;
 332
 333         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 334                 kvmppc_core_start_stolen(vc);
 335
 336         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 337         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 338                 vcpu->arch.busy_preempt = mftb();
 339         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 340 }
 341
 342 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 343 {
 344         vcpu->arch.pvr = pvr;
 345 }
 346
 347 /* Dummy value used in computing PCR value below */
 348 #define PCR_ARCH_31    (PCR_ARCH_300 << 1)
 349
 350 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 351 {
 352         unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
 353         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 354
 355         /* We can (emulate) our own architecture version and anything older */
 356         if (cpu_has_feature(CPU_FTR_ARCH_31))
 357                 host_pcr_bit = PCR_ARCH_31;
 358         else if (cpu_has_feature(CPU_FTR_ARCH_300))
 359                 host_pcr_bit = PCR_ARCH_300;
 360         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
 361                 host_pcr_bit = PCR_ARCH_207;
 362         else if (cpu_has_feature(CPU_FTR_ARCH_206))
 363                 host_pcr_bit = PCR_ARCH_206;
 364         else
 365                 host_pcr_bit = PCR_ARCH_205;
 366
 367         /* Determine lowest PCR bit needed to run guest in given PVR level */
 368         guest_pcr_bit = host_pcr_bit;
 369         if (arch_compat) {
 370                 switch (arch_compat) {
 371                 case PVR_ARCH_205:
 372                         guest_pcr_bit = PCR_ARCH_205;
 373                         break;
 374                 case PVR_ARCH_206:
 375                 case PVR_ARCH_206p:
 376                         guest_pcr_bit = PCR_ARCH_206;
 377                         break;
 378                 case PVR_ARCH_207:
 379                         guest_pcr_bit = PCR_ARCH_207;
 380                         break;
 381                 case PVR_ARCH_300:
 382                         guest_pcr_bit = PCR_ARCH_300;
 383                         break;
 384                 case PVR_ARCH_31:
 385                         guest_pcr_bit = PCR_ARCH_31;
 386                         break;
 387                 default:
 388                         return -EINVAL;
 389                 }
 390         }
 391
 392         /* Check requested PCR bits don't exceed our capabilities */
 393         if (guest_pcr_bit > host_pcr_bit)
 394                 return -EINVAL;
 395
 396         spin_lock(&vc->lock);
 397         vc->arch_compat = arch_compat;
 398         /*
 399          * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
 400          * Also set all reserved PCR bits
 401          */
 402         vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
 403         spin_unlock(&vc->lock);
 404
 405         return 0;
 406 }
 407
 408 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 409 {
 410         int r;
 411
 412         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 413         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 414                vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
 415         for (r = 0; r < 16; ++r)
 416                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 417                        r, kvmppc_get_gpr(vcpu, r),
 418                        r+16, kvmppc_get_gpr(vcpu, r+16));
 419         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 420                vcpu->arch.regs.ctr, vcpu->arch.regs.link);
 421         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 422                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 423         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 424                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 425         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 426                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 427         pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
 428                vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
 429         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 430         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 431                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 432         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 433         for (r = 0; r < vcpu->arch.slb_max; ++r)
 434                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 435                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 436         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 437                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 438                vcpu->arch.last_inst);
 439 }
 440
 441 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 442 {
 443         return kvm_get_vcpu_by_id(kvm, id);
 444 }
 445
 446 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 447 {
 448         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 449         vpa->yield_count = cpu_to_be32(1);
 450 }
 451
 452 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 453                    unsigned long addr, unsigned long len)
 454 {
 455         /* check address is cacheline aligned */
 456         if (addr & (L1_CACHE_BYTES - 1))
 457                 return -EINVAL;
 458         spin_lock(&vcpu->arch.vpa_update_lock);
 459         if (v->next_gpa != addr || v->len != len) {
 460                 v->next_gpa = addr;
 461                 v->len = addr ? len : 0;
 462                 v->update_pending = 1;
 463         }
 464         spin_unlock(&vcpu->arch.vpa_update_lock);
 465         return 0;
 466 }
 467
 468 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 469 struct reg_vpa {
 470         u32 dummy;
 471         union {
 472                 __be16 hword;
 473                 __be32 word;
 474         } length;
 475 };
 476
 477 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 478 {
 479         if (vpap->update_pending)
 480                 return vpap->next_gpa != 0;
 481         return vpap->pinned_addr != NULL;
 482 }
 483
 484 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 485                                        unsigned long flags,
 486                                        unsigned long vcpuid, unsigned long vpa)
 487 {
 488         struct kvm *kvm = vcpu->kvm;
 489         unsigned long len, nb;
 490         void *va;
 491         struct kvm_vcpu *tvcpu;
 492         int err;
 493         int subfunc;
 494         struct kvmppc_vpa *vpap;
 495
 496         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 497         if (!tvcpu)
 498                 return H_PARAMETER;
 499
 500         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 501         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 502             subfunc == H_VPA_REG_SLB) {
 503                 /* Registering new area - address must be cache-line aligned */
 504                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 505                         return H_PARAMETER;
 506
 507                 /* convert logical addr to kernel addr and read length */
 508                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 509                 if (va == NULL)
 510                         return H_PARAMETER;
 511                 if (subfunc == H_VPA_REG_VPA)
 512                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 513                 else
 514                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 515                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 516
 517                 /* Check length */
 518                 if (len > nb || len < sizeof(struct reg_vpa))
 519                         return H_PARAMETER;
 520         } else {
 521                 vpa = 0;
 522                 len = 0;
 523         }
 524
 525         err = H_PARAMETER;
 526         vpap = NULL;
 527         spin_lock(&tvcpu->arch.vpa_update_lock);
 528
 529         switch (subfunc) {
 530         case H_VPA_REG_VPA:             /* register VPA */
 531                 /*
 532                  * The size of our lppaca is 1kB because of the way we align
 533                  * it for the guest to avoid crossing a 4kB boundary. We only
 534                  * use 640 bytes of the structure though, so we should accept
 535                  * clients that set a size of 640.
 536                  */
 537                 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
 538                 if (len < sizeof(struct lppaca))
 539                         break;
 540                 vpap = &tvcpu->arch.vpa;
 541                 err = 0;
 542                 break;
 543
 544         case H_VPA_REG_DTL:             /* register DTL */
 545                 if (len < sizeof(struct dtl_entry))
 546                         break;
 547                 len -= len % sizeof(struct dtl_entry);
 548
 549                 /* Check that they have previously registered a VPA */
 550                 err = H_RESOURCE;
 551                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 552                         break;
 553
 554                 vpap = &tvcpu->arch.dtl;
 555                 err = 0;
 556                 break;
 557
 558         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 559                 /* Check that they have previously registered a VPA */
 560                 err = H_RESOURCE;
 561                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 562                         break;
 563
 564                 vpap = &tvcpu->arch.slb_shadow;
 565                 err = 0;
 566                 break;
 567
 568         case H_VPA_DEREG_VPA:           /* deregister VPA */
 569                 /* Check they don't still have a DTL or SLB buf registered */
 570                 err = H_RESOURCE;
 571                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 572                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 573                         break;
 574
 575                 vpap = &tvcpu->arch.vpa;
 576                 err = 0;
 577                 break;
 578
 579         case H_VPA_DEREG_DTL:           /* deregister DTL */
 580                 vpap = &tvcpu->arch.dtl;
 581                 err = 0;
 582                 break;
 583
 584         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 585                 vpap = &tvcpu->arch.slb_shadow;
 586                 err = 0;
 587                 break;
 588         }
 589
 590         if (vpap) {
 591                 vpap->next_gpa = vpa;
 592                 vpap->len = len;
 593                 vpap->update_pending = 1;
 594         }
 595
 596         spin_unlock(&tvcpu->arch.vpa_update_lock);
 597
 598         return err;
 599 }
 600
 601 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 602 {
 603         struct kvm *kvm = vcpu->kvm;
 604         void *va;
 605         unsigned long nb;
 606         unsigned long gpa;
 607
 608         /*
 609          * We need to pin the page pointed to by vpap->next_gpa,
 610          * but we can't call kvmppc_pin_guest_page under the lock
 611          * as it does get_user_pages() and down_read().  So we
 612          * have to drop the lock, pin the page, then get the lock
 613          * again and check that a new area didn't get registered
 614          * in the meantime.
 615          */
 616         for (;;) {
 617                 gpa = vpap->next_gpa;
 618                 spin_unlock(&vcpu->arch.vpa_update_lock);
 619                 va = NULL;
 620                 nb = 0;
 621                 if (gpa)
 622                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 623                 spin_lock(&vcpu->arch.vpa_update_lock);
 624                 if (gpa == vpap->next_gpa)
 625                         break;
 626                 /* sigh... unpin that one and try again */
 627                 if (va)
 628                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 629         }
 630
 631         vpap->update_pending = 0;
 632         if (va && nb < vpap->len) {
 633                 /*
 634                  * If it's now too short, it must be that userspace
 635                  * has changed the mappings underlying guest memory,
 636                  * so unregister the region.
 637                  */
 638                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 639                 va = NULL;
 640         }
 641         if (vpap->pinned_addr)
 642                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 643                                         vpap->dirty);
 644         vpap->gpa = gpa;
 645         vpap->pinned_addr = va;
 646         vpap->dirty = false;
 647         if (va)
 648                 vpap->pinned_end = va + vpap->len;
 649 }
 650
 651 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 652 {
 653         if (!(vcpu->arch.vpa.update_pending ||
 654               vcpu->arch.slb_shadow.update_pending ||
 655               vcpu->arch.dtl.update_pending))
 656                 return;
 657
 658         spin_lock(&vcpu->arch.vpa_update_lock);
 659         if (vcpu->arch.vpa.update_pending) {
 660                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 661                 if (vcpu->arch.vpa.pinned_addr)
 662                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 663         }
 664         if (vcpu->arch.dtl.update_pending) {
 665                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 666                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 667                 vcpu->arch.dtl_index = 0;
 668         }
 669         if (vcpu->arch.slb_shadow.update_pending)
 670                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 671         spin_unlock(&vcpu->arch.vpa_update_lock);
 672 }
 673
 674 /*
 675  * Return the accumulated stolen time for the vcore up until `now'.
 676  * The caller should hold the vcore lock.
 677  */
 678 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 679 {
 680         u64 p;
 681         unsigned long flags;
 682
 683         spin_lock_irqsave(&vc->stoltb_lock, flags);
 684         p = vc->stolen_tb;
 685         if (vc->vcore_state != VCORE_INACTIVE &&
 686             vc->preempt_tb != TB_NIL)
 687                 p += now - vc->preempt_tb;
 688         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 689         return p;
 690 }
 691
 692 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 693                                     struct kvmppc_vcore *vc)
 694 {
 695         struct dtl_entry *dt;
 696         struct lppaca *vpa;
 697         unsigned long stolen;
 698         unsigned long core_stolen;
 699         u64 now;
 700         unsigned long flags;
 701
 702         dt = vcpu->arch.dtl_ptr;
 703         vpa = vcpu->arch.vpa.pinned_addr;
 704         now = mftb();
 705         core_stolen = vcore_stolen_time(vc, now);
 706         stolen = core_stolen - vcpu->arch.stolen_logged;
 707         vcpu->arch.stolen_logged = core_stolen;
 708         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 709         stolen += vcpu->arch.busy_stolen;
 710         vcpu->arch.busy_stolen = 0;
 711         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 712         if (!dt || !vpa)
 713                 return;
 714         memset(dt, 0, sizeof(struct dtl_entry));
 715         dt->dispatch_reason = 7;
 716         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 717         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 718         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 719         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 720         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 721         ++dt;
 722         if (dt == vcpu->arch.dtl.pinned_end)
 723                 dt = vcpu->arch.dtl.pinned_addr;
 724         vcpu->arch.dtl_ptr = dt;
 725         /* order writing *dt vs. writing vpa->dtl_idx */
 726         smp_wmb();
 727         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 728         vcpu->arch.dtl.dirty = true;
 729 }
 730
 731 /* See if there is a doorbell interrupt pending for a vcpu */
 732 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
 733 {
 734         int thr;
 735         struct kvmppc_vcore *vc;
 736
 737         if (vcpu->arch.doorbell_request)
 738                 return true;
 739         /*
 740          * Ensure that the read of vcore->dpdes comes after the read
 741          * of vcpu->doorbell_request.  This barrier matches the
 742          * smp_wmb() in kvmppc_guest_entry_inject().
 743          */
 744         smp_rmb();
 745         vc = vcpu->arch.vcore;
 746         thr = vcpu->vcpu_id - vc->first_vcpuid;
 747         return !!(vc->dpdes & (1 << thr));
 748 }
 749
 750 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 751 {
 752         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 753                 return true;
 754         if ((!vcpu->arch.vcore->arch_compat) &&
 755             cpu_has_feature(CPU_FTR_ARCH_207S))
 756                 return true;
 757         return false;
 758 }
 759
 760 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 761                              unsigned long resource, unsigned long value1,
 762                              unsigned long value2)
 763 {
 764         switch (resource) {
 765         case H_SET_MODE_RESOURCE_SET_CIABR:
 766                 if (!kvmppc_power8_compatible(vcpu))
 767                         return H_P2;
 768                 if (value2)
 769                         return H_P4;
 770                 if (mflags)
 771                         return H_UNSUPPORTED_FLAG_START;
 772                 /* Guests can't breakpoint the hypervisor */
 773                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 774                         return H_P3;
 775                 vcpu->arch.ciabr  = value1;
 776                 return H_SUCCESS;
 777         case H_SET_MODE_RESOURCE_SET_DAWR0:
 778                 if (!kvmppc_power8_compatible(vcpu))
 779                         return H_P2;
 780                 if (!ppc_breakpoint_available())
 781                         return H_P2;
 782                 if (mflags)
 783                         return H_UNSUPPORTED_FLAG_START;
 784                 if (value2 & DABRX_HYP)
 785                         return H_P4;
 786                 vcpu->arch.dawr  = value1;
 787                 vcpu->arch.dawrx = value2;
 788                 return H_SUCCESS;
 789         case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
 790                 /* KVM does not support mflags=2 (AIL=2) */
 791                 if (mflags != 0 && mflags != 3)
 792                         return H_UNSUPPORTED_FLAG_START;
 793                 return H_TOO_HARD;
 794         default:
 795                 return H_TOO_HARD;
 796         }
 797 }
 798
 799 /* Copy guest memory in place - must reside within a single memslot */
 800 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
 801                                   unsigned long len)
 802 {
 803         struct kvm_memory_slot *to_memslot = NULL;
 804         struct kvm_memory_slot *from_memslot = NULL;
 805         unsigned long to_addr, from_addr;
 806         int r;
 807
 808         /* Get HPA for from address */
 809         from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
 810         if (!from_memslot)
 811                 return -EFAULT;
 812         if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
 813                              << PAGE_SHIFT))
 814                 return -EINVAL;
 815         from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
 816         if (kvm_is_error_hva(from_addr))
 817                 return -EFAULT;
 818         from_addr |= (from & (PAGE_SIZE - 1));
 819
 820         /* Get HPA for to address */
 821         to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
 822         if (!to_memslot)
 823                 return -EFAULT;
 824         if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
 825                            << PAGE_SHIFT))
 826                 return -EINVAL;
 827         to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
 828         if (kvm_is_error_hva(to_addr))
 829                 return -EFAULT;
 830         to_addr |= (to & (PAGE_SIZE - 1));
 831
 832         /* Perform copy */
 833         r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
 834                              len);
 835         if (r)
 836                 return -EFAULT;
 837         mark_page_dirty(kvm, to >> PAGE_SHIFT);
 838         return 0;
 839 }
 840
 841 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
 842                                unsigned long dest, unsigned long src)
 843 {
 844         u64 pg_sz = SZ_4K;              /* 4K page size */
 845         u64 pg_mask = SZ_4K - 1;
 846         int ret;
 847
 848         /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
 849         if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
 850                       H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
 851                 return H_PARAMETER;
 852
 853         /* dest (and src if copy_page flag set) must be page aligned */
 854         if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
 855                 return H_PARAMETER;
 856
 857         /* zero and/or copy the page as determined by the flags */
 858         if (flags & H_COPY_PAGE) {
 859                 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
 860                 if (ret < 0)
 861                         return H_PARAMETER;
 862         } else if (flags & H_ZERO_PAGE) {
 863                 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
 864                 if (ret < 0)
 865                         return H_PARAMETER;
 866         }
 867
 868         /* We can ignore the remaining flags */
 869
 870         return H_SUCCESS;
 871 }
 872
 873 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 874 {
 875         struct kvmppc_vcore *vcore = target->arch.vcore;
 876
 877         /*
 878          * We expect to have been called by the real mode handler
 879          * (kvmppc_rm_h_confer()) which would have directly returned
 880          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 881          * have useful work to do and should not confer) so we don't
 882          * recheck that here.
 883          */
 884
 885         spin_lock(&vcore->lock);
 886         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 887             vcore->vcore_state != VCORE_INACTIVE &&
 888             vcore->runner)
 889                 target = vcore->runner;
 890         spin_unlock(&vcore->lock);
 891
 892         return kvm_vcpu_yield_to(target);
 893 }
 894
 895 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 896 {
 897         int yield_count = 0;
 898         struct lppaca *lppaca;
 899
 900         spin_lock(&vcpu->arch.vpa_update_lock);
 901         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 902         if (lppaca)
 903                 yield_count = be32_to_cpu(lppaca->yield_count);
 904         spin_unlock(&vcpu->arch.vpa_update_lock);
 905         return yield_count;
 906 }
 907
 908 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 909 {
 910         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 911         unsigned long target, ret = H_SUCCESS;
 912         int yield_count;
 913         struct kvm_vcpu *tvcpu;
 914         int idx, rc;
 915
 916         if (req <= MAX_HCALL_OPCODE &&
 917             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 918                 return RESUME_HOST;
 919
 920         switch (req) {
 921         case H_CEDE:
 922                 break;
 923         case H_PROD:
 924                 target = kvmppc_get_gpr(vcpu, 4);
 925                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 926                 if (!tvcpu) {
 927                         ret = H_PARAMETER;
 928                         break;
 929                 }
 930                 tvcpu->arch.prodded = 1;
 931                 smp_mb();
 932                 if (tvcpu->arch.ceded)
 933                         kvmppc_fast_vcpu_kick_hv(tvcpu);
 934                 break;
 935         case H_CONFER:
 936                 target = kvmppc_get_gpr(vcpu, 4);
 937                 if (target == -1)
 938                         break;
 939                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 940                 if (!tvcpu) {
 941                         ret = H_PARAMETER;
 942                         break;
 943                 }
 944                 yield_count = kvmppc_get_gpr(vcpu, 5);
 945                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 946                         break;
 947                 kvm_arch_vcpu_yield_to(tvcpu);
 948                 break;
 949         case H_REGISTER_VPA:
 950                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 951                                         kvmppc_get_gpr(vcpu, 5),
 952                                         kvmppc_get_gpr(vcpu, 6));
 953                 break;
 954         case H_RTAS:
 955                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 956                         return RESUME_HOST;
 957
 958                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 959                 rc = kvmppc_rtas_hcall(vcpu);
 960                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 961
 962                 if (rc == -ENOENT)
 963                         return RESUME_HOST;
 964                 else if (rc == 0)
 965                         break;
 966
 967                 /* Send the error out to userspace via KVM_RUN */
 968                 return rc;
 969         case H_LOGICAL_CI_LOAD:
 970                 ret = kvmppc_h_logical_ci_load(vcpu);
 971                 if (ret == H_TOO_HARD)
 972                         return RESUME_HOST;
 973                 break;
 974         case H_LOGICAL_CI_STORE:
 975                 ret = kvmppc_h_logical_ci_store(vcpu);
 976                 if (ret == H_TOO_HARD)
 977                         return RESUME_HOST;
 978                 break;
 979         case H_SET_MODE:
 980                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 981                                         kvmppc_get_gpr(vcpu, 5),
 982                                         kvmppc_get_gpr(vcpu, 6),
 983                                         kvmppc_get_gpr(vcpu, 7));
 984                 if (ret == H_TOO_HARD)
 985                         return RESUME_HOST;
 986                 break;
 987         case H_XIRR:
 988         case H_CPPR:
 989         case H_EOI:
 990         case H_IPI:
 991         case H_IPOLL:
 992         case H_XIRR_X:
 993                 if (kvmppc_xics_enabled(vcpu)) {
 994                         if (xics_on_xive()) {
 995                                 ret = H_NOT_AVAILABLE;
 996                                 return RESUME_GUEST;
 997                         }
 998                         ret = kvmppc_xics_hcall(vcpu, req);
 999                         break;
1000                 }
1001                 return RESUME_HOST;
1002         case H_SET_DABR:
1003                 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1004                 break;
1005         case H_SET_XDABR:
1006                 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1007                                                 kvmppc_get_gpr(vcpu, 5));
1008                 break;
1009 #ifdef CONFIG_SPAPR_TCE_IOMMU
1010         case H_GET_TCE:
1011                 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1012                                                 kvmppc_get_gpr(vcpu, 5));
1013                 if (ret == H_TOO_HARD)
1014                         return RESUME_HOST;
1015                 break;
1016         case H_PUT_TCE:
1017                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1018                                                 kvmppc_get_gpr(vcpu, 5),
1019                                                 kvmppc_get_gpr(vcpu, 6));
1020                 if (ret == H_TOO_HARD)
1021                         return RESUME_HOST;
1022                 break;
1023         case H_PUT_TCE_INDIRECT:
1024                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1025                                                 kvmppc_get_gpr(vcpu, 5),
1026                                                 kvmppc_get_gpr(vcpu, 6),
1027                                                 kvmppc_get_gpr(vcpu, 7));
1028                 if (ret == H_TOO_HARD)
1029                         return RESUME_HOST;
1030                 break;
1031         case H_STUFF_TCE:
1032                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1033                                                 kvmppc_get_gpr(vcpu, 5),
1034                                                 kvmppc_get_gpr(vcpu, 6),
1035                                                 kvmppc_get_gpr(vcpu, 7));
1036                 if (ret == H_TOO_HARD)
1037                         return RESUME_HOST;
1038                 break;
1039 #endif
1040         case H_RANDOM:
1041                 if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1042                         ret = H_HARDWARE;
1043                 break;
1044
1045         case H_SET_PARTITION_TABLE:
1046                 ret = H_FUNCTION;
1047                 if (nesting_enabled(vcpu->kvm))
1048                         ret = kvmhv_set_partition_table(vcpu);
1049                 break;
1050         case H_ENTER_NESTED:
1051                 ret = H_FUNCTION;
1052                 if (!nesting_enabled(vcpu->kvm))
1053                         break;
1054                 ret = kvmhv_enter_nested_guest(vcpu);
1055                 if (ret == H_INTERRUPT) {
1056                         kvmppc_set_gpr(vcpu, 3, 0);
1057                         vcpu->arch.hcall_needed = 0;
1058                         return -EINTR;
1059                 } else if (ret == H_TOO_HARD) {
1060                         kvmppc_set_gpr(vcpu, 3, 0);
1061                         vcpu->arch.hcall_needed = 0;
1062                         return RESUME_HOST;
1063                 }
1064                 break;
1065         case H_TLB_INVALIDATE:
1066                 ret = H_FUNCTION;
1067                 if (nesting_enabled(vcpu->kvm))
1068                         ret = kvmhv_do_nested_tlbie(vcpu);
1069                 break;
1070         case H_COPY_TOFROM_GUEST:
1071                 ret = H_FUNCTION;
1072                 if (nesting_enabled(vcpu->kvm))
1073                         ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1074                 break;
1075         case H_PAGE_INIT:
1076                 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1077                                          kvmppc_get_gpr(vcpu, 5),
1078                                          kvmppc_get_gpr(vcpu, 6));
1079                 break;
1080         case H_SVM_PAGE_IN:
1081                 ret = H_UNSUPPORTED;
1082                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1083                         ret = kvmppc_h_svm_page_in(vcpu->kvm,
1084                                                    kvmppc_get_gpr(vcpu, 4),
1085                                                    kvmppc_get_gpr(vcpu, 5),
1086                                                    kvmppc_get_gpr(vcpu, 6));
1087                 break;
1088         case H_SVM_PAGE_OUT:
1089                 ret = H_UNSUPPORTED;
1090                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1091                         ret = kvmppc_h_svm_page_out(vcpu->kvm,
1092                                                     kvmppc_get_gpr(vcpu, 4),
1093                                                     kvmppc_get_gpr(vcpu, 5),
1094                                                     kvmppc_get_gpr(vcpu, 6));
1095                 break;
1096         case H_SVM_INIT_START:
1097                 ret = H_UNSUPPORTED;
1098                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1099                         ret = kvmppc_h_svm_init_start(vcpu->kvm);
1100                 break;
1101         case H_SVM_INIT_DONE:
1102                 ret = H_UNSUPPORTED;
1103                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1104                         ret = kvmppc_h_svm_init_done(vcpu->kvm);
1105                 break;
1106         case H_SVM_INIT_ABORT:
1107                 /*
1108                  * Even if that call is made by the Ultravisor, the SSR1 value
1109                  * is the guest context one, with the secure bit clear as it has
1110                  * not yet been secured. So we can't check it here.
1111                  * Instead the kvm->arch.secure_guest flag is checked inside
1112                  * kvmppc_h_svm_init_abort().
1113                  */
1114                 ret = kvmppc_h_svm_init_abort(vcpu->kvm);
1115                 break;
1116
1117         default:
1118                 return RESUME_HOST;
1119         }
1120         kvmppc_set_gpr(vcpu, 3, ret);
1121         vcpu->arch.hcall_needed = 0;
1122         return RESUME_GUEST;
1123 }
1124
1125 /*
1126  * Handle H_CEDE in the nested virtualization case where we haven't
1127  * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1128  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1129  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1130  */
1131 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1132 {
1133         vcpu->arch.shregs.msr |= MSR_EE;
1134         vcpu->arch.ceded = 1;
1135         smp_mb();
1136         if (vcpu->arch.prodded) {
1137                 vcpu->arch.prodded = 0;
1138                 smp_mb();
1139                 vcpu->arch.ceded = 0;
1140         }
1141 }
1142
1143 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1144 {
1145         switch (cmd) {
1146         case H_CEDE:
1147         case H_PROD:
1148         case H_CONFER:
1149         case H_REGISTER_VPA:
1150         case H_SET_MODE:
1151         case H_LOGICAL_CI_LOAD:
1152         case H_LOGICAL_CI_STORE:
1153 #ifdef CONFIG_KVM_XICS
1154         case H_XIRR:
1155         case H_CPPR:
1156         case H_EOI:
1157         case H_IPI:
1158         case H_IPOLL:
1159         case H_XIRR_X:
1160 #endif
1161         case H_PAGE_INIT:
1162                 return 1;
1163         }
1164
1165         /* See if it's in the real-mode table */
1166         return kvmppc_hcall_impl_hv_realmode(cmd);
1167 }
1168
1169 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1170 {
1171         u32 last_inst;
1172
1173         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1174                                         EMULATE_DONE) {
1175                 /*
1176                  * Fetch failed, so return to guest and
1177                  * try executing it again.
1178                  */
1179                 return RESUME_GUEST;
1180         }
1181
1182         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1183                 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1184                 vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1185                 return RESUME_HOST;
1186         } else {
1187                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1188                 return RESUME_GUEST;
1189         }
1190 }
1191
1192 static void do_nothing(void *x)
1193 {
1194 }
1195
1196 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1197 {
1198         int thr, cpu, pcpu, nthreads;
1199         struct kvm_vcpu *v;
1200         unsigned long dpdes;
1201
1202         nthreads = vcpu->kvm->arch.emul_smt_mode;
1203         dpdes = 0;
1204         cpu = vcpu->vcpu_id & ~(nthreads - 1);
1205         for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1206                 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1207                 if (!v)
1208                         continue;
1209                 /*
1210                  * If the vcpu is currently running on a physical cpu thread,
1211                  * interrupt it in order to pull it out of the guest briefly,
1212                  * which will update its vcore->dpdes value.
1213                  */
1214                 pcpu = READ_ONCE(v->cpu);
1215                 if (pcpu >= 0)
1216                         smp_call_function_single(pcpu, do_nothing, NULL, 1);
1217                 if (kvmppc_doorbell_pending(v))
1218                         dpdes |= 1 << thr;
1219         }
1220         return dpdes;
1221 }
1222
1223 /*
1224  * On POWER9, emulate doorbell-related instructions in order to
1225  * give the guest the illusion of running on a multi-threaded core.
1226  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1227  * and mfspr DPDES.
1228  */
1229 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1230 {
1231         u32 inst, rb, thr;
1232         unsigned long arg;
1233         struct kvm *kvm = vcpu->kvm;
1234         struct kvm_vcpu *tvcpu;
1235
1236         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1237                 return RESUME_GUEST;
1238         if (get_op(inst) != 31)
1239                 return EMULATE_FAIL;
1240         rb = get_rb(inst);
1241         thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1242         switch (get_xop(inst)) {
1243         case OP_31_XOP_MSGSNDP:
1244                 arg = kvmppc_get_gpr(vcpu, rb);
1245                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1246                         break;
1247                 arg &= 0x3f;
1248                 if (arg >= kvm->arch.emul_smt_mode)
1249                         break;
1250                 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1251                 if (!tvcpu)
1252                         break;
1253                 if (!tvcpu->arch.doorbell_request) {
1254                         tvcpu->arch.doorbell_request = 1;
1255                         kvmppc_fast_vcpu_kick_hv(tvcpu);
1256                 }
1257                 break;
1258         case OP_31_XOP_MSGCLRP:
1259                 arg = kvmppc_get_gpr(vcpu, rb);
1260                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1261                         break;
1262                 vcpu->arch.vcore->dpdes = 0;
1263                 vcpu->arch.doorbell_request = 0;
1264                 break;
1265         case OP_31_XOP_MFSPR:
1266                 switch (get_sprn(inst)) {
1267                 case SPRN_TIR:
1268                         arg = thr;
1269                         break;
1270                 case SPRN_DPDES:
1271                         arg = kvmppc_read_dpdes(vcpu);
1272                         break;
1273                 default:
1274                         return EMULATE_FAIL;
1275                 }
1276                 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1277                 break;
1278         default:
1279                 return EMULATE_FAIL;
1280         }
1281         kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1282         return RESUME_GUEST;
1283 }
1284
1285 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1286                                  struct task_struct *tsk)
1287 {
1288         struct kvm_run *run = vcpu->run;
1289         int r = RESUME_HOST;
1290
1291         vcpu->stat.sum_exits++;
1292
1293         /*
1294          * This can happen if an interrupt occurs in the last stages
1295          * of guest entry or the first stages of guest exit (i.e. after
1296          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1297          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1298          * That can happen due to a bug, or due to a machine check
1299          * occurring at just the wrong time.
1300          */
1301         if (vcpu->arch.shregs.msr & MSR_HV) {
1302                 printk(KERN_EMERG "KVM trap in HV mode!\n");
1303                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1304                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1305                         vcpu->arch.shregs.msr);
1306                 kvmppc_dump_regs(vcpu);
1307                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1308                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1309                 return RESUME_HOST;
1310         }
1311         run->exit_reason = KVM_EXIT_UNKNOWN;
1312         run->ready_for_interrupt_injection = 1;
1313         switch (vcpu->arch.trap) {
1314         /* We're good on these - the host merely wanted to get our attention */
1315         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1316                 vcpu->stat.dec_exits++;
1317                 r = RESUME_GUEST;
1318                 break;
1319         case BOOK3S_INTERRUPT_EXTERNAL:
1320         case BOOK3S_INTERRUPT_H_DOORBELL:
1321         case BOOK3S_INTERRUPT_H_VIRT:
1322                 vcpu->stat.ext_intr_exits++;
1323                 r = RESUME_GUEST;
1324                 break;
1325         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1326         case BOOK3S_INTERRUPT_HMI:
1327         case BOOK3S_INTERRUPT_PERFMON:
1328         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1329                 r = RESUME_GUEST;
1330                 break;
1331         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1332                 /* Print the MCE event to host console. */
1333                 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1334
1335                 /*
1336                  * If the guest can do FWNMI, exit to userspace so it can
1337                  * deliver a FWNMI to the guest.
1338                  * Otherwise we synthesize a machine check for the guest
1339                  * so that it knows that the machine check occurred.
1340                  */
1341                 if (!vcpu->kvm->arch.fwnmi_enabled) {
1342                         ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1343                         kvmppc_core_queue_machine_check(vcpu, flags);
1344                         r = RESUME_GUEST;
1345                         break;
1346                 }
1347
1348                 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1349                 run->exit_reason = KVM_EXIT_NMI;
1350                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1351                 /* Clear out the old NMI status from run->flags */
1352                 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1353                 /* Now set the NMI status */
1354                 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1355                         run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1356                 else
1357                         run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1358
1359                 r = RESUME_HOST;
1360                 break;
1361         case BOOK3S_INTERRUPT_PROGRAM:
1362         {
1363                 ulong flags;
1364                 /*
1365                  * Normally program interrupts are delivered directly
1366                  * to the guest by the hardware, but we can get here
1367                  * as a result of a hypervisor emulation interrupt
1368                  * (e40) getting turned into a 700 by BML RTAS.
1369                  */
1370                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1371                 kvmppc_core_queue_program(vcpu, flags);
1372                 r = RESUME_GUEST;
1373                 break;
1374         }
1375         case BOOK3S_INTERRUPT_SYSCALL:
1376         {
1377                 /* hcall - punt to userspace */
1378                 int i;
1379
1380                 /* hypercall with MSR_PR has already been handled in rmode,
1381                  * and never reaches here.
1382                  */
1383
1384                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1385                 for (i = 0; i < 9; ++i)
1386                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1387                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1388                 vcpu->arch.hcall_needed = 1;
1389                 r = RESUME_HOST;
1390                 break;
1391         }
1392         /*
1393          * We get these next two if the guest accesses a page which it thinks
1394          * it has mapped but which is not actually present, either because
1395          * it is for an emulated I/O device or because the corresonding
1396          * host page has been paged out.  Any other HDSI/HISI interrupts
1397          * have been handled already.
1398          */
1399         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1400                 r = RESUME_PAGE_FAULT;
1401                 break;
1402         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1403                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1404                 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1405                         DSISR_SRR1_MATCH_64S;
1406                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1407                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1408                 r = RESUME_PAGE_FAULT;
1409                 break;
1410         /*
1411          * This occurs if the guest executes an illegal instruction.
1412          * If the guest debug is disabled, generate a program interrupt
1413          * to the guest. If guest debug is enabled, we need to check
1414          * whether the instruction is a software breakpoint instruction.
1415          * Accordingly return to Guest or Host.
1416          */
1417         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1418                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1419                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1420                                 swab32(vcpu->arch.emul_inst) :
1421                                 vcpu->arch.emul_inst;
1422                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1423                         r = kvmppc_emulate_debug_inst(vcpu);
1424                 } else {
1425                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1426                         r = RESUME_GUEST;
1427                 }
1428                 break;
1429         /*
1430          * This occurs if the guest (kernel or userspace), does something that
1431          * is prohibited by HFSCR.
1432          * On POWER9, this could be a doorbell instruction that we need
1433          * to emulate.
1434          * Otherwise, we just generate a program interrupt to the guest.
1435          */
1436         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1437                 r = EMULATE_FAIL;
1438                 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1439                     cpu_has_feature(CPU_FTR_ARCH_300))
1440                         r = kvmppc_emulate_doorbell_instr(vcpu);
1441                 if (r == EMULATE_FAIL) {
1442                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1443                         r = RESUME_GUEST;
1444                 }
1445                 break;
1446
1447 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1448         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1449                 /*
1450                  * This occurs for various TM-related instructions that
1451                  * we need to emulate on POWER9 DD2.2.  We have already
1452                  * handled the cases where the guest was in real-suspend
1453                  * mode and was transitioning to transactional state.
1454                  */
1455                 r = kvmhv_p9_tm_emulation(vcpu);
1456                 break;
1457 #endif
1458
1459         case BOOK3S_INTERRUPT_HV_RM_HARD:
1460                 r = RESUME_PASSTHROUGH;
1461                 break;
1462         default:
1463                 kvmppc_dump_regs(vcpu);
1464                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1465                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1466                         vcpu->arch.shregs.msr);
1467                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1468                 r = RESUME_HOST;
1469                 break;
1470         }
1471
1472         return r;
1473 }
1474
1475 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1476 {
1477         int r;
1478         int srcu_idx;
1479
1480         vcpu->stat.sum_exits++;
1481
1482         /*
1483          * This can happen if an interrupt occurs in the last stages
1484          * of guest entry or the first stages of guest exit (i.e. after
1485          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1486          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1487          * That can happen due to a bug, or due to a machine check
1488          * occurring at just the wrong time.
1489          */
1490         if (vcpu->arch.shregs.msr & MSR_HV) {
1491                 pr_emerg("KVM trap in HV mode while nested!\n");
1492                 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1493                          vcpu->arch.trap, kvmppc_get_pc(vcpu),
1494                          vcpu->arch.shregs.msr);
1495                 kvmppc_dump_regs(vcpu);
1496                 return RESUME_HOST;
1497         }
1498         switch (vcpu->arch.trap) {
1499         /* We're good on these - the host merely wanted to get our attention */
1500         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1501                 vcpu->stat.dec_exits++;
1502                 r = RESUME_GUEST;
1503                 break;
1504         case BOOK3S_INTERRUPT_EXTERNAL:
1505                 vcpu->stat.ext_intr_exits++;
1506                 r = RESUME_HOST;
1507                 break;
1508         case BOOK3S_INTERRUPT_H_DOORBELL:
1509         case BOOK3S_INTERRUPT_H_VIRT:
1510                 vcpu->stat.ext_intr_exits++;
1511                 r = RESUME_GUEST;
1512                 break;
1513         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1514         case BOOK3S_INTERRUPT_HMI:
1515         case BOOK3S_INTERRUPT_PERFMON:
1516         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1517                 r = RESUME_GUEST;
1518                 break;
1519         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1520                 /* Pass the machine check to the L1 guest */
1521                 r = RESUME_HOST;
1522                 /* Print the MCE event to host console. */
1523                 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1524                 break;
1525         /*
1526          * We get these next two if the guest accesses a page which it thinks
1527          * it has mapped but which is not actually present, either because
1528          * it is for an emulated I/O device or because the corresonding
1529          * host page has been paged out.
1530          */
1531         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1532                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1533                 r = kvmhv_nested_page_fault(vcpu);
1534                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1535                 break;
1536         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1537                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1538                 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1539                                          DSISR_SRR1_MATCH_64S;
1540                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1541                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1542                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1543                 r = kvmhv_nested_page_fault(vcpu);
1544                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1545                 break;
1546
1547 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1548         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1549                 /*
1550                  * This occurs for various TM-related instructions that
1551                  * we need to emulate on POWER9 DD2.2.  We have already
1552                  * handled the cases where the guest was in real-suspend
1553                  * mode and was transitioning to transactional state.
1554                  */
1555                 r = kvmhv_p9_tm_emulation(vcpu);
1556                 break;
1557 #endif
1558
1559         case BOOK3S_INTERRUPT_HV_RM_HARD:
1560                 vcpu->arch.trap = 0;
1561                 r = RESUME_GUEST;
1562                 if (!xics_on_xive())
1563                         kvmppc_xics_rm_complete(vcpu, 0);
1564                 break;
1565         default:
1566                 r = RESUME_HOST;
1567                 break;
1568         }
1569
1570         return r;
1571 }
1572
1573 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1574                                             struct kvm_sregs *sregs)
1575 {
1576         int i;
1577
1578         memset(sregs, 0, sizeof(struct kvm_sregs));
1579         sregs->pvr = vcpu->arch.pvr;
1580         for (i = 0; i < vcpu->arch.slb_max; i++) {
1581                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1582                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1583         }
1584
1585         return 0;
1586 }
1587
1588 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1589                                             struct kvm_sregs *sregs)
1590 {
1591         int i, j;
1592
1593         /* Only accept the same PVR as the host's, since we can't spoof it */
1594         if (sregs->pvr != vcpu->arch.pvr)
1595                 return -EINVAL;
1596
1597         j = 0;
1598         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1599                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1600                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1601                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1602                         ++j;
1603                 }
1604         }
1605         vcpu->arch.slb_max = j;
1606
1607         return 0;
1608 }
1609
1610 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1611                 bool preserve_top32)
1612 {
1613         struct kvm *kvm = vcpu->kvm;
1614         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1615         u64 mask;
1616
1617         spin_lock(&vc->lock);
1618         /*
1619          * If ILE (interrupt little-endian) has changed, update the
1620          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1621          */
1622         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1623                 struct kvm_vcpu *vcpu;
1624                 int i;
1625
1626                 kvm_for_each_vcpu(i, vcpu, kvm) {
1627                         if (vcpu->arch.vcore != vc)
1628                                 continue;
1629                         if (new_lpcr & LPCR_ILE)
1630                                 vcpu->arch.intr_msr |= MSR_LE;
1631                         else
1632                                 vcpu->arch.intr_msr &= ~MSR_LE;
1633                 }
1634         }
1635
1636         /*
1637          * Userspace can only modify DPFD (default prefetch depth),
1638          * ILE (interrupt little-endian) and TC (translation control).
1639          * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1640          */
1641         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1642         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1643                 mask |= LPCR_AIL;
1644         /*
1645          * On POWER9, allow userspace to enable large decrementer for the
1646          * guest, whether or not the host has it enabled.
1647          */
1648         if (cpu_has_feature(CPU_FTR_ARCH_300))
1649                 mask |= LPCR_LD;
1650
1651         /* Broken 32-bit version of LPCR must not clear top bits */
1652         if (preserve_top32)
1653                 mask &= 0xFFFFFFFF;
1654         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1655         spin_unlock(&vc->lock);
1656 }
1657
1658 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1659                                  union kvmppc_one_reg *val)
1660 {
1661         int r = 0;
1662         long int i;
1663
1664         switch (id) {
1665         case KVM_REG_PPC_DEBUG_INST:
1666                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1667                 break;
1668         case KVM_REG_PPC_HIOR:
1669                 *val = get_reg_val(id, 0);
1670                 break;
1671         case KVM_REG_PPC_DABR:
1672                 *val = get_reg_val(id, vcpu->arch.dabr);
1673                 break;
1674         case KVM_REG_PPC_DABRX:
1675                 *val = get_reg_val(id, vcpu->arch.dabrx);
1676                 break;
1677         case KVM_REG_PPC_DSCR:
1678                 *val = get_reg_val(id, vcpu->arch.dscr);
1679                 break;
1680         case KVM_REG_PPC_PURR:
1681                 *val = get_reg_val(id, vcpu->arch.purr);
1682                 break;
1683         case KVM_REG_PPC_SPURR:
1684                 *val = get_reg_val(id, vcpu->arch.spurr);
1685                 break;
1686         case KVM_REG_PPC_AMR:
1687                 *val = get_reg_val(id, vcpu->arch.amr);
1688                 break;
1689         case KVM_REG_PPC_UAMOR:
1690                 *val = get_reg_val(id, vcpu->arch.uamor);
1691                 break;
1692         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
1693                 i = id - KVM_REG_PPC_MMCR0;
1694                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1695                 break;
1696         case KVM_REG_PPC_MMCR2:
1697                 *val = get_reg_val(id, vcpu->arch.mmcr[2]);
1698                 break;
1699         case KVM_REG_PPC_MMCRA:
1700                 *val = get_reg_val(id, vcpu->arch.mmcra);
1701                 break;
1702         case KVM_REG_PPC_MMCRS:
1703                 *val = get_reg_val(id, vcpu->arch.mmcrs);
1704                 break;
1705         case KVM_REG_PPC_MMCR3:
1706                 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
1707                 break;
1708         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1709                 i = id - KVM_REG_PPC_PMC1;
1710                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1711                 break;
1712         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1713                 i = id - KVM_REG_PPC_SPMC1;
1714                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1715                 break;
1716         case KVM_REG_PPC_SIAR:
1717                 *val = get_reg_val(id, vcpu->arch.siar);
1718                 break;
1719         case KVM_REG_PPC_SDAR:
1720                 *val = get_reg_val(id, vcpu->arch.sdar);
1721                 break;
1722         case KVM_REG_PPC_SIER:
1723                 *val = get_reg_val(id, vcpu->arch.sier[0]);
1724                 break;
1725         case KVM_REG_PPC_SIER2:
1726                 *val = get_reg_val(id, vcpu->arch.sier[1]);
1727                 break;
1728         case KVM_REG_PPC_SIER3:
1729                 *val = get_reg_val(id, vcpu->arch.sier[2]);
1730                 break;
1731         case KVM_REG_PPC_IAMR:
1732                 *val = get_reg_val(id, vcpu->arch.iamr);
1733                 break;
1734         case KVM_REG_PPC_PSPB:
1735                 *val = get_reg_val(id, vcpu->arch.pspb);
1736                 break;
1737         case KVM_REG_PPC_DPDES:
1738                 /*
1739                  * On POWER9, where we are emulating msgsndp etc.,
1740                  * we return 1 bit for each vcpu, which can come from
1741                  * either vcore->dpdes or doorbell_request.
1742                  * On POWER8, doorbell_request is 0.
1743                  */
1744                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes |
1745                                    vcpu->arch.doorbell_request);
1746                 break;
1747         case KVM_REG_PPC_VTB:
1748                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1749                 break;
1750         case KVM_REG_PPC_DAWR:
1751                 *val = get_reg_val(id, vcpu->arch.dawr);
1752                 break;
1753         case KVM_REG_PPC_DAWRX:
1754                 *val = get_reg_val(id, vcpu->arch.dawrx);
1755                 break;
1756         case KVM_REG_PPC_CIABR:
1757                 *val = get_reg_val(id, vcpu->arch.ciabr);
1758                 break;
1759         case KVM_REG_PPC_CSIGR:
1760                 *val = get_reg_val(id, vcpu->arch.csigr);
1761                 break;
1762         case KVM_REG_PPC_TACR:
1763                 *val = get_reg_val(id, vcpu->arch.tacr);
1764                 break;
1765         case KVM_REG_PPC_TCSCR:
1766                 *val = get_reg_val(id, vcpu->arch.tcscr);
1767                 break;
1768         case KVM_REG_PPC_PID:
1769                 *val = get_reg_val(id, vcpu->arch.pid);
1770                 break;
1771         case KVM_REG_PPC_ACOP:
1772                 *val = get_reg_val(id, vcpu->arch.acop);
1773                 break;
1774         case KVM_REG_PPC_WORT:
1775                 *val = get_reg_val(id, vcpu->arch.wort);
1776                 break;
1777         case KVM_REG_PPC_TIDR:
1778                 *val = get_reg_val(id, vcpu->arch.tid);
1779                 break;
1780         case KVM_REG_PPC_PSSCR:
1781                 *val = get_reg_val(id, vcpu->arch.psscr);
1782                 break;
1783         case KVM_REG_PPC_VPA_ADDR:
1784                 spin_lock(&vcpu->arch.vpa_update_lock);
1785                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1786                 spin_unlock(&vcpu->arch.vpa_update_lock);
1787                 break;
1788         case KVM_REG_PPC_VPA_SLB:
1789                 spin_lock(&vcpu->arch.vpa_update_lock);
1790                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1791                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1792                 spin_unlock(&vcpu->arch.vpa_update_lock);
1793                 break;
1794         case KVM_REG_PPC_VPA_DTL:
1795                 spin_lock(&vcpu->arch.vpa_update_lock);
1796                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1797                 val->vpaval.length = vcpu->arch.dtl.len;
1798                 spin_unlock(&vcpu->arch.vpa_update_lock);
1799                 break;
1800         case KVM_REG_PPC_TB_OFFSET:
1801                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1802                 break;
1803         case KVM_REG_PPC_LPCR:
1804         case KVM_REG_PPC_LPCR_64:
1805                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1806                 break;
1807         case KVM_REG_PPC_PPR:
1808                 *val = get_reg_val(id, vcpu->arch.ppr);
1809                 break;
1810 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1811         case KVM_REG_PPC_TFHAR:
1812                 *val = get_reg_val(id, vcpu->arch.tfhar);
1813                 break;
1814         case KVM_REG_PPC_TFIAR:
1815                 *val = get_reg_val(id, vcpu->arch.tfiar);
1816                 break;
1817         case KVM_REG_PPC_TEXASR:
1818                 *val = get_reg_val(id, vcpu->arch.texasr);
1819                 break;
1820         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1821                 i = id - KVM_REG_PPC_TM_GPR0;
1822                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1823                 break;
1824         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1825         {
1826                 int j;
1827                 i = id - KVM_REG_PPC_TM_VSR0;
1828                 if (i < 32)
1829                         for (j = 0; j < TS_FPRWIDTH; j++)
1830                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1831                 else {
1832                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1833                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1834                         else
1835                                 r = -ENXIO;
1836                 }
1837                 break;
1838         }
1839         case KVM_REG_PPC_TM_CR:
1840                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1841                 break;
1842         case KVM_REG_PPC_TM_XER:
1843                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1844                 break;
1845         case KVM_REG_PPC_TM_LR:
1846                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1847                 break;
1848         case KVM_REG_PPC_TM_CTR:
1849                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1850                 break;
1851         case KVM_REG_PPC_TM_FPSCR:
1852                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1853                 break;
1854         case KVM_REG_PPC_TM_AMR:
1855                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1856                 break;
1857         case KVM_REG_PPC_TM_PPR:
1858                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1859                 break;
1860         case KVM_REG_PPC_TM_VRSAVE:
1861                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1862                 break;
1863         case KVM_REG_PPC_TM_VSCR:
1864                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1865                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1866                 else
1867                         r = -ENXIO;
1868                 break;
1869         case KVM_REG_PPC_TM_DSCR:
1870                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1871                 break;
1872         case KVM_REG_PPC_TM_TAR:
1873                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1874                 break;
1875 #endif
1876         case KVM_REG_PPC_ARCH_COMPAT:
1877                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1878                 break;
1879         case KVM_REG_PPC_DEC_EXPIRY:
1880                 *val = get_reg_val(id, vcpu->arch.dec_expires +
1881                                    vcpu->arch.vcore->tb_offset);
1882                 break;
1883         case KVM_REG_PPC_ONLINE:
1884                 *val = get_reg_val(id, vcpu->arch.online);
1885                 break;
1886         case KVM_REG_PPC_PTCR:
1887                 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1888                 break;
1889         default:
1890                 r = -EINVAL;
1891                 break;
1892         }
1893
1894         return r;
1895 }
1896
1897 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1898                                  union kvmppc_one_reg *val)
1899 {
1900         int r = 0;
1901         long int i;
1902         unsigned long addr, len;
1903
1904         switch (id) {
1905         case KVM_REG_PPC_HIOR:
1906                 /* Only allow this to be set to zero */
1907                 if (set_reg_val(id, *val))
1908                         r = -EINVAL;
1909                 break;
1910         case KVM_REG_PPC_DABR:
1911                 vcpu->arch.dabr = set_reg_val(id, *val);
1912                 break;
1913         case KVM_REG_PPC_DABRX:
1914                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1915                 break;
1916         case KVM_REG_PPC_DSCR:
1917                 vcpu->arch.dscr = set_reg_val(id, *val);
1918                 break;
1919         case KVM_REG_PPC_PURR:
1920                 vcpu->arch.purr = set_reg_val(id, *val);
1921                 break;
1922         case KVM_REG_PPC_SPURR:
1923                 vcpu->arch.spurr = set_reg_val(id, *val);
1924                 break;
1925         case KVM_REG_PPC_AMR:
1926                 vcpu->arch.amr = set_reg_val(id, *val);
1927                 break;
1928         case KVM_REG_PPC_UAMOR:
1929                 vcpu->arch.uamor = set_reg_val(id, *val);
1930                 break;
1931         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
1932                 i = id - KVM_REG_PPC_MMCR0;
1933                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1934                 break;
1935         case KVM_REG_PPC_MMCR2:
1936                 vcpu->arch.mmcr[2] = set_reg_val(id, *val);
1937                 break;
1938         case KVM_REG_PPC_MMCRA:
1939                 vcpu->arch.mmcra = set_reg_val(id, *val);
1940                 break;
1941         case KVM_REG_PPC_MMCRS:
1942                 vcpu->arch.mmcrs = set_reg_val(id, *val);
1943                 break;
1944         case KVM_REG_PPC_MMCR3:
1945                 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
1946                 break;
1947         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1948                 i = id - KVM_REG_PPC_PMC1;
1949                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1950                 break;
1951         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1952                 i = id - KVM_REG_PPC_SPMC1;
1953                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1954                 break;
1955         case KVM_REG_PPC_SIAR:
1956                 vcpu->arch.siar = set_reg_val(id, *val);
1957                 break;
1958         case KVM_REG_PPC_SDAR:
1959                 vcpu->arch.sdar = set_reg_val(id, *val);
1960                 break;
1961         case KVM_REG_PPC_SIER:
1962                 vcpu->arch.sier[0] = set_reg_val(id, *val);
1963                 break;
1964         case KVM_REG_PPC_SIER2:
1965                 vcpu->arch.sier[1] = set_reg_val(id, *val);
1966                 break;
1967         case KVM_REG_PPC_SIER3:
1968                 vcpu->arch.sier[2] = set_reg_val(id, *val);
1969                 break;
1970         case KVM_REG_PPC_IAMR:
1971                 vcpu->arch.iamr = set_reg_val(id, *val);
1972                 break;
1973         case KVM_REG_PPC_PSPB:
1974                 vcpu->arch.pspb = set_reg_val(id, *val);
1975                 break;
1976         case KVM_REG_PPC_DPDES:
1977                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1978                 break;
1979         case KVM_REG_PPC_VTB:
1980                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1981                 break;
1982         case KVM_REG_PPC_DAWR:
1983                 vcpu->arch.dawr = set_reg_val(id, *val);
1984                 break;
1985         case KVM_REG_PPC_DAWRX:
1986                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1987                 break;
1988         case KVM_REG_PPC_CIABR:
1989                 vcpu->arch.ciabr = set_reg_val(id, *val);
1990                 /* Don't allow setting breakpoints in hypervisor code */
1991                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1992                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1993                 break;
1994         case KVM_REG_PPC_CSIGR:
1995                 vcpu->arch.csigr = set_reg_val(id, *val);
1996                 break;
1997         case KVM_REG_PPC_TACR:
1998                 vcpu->arch.tacr = set_reg_val(id, *val);
1999                 break;
2000         case KVM_REG_PPC_TCSCR:
2001                 vcpu->arch.tcscr = set_reg_val(id, *val);
2002                 break;
2003         case KVM_REG_PPC_PID:
2004                 vcpu->arch.pid = set_reg_val(id, *val);
2005                 break;
2006         case KVM_REG_PPC_ACOP:
2007                 vcpu->arch.acop = set_reg_val(id, *val);
2008                 break;
2009         case KVM_REG_PPC_WORT:
2010                 vcpu->arch.wort = set_reg_val(id, *val);
2011                 break;
2012         case KVM_REG_PPC_TIDR:
2013                 vcpu->arch.tid = set_reg_val(id, *val);
2014                 break;
2015         case KVM_REG_PPC_PSSCR:
2016                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2017                 break;
2018         case KVM_REG_PPC_VPA_ADDR:
2019                 addr = set_reg_val(id, *val);
2020                 r = -EINVAL;
2021                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2022                               vcpu->arch.dtl.next_gpa))
2023                         break;
2024                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2025                 break;
2026         case KVM_REG_PPC_VPA_SLB:
2027                 addr = val->vpaval.addr;
2028                 len = val->vpaval.length;
2029                 r = -EINVAL;
2030                 if (addr && !vcpu->arch.vpa.next_gpa)
2031                         break;
2032                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2033                 break;
2034         case KVM_REG_PPC_VPA_DTL:
2035                 addr = val->vpaval.addr;
2036                 len = val->vpaval.length;
2037                 r = -EINVAL;
2038                 if (addr && (len < sizeof(struct dtl_entry) ||
2039                              !vcpu->arch.vpa.next_gpa))
2040                         break;
2041                 len -= len % sizeof(struct dtl_entry);
2042                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2043                 break;
2044         case KVM_REG_PPC_TB_OFFSET:
2045                 /* round up to multiple of 2^24 */
2046                 vcpu->arch.vcore->tb_offset =
2047                         ALIGN(set_reg_val(id, *val), 1UL << 24);
2048                 break;
2049         case KVM_REG_PPC_LPCR:
2050                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2051                 break;
2052         case KVM_REG_PPC_LPCR_64:
2053                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2054                 break;
2055         case KVM_REG_PPC_PPR:
2056                 vcpu->arch.ppr = set_reg_val(id, *val);
2057                 break;
2058 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2059         case KVM_REG_PPC_TFHAR:
2060                 vcpu->arch.tfhar = set_reg_val(id, *val);
2061                 break;
2062         case KVM_REG_PPC_TFIAR:
2063                 vcpu->arch.tfiar = set_reg_val(id, *val);
2064                 break;
2065         case KVM_REG_PPC_TEXASR:
2066                 vcpu->arch.texasr = set_reg_val(id, *val);
2067                 break;
2068         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2069                 i = id - KVM_REG_PPC_TM_GPR0;
2070                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2071                 break;
2072         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2073         {
2074                 int j;
2075                 i = id - KVM_REG_PPC_TM_VSR0;
2076                 if (i < 32)
2077                         for (j = 0; j < TS_FPRWIDTH; j++)
2078                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2079                 else
2080                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
2081                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2082                         else
2083                                 r = -ENXIO;
2084                 break;
2085         }
2086         case KVM_REG_PPC_TM_CR:
2087                 vcpu->arch.cr_tm = set_reg_val(id, *val);
2088                 break;
2089         case KVM_REG_PPC_TM_XER:
2090                 vcpu->arch.xer_tm = set_reg_val(id, *val);
2091                 break;
2092         case KVM_REG_PPC_TM_LR:
2093                 vcpu->arch.lr_tm = set_reg_val(id, *val);
2094                 break;
2095         case KVM_REG_PPC_TM_CTR:
2096                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2097                 break;
2098         case KVM_REG_PPC_TM_FPSCR:
2099                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2100                 break;
2101         case KVM_REG_PPC_TM_AMR:
2102                 vcpu->arch.amr_tm = set_reg_val(id, *val);
2103                 break;
2104         case KVM_REG_PPC_TM_PPR:
2105                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2106                 break;
2107         case KVM_REG_PPC_TM_VRSAVE:
2108                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2109                 break;
2110         case KVM_REG_PPC_TM_VSCR:
2111                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2112                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2113                 else
2114                         r = - ENXIO;
2115                 break;
2116         case KVM_REG_PPC_TM_DSCR:
2117                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2118                 break;
2119         case KVM_REG_PPC_TM_TAR:
2120                 vcpu->arch.tar_tm = set_reg_val(id, *val);
2121                 break;
2122 #endif
2123         case KVM_REG_PPC_ARCH_COMPAT:
2124                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2125                 break;
2126         case KVM_REG_PPC_DEC_EXPIRY:
2127                 vcpu->arch.dec_expires = set_reg_val(id, *val) -
2128                         vcpu->arch.vcore->tb_offset;
2129                 break;
2130         case KVM_REG_PPC_ONLINE:
2131                 i = set_reg_val(id, *val);
2132                 if (i && !vcpu->arch.online)
2133                         atomic_inc(&vcpu->arch.vcore->online_count);
2134                 else if (!i && vcpu->arch.online)
2135                         atomic_dec(&vcpu->arch.vcore->online_count);
2136                 vcpu->arch.online = i;
2137                 break;
2138         case KVM_REG_PPC_PTCR:
2139                 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2140                 break;
2141         default:
2142                 r = -EINVAL;
2143                 break;
2144         }
2145
2146         return r;
2147 }
2148
2149 /*
2150  * On POWER9, threads are independent and can be in different partitions.
2151  * Therefore we consider each thread to be a subcore.
2152  * There is a restriction that all threads have to be in the same
2153  * MMU mode (radix or HPT), unfortunately, but since we only support
2154  * HPT guests on a HPT host so far, that isn't an impediment yet.
2155  */
2156 static int threads_per_vcore(struct kvm *kvm)
2157 {
2158         if (kvm->arch.threads_indep)
2159                 return 1;
2160         return threads_per_subcore;
2161 }
2162
2163 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2164 {
2165         struct kvmppc_vcore *vcore;
2166
2167         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2168
2169         if (vcore == NULL)
2170                 return NULL;
2171
2172         spin_lock_init(&vcore->lock);
2173         spin_lock_init(&vcore->stoltb_lock);
2174         rcuwait_init(&vcore->wait);
2175         vcore->preempt_tb = TB_NIL;
2176         vcore->lpcr = kvm->arch.lpcr;
2177         vcore->first_vcpuid = id;
2178         vcore->kvm = kvm;
2179         INIT_LIST_HEAD(&vcore->preempt_list);
2180
2181         return vcore;
2182 }
2183
2184 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2185 static struct debugfs_timings_element {
2186         const char *name;
2187         size_t offset;
2188 } timings[] = {
2189         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
2190         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
2191         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
2192         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
2193         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
2194 };
2195
2196 #define N_TIMINGS       (ARRAY_SIZE(timings))
2197
2198 struct debugfs_timings_state {
2199         struct kvm_vcpu *vcpu;
2200         unsigned int    buflen;
2201         char            buf[N_TIMINGS * 100];
2202 };
2203
2204 static int debugfs_timings_open(struct inode *inode, struct file *file)
2205 {
2206         struct kvm_vcpu *vcpu = inode->i_private;
2207         struct debugfs_timings_state *p;
2208
2209         p = kzalloc(sizeof(*p), GFP_KERNEL);
2210         if (!p)
2211                 return -ENOMEM;
2212
2213         kvm_get_kvm(vcpu->kvm);
2214         p->vcpu = vcpu;
2215         file->private_data = p;
2216
2217         return nonseekable_open(inode, file);
2218 }
2219
2220 static int debugfs_timings_release(struct inode *inode, struct file *file)
2221 {
2222         struct debugfs_timings_state *p = file->private_data;
2223
2224         kvm_put_kvm(p->vcpu->kvm);
2225         kfree(p);
2226         return 0;
2227 }
2228
2229 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2230                                     size_t len, loff_t *ppos)
2231 {
2232         struct debugfs_timings_state *p = file->private_data;
2233         struct kvm_vcpu *vcpu = p->vcpu;
2234         char *s, *buf_end;
2235         struct kvmhv_tb_accumulator tb;
2236         u64 count;
2237         loff_t pos;
2238         ssize_t n;
2239         int i, loops;
2240         bool ok;
2241
2242         if (!p->buflen) {
2243                 s = p->buf;
2244                 buf_end = s + sizeof(p->buf);
2245                 for (i = 0; i < N_TIMINGS; ++i) {
2246                         struct kvmhv_tb_accumulator *acc;
2247
2248                         acc = (struct kvmhv_tb_accumulator *)
2249                                 ((unsigned long)vcpu + timings[i].offset);
2250                         ok = false;
2251                         for (loops = 0; loops < 1000; ++loops) {
2252                                 count = acc->seqcount;
2253                                 if (!(count & 1)) {
2254                                         smp_rmb();
2255                                         tb = *acc;
2256                                         smp_rmb();
2257                                         if (count == acc->seqcount) {
2258                                                 ok = true;
2259                                                 break;
2260                                         }
2261                                 }
2262                                 udelay(1);
2263                         }
2264                         if (!ok)
2265                                 snprintf(s, buf_end - s, "%s: stuck\n",
2266                                         timings[i].name);
2267                         else
2268                                 snprintf(s, buf_end - s,
2269                                         "%s: %llu %llu %llu %llu\n",
2270                                         timings[i].name, count / 2,
2271                                         tb_to_ns(tb.tb_total),
2272                                         tb_to_ns(tb.tb_min),
2273                                         tb_to_ns(tb.tb_max));
2274                         s += strlen(s);
2275                 }
2276                 p->buflen = s - p->buf;
2277         }
2278
2279         pos = *ppos;
2280         if (pos >= p->buflen)
2281                 return 0;
2282         if (len > p->buflen - pos)
2283                 len = p->buflen - pos;
2284         n = copy_to_user(buf, p->buf + pos, len);
2285         if (n) {
2286                 if (n == len)
2287                         return -EFAULT;
2288                 len -= n;
2289         }
2290         *ppos = pos + len;
2291         return len;
2292 }
2293
2294 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2295                                      size_t len, loff_t *ppos)
2296 {
2297         return -EACCES;
2298 }
2299
2300 static const struct file_operations debugfs_timings_ops = {
2301         .owner   = THIS_MODULE,
2302         .open    = debugfs_timings_open,
2303         .release = debugfs_timings_release,
2304         .read    = debugfs_timings_read,
2305         .write   = debugfs_timings_write,
2306         .llseek  = generic_file_llseek,
2307 };
2308
2309 /* Create a debugfs directory for the vcpu */
2310 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2311 {
2312         char buf[16];
2313         struct kvm *kvm = vcpu->kvm;
2314
2315         snprintf(buf, sizeof(buf), "vcpu%u", id);
2316         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2317         debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir, vcpu,
2318                             &debugfs_timings_ops);
2319 }
2320
2321 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2322 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2323 {
2324 }
2325 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2326
2327 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2328 {
2329         int err;
2330         int core;
2331         struct kvmppc_vcore *vcore;
2332         struct kvm *kvm;
2333         unsigned int id;
2334
2335         kvm = vcpu->kvm;
2336         id = vcpu->vcpu_id;
2337
2338         vcpu->arch.shared = &vcpu->arch.shregs;
2339 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2340         /*
2341          * The shared struct is never shared on HV,
2342          * so we can always use host endianness
2343          */
2344 #ifdef __BIG_ENDIAN__
2345         vcpu->arch.shared_big_endian = true;
2346 #else
2347         vcpu->arch.shared_big_endian = false;
2348 #endif
2349 #endif
2350         vcpu->arch.mmcr[0] = MMCR0_FC;
2351         vcpu->arch.ctrl = CTRL_RUNLATCH;
2352         /* default to host PVR, since we can't spoof it */
2353         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2354         spin_lock_init(&vcpu->arch.vpa_update_lock);
2355         spin_lock_init(&vcpu->arch.tbacct_lock);
2356         vcpu->arch.busy_preempt = TB_NIL;
2357         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2358
2359         /*
2360          * Set the default HFSCR for the guest from the host value.
2361          * This value is only used on POWER9.
2362          * On POWER9, we want to virtualize the doorbell facility, so we
2363          * don't set the HFSCR_MSGP bit, and that causes those instructions
2364          * to trap and then we emulate them.
2365          */
2366         vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2367                 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP | HFSCR_PREFIX;
2368         if (cpu_has_feature(CPU_FTR_HVMODE)) {
2369                 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2370 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2371                 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2372                         vcpu->arch.hfscr |= HFSCR_TM;
2373 #endif
2374         }
2375         if (cpu_has_feature(CPU_FTR_TM_COMP))
2376                 vcpu->arch.hfscr |= HFSCR_TM;
2377
2378         kvmppc_mmu_book3s_hv_init(vcpu);
2379
2380         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2381
2382         init_waitqueue_head(&vcpu->arch.cpu_run);
2383
2384         mutex_lock(&kvm->lock);
2385         vcore = NULL;
2386         err = -EINVAL;
2387         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2388                 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2389                         pr_devel("KVM: VCPU ID too high\n");
2390                         core = KVM_MAX_VCORES;
2391                 } else {
2392                         BUG_ON(kvm->arch.smt_mode != 1);
2393                         core = kvmppc_pack_vcpu_id(kvm, id);
2394                 }
2395         } else {
2396                 core = id / kvm->arch.smt_mode;
2397         }
2398         if (core < KVM_MAX_VCORES) {
2399                 vcore = kvm->arch.vcores[core];
2400                 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2401                         pr_devel("KVM: collision on id %u", id);
2402                         vcore = NULL;
2403                 } else if (!vcore) {
2404                         /*
2405                          * Take mmu_setup_lock for mutual exclusion
2406                          * with kvmppc_update_lpcr().
2407                          */
2408                         err = -ENOMEM;
2409                         vcore = kvmppc_vcore_create(kvm,
2410                                         id & ~(kvm->arch.smt_mode - 1));
2411                         mutex_lock(&kvm->arch.mmu_setup_lock);
2412                         kvm->arch.vcores[core] = vcore;
2413                         kvm->arch.online_vcores++;
2414                         mutex_unlock(&kvm->arch.mmu_setup_lock);
2415                 }
2416         }
2417         mutex_unlock(&kvm->lock);
2418
2419         if (!vcore)
2420                 return err;
2421
2422         spin_lock(&vcore->lock);
2423         ++vcore->num_threads;
2424         spin_unlock(&vcore->lock);
2425         vcpu->arch.vcore = vcore;
2426         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2427         vcpu->arch.thread_cpu = -1;
2428         vcpu->arch.prev_cpu = -1;
2429
2430         vcpu->arch.cpu_type = KVM_CPU_3S_64;
2431         kvmppc_sanity_check(vcpu);
2432
2433         debugfs_vcpu_init(vcpu, id);
2434
2435         return 0;
2436 }
2437
2438 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2439                               unsigned long flags)
2440 {
2441         int err;
2442         int esmt = 0;
2443
2444         if (flags)
2445                 return -EINVAL;
2446         if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2447                 return -EINVAL;
2448         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2449                 /*
2450                  * On POWER8 (or POWER7), the threading mode is "strict",
2451                  * so we pack smt_mode vcpus per vcore.
2452                  */
2453                 if (smt_mode > threads_per_subcore)
2454                         return -EINVAL;
2455         } else {
2456                 /*
2457                  * On POWER9, the threading mode is "loose",
2458                  * so each vcpu gets its own vcore.
2459                  */
2460                 esmt = smt_mode;
2461                 smt_mode = 1;
2462         }
2463         mutex_lock(&kvm->lock);
2464         err = -EBUSY;
2465         if (!kvm->arch.online_vcores) {
2466                 kvm->arch.smt_mode = smt_mode;
2467                 kvm->arch.emul_smt_mode = esmt;
2468                 err = 0;
2469         }
2470         mutex_unlock(&kvm->lock);
2471
2472         return err;
2473 }
2474
2475 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2476 {
2477         if (vpa->pinned_addr)
2478                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2479                                         vpa->dirty);
2480 }
2481
2482 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2483 {
2484         spin_lock(&vcpu->arch.vpa_update_lock);
2485         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2486         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2487         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2488         spin_unlock(&vcpu->arch.vpa_update_lock);
2489 }
2490
2491 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2492 {
2493         /* Indicate we want to get back into the guest */
2494         return 1;
2495 }
2496
2497 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2498 {
2499         unsigned long dec_nsec, now;
2500
2501         now = get_tb();
2502         if (now > vcpu->arch.dec_expires) {
2503                 /* decrementer has already gone negative */
2504                 kvmppc_core_queue_dec(vcpu);
2505                 kvmppc_core_prepare_to_enter(vcpu);
2506                 return;
2507         }
2508         dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2509         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2510         vcpu->arch.timer_running = 1;
2511 }
2512
2513 extern int __kvmppc_vcore_entry(void);
2514
2515 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2516                                    struct kvm_vcpu *vcpu)
2517 {
2518         u64 now;
2519
2520         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2521                 return;
2522         spin_lock_irq(&vcpu->arch.tbacct_lock);
2523         now = mftb();
2524         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2525                 vcpu->arch.stolen_logged;
2526         vcpu->arch.busy_preempt = now;
2527         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2528         spin_unlock_irq(&vcpu->arch.tbacct_lock);
2529         --vc->n_runnable;
2530         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2531 }
2532
2533 static int kvmppc_grab_hwthread(int cpu)
2534 {
2535         struct paca_struct *tpaca;
2536         long timeout = 10000;
2537
2538         tpaca = paca_ptrs[cpu];
2539
2540         /* Ensure the thread won't go into the kernel if it wakes */
2541         tpaca->kvm_hstate.kvm_vcpu = NULL;
2542         tpaca->kvm_hstate.kvm_vcore = NULL;
2543         tpaca->kvm_hstate.napping = 0;
2544         smp_wmb();
2545         tpaca->kvm_hstate.hwthread_req = 1;
2546
2547         /*
2548          * If the thread is already executing in the kernel (e.g. handling
2549          * a stray interrupt), wait for it to get back to nap mode.
2550          * The smp_mb() is to ensure that our setting of hwthread_req
2551          * is visible before we look at hwthread_state, so if this
2552          * races with the code at system_reset_pSeries and the thread
2553          * misses our setting of hwthread_req, we are sure to see its
2554          * setting of hwthread_state, and vice versa.
2555          */
2556         smp_mb();
2557         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2558                 if (--timeout <= 0) {
2559                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
2560                         return -EBUSY;
2561                 }
2562                 udelay(1);
2563         }
2564         return 0;
2565 }
2566
2567 static void kvmppc_release_hwthread(int cpu)
2568 {
2569         struct paca_struct *tpaca;
2570
2571         tpaca = paca_ptrs[cpu];
2572         tpaca->kvm_hstate.hwthread_req = 0;
2573         tpaca->kvm_hstate.kvm_vcpu = NULL;
2574         tpaca->kvm_hstate.kvm_vcore = NULL;
2575         tpaca->kvm_hstate.kvm_split_mode = NULL;
2576 }
2577
2578 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2579 {
2580         struct kvm_nested_guest *nested = vcpu->arch.nested;
2581         cpumask_t *cpu_in_guest;
2582         int i;
2583
2584         cpu = cpu_first_tlb_thread_sibling(cpu);
2585         if (nested) {
2586                 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2587                 cpu_in_guest = &nested->cpu_in_guest;
2588         } else {
2589                 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2590                 cpu_in_guest = &kvm->arch.cpu_in_guest;
2591         }
2592         /*
2593          * Make sure setting of bit in need_tlb_flush precedes
2594          * testing of cpu_in_guest bits.  The matching barrier on
2595          * the other side is the first smp_mb() in kvmppc_run_core().
2596          */
2597         smp_mb();
2598         for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
2599                                         i += cpu_tlb_thread_sibling_step())
2600                 if (cpumask_test_cpu(i, cpu_in_guest))
2601                         smp_call_function_single(i, do_nothing, NULL, 1);
2602 }
2603
2604 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2605 {
2606         struct kvm_nested_guest *nested = vcpu->arch.nested;
2607         struct kvm *kvm = vcpu->kvm;
2608         int prev_cpu;
2609
2610         if (!cpu_has_feature(CPU_FTR_HVMODE))
2611                 return;
2612
2613         if (nested)
2614                 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2615         else
2616                 prev_cpu = vcpu->arch.prev_cpu;
2617
2618         /*
2619          * With radix, the guest can do TLB invalidations itself,
2620          * and it could choose to use the local form (tlbiel) if
2621          * it is invalidating a translation that has only ever been
2622          * used on one vcpu.  However, that doesn't mean it has
2623          * only ever been used on one physical cpu, since vcpus
2624          * can move around between pcpus.  To cope with this, when
2625          * a vcpu moves from one pcpu to another, we need to tell
2626          * any vcpus running on the same core as this vcpu previously
2627          * ran to flush the TLB.  The TLB is shared between threads,
2628          * so we use a single bit in .need_tlb_flush for all 4 threads.
2629          */
2630         if (prev_cpu != pcpu) {
2631                 if (prev_cpu >= 0 &&
2632                     cpu_first_tlb_thread_sibling(prev_cpu) !=
2633                     cpu_first_tlb_thread_sibling(pcpu))
2634                         radix_flush_cpu(kvm, prev_cpu, vcpu);
2635                 if (nested)
2636                         nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2637                 else
2638                         vcpu->arch.prev_cpu = pcpu;
2639         }
2640 }
2641
2642 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2643 {
2644         int cpu;
2645         struct paca_struct *tpaca;
2646         struct kvm *kvm = vc->kvm;
2647
2648         cpu = vc->pcpu;
2649         if (vcpu) {
2650                 if (vcpu->arch.timer_running) {
2651                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2652                         vcpu->arch.timer_running = 0;
2653                 }
2654                 cpu += vcpu->arch.ptid;
2655                 vcpu->cpu = vc->pcpu;
2656                 vcpu->arch.thread_cpu = cpu;
2657                 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2658         }
2659         tpaca = paca_ptrs[cpu];
2660         tpaca->kvm_hstate.kvm_vcpu = vcpu;
2661         tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2662         tpaca->kvm_hstate.fake_suspend = 0;
2663         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2664         smp_wmb();
2665         tpaca->kvm_hstate.kvm_vcore = vc;
2666         if (cpu != smp_processor_id())
2667                 kvmppc_ipi_thread(cpu);
2668 }
2669
2670 static void kvmppc_wait_for_nap(int n_threads)
2671 {
2672         int cpu = smp_processor_id();
2673         int i, loops;
2674
2675         if (n_threads <= 1)
2676                 return;
2677         for (loops = 0; loops < 1000000; ++loops) {
2678                 /*
2679                  * Check if all threads are finished.
2680                  * We set the vcore pointer when starting a thread
2681                  * and the thread clears it when finished, so we look
2682                  * for any threads that still have a non-NULL vcore ptr.
2683                  */
2684                 for (i = 1; i < n_threads; ++i)
2685                         if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2686                                 break;
2687                 if (i == n_threads) {
2688                         HMT_medium();
2689                         return;
2690                 }
2691                 HMT_low();
2692         }
2693         HMT_medium();
2694         for (i = 1; i < n_threads; ++i)
2695                 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2696                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2697 }
2698
2699 /*
2700  * Check that we are on thread 0 and that any other threads in
2701  * this core are off-line.  Then grab the threads so they can't
2702  * enter the kernel.
2703  */
2704 static int on_primary_thread(void)
2705 {
2706         int cpu = smp_processor_id();
2707         int thr;
2708
2709         /* Are we on a primary subcore? */
2710         if (cpu_thread_in_subcore(cpu))
2711                 return 0;
2712
2713         thr = 0;
2714         while (++thr < threads_per_subcore)
2715                 if (cpu_online(cpu + thr))
2716                         return 0;
2717
2718         /* Grab all hw threads so they can't go into the kernel */
2719         for (thr = 1; thr < threads_per_subcore; ++thr) {
2720                 if (kvmppc_grab_hwthread(cpu + thr)) {
2721                         /* Couldn't grab one; let the others go */
2722                         do {
2723                                 kvmppc_release_hwthread(cpu + thr);
2724                         } while (--thr > 0);
2725                         return 0;
2726                 }
2727         }
2728         return 1;
2729 }
2730
2731 /*
2732  * A list of virtual cores for each physical CPU.
2733  * These are vcores that could run but their runner VCPU tasks are
2734  * (or may be) preempted.
2735  */
2736 struct preempted_vcore_list {
2737         struct list_head        list;
2738         spinlock_t              lock;
2739 };
2740
2741 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2742
2743 static void init_vcore_lists(void)
2744 {
2745         int cpu;
2746
2747         for_each_possible_cpu(cpu) {
2748                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2749                 spin_lock_init(&lp->lock);
2750                 INIT_LIST_HEAD(&lp->list);
2751         }
2752 }
2753
2754 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2755 {
2756         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2757
2758         vc->vcore_state = VCORE_PREEMPT;
2759         vc->pcpu = smp_processor_id();
2760         if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2761                 spin_lock(&lp->lock);
2762                 list_add_tail(&vc->preempt_list, &lp->list);
2763                 spin_unlock(&lp->lock);
2764         }
2765
2766         /* Start accumulating stolen time */
2767         kvmppc_core_start_stolen(vc);
2768 }
2769
2770 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2771 {
2772         struct preempted_vcore_list *lp;
2773
2774         kvmppc_core_end_stolen(vc);
2775         if (!list_empty(&vc->preempt_list)) {
2776                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2777                 spin_lock(&lp->lock);
2778                 list_del_init(&vc->preempt_list);
2779                 spin_unlock(&lp->lock);
2780         }
2781         vc->vcore_state = VCORE_INACTIVE;
2782 }
2783
2784 /*
2785  * This stores information about the virtual cores currently
2786  * assigned to a physical core.
2787  */
2788 struct core_info {
2789         int             n_subcores;
2790         int             max_subcore_threads;
2791         int             total_threads;
2792         int             subcore_threads[MAX_SUBCORES];
2793         struct kvmppc_vcore *vc[MAX_SUBCORES];
2794 };
2795
2796 /*
2797  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2798  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2799  */
2800 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2801
2802 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2803 {
2804         memset(cip, 0, sizeof(*cip));
2805         cip->n_subcores = 1;
2806         cip->max_subcore_threads = vc->num_threads;
2807         cip->total_threads = vc->num_threads;
2808         cip->subcore_threads[0] = vc->num_threads;
2809         cip->vc[0] = vc;
2810 }
2811
2812 static bool subcore_config_ok(int n_subcores, int n_threads)
2813 {
2814         /*
2815          * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2816          * split-core mode, with one thread per subcore.
2817          */
2818         if (cpu_has_feature(CPU_FTR_ARCH_300))
2819                 return n_subcores <= 4 && n_threads == 1;
2820
2821         /* On POWER8, can only dynamically split if unsplit to begin with */
2822         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2823                 return false;
2824         if (n_subcores > MAX_SUBCORES)
2825                 return false;
2826         if (n_subcores > 1) {
2827                 if (!(dynamic_mt_modes & 2))
2828                         n_subcores = 4;
2829                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2830                         return false;
2831         }
2832
2833         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2834 }
2835
2836 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2837 {
2838         vc->entry_exit_map = 0;
2839         vc->in_guest = 0;
2840         vc->napping_threads = 0;
2841         vc->conferring_threads = 0;
2842         vc->tb_offset_applied = 0;
2843 }
2844
2845 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2846 {
2847         int n_threads = vc->num_threads;
2848         int sub;
2849
2850         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2851                 return false;
2852
2853         /* In one_vm_per_core mode, require all vcores to be from the same vm */
2854         if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2855                 return false;
2856
2857         /* Some POWER9 chips require all threads to be in the same MMU mode */
2858         if (no_mixing_hpt_and_radix &&
2859             kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2860                 return false;
2861
2862         if (n_threads < cip->max_subcore_threads)
2863                 n_threads = cip->max_subcore_threads;
2864         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2865                 return false;
2866         cip->max_subcore_threads = n_threads;
2867
2868         sub = cip->n_subcores;
2869         ++cip->n_subcores;
2870         cip->total_threads += vc->num_threads;
2871         cip->subcore_threads[sub] = vc->num_threads;
2872         cip->vc[sub] = vc;
2873         init_vcore_to_run(vc);
2874         list_del_init(&vc->preempt_list);
2875
2876         return true;
2877 }
2878
2879 /*
2880  * Work out whether it is possible to piggyback the execution of
2881  * vcore *pvc onto the execution of the other vcores described in *cip.
2882  */
2883 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2884                           int target_threads)
2885 {
2886         if (cip->total_threads + pvc->num_threads > target_threads)
2887                 return false;
2888
2889         return can_dynamic_split(pvc, cip);
2890 }
2891
2892 static void prepare_threads(struct kvmppc_vcore *vc)
2893 {
2894         int i;
2895         struct kvm_vcpu *vcpu;
2896
2897         for_each_runnable_thread(i, vcpu, vc) {
2898                 if (signal_pending(vcpu->arch.run_task))
2899                         vcpu->arch.ret = -EINTR;
2900                 else if (vcpu->arch.vpa.update_pending ||
2901                          vcpu->arch.slb_shadow.update_pending ||
2902                          vcpu->arch.dtl.update_pending)
2903                         vcpu->arch.ret = RESUME_GUEST;
2904                 else
2905                         continue;
2906                 kvmppc_remove_runnable(vc, vcpu);
2907                 wake_up(&vcpu->arch.cpu_run);
2908         }
2909 }
2910
2911 static void collect_piggybacks(struct core_info *cip, int target_threads)
2912 {
2913         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2914         struct kvmppc_vcore *pvc, *vcnext;
2915
2916         spin_lock(&lp->lock);
2917         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2918                 if (!spin_trylock(&pvc->lock))
2919                         continue;
2920                 prepare_threads(pvc);
2921                 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
2922                         list_del_init(&pvc->preempt_list);
2923                         if (pvc->runner == NULL) {
2924                                 pvc->vcore_state = VCORE_INACTIVE;
2925                                 kvmppc_core_end_stolen(pvc);
2926                         }
2927                         spin_unlock(&pvc->lock);
2928                         continue;
2929                 }
2930                 if (!can_piggyback(pvc, cip, target_threads)) {
2931                         spin_unlock(&pvc->lock);
2932                         continue;
2933                 }
2934                 kvmppc_core_end_stolen(pvc);
2935                 pvc->vcore_state = VCORE_PIGGYBACK;
2936                 if (cip->total_threads >= target_threads)
2937                         break;
2938         }
2939         spin_unlock(&lp->lock);
2940 }
2941
2942 static bool recheck_signals_and_mmu(struct core_info *cip)
2943 {
2944         int sub, i;
2945         struct kvm_vcpu *vcpu;
2946         struct kvmppc_vcore *vc;
2947
2948         for (sub = 0; sub < cip->n_subcores; ++sub) {
2949                 vc = cip->vc[sub];
2950                 if (!vc->kvm->arch.mmu_ready)
2951                         return true;
2952                 for_each_runnable_thread(i, vcpu, vc)
2953                         if (signal_pending(vcpu->arch.run_task))
2954                                 return true;
2955         }
2956         return false;
2957 }
2958
2959 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2960 {
2961         int still_running = 0, i;
2962         u64 now;
2963         long ret;
2964         struct kvm_vcpu *vcpu;
2965
2966         spin_lock(&vc->lock);
2967         now = get_tb();
2968         for_each_runnable_thread(i, vcpu, vc) {
2969                 /*
2970                  * It's safe to unlock the vcore in the loop here, because
2971                  * for_each_runnable_thread() is safe against removal of
2972                  * the vcpu, and the vcore state is VCORE_EXITING here,
2973                  * so any vcpus becoming runnable will have their arch.trap
2974                  * set to zero and can't actually run in the guest.
2975                  */
2976                 spin_unlock(&vc->lock);
2977                 /* cancel pending dec exception if dec is positive */
2978                 if (now < vcpu->arch.dec_expires &&
2979                     kvmppc_core_pending_dec(vcpu))
2980                         kvmppc_core_dequeue_dec(vcpu);
2981
2982                 trace_kvm_guest_exit(vcpu);
2983
2984                 ret = RESUME_GUEST;
2985                 if (vcpu->arch.trap)
2986                         ret = kvmppc_handle_exit_hv(vcpu,
2987                                                     vcpu->arch.run_task);
2988
2989                 vcpu->arch.ret = ret;
2990                 vcpu->arch.trap = 0;
2991
2992                 spin_lock(&vc->lock);
2993                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2994                         if (vcpu->arch.pending_exceptions)
2995                                 kvmppc_core_prepare_to_enter(vcpu);
2996                         if (vcpu->arch.ceded)
2997                                 kvmppc_set_timer(vcpu);
2998                         else
2999                                 ++still_running;
3000                 } else {
3001                         kvmppc_remove_runnable(vc, vcpu);
3002                         wake_up(&vcpu->arch.cpu_run);
3003                 }
3004         }
3005         if (!is_master) {
3006                 if (still_running > 0) {
3007                         kvmppc_vcore_preempt(vc);
3008                 } else if (vc->runner) {
3009                         vc->vcore_state = VCORE_PREEMPT;
3010                         kvmppc_core_start_stolen(vc);
3011                 } else {
3012                         vc->vcore_state = VCORE_INACTIVE;
3013                 }
3014                 if (vc->n_runnable > 0 && vc->runner == NULL) {
3015                         /* make sure there's a candidate runner awake */
3016                         i = -1;
3017                         vcpu = next_runnable_thread(vc, &i);
3018                         wake_up(&vcpu->arch.cpu_run);
3019                 }
3020         }
3021         spin_unlock(&vc->lock);
3022 }
3023
3024 /*
3025  * Clear core from the list of active host cores as we are about to
3026  * enter the guest. Only do this if it is the primary thread of the
3027  * core (not if a subcore) that is entering the guest.
3028  */
3029 static inline int kvmppc_clear_host_core(unsigned int cpu)
3030 {
3031         int core;
3032
3033         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3034                 return 0;
3035         /*
3036          * Memory barrier can be omitted here as we will do a smp_wmb()
3037          * later in kvmppc_start_thread and we need ensure that state is
3038          * visible to other CPUs only after we enter guest.
3039          */
3040         core = cpu >> threads_shift;
3041         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3042         return 0;
3043 }
3044
3045 /*
3046  * Advertise this core as an active host core since we exited the guest
3047  * Only need to do this if it is the primary thread of the core that is
3048  * exiting.
3049  */
3050 static inline int kvmppc_set_host_core(unsigned int cpu)
3051 {
3052         int core;
3053
3054         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3055                 return 0;
3056
3057         /*
3058          * Memory barrier can be omitted here because we do a spin_unlock
3059          * immediately after this which provides the memory barrier.
3060          */
3061         core = cpu >> threads_shift;
3062         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3063         return 0;
3064 }
3065
3066 static void set_irq_happened(int trap)
3067 {
3068         switch (trap) {
3069         case BOOK3S_INTERRUPT_EXTERNAL:
3070                 local_paca->irq_happened |= PACA_IRQ_EE;
3071                 break;
3072         case BOOK3S_INTERRUPT_H_DOORBELL:
3073                 local_paca->irq_happened |= PACA_IRQ_DBELL;
3074                 break;
3075         case BOOK3S_INTERRUPT_HMI:
3076                 local_paca->irq_happened |= PACA_IRQ_HMI;
3077                 break;
3078         case BOOK3S_INTERRUPT_SYSTEM_RESET:
3079                 replay_system_reset();
3080                 break;
3081         }
3082 }
3083
3084 /*
3085  * Run a set of guest threads on a physical core.
3086  * Called with vc->lock held.
3087  */
3088 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3089 {
3090         struct kvm_vcpu *vcpu;
3091         int i;
3092         int srcu_idx;
3093         struct core_info core_info;
3094         struct kvmppc_vcore *pvc;
3095         struct kvm_split_mode split_info, *sip;
3096         int split, subcore_size, active;
3097         int sub;
3098         bool thr0_done;
3099         unsigned long cmd_bit, stat_bit;
3100         int pcpu, thr;
3101         int target_threads;
3102         int controlled_threads;
3103         int trap;
3104         bool is_power8;
3105         bool hpt_on_radix;
3106
3107         /*
3108          * Remove from the list any threads that have a signal pending
3109          * or need a VPA update done
3110          */
3111         prepare_threads(vc);
3112
3113         /* if the runner is no longer runnable, let the caller pick a new one */
3114         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3115                 return;
3116
3117         /*
3118          * Initialize *vc.
3119          */
3120         init_vcore_to_run(vc);
3121         vc->preempt_tb = TB_NIL;
3122
3123         /*
3124          * Number of threads that we will be controlling: the same as
3125          * the number of threads per subcore, except on POWER9,
3126          * where it's 1 because the threads are (mostly) independent.
3127          */
3128         controlled_threads = threads_per_vcore(vc->kvm);
3129
3130         /*
3131          * Make sure we are running on primary threads, and that secondary
3132          * threads are offline.  Also check if the number of threads in this
3133          * guest are greater than the current system threads per guest.
3134          * On POWER9, we need to be not in independent-threads mode if
3135          * this is a HPT guest on a radix host machine where the
3136          * CPU threads may not be in different MMU modes.
3137          */
3138         hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3139                 !kvm_is_radix(vc->kvm);
3140         if (((controlled_threads > 1) &&
3141              ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3142             (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3143                 for_each_runnable_thread(i, vcpu, vc) {
3144                         vcpu->arch.ret = -EBUSY;
3145                         kvmppc_remove_runnable(vc, vcpu);
3146                         wake_up(&vcpu->arch.cpu_run);
3147                 }
3148                 goto out;
3149         }
3150
3151         /*
3152          * See if we could run any other vcores on the physical core
3153          * along with this one.
3154          */
3155         init_core_info(&core_info, vc);
3156         pcpu = smp_processor_id();
3157         target_threads = controlled_threads;
3158         if (target_smt_mode && target_smt_mode < target_threads)
3159                 target_threads = target_smt_mode;
3160         if (vc->num_threads < target_threads)
3161                 collect_piggybacks(&core_info, target_threads);
3162
3163         /*
3164          * On radix, arrange for TLB flushing if necessary.
3165          * This has to be done before disabling interrupts since
3166          * it uses smp_call_function().
3167          */
3168         pcpu = smp_processor_id();
3169         if (kvm_is_radix(vc->kvm)) {
3170                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3171                         for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3172                                 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3173         }
3174
3175         /*
3176          * Hard-disable interrupts, and check resched flag and signals.
3177          * If we need to reschedule or deliver a signal, clean up
3178          * and return without going into the guest(s).
3179          * If the mmu_ready flag has been cleared, don't go into the
3180          * guest because that means a HPT resize operation is in progress.
3181          */
3182         local_irq_disable();
3183         hard_irq_disable();
3184         if (lazy_irq_pending() || need_resched() ||
3185             recheck_signals_and_mmu(&core_info)) {
3186                 local_irq_enable();
3187                 vc->vcore_state = VCORE_INACTIVE;
3188                 /* Unlock all except the primary vcore */
3189                 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3190                         pvc = core_info.vc[sub];
3191                         /* Put back on to the preempted vcores list */
3192                         kvmppc_vcore_preempt(pvc);
3193                         spin_unlock(&pvc->lock);
3194                 }
3195                 for (i = 0; i < controlled_threads; ++i)
3196                         kvmppc_release_hwthread(pcpu + i);
3197                 return;
3198         }
3199
3200         kvmppc_clear_host_core(pcpu);
3201
3202         /* Decide on micro-threading (split-core) mode */
3203         subcore_size = threads_per_subcore;
3204         cmd_bit = stat_bit = 0;
3205         split = core_info.n_subcores;
3206         sip = NULL;
3207         is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3208                 && !cpu_has_feature(CPU_FTR_ARCH_300);
3209
3210         if (split > 1 || hpt_on_radix) {
3211                 sip = &split_info;
3212                 memset(&split_info, 0, sizeof(split_info));
3213                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3214                         split_info.vc[sub] = core_info.vc[sub];
3215
3216                 if (is_power8) {
3217                         if (split == 2 && (dynamic_mt_modes & 2)) {
3218                                 cmd_bit = HID0_POWER8_1TO2LPAR;
3219                                 stat_bit = HID0_POWER8_2LPARMODE;
3220                         } else {
3221                                 split = 4;
3222                                 cmd_bit = HID0_POWER8_1TO4LPAR;
3223                                 stat_bit = HID0_POWER8_4LPARMODE;
3224                         }
3225                         subcore_size = MAX_SMT_THREADS / split;
3226                         split_info.rpr = mfspr(SPRN_RPR);
3227                         split_info.pmmar = mfspr(SPRN_PMMAR);
3228                         split_info.ldbar = mfspr(SPRN_LDBAR);
3229                         split_info.subcore_size = subcore_size;
3230                 } else {
3231                         split_info.subcore_size = 1;
3232                         if (hpt_on_radix) {
3233                                 /* Use the split_info for LPCR/LPIDR changes */
3234                                 split_info.lpcr_req = vc->lpcr;
3235                                 split_info.lpidr_req = vc->kvm->arch.lpid;
3236                                 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3237                                 split_info.do_set = 1;
3238                         }
3239                 }
3240
3241                 /* order writes to split_info before kvm_split_mode pointer */
3242                 smp_wmb();
3243         }
3244
3245         for (thr = 0; thr < controlled_threads; ++thr) {
3246                 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3247
3248                 paca->kvm_hstate.tid = thr;
3249                 paca->kvm_hstate.napping = 0;
3250                 paca->kvm_hstate.kvm_split_mode = sip;
3251         }
3252
3253         /* Initiate micro-threading (split-core) on POWER8 if required */
3254         if (cmd_bit) {
3255                 unsigned long hid0 = mfspr(SPRN_HID0);
3256
3257                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3258                 mb();
3259                 mtspr(SPRN_HID0, hid0);
3260                 isync();
3261                 for (;;) {
3262                         hid0 = mfspr(SPRN_HID0);
3263                         if (hid0 & stat_bit)
3264                                 break;
3265                         cpu_relax();
3266                 }
3267         }
3268
3269         /*
3270          * On POWER8, set RWMR register.
3271          * Since it only affects PURR and SPURR, it doesn't affect
3272          * the host, so we don't save/restore the host value.
3273          */
3274         if (is_power8) {
3275                 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3276                 int n_online = atomic_read(&vc->online_count);
3277
3278                 /*
3279                  * Use the 8-thread value if we're doing split-core
3280                  * or if the vcore's online count looks bogus.
3281                  */
3282                 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3283                     n_online >= 1 && n_online <= MAX_SMT_THREADS)
3284                         rwmr_val = p8_rwmr_values[n_online];
3285                 mtspr(SPRN_RWMR, rwmr_val);
3286         }
3287
3288         /* Start all the threads */
3289         active = 0;
3290         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3291                 thr = is_power8 ? subcore_thread_map[sub] : sub;
3292                 thr0_done = false;
3293                 active |= 1 << thr;
3294                 pvc = core_info.vc[sub];
3295                 pvc->pcpu = pcpu + thr;
3296                 for_each_runnable_thread(i, vcpu, pvc) {
3297                         kvmppc_start_thread(vcpu, pvc);
3298                         kvmppc_create_dtl_entry(vcpu, pvc);
3299                         trace_kvm_guest_enter(vcpu);
3300                         if (!vcpu->arch.ptid)
3301                                 thr0_done = true;
3302                         active |= 1 << (thr + vcpu->arch.ptid);
3303                 }
3304                 /*
3305                  * We need to start the first thread of each subcore
3306                  * even if it doesn't have a vcpu.
3307                  */
3308                 if (!thr0_done)
3309                         kvmppc_start_thread(NULL, pvc);
3310         }
3311
3312         /*
3313          * Ensure that split_info.do_nap is set after setting
3314          * the vcore pointer in the PACA of the secondaries.
3315          */
3316         smp_mb();
3317
3318         /*
3319          * When doing micro-threading, poke the inactive threads as well.
3320          * This gets them to the nap instruction after kvm_do_nap,
3321          * which reduces the time taken to unsplit later.
3322          * For POWER9 HPT guest on radix host, we need all the secondary
3323          * threads woken up so they can do the LPCR/LPIDR change.
3324          */
3325         if (cmd_bit || hpt_on_radix) {
3326                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
3327                 for (thr = 1; thr < threads_per_subcore; ++thr)
3328                         if (!(active & (1 << thr)))
3329                                 kvmppc_ipi_thread(pcpu + thr);
3330         }
3331
3332         vc->vcore_state = VCORE_RUNNING;
3333         preempt_disable();
3334
3335         trace_kvmppc_run_core(vc, 0);
3336
3337         for (sub = 0; sub < core_info.n_subcores; ++sub)
3338                 spin_unlock(&core_info.vc[sub]->lock);
3339
3340         guest_enter_irqoff();
3341
3342         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3343
3344         this_cpu_disable_ftrace();
3345
3346         /*
3347          * Interrupts will be enabled once we get into the guest,
3348          * so tell lockdep that we're about to enable interrupts.
3349          */
3350         trace_hardirqs_on();
3351
3352         trap = __kvmppc_vcore_entry();
3353
3354         trace_hardirqs_off();
3355
3356         this_cpu_enable_ftrace();
3357
3358         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3359
3360         set_irq_happened(trap);
3361
3362         spin_lock(&vc->lock);
3363         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3364         vc->vcore_state = VCORE_EXITING;
3365
3366         /* wait for secondary threads to finish writing their state to memory */
3367         kvmppc_wait_for_nap(controlled_threads);
3368
3369         /* Return to whole-core mode if we split the core earlier */
3370         if (cmd_bit) {
3371                 unsigned long hid0 = mfspr(SPRN_HID0);
3372                 unsigned long loops = 0;
3373
3374                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3375                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3376                 mb();
3377                 mtspr(SPRN_HID0, hid0);
3378                 isync();
3379                 for (;;) {
3380                         hid0 = mfspr(SPRN_HID0);
3381                         if (!(hid0 & stat_bit))
3382                                 break;
3383                         cpu_relax();
3384                         ++loops;
3385                 }
3386         } else if (hpt_on_radix) {
3387                 /* Wait for all threads to have seen final sync */
3388                 for (thr = 1; thr < controlled_threads; ++thr) {
3389                         struct paca_struct *paca = paca_ptrs[pcpu + thr];
3390
3391                         while (paca->kvm_hstate.kvm_split_mode) {
3392                                 HMT_low();
3393                                 barrier();
3394                         }
3395                         HMT_medium();
3396                 }
3397         }
3398         split_info.do_nap = 0;
3399
3400         kvmppc_set_host_core(pcpu);
3401
3402         context_tracking_guest_exit();
3403         if (!vtime_accounting_enabled_this_cpu()) {
3404                 local_irq_enable();
3405                 /*
3406                  * Service IRQs here before vtime_account_guest_exit() so any
3407                  * ticks that occurred while running the guest are accounted to
3408                  * the guest. If vtime accounting is enabled, accounting uses
3409                  * TB rather than ticks, so it can be done without enabling
3410                  * interrupts here, which has the problem that it accounts
3411                  * interrupt processing overhead to the host.
3412                  */
3413                 local_irq_disable();
3414         }
3415         vtime_account_guest_exit();
3416
3417         local_irq_enable();
3418
3419         /* Let secondaries go back to the offline loop */
3420         for (i = 0; i < controlled_threads; ++i) {
3421                 kvmppc_release_hwthread(pcpu + i);
3422                 if (sip && sip->napped[i])
3423                         kvmppc_ipi_thread(pcpu + i);
3424                 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3425         }
3426
3427         spin_unlock(&vc->lock);
3428
3429         /* make sure updates to secondary vcpu structs are visible now */
3430         smp_mb();
3431
3432         preempt_enable();
3433
3434         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3435                 pvc = core_info.vc[sub];
3436                 post_guest_process(pvc, pvc == vc);
3437         }
3438
3439         spin_lock(&vc->lock);
3440
3441  out:
3442         vc->vcore_state = VCORE_INACTIVE;
3443         trace_kvmppc_run_core(vc, 1);
3444 }
3445
3446 /*
3447  * Load up hypervisor-mode registers on P9.
3448  */
3449 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3450                                      unsigned long lpcr)
3451 {
3452         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3453         s64 hdec;
3454         u64 tb, purr, spurr;
3455         int trap;
3456         unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3457         unsigned long host_ciabr = mfspr(SPRN_CIABR);
3458         unsigned long host_dawr = mfspr(SPRN_DAWR0);
3459         unsigned long host_dawrx = mfspr(SPRN_DAWRX0);
3460         unsigned long host_psscr = mfspr(SPRN_PSSCR);
3461         unsigned long host_pidr = mfspr(SPRN_PID);
3462
3463         /*
3464          * P8 and P9 suppress the HDEC exception when LPCR[HDICE] = 0,
3465          * so set HDICE before writing HDEC.
3466          */
3467         mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr | LPCR_HDICE);
3468         isync();
3469
3470         hdec = time_limit - mftb();
3471         if (hdec < 0) {
3472                 mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3473                 isync();
3474                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3475         }
3476         mtspr(SPRN_HDEC, hdec);
3477
3478         if (vc->tb_offset) {
3479                 u64 new_tb = mftb() + vc->tb_offset;
3480                 mtspr(SPRN_TBU40, new_tb);
3481                 tb = mftb();
3482                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3483                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3484                 vc->tb_offset_applied = vc->tb_offset;
3485         }
3486
3487         if (vc->pcr)
3488                 mtspr(SPRN_PCR, vc->pcr | PCR_MASK);
3489         mtspr(SPRN_DPDES, vc->dpdes);
3490         mtspr(SPRN_VTB, vc->vtb);
3491
3492         local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3493         local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3494         mtspr(SPRN_PURR, vcpu->arch.purr);
3495         mtspr(SPRN_SPURR, vcpu->arch.spurr);
3496
3497         if (dawr_enabled()) {
3498                 mtspr(SPRN_DAWR0, vcpu->arch.dawr);
3499                 mtspr(SPRN_DAWRX0, vcpu->arch.dawrx);
3500         }
3501         mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3502         mtspr(SPRN_IC, vcpu->arch.ic);
3503         mtspr(SPRN_PID, vcpu->arch.pid);
3504
3505         mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3506               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3507
3508         mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3509
3510         mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3511         mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3512         mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3513         mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3514
3515         mtspr(SPRN_AMOR, ~0UL);
3516
3517         mtspr(SPRN_LPCR, lpcr);
3518         isync();
3519
3520         kvmppc_xive_push_vcpu(vcpu);
3521
3522         mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3523         mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3524
3525         trap = __kvmhv_vcpu_entry_p9(vcpu);
3526
3527         /* Advance host PURR/SPURR by the amount used by guest */
3528         purr = mfspr(SPRN_PURR);
3529         spurr = mfspr(SPRN_SPURR);
3530         mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3531               purr - vcpu->arch.purr);
3532         mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3533               spurr - vcpu->arch.spurr);
3534         vcpu->arch.purr = purr;
3535         vcpu->arch.spurr = spurr;
3536
3537         vcpu->arch.ic = mfspr(SPRN_IC);
3538         vcpu->arch.pid = mfspr(SPRN_PID);
3539         vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3540
3541         vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3542         vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3543         vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3544         vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3545
3546         /* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3547         mtspr(SPRN_PSSCR, host_psscr |
3548               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3549         mtspr(SPRN_HFSCR, host_hfscr);
3550         mtspr(SPRN_CIABR, host_ciabr);
3551         mtspr(SPRN_DAWR0, host_dawr);
3552         mtspr(SPRN_DAWRX0, host_dawrx);
3553         mtspr(SPRN_PID, host_pidr);
3554
3555         /*
3556          * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3557          * case we interrupted the guest between a tlbie and a ptesync.
3558          */
3559         asm volatile("eieio; tlbsync; ptesync");
3560
3561         /*
3562          * cp_abort is required if the processor supports local copy-paste
3563          * to clear the copy buffer that was under control of the guest.
3564          */
3565         if (cpu_has_feature(CPU_FTR_ARCH_31))
3566                 asm volatile(PPC_CP_ABORT);
3567
3568         mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid);    /* restore host LPID */
3569         isync();
3570
3571         vc->dpdes = mfspr(SPRN_DPDES);
3572         vc->vtb = mfspr(SPRN_VTB);
3573         mtspr(SPRN_DPDES, 0);
3574         if (vc->pcr)
3575                 mtspr(SPRN_PCR, PCR_MASK);
3576
3577         if (vc->tb_offset_applied) {
3578                 u64 new_tb = mftb() - vc->tb_offset_applied;
3579                 mtspr(SPRN_TBU40, new_tb);
3580                 tb = mftb();
3581                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3582                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3583                 vc->tb_offset_applied = 0;
3584         }
3585
3586         mtspr(SPRN_HDEC, 0x7fffffff);
3587         mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3588
3589         return trap;
3590 }
3591
3592 /*
3593  * Virtual-mode guest entry for POWER9 and later when the host and
3594  * guest are both using the radix MMU.  The LPIDR has already been set.
3595  */
3596 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3597                          unsigned long lpcr)
3598 {
3599         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3600         unsigned long host_dscr = mfspr(SPRN_DSCR);
3601         unsigned long host_tidr = mfspr(SPRN_TIDR);
3602         unsigned long host_iamr = mfspr(SPRN_IAMR);
3603         unsigned long host_amr = mfspr(SPRN_AMR);
3604         unsigned long host_fscr = mfspr(SPRN_FSCR);
3605         s64 dec;
3606         u64 tb;
3607         int trap, save_pmu;
3608
3609         dec = mfspr(SPRN_DEC);
3610         tb = mftb();
3611         if (dec < 0)
3612                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3613         local_paca->kvm_hstate.dec_expires = dec + tb;
3614         if (local_paca->kvm_hstate.dec_expires < time_limit)
3615                 time_limit = local_paca->kvm_hstate.dec_expires;
3616
3617         vcpu->arch.ceded = 0;
3618
3619         kvmhv_save_host_pmu();          /* saves it to PACA kvm_hstate */
3620
3621         kvmppc_subcore_enter_guest();
3622
3623         vc->entry_exit_map = 1;
3624         vc->in_guest = 1;
3625
3626         if (vcpu->arch.vpa.pinned_addr) {
3627                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3628                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3629                 lp->yield_count = cpu_to_be32(yield_count);
3630                 vcpu->arch.vpa.dirty = 1;
3631         }
3632
3633         if (cpu_has_feature(CPU_FTR_TM) ||
3634             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3635                 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3636
3637 #ifdef CONFIG_PPC_PSERIES
3638         if (kvmhv_on_pseries()) {
3639                 barrier();
3640                 if (vcpu->arch.vpa.pinned_addr) {
3641                         struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3642                         get_lppaca()->pmcregs_in_use = lp->pmcregs_in_use;
3643                 } else {
3644                         get_lppaca()->pmcregs_in_use = 1;
3645                 }
3646                 barrier();
3647         }
3648 #endif
3649         kvmhv_load_guest_pmu(vcpu);
3650
3651         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3652         load_fp_state(&vcpu->arch.fp);
3653 #ifdef CONFIG_ALTIVEC
3654         load_vr_state(&vcpu->arch.vr);
3655 #endif
3656         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3657
3658         mtspr(SPRN_DSCR, vcpu->arch.dscr);
3659         mtspr(SPRN_IAMR, vcpu->arch.iamr);
3660         mtspr(SPRN_PSPB, vcpu->arch.pspb);
3661         mtspr(SPRN_FSCR, vcpu->arch.fscr);
3662         mtspr(SPRN_TAR, vcpu->arch.tar);
3663         mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3664         mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3665         mtspr(SPRN_BESCR, vcpu->arch.bescr);
3666         mtspr(SPRN_WORT, vcpu->arch.wort);
3667         mtspr(SPRN_TIDR, vcpu->arch.tid);
3668         mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3669         mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3670         mtspr(SPRN_AMR, vcpu->arch.amr);
3671         mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3672
3673         if (!(vcpu->arch.ctrl & 1))
3674                 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3675
3676         mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3677
3678         if (kvmhv_on_pseries()) {
3679                 /*
3680                  * We need to save and restore the guest visible part of the
3681                  * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3682                  * doesn't do this for us. Note only required if pseries since
3683                  * this is done in kvmhv_load_hv_regs_and_go() below otherwise.
3684                  */
3685                 unsigned long host_psscr;
3686                 /* call our hypervisor to load up HV regs and go */
3687                 struct hv_guest_state hvregs;
3688
3689                 host_psscr = mfspr(SPRN_PSSCR_PR);
3690                 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3691                 kvmhv_save_hv_regs(vcpu, &hvregs);
3692                 hvregs.lpcr = lpcr;
3693                 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3694                 hvregs.version = HV_GUEST_STATE_VERSION;
3695                 if (vcpu->arch.nested) {
3696                         hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3697                         hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3698                 } else {
3699                         hvregs.lpid = vcpu->kvm->arch.lpid;
3700                         hvregs.vcpu_token = vcpu->vcpu_id;
3701                 }
3702                 hvregs.hdec_expiry = time_limit;
3703                 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3704                                           __pa(&vcpu->arch.regs));
3705                 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3706                 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3707                 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3708                 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3709                 vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3710                 mtspr(SPRN_PSSCR_PR, host_psscr);
3711
3712                 /* H_CEDE has to be handled now, not later */
3713                 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3714                     kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3715                         kvmppc_nested_cede(vcpu);
3716                         kvmppc_set_gpr(vcpu, 3, 0);
3717                         trap = 0;
3718                 }
3719         } else {
3720                 trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3721         }
3722
3723         vcpu->arch.slb_max = 0;
3724         dec = mfspr(SPRN_DEC);
3725         if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
3726                 dec = (s32) dec;
3727         tb = mftb();
3728         vcpu->arch.dec_expires = dec + tb;
3729         vcpu->cpu = -1;
3730         vcpu->arch.thread_cpu = -1;
3731         /* Save guest CTRL register, set runlatch to 1 */
3732         vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3733         if (!(vcpu->arch.ctrl & 1))
3734                 mtspr(SPRN_CTRLT, vcpu->arch.ctrl | 1);
3735
3736         vcpu->arch.iamr = mfspr(SPRN_IAMR);
3737         vcpu->arch.pspb = mfspr(SPRN_PSPB);
3738         vcpu->arch.fscr = mfspr(SPRN_FSCR);
3739         vcpu->arch.tar = mfspr(SPRN_TAR);
3740         vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3741         vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3742         vcpu->arch.bescr = mfspr(SPRN_BESCR);
3743         vcpu->arch.wort = mfspr(SPRN_WORT);
3744         vcpu->arch.tid = mfspr(SPRN_TIDR);
3745         vcpu->arch.amr = mfspr(SPRN_AMR);
3746         vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3747         vcpu->arch.dscr = mfspr(SPRN_DSCR);
3748
3749         mtspr(SPRN_PSPB, 0);
3750         mtspr(SPRN_WORT, 0);
3751         mtspr(SPRN_UAMOR, 0);
3752         mtspr(SPRN_DSCR, host_dscr);
3753         mtspr(SPRN_TIDR, host_tidr);
3754         mtspr(SPRN_IAMR, host_iamr);
3755         mtspr(SPRN_PSPB, 0);
3756
3757         if (host_amr != vcpu->arch.amr)
3758                 mtspr(SPRN_AMR, host_amr);
3759
3760         if (host_fscr != vcpu->arch.fscr)
3761                 mtspr(SPRN_FSCR, host_fscr);
3762
3763         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3764         store_fp_state(&vcpu->arch.fp);
3765 #ifdef CONFIG_ALTIVEC
3766         store_vr_state(&vcpu->arch.vr);
3767 #endif
3768         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3769
3770         if (cpu_has_feature(CPU_FTR_TM) ||
3771             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3772                 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3773
3774         save_pmu = 1;
3775         if (vcpu->arch.vpa.pinned_addr) {
3776                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3777                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3778                 lp->yield_count = cpu_to_be32(yield_count);
3779                 vcpu->arch.vpa.dirty = 1;
3780                 save_pmu = lp->pmcregs_in_use;
3781         }
3782         /* Must save pmu if this guest is capable of running nested guests */
3783         save_pmu |= nesting_enabled(vcpu->kvm);
3784
3785         kvmhv_save_guest_pmu(vcpu, save_pmu);
3786 #ifdef CONFIG_PPC_PSERIES
3787         if (kvmhv_on_pseries()) {
3788                 barrier();
3789                 get_lppaca()->pmcregs_in_use = ppc_get_pmu_inuse();
3790                 barrier();
3791         }
3792 #endif
3793
3794         vc->entry_exit_map = 0x101;
3795         vc->in_guest = 0;
3796
3797         mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3798         mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3799
3800         kvmhv_load_host_pmu();
3801
3802         kvmppc_subcore_exit_guest();
3803
3804         return trap;
3805 }
3806
3807 /*
3808  * Wait for some other vcpu thread to execute us, and
3809  * wake us up when we need to handle something in the host.
3810  */
3811 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3812                                  struct kvm_vcpu *vcpu, int wait_state)
3813 {
3814         DEFINE_WAIT(wait);
3815
3816         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3817         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3818                 spin_unlock(&vc->lock);
3819                 schedule();
3820                 spin_lock(&vc->lock);
3821         }
3822         finish_wait(&vcpu->arch.cpu_run, &wait);
3823 }
3824
3825 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3826 {
3827         if (!halt_poll_ns_grow)
3828                 return;
3829
3830         vc->halt_poll_ns *= halt_poll_ns_grow;
3831         if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3832                 vc->halt_poll_ns = halt_poll_ns_grow_start;
3833 }
3834
3835 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3836 {
3837         if (halt_poll_ns_shrink == 0)
3838                 vc->halt_poll_ns = 0;
3839         else
3840                 vc->halt_poll_ns /= halt_poll_ns_shrink;
3841 }
3842
3843 #ifdef CONFIG_KVM_XICS
3844 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3845 {
3846         if (!xics_on_xive())
3847                 return false;
3848         return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3849                 vcpu->arch.xive_saved_state.cppr;
3850 }
3851 #else
3852 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3853 {
3854         return false;
3855 }
3856 #endif /* CONFIG_KVM_XICS */
3857
3858 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3859 {
3860         if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3861             kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3862                 return true;
3863
3864         return false;
3865 }
3866
3867 /*
3868  * Check to see if any of the runnable vcpus on the vcore have pending
3869  * exceptions or are no longer ceded
3870  */
3871 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3872 {
3873         struct kvm_vcpu *vcpu;
3874         int i;
3875
3876         for_each_runnable_thread(i, vcpu, vc) {
3877                 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3878                         return 1;
3879         }
3880
3881         return 0;
3882 }
3883
3884 /*
3885  * All the vcpus in this vcore are idle, so wait for a decrementer
3886  * or external interrupt to one of the vcpus.  vc->lock is held.
3887  */
3888 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3889 {
3890         ktime_t cur, start_poll, start_wait;
3891         int do_sleep = 1;
3892         u64 block_ns;
3893
3894         /* Poll for pending exceptions and ceded state */
3895         cur = start_poll = ktime_get();
3896         if (vc->halt_poll_ns) {
3897                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3898                 ++vc->runner->stat.halt_attempted_poll;
3899
3900                 vc->vcore_state = VCORE_POLLING;
3901                 spin_unlock(&vc->lock);
3902
3903                 do {
3904                         if (kvmppc_vcore_check_block(vc)) {
3905                                 do_sleep = 0;
3906                                 break;
3907                         }
3908                         cur = ktime_get();
3909                 } while (single_task_running() && ktime_before(cur, stop));
3910
3911                 spin_lock(&vc->lock);
3912                 vc->vcore_state = VCORE_INACTIVE;
3913
3914                 if (!do_sleep) {
3915                         ++vc->runner->stat.halt_successful_poll;
3916                         goto out;
3917                 }
3918         }
3919
3920         prepare_to_rcuwait(&vc->wait);
3921         set_current_state(TASK_INTERRUPTIBLE);
3922         if (kvmppc_vcore_check_block(vc)) {
3923                 finish_rcuwait(&vc->wait);
3924                 do_sleep = 0;
3925                 /* If we polled, count this as a successful poll */
3926                 if (vc->halt_poll_ns)
3927                         ++vc->runner->stat.halt_successful_poll;
3928                 goto out;
3929         }
3930
3931         start_wait = ktime_get();
3932
3933         vc->vcore_state = VCORE_SLEEPING;
3934         trace_kvmppc_vcore_blocked(vc, 0);
3935         spin_unlock(&vc->lock);
3936         schedule();
3937         finish_rcuwait(&vc->wait);
3938         spin_lock(&vc->lock);
3939         vc->vcore_state = VCORE_INACTIVE;
3940         trace_kvmppc_vcore_blocked(vc, 1);
3941         ++vc->runner->stat.halt_successful_wait;
3942
3943         cur = ktime_get();
3944
3945 out:
3946         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3947
3948         /* Attribute wait time */
3949         if (do_sleep) {
3950                 vc->runner->stat.halt_wait_ns +=
3951                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
3952                 /* Attribute failed poll time */
3953                 if (vc->halt_poll_ns)
3954                         vc->runner->stat.halt_poll_fail_ns +=
3955                                 ktime_to_ns(start_wait) -
3956                                 ktime_to_ns(start_poll);
3957         } else {
3958                 /* Attribute successful poll time */
3959                 if (vc->halt_poll_ns)
3960                         vc->runner->stat.halt_poll_success_ns +=
3961                                 ktime_to_ns(cur) -
3962                                 ktime_to_ns(start_poll);
3963         }
3964
3965         /* Adjust poll time */
3966         if (halt_poll_ns) {
3967                 if (block_ns <= vc->halt_poll_ns)
3968                         ;
3969                 /* We slept and blocked for longer than the max halt time */
3970                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3971                         shrink_halt_poll_ns(vc);
3972                 /* We slept and our poll time is too small */
3973                 else if (vc->halt_poll_ns < halt_poll_ns &&
3974                                 block_ns < halt_poll_ns)
3975                         grow_halt_poll_ns(vc);
3976                 if (vc->halt_poll_ns > halt_poll_ns)
3977                         vc->halt_poll_ns = halt_poll_ns;
3978         } else
3979                 vc->halt_poll_ns = 0;
3980
3981         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3982 }
3983
3984 /*
3985  * This never fails for a radix guest, as none of the operations it does
3986  * for a radix guest can fail or have a way to report failure.
3987  * kvmhv_run_single_vcpu() relies on this fact.
3988  */
3989 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3990 {
3991         int r = 0;
3992         struct kvm *kvm = vcpu->kvm;
3993
3994         mutex_lock(&kvm->arch.mmu_setup_lock);
3995         if (!kvm->arch.mmu_ready) {
3996                 if (!kvm_is_radix(kvm))
3997                         r = kvmppc_hv_setup_htab_rma(vcpu);
3998                 if (!r) {
3999                         if (cpu_has_feature(CPU_FTR_ARCH_300))
4000                                 kvmppc_setup_partition_table(kvm);
4001                         kvm->arch.mmu_ready = 1;
4002                 }
4003         }
4004         mutex_unlock(&kvm->arch.mmu_setup_lock);
4005         return r;
4006 }
4007
4008 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
4009 {
4010         struct kvm_run *run = vcpu->run;
4011         int n_ceded, i, r;
4012         struct kvmppc_vcore *vc;
4013         struct kvm_vcpu *v;
4014
4015         trace_kvmppc_run_vcpu_enter(vcpu);
4016
4017         run->exit_reason = 0;
4018         vcpu->arch.ret = RESUME_GUEST;
4019         vcpu->arch.trap = 0;
4020         kvmppc_update_vpas(vcpu);
4021
4022         /*
4023          * Synchronize with other threads in this virtual core
4024          */
4025         vc = vcpu->arch.vcore;
4026         spin_lock(&vc->lock);
4027         vcpu->arch.ceded = 0;
4028         vcpu->arch.run_task = current;
4029         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4030         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4031         vcpu->arch.busy_preempt = TB_NIL;
4032         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4033         ++vc->n_runnable;
4034
4035         /*
4036          * This happens the first time this is called for a vcpu.
4037          * If the vcore is already running, we may be able to start
4038          * this thread straight away and have it join in.
4039          */
4040         if (!signal_pending(current)) {
4041                 if ((vc->vcore_state == VCORE_PIGGYBACK ||
4042                      vc->vcore_state == VCORE_RUNNING) &&
4043                            !VCORE_IS_EXITING(vc)) {
4044                         kvmppc_create_dtl_entry(vcpu, vc);
4045                         kvmppc_start_thread(vcpu, vc);
4046                         trace_kvm_guest_enter(vcpu);
4047                 } else if (vc->vcore_state == VCORE_SLEEPING) {
4048                         rcuwait_wake_up(&vc->wait);
4049                 }
4050
4051         }
4052
4053         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4054                !signal_pending(current)) {
4055                 /* See if the MMU is ready to go */
4056                 if (!vcpu->kvm->arch.mmu_ready) {
4057                         spin_unlock(&vc->lock);
4058                         r = kvmhv_setup_mmu(vcpu);
4059                         spin_lock(&vc->lock);
4060                         if (r) {
4061                                 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4062                                 run->fail_entry.
4063                                         hardware_entry_failure_reason = 0;
4064                                 vcpu->arch.ret = r;
4065                                 break;
4066                         }
4067                 }
4068
4069                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4070                         kvmppc_vcore_end_preempt(vc);
4071
4072                 if (vc->vcore_state != VCORE_INACTIVE) {
4073                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4074                         continue;
4075                 }
4076                 for_each_runnable_thread(i, v, vc) {
4077                         kvmppc_core_prepare_to_enter(v);
4078                         if (signal_pending(v->arch.run_task)) {
4079                                 kvmppc_remove_runnable(vc, v);
4080                                 v->stat.signal_exits++;
4081                                 v->run->exit_reason = KVM_EXIT_INTR;
4082                                 v->arch.ret = -EINTR;
4083                                 wake_up(&v->arch.cpu_run);
4084                         }
4085                 }
4086                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4087                         break;
4088                 n_ceded = 0;
4089                 for_each_runnable_thread(i, v, vc) {
4090                         if (!kvmppc_vcpu_woken(v))
4091                                 n_ceded += v->arch.ceded;
4092                         else
4093                                 v->arch.ceded = 0;
4094                 }
4095                 vc->runner = vcpu;
4096                 if (n_ceded == vc->n_runnable) {
4097                         kvmppc_vcore_blocked(vc);
4098                 } else if (need_resched()) {
4099                         kvmppc_vcore_preempt(vc);
4100                         /* Let something else run */
4101                         cond_resched_lock(&vc->lock);
4102                         if (vc->vcore_state == VCORE_PREEMPT)
4103                                 kvmppc_vcore_end_preempt(vc);
4104                 } else {
4105                         kvmppc_run_core(vc);
4106                 }
4107                 vc->runner = NULL;
4108         }
4109
4110         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4111                (vc->vcore_state == VCORE_RUNNING ||
4112                 vc->vcore_state == VCORE_EXITING ||
4113                 vc->vcore_state == VCORE_PIGGYBACK))
4114                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4115
4116         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4117                 kvmppc_vcore_end_preempt(vc);
4118
4119         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4120                 kvmppc_remove_runnable(vc, vcpu);
4121                 vcpu->stat.signal_exits++;
4122                 run->exit_reason = KVM_EXIT_INTR;
4123                 vcpu->arch.ret = -EINTR;
4124         }
4125
4126         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4127                 /* Wake up some vcpu to run the core */
4128                 i = -1;
4129                 v = next_runnable_thread(vc, &i);
4130                 wake_up(&v->arch.cpu_run);
4131         }
4132
4133         trace_kvmppc_run_vcpu_exit(vcpu);
4134         spin_unlock(&vc->lock);
4135         return vcpu->arch.ret;
4136 }
4137
4138 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4139                           unsigned long lpcr)
4140 {
4141         struct kvm_run *run = vcpu->run;
4142         int trap, r, pcpu;
4143         int srcu_idx, lpid;
4144         struct kvmppc_vcore *vc;
4145         struct kvm *kvm = vcpu->kvm;
4146         struct kvm_nested_guest *nested = vcpu->arch.nested;
4147
4148         trace_kvmppc_run_vcpu_enter(vcpu);
4149
4150         run->exit_reason = 0;
4151         vcpu->arch.ret = RESUME_GUEST;
4152         vcpu->arch.trap = 0;
4153
4154         vc = vcpu->arch.vcore;
4155         vcpu->arch.ceded = 0;
4156         vcpu->arch.run_task = current;
4157         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4158         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4159         vcpu->arch.busy_preempt = TB_NIL;
4160         vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4161         vc->runnable_threads[0] = vcpu;
4162         vc->n_runnable = 1;
4163         vc->runner = vcpu;
4164
4165         /* See if the MMU is ready to go */
4166         if (!kvm->arch.mmu_ready)
4167                 kvmhv_setup_mmu(vcpu);
4168
4169         if (need_resched())
4170                 cond_resched();
4171
4172         kvmppc_update_vpas(vcpu);
4173
4174         init_vcore_to_run(vc);
4175         vc->preempt_tb = TB_NIL;
4176
4177         preempt_disable();
4178         pcpu = smp_processor_id();
4179         vc->pcpu = pcpu;
4180         kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4181
4182         local_irq_disable();
4183         hard_irq_disable();
4184         if (signal_pending(current))
4185                 goto sigpend;
4186         if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4187                 goto out;
4188
4189         if (!nested) {
4190                 kvmppc_core_prepare_to_enter(vcpu);
4191                 if (vcpu->arch.doorbell_request) {
4192                         vc->dpdes = 1;
4193                         smp_wmb();
4194                         vcpu->arch.doorbell_request = 0;
4195                 }
4196                 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4197                              &vcpu->arch.pending_exceptions))
4198                         lpcr |= LPCR_MER;
4199         } else if (vcpu->arch.pending_exceptions ||
4200                    vcpu->arch.doorbell_request ||
4201                    xive_interrupt_pending(vcpu)) {
4202                 vcpu->arch.ret = RESUME_HOST;
4203                 goto out;
4204         }
4205
4206         kvmppc_clear_host_core(pcpu);
4207
4208         local_paca->kvm_hstate.tid = 0;
4209         local_paca->kvm_hstate.napping = 0;
4210         local_paca->kvm_hstate.kvm_split_mode = NULL;
4211         kvmppc_start_thread(vcpu, vc);
4212         kvmppc_create_dtl_entry(vcpu, vc);
4213         trace_kvm_guest_enter(vcpu);
4214
4215         vc->vcore_state = VCORE_RUNNING;
4216         trace_kvmppc_run_core(vc, 0);
4217
4218         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4219                 lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4220                 mtspr(SPRN_LPID, lpid);
4221                 isync();
4222                 kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4223         }
4224
4225         guest_enter_irqoff();
4226
4227         srcu_idx = srcu_read_lock(&kvm->srcu);
4228
4229         this_cpu_disable_ftrace();
4230
4231         /* Tell lockdep that we're about to enable interrupts */
4232         trace_hardirqs_on();
4233
4234         trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4235         vcpu->arch.trap = trap;
4236
4237         trace_hardirqs_off();
4238
4239         this_cpu_enable_ftrace();
4240
4241         srcu_read_unlock(&kvm->srcu, srcu_idx);
4242
4243         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4244                 mtspr(SPRN_LPID, kvm->arch.host_lpid);
4245                 isync();
4246         }
4247
4248         set_irq_happened(trap);
4249
4250         kvmppc_set_host_core(pcpu);
4251
4252         context_tracking_guest_exit();
4253         if (!vtime_accounting_enabled_this_cpu()) {
4254                 local_irq_enable();
4255                 /*
4256                  * Service IRQs here before vtime_account_guest_exit() so any
4257                  * ticks that occurred while running the guest are accounted to
4258                  * the guest. If vtime accounting is enabled, accounting uses
4259                  * TB rather than ticks, so it can be done without enabling
4260                  * interrupts here, which has the problem that it accounts
4261                  * interrupt processing overhead to the host.
4262                  */
4263                 local_irq_disable();
4264         }
4265         vtime_account_guest_exit();
4266
4267         local_irq_enable();
4268
4269         cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4270
4271         preempt_enable();
4272
4273         /*
4274          * cancel pending decrementer exception if DEC is now positive, or if
4275          * entering a nested guest in which case the decrementer is now owned
4276          * by L2 and the L1 decrementer is provided in hdec_expires
4277          */
4278         if (kvmppc_core_pending_dec(vcpu) &&
4279                         ((get_tb() < vcpu->arch.dec_expires) ||
4280                          (trap == BOOK3S_INTERRUPT_SYSCALL &&
4281                           kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4282                 kvmppc_core_dequeue_dec(vcpu);
4283
4284         trace_kvm_guest_exit(vcpu);
4285         r = RESUME_GUEST;
4286         if (trap) {
4287                 if (!nested)
4288                         r = kvmppc_handle_exit_hv(vcpu, current);
4289                 else
4290                         r = kvmppc_handle_nested_exit(vcpu);
4291         }
4292         vcpu->arch.ret = r;
4293
4294         if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4295             !kvmppc_vcpu_woken(vcpu)) {
4296                 kvmppc_set_timer(vcpu);
4297                 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4298                         if (signal_pending(current)) {
4299                                 vcpu->stat.signal_exits++;
4300                                 run->exit_reason = KVM_EXIT_INTR;
4301                                 vcpu->arch.ret = -EINTR;
4302                                 break;
4303                         }
4304                         spin_lock(&vc->lock);
4305                         kvmppc_vcore_blocked(vc);
4306                         spin_unlock(&vc->lock);
4307                 }
4308         }
4309         vcpu->arch.ceded = 0;
4310
4311         vc->vcore_state = VCORE_INACTIVE;
4312         trace_kvmppc_run_core(vc, 1);
4313
4314  done:
4315         kvmppc_remove_runnable(vc, vcpu);
4316         trace_kvmppc_run_vcpu_exit(vcpu);
4317
4318         return vcpu->arch.ret;
4319
4320  sigpend:
4321         vcpu->stat.signal_exits++;
4322         run->exit_reason = KVM_EXIT_INTR;
4323         vcpu->arch.ret = -EINTR;
4324  out:
4325         local_irq_enable();
4326         preempt_enable();
4327         goto done;
4328 }
4329
4330 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
4331 {
4332         struct kvm_run *run = vcpu->run;
4333         int r;
4334         int srcu_idx;
4335         unsigned long ebb_regs[3] = {}; /* shut up GCC */
4336         unsigned long user_tar = 0;
4337         unsigned int user_vrsave;
4338         struct kvm *kvm;
4339
4340         if (!vcpu->arch.sane) {
4341                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4342                 return -EINVAL;
4343         }
4344
4345         /*
4346          * Don't allow entry with a suspended transaction, because
4347          * the guest entry/exit code will lose it.
4348          * If the guest has TM enabled, save away their TM-related SPRs
4349          * (they will get restored by the TM unavailable interrupt).
4350          */
4351 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4352         if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4353             (current->thread.regs->msr & MSR_TM)) {
4354                 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4355                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4356                         run->fail_entry.hardware_entry_failure_reason = 0;
4357                         return -EINVAL;
4358                 }
4359                 /* Enable TM so we can read the TM SPRs */
4360                 mtmsr(mfmsr() | MSR_TM);
4361                 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4362                 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4363                 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4364                 current->thread.regs->msr &= ~MSR_TM;
4365         }
4366 #endif
4367
4368         /*
4369          * Force online to 1 for the sake of old userspace which doesn't
4370          * set it.
4371          */
4372         if (!vcpu->arch.online) {
4373                 atomic_inc(&vcpu->arch.vcore->online_count);
4374                 vcpu->arch.online = 1;
4375         }
4376
4377         kvmppc_core_prepare_to_enter(vcpu);
4378
4379         /* No need to go into the guest when all we'll do is come back out */
4380         if (signal_pending(current)) {
4381                 run->exit_reason = KVM_EXIT_INTR;
4382                 return -EINTR;
4383         }
4384
4385         kvm = vcpu->kvm;
4386         atomic_inc(&kvm->arch.vcpus_running);
4387         /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4388         smp_mb();
4389
4390         flush_all_to_thread(current);
4391
4392         /* Save userspace EBB and other register values */
4393         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4394                 ebb_regs[0] = mfspr(SPRN_EBBHR);
4395                 ebb_regs[1] = mfspr(SPRN_EBBRR);
4396                 ebb_regs[2] = mfspr(SPRN_BESCR);
4397                 user_tar = mfspr(SPRN_TAR);
4398         }
4399         user_vrsave = mfspr(SPRN_VRSAVE);
4400
4401         vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4402         vcpu->arch.pgdir = kvm->mm->pgd;
4403         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4404
4405         do {
4406                 /*
4407                  * The early POWER9 chips that can't mix radix and HPT threads
4408                  * on the same core also need the workaround for the problem
4409                  * where the TLB would prefetch entries in the guest exit path
4410                  * for radix guests using the guest PIDR value and LPID 0.
4411                  * The workaround is in the old path (kvmppc_run_vcpu())
4412                  * but not the new path (kvmhv_run_single_vcpu()).
4413                  */
4414                 if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4415                     !no_mixing_hpt_and_radix)
4416                         r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
4417                                                   vcpu->arch.vcore->lpcr);
4418                 else
4419                         r = kvmppc_run_vcpu(vcpu);
4420
4421                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4422                     !(vcpu->arch.shregs.msr & MSR_PR)) {
4423                         trace_kvm_hcall_enter(vcpu);
4424                         r = kvmppc_pseries_do_hcall(vcpu);
4425                         trace_kvm_hcall_exit(vcpu, r);
4426                         kvmppc_core_prepare_to_enter(vcpu);
4427                 } else if (r == RESUME_PAGE_FAULT) {
4428                         srcu_idx = srcu_read_lock(&kvm->srcu);
4429                         r = kvmppc_book3s_hv_page_fault(vcpu,
4430                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4431                         srcu_read_unlock(&kvm->srcu, srcu_idx);
4432                 } else if (r == RESUME_PASSTHROUGH) {
4433                         if (WARN_ON(xics_on_xive()))
4434                                 r = H_SUCCESS;
4435                         else
4436                                 r = kvmppc_xics_rm_complete(vcpu, 0);
4437                 }
4438         } while (is_kvmppc_resume_guest(r));
4439
4440         /* Restore userspace EBB and other register values */
4441         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4442                 mtspr(SPRN_EBBHR, ebb_regs[0]);
4443                 mtspr(SPRN_EBBRR, ebb_regs[1]);
4444                 mtspr(SPRN_BESCR, ebb_regs[2]);
4445                 mtspr(SPRN_TAR, user_tar);
4446                 mtspr(SPRN_FSCR, current->thread.fscr);
4447         }
4448         mtspr(SPRN_VRSAVE, user_vrsave);
4449
4450         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4451         atomic_dec(&kvm->arch.vcpus_running);
4452         return r;
4453 }
4454
4455 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4456                                      int shift, int sllp)
4457 {
4458         (*sps)->page_shift = shift;
4459         (*sps)->slb_enc = sllp;
4460         (*sps)->enc[0].page_shift = shift;
4461         (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4462         /*
4463          * Add 16MB MPSS support (may get filtered out by userspace)
4464          */
4465         if (shift != 24) {
4466                 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4467                 if (penc != -1) {
4468                         (*sps)->enc[1].page_shift = 24;
4469                         (*sps)->enc[1].pte_enc = penc;
4470                 }
4471         }
4472         (*sps)++;
4473 }
4474
4475 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4476                                          struct kvm_ppc_smmu_info *info)
4477 {
4478         struct kvm_ppc_one_seg_page_size *sps;
4479
4480         /*
4481          * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4482          * POWER7 doesn't support keys for instruction accesses,
4483          * POWER8 and POWER9 do.
4484          */
4485         info->data_keys = 32;
4486         info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4487
4488         /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4489         info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4490         info->slb_size = 32;
4491
4492         /* We only support these sizes for now, and no muti-size segments */
4493         sps = &info->sps[0];
4494         kvmppc_add_seg_page_size(&sps, 12, 0);
4495         kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4496         kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4497
4498         /* If running as a nested hypervisor, we don't support HPT guests */
4499         if (kvmhv_on_pseries())
4500                 info->flags |= KVM_PPC_NO_HASH;
4501
4502         return 0;
4503 }
4504
4505 /*
4506  * Get (and clear) the dirty memory log for a memory slot.
4507  */
4508 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4509                                          struct kvm_dirty_log *log)
4510 {
4511         struct kvm_memslots *slots;
4512         struct kvm_memory_slot *memslot;
4513         int i, r;
4514         unsigned long n;
4515         unsigned long *buf, *p;
4516         struct kvm_vcpu *vcpu;
4517
4518         mutex_lock(&kvm->slots_lock);
4519
4520         r = -EINVAL;
4521         if (log->slot >= KVM_USER_MEM_SLOTS)
4522                 goto out;
4523
4524         slots = kvm_memslots(kvm);
4525         memslot = id_to_memslot(slots, log->slot);
4526         r = -ENOENT;
4527         if (!memslot || !memslot->dirty_bitmap)
4528                 goto out;
4529
4530         /*
4531          * Use second half of bitmap area because both HPT and radix
4532          * accumulate bits in the first half.
4533          */
4534         n = kvm_dirty_bitmap_bytes(memslot);
4535         buf = memslot->dirty_bitmap + n / sizeof(long);
4536         memset(buf, 0, n);
4537
4538         if (kvm_is_radix(kvm))
4539                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4540         else
4541                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4542         if (r)
4543                 goto out;
4544
4545         /*
4546          * We accumulate dirty bits in the first half of the
4547          * memslot's dirty_bitmap area, for when pages are paged
4548          * out or modified by the host directly.  Pick up these
4549          * bits and add them to the map.
4550          */
4551         p = memslot->dirty_bitmap;
4552         for (i = 0; i < n / sizeof(long); ++i)
4553                 buf[i] |= xchg(&p[i], 0);
4554
4555         /* Harvest dirty bits from VPA and DTL updates */
4556         /* Note: we never modify the SLB shadow buffer areas */
4557         kvm_for_each_vcpu(i, vcpu, kvm) {
4558                 spin_lock(&vcpu->arch.vpa_update_lock);
4559                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4560                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4561                 spin_unlock(&vcpu->arch.vpa_update_lock);
4562         }
4563
4564         r = -EFAULT;
4565         if (copy_to_user(log->dirty_bitmap, buf, n))
4566                 goto out;
4567
4568         r = 0;
4569 out:
4570         mutex_unlock(&kvm->slots_lock);
4571         return r;
4572 }
4573
4574 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
4575 {
4576         vfree(slot->arch.rmap);
4577         slot->arch.rmap = NULL;
4578 }
4579
4580 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4581                                         struct kvm_memory_slot *slot,
4582                                         const struct kvm_userspace_memory_region *mem,
4583                                         enum kvm_mr_change change)
4584 {
4585         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4586
4587         if (change == KVM_MR_CREATE) {
4588                 unsigned long size = array_size(npages, sizeof(*slot->arch.rmap));
4589
4590                 if ((size >> PAGE_SHIFT) > totalram_pages())
4591                         return -ENOMEM;
4592
4593                 slot->arch.rmap = vzalloc(size);
4594                 if (!slot->arch.rmap)
4595                         return -ENOMEM;
4596         }
4597
4598         return 0;
4599 }
4600
4601 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4602                                 const struct kvm_userspace_memory_region *mem,
4603                                 const struct kvm_memory_slot *old,
4604                                 const struct kvm_memory_slot *new,
4605                                 enum kvm_mr_change change)
4606 {
4607         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4608
4609         /*
4610          * If we are making a new memslot, it might make
4611          * some address that was previously cached as emulated
4612          * MMIO be no longer emulated MMIO, so invalidate
4613          * all the caches of emulated MMIO translations.
4614          */
4615         if (npages)
4616                 atomic64_inc(&kvm->arch.mmio_update);
4617
4618         /*
4619          * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4620          * have already called kvm_arch_flush_shadow_memslot() to
4621          * flush shadow mappings.  For KVM_MR_CREATE we have no
4622          * previous mappings.  So the only case to handle is
4623          * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4624          * has been changed.
4625          * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4626          * to get rid of any THP PTEs in the partition-scoped page tables
4627          * so we can track dirtiness at the page level; we flush when
4628          * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4629          * using THP PTEs.
4630          */
4631         if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4632             ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4633                 kvmppc_radix_flush_memslot(kvm, old);
4634         /*
4635          * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4636          */
4637         if (!kvm->arch.secure_guest)
4638                 return;
4639
4640         switch (change) {
4641         case KVM_MR_CREATE:
4642                 /*
4643                  * @TODO kvmppc_uvmem_memslot_create() can fail and
4644                  * return error. Fix this.
4645                  */
4646                 kvmppc_uvmem_memslot_create(kvm, new);
4647                 break;
4648         case KVM_MR_DELETE:
4649                 kvmppc_uvmem_memslot_delete(kvm, old);
4650                 break;
4651         default:
4652                 /* TODO: Handle KVM_MR_MOVE */
4653                 break;
4654         }
4655 }
4656
4657 /*
4658  * Update LPCR values in kvm->arch and in vcores.
4659  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4660  * of kvm->arch.lpcr update).
4661  */
4662 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4663 {
4664         long int i;
4665         u32 cores_done = 0;
4666
4667         if ((kvm->arch.lpcr & mask) == lpcr)
4668                 return;
4669
4670         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4671
4672         for (i = 0; i < KVM_MAX_VCORES; ++i) {
4673                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4674                 if (!vc)
4675                         continue;
4676                 spin_lock(&vc->lock);
4677                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4678                 spin_unlock(&vc->lock);
4679                 if (++cores_done >= kvm->arch.online_vcores)
4680                         break;
4681         }
4682 }
4683
4684 void kvmppc_setup_partition_table(struct kvm *kvm)
4685 {
4686         unsigned long dw0, dw1;
4687
4688         if (!kvm_is_radix(kvm)) {
4689                 /* PS field - page size for VRMA */
4690                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4691                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4692                 /* HTABSIZE and HTABORG fields */
4693                 dw0 |= kvm->arch.sdr1;
4694
4695                 /* Second dword as set by userspace */
4696                 dw1 = kvm->arch.process_table;
4697         } else {
4698                 dw0 = PATB_HR | radix__get_tree_size() |
4699                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4700                 dw1 = PATB_GR | kvm->arch.process_table;
4701         }
4702         kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4703 }
4704
4705 /*
4706  * Set up HPT (hashed page table) and RMA (real-mode area).
4707  * Must be called with kvm->arch.mmu_setup_lock held.
4708  */
4709 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4710 {
4711         int err = 0;
4712         struct kvm *kvm = vcpu->kvm;
4713         unsigned long hva;
4714         struct kvm_memory_slot *memslot;
4715         struct vm_area_struct *vma;
4716         unsigned long lpcr = 0, senc;
4717         unsigned long psize, porder;
4718         int srcu_idx;
4719
4720         /* Allocate hashed page table (if not done already) and reset it */
4721         if (!kvm->arch.hpt.virt) {
4722                 int order = KVM_DEFAULT_HPT_ORDER;
4723                 struct kvm_hpt_info info;
4724
4725                 err = kvmppc_allocate_hpt(&info, order);
4726                 /* If we get here, it means userspace didn't specify a
4727                  * size explicitly.  So, try successively smaller
4728                  * sizes if the default failed. */
4729                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4730                         err  = kvmppc_allocate_hpt(&info, order);
4731
4732                 if (err < 0) {
4733                         pr_err("KVM: Couldn't alloc HPT\n");
4734                         goto out;
4735                 }
4736
4737                 kvmppc_set_hpt(kvm, &info);
4738         }
4739
4740         /* Look up the memslot for guest physical address 0 */
4741         srcu_idx = srcu_read_lock(&kvm->srcu);
4742         memslot = gfn_to_memslot(kvm, 0);
4743
4744         /* We must have some memory at 0 by now */
4745         err = -EINVAL;
4746         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4747                 goto out_srcu;
4748
4749         /* Look up the VMA for the start of this memory slot */
4750         hva = memslot->userspace_addr;
4751         mmap_read_lock(kvm->mm);
4752         vma = find_vma(kvm->mm, hva);
4753         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4754                 goto up_out;
4755
4756         psize = vma_kernel_pagesize(vma);
4757
4758         mmap_read_unlock(kvm->mm);
4759
4760         /* We can handle 4k, 64k or 16M pages in the VRMA */
4761         if (psize >= 0x1000000)
4762                 psize = 0x1000000;
4763         else if (psize >= 0x10000)
4764                 psize = 0x10000;
4765         else
4766                 psize = 0x1000;
4767         porder = __ilog2(psize);
4768
4769         senc = slb_pgsize_encoding(psize);
4770         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4771                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4772         /* Create HPTEs in the hash page table for the VRMA */
4773         kvmppc_map_vrma(vcpu, memslot, porder);
4774
4775         /* Update VRMASD field in the LPCR */
4776         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4777                 /* the -4 is to account for senc values starting at 0x10 */
4778                 lpcr = senc << (LPCR_VRMASD_SH - 4);
4779                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4780         }
4781
4782         /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4783         smp_wmb();
4784         err = 0;
4785  out_srcu:
4786         srcu_read_unlock(&kvm->srcu, srcu_idx);
4787  out:
4788         return err;
4789
4790  up_out:
4791         mmap_read_unlock(kvm->mm);
4792         goto out_srcu;
4793 }
4794
4795 /*
4796  * Must be called with kvm->arch.mmu_setup_lock held and
4797  * mmu_ready = 0 and no vcpus running.
4798  */
4799 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4800 {
4801         if (nesting_enabled(kvm))
4802                 kvmhv_release_all_nested(kvm);
4803         kvmppc_rmap_reset(kvm);
4804         kvm->arch.process_table = 0;
4805         /* Mutual exclusion with kvm_unmap_hva_range etc. */
4806         spin_lock(&kvm->mmu_lock);
4807         kvm->arch.radix = 0;
4808         spin_unlock(&kvm->mmu_lock);
4809         kvmppc_free_radix(kvm);
4810         kvmppc_update_lpcr(kvm, LPCR_VPM1,
4811                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4812         return 0;
4813 }
4814
4815 /*
4816  * Must be called with kvm->arch.mmu_setup_lock held and
4817  * mmu_ready = 0 and no vcpus running.
4818  */
4819 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4820 {
4821         int err;
4822
4823         err = kvmppc_init_vm_radix(kvm);
4824         if (err)
4825                 return err;
4826         kvmppc_rmap_reset(kvm);
4827         /* Mutual exclusion with kvm_unmap_hva_range etc. */
4828         spin_lock(&kvm->mmu_lock);
4829         kvm->arch.radix = 1;
4830         spin_unlock(&kvm->mmu_lock);
4831         kvmppc_free_hpt(&kvm->arch.hpt);
4832         kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4833                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4834         return 0;
4835 }
4836
4837 #ifdef CONFIG_KVM_XICS
4838 /*
4839  * Allocate a per-core structure for managing state about which cores are
4840  * running in the host versus the guest and for exchanging data between
4841  * real mode KVM and CPU running in the host.
4842  * This is only done for the first VM.
4843  * The allocated structure stays even if all VMs have stopped.
4844  * It is only freed when the kvm-hv module is unloaded.
4845  * It's OK for this routine to fail, we just don't support host
4846  * core operations like redirecting H_IPI wakeups.
4847  */
4848 void kvmppc_alloc_host_rm_ops(void)
4849 {
4850         struct kvmppc_host_rm_ops *ops;
4851         unsigned long l_ops;
4852         int cpu, core;
4853         int size;
4854
4855         /* Not the first time here ? */
4856         if (kvmppc_host_rm_ops_hv != NULL)
4857                 return;
4858
4859         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4860         if (!ops)
4861                 return;
4862
4863         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4864         ops->rm_core = kzalloc(size, GFP_KERNEL);
4865
4866         if (!ops->rm_core) {
4867                 kfree(ops);
4868                 return;
4869         }
4870
4871         cpus_read_lock();
4872
4873         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4874                 if (!cpu_online(cpu))
4875                         continue;
4876
4877                 core = cpu >> threads_shift;
4878                 ops->rm_core[core].rm_state.in_host = 1;
4879         }
4880
4881         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4882
4883         /*
4884          * Make the contents of the kvmppc_host_rm_ops structure visible
4885          * to other CPUs before we assign it to the global variable.
4886          * Do an atomic assignment (no locks used here), but if someone
4887          * beats us to it, just free our copy and return.
4888          */
4889         smp_wmb();
4890         l_ops = (unsigned long) ops;
4891
4892         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4893                 cpus_read_unlock();
4894                 kfree(ops->rm_core);
4895                 kfree(ops);
4896                 return;
4897         }
4898
4899         cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4900                                              "ppc/kvm_book3s:prepare",
4901                                              kvmppc_set_host_core,
4902                                              kvmppc_clear_host_core);
4903         cpus_read_unlock();
4904 }
4905
4906 void kvmppc_free_host_rm_ops(void)
4907 {
4908         if (kvmppc_host_rm_ops_hv) {
4909                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4910                 kfree(kvmppc_host_rm_ops_hv->rm_core);
4911                 kfree(kvmppc_host_rm_ops_hv);
4912                 kvmppc_host_rm_ops_hv = NULL;
4913         }
4914 }
4915 #endif
4916
4917 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4918 {
4919         unsigned long lpcr, lpid;
4920         char buf[32];
4921         int ret;
4922
4923         mutex_init(&kvm->arch.uvmem_lock);
4924         INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
4925         mutex_init(&kvm->arch.mmu_setup_lock);
4926
4927         /* Allocate the guest's logical partition ID */
4928
4929         lpid = kvmppc_alloc_lpid();
4930         if ((long)lpid < 0)
4931                 return -ENOMEM;
4932         kvm->arch.lpid = lpid;
4933
4934         kvmppc_alloc_host_rm_ops();
4935
4936         kvmhv_vm_nested_init(kvm);
4937
4938         /*
4939          * Since we don't flush the TLB when tearing down a VM,
4940          * and this lpid might have previously been used,
4941          * make sure we flush on each core before running the new VM.
4942          * On POWER9, the tlbie in mmu_partition_table_set_entry()
4943          * does this flush for us.
4944          */
4945         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4946                 cpumask_setall(&kvm->arch.need_tlb_flush);
4947
4948         /* Start out with the default set of hcalls enabled */
4949         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4950                sizeof(kvm->arch.enabled_hcalls));
4951
4952         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4953                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4954
4955         /* Init LPCR for virtual RMA mode */
4956         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4957                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
4958                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4959                 lpcr &= LPCR_PECE | LPCR_LPES;
4960         } else {
4961                 lpcr = 0;
4962         }
4963         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4964                 LPCR_VPM0 | LPCR_VPM1;
4965         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4966                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4967         /* On POWER8 turn on online bit to enable PURR/SPURR */
4968         if (cpu_has_feature(CPU_FTR_ARCH_207S))
4969                 lpcr |= LPCR_ONL;
4970         /*
4971          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4972          * Set HVICE bit to enable hypervisor virtualization interrupts.
4973          * Set HEIC to prevent OS interrupts to go to hypervisor (should
4974          * be unnecessary but better safe than sorry in case we re-enable
4975          * EE in HV mode with this LPCR still set)
4976          */
4977         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4978                 lpcr &= ~LPCR_VPM0;
4979                 lpcr |= LPCR_HVICE | LPCR_HEIC;
4980
4981                 /*
4982                  * If xive is enabled, we route 0x500 interrupts directly
4983                  * to the guest.
4984                  */
4985                 if (xics_on_xive())
4986                         lpcr |= LPCR_LPES;
4987         }
4988
4989         /*
4990          * If the host uses radix, the guest starts out as radix.
4991          */
4992         if (radix_enabled()) {
4993                 kvm->arch.radix = 1;
4994                 kvm->arch.mmu_ready = 1;
4995                 lpcr &= ~LPCR_VPM1;
4996                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4997                 ret = kvmppc_init_vm_radix(kvm);
4998                 if (ret) {
4999                         kvmppc_free_lpid(kvm->arch.lpid);
5000                         return ret;
5001                 }
5002                 kvmppc_setup_partition_table(kvm);
5003         }
5004
5005         kvm->arch.lpcr = lpcr;
5006
5007         /* Initialization for future HPT resizes */
5008         kvm->arch.resize_hpt = NULL;
5009
5010         /*
5011          * Work out how many sets the TLB has, for the use of
5012          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
5013          */
5014         if (radix_enabled())
5015                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
5016         else if (cpu_has_feature(CPU_FTR_ARCH_300))
5017                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
5018         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
5019                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
5020         else
5021                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
5022
5023         /*
5024          * Track that we now have a HV mode VM active. This blocks secondary
5025          * CPU threads from coming online.
5026          * On POWER9, we only need to do this if the "indep_threads_mode"
5027          * module parameter has been set to N.
5028          */
5029         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5030                 if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
5031                         pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
5032                         kvm->arch.threads_indep = true;
5033                 } else {
5034                         kvm->arch.threads_indep = indep_threads_mode;
5035                 }
5036         }
5037         if (!kvm->arch.threads_indep)
5038                 kvm_hv_vm_activated();
5039
5040         /*
5041          * Initialize smt_mode depending on processor.
5042          * POWER8 and earlier have to use "strict" threading, where
5043          * all vCPUs in a vcore have to run on the same (sub)core,
5044          * whereas on POWER9 the threads can each run a different
5045          * guest.
5046          */
5047         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5048                 kvm->arch.smt_mode = threads_per_subcore;
5049         else
5050                 kvm->arch.smt_mode = 1;
5051         kvm->arch.emul_smt_mode = 1;
5052
5053         /*
5054          * Create a debugfs directory for the VM
5055          */
5056         snprintf(buf, sizeof(buf), "vm%d", current->pid);
5057         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
5058         kvmppc_mmu_debugfs_init(kvm);
5059         if (radix_enabled())
5060                 kvmhv_radix_debugfs_init(kvm);
5061
5062         return 0;
5063 }
5064
5065 static void kvmppc_free_vcores(struct kvm *kvm)
5066 {
5067         long int i;
5068
5069         for (i = 0; i < KVM_MAX_VCORES; ++i)
5070                 kfree(kvm->arch.vcores[i]);
5071         kvm->arch.online_vcores = 0;
5072 }
5073
5074 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5075 {
5076         debugfs_remove_recursive(kvm->arch.debugfs_dir);
5077
5078         if (!kvm->arch.threads_indep)
5079                 kvm_hv_vm_deactivated();
5080
5081         kvmppc_free_vcores(kvm);
5082
5083
5084         if (kvm_is_radix(kvm))
5085                 kvmppc_free_radix(kvm);
5086         else
5087                 kvmppc_free_hpt(&kvm->arch.hpt);
5088
5089         /* Perform global invalidation and return lpid to the pool */
5090         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5091                 if (nesting_enabled(kvm))
5092                         kvmhv_release_all_nested(kvm);
5093                 kvm->arch.process_table = 0;
5094                 if (kvm->arch.secure_guest)
5095                         uv_svm_terminate(kvm->arch.lpid);
5096                 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5097         }
5098
5099         kvmppc_free_lpid(kvm->arch.lpid);
5100
5101         kvmppc_free_pimap(kvm);
5102 }
5103
5104 /* We don't need to emulate any privileged instructions or dcbz */
5105 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5106                                      unsigned int inst, int *advance)
5107 {
5108         return EMULATE_FAIL;
5109 }
5110
5111 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5112                                         ulong spr_val)
5113 {
5114         return EMULATE_FAIL;
5115 }
5116
5117 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5118                                         ulong *spr_val)
5119 {
5120         return EMULATE_FAIL;
5121 }
5122
5123 static int kvmppc_core_check_processor_compat_hv(void)
5124 {
5125         if (cpu_has_feature(CPU_FTR_HVMODE) &&
5126             cpu_has_feature(CPU_FTR_ARCH_206))
5127                 return 0;
5128
5129         /* POWER9 in radix mode is capable of being a nested hypervisor. */
5130         if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5131                 return 0;
5132
5133         return -EIO;
5134 }
5135
5136 #ifdef CONFIG_KVM_XICS
5137
5138 void kvmppc_free_pimap(struct kvm *kvm)
5139 {
5140         kfree(kvm->arch.pimap);
5141 }
5142
5143 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5144 {
5145         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5146 }
5147
5148 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5149 {
5150         struct irq_desc *desc;
5151         struct kvmppc_irq_map *irq_map;
5152         struct kvmppc_passthru_irqmap *pimap;
5153         struct irq_chip *chip;
5154         int i, rc = 0;
5155
5156         if (!kvm_irq_bypass)
5157                 return 1;
5158
5159         desc = irq_to_desc(host_irq);
5160         if (!desc)
5161                 return -EIO;
5162
5163         mutex_lock(&kvm->lock);
5164
5165         pimap = kvm->arch.pimap;
5166         if (pimap == NULL) {
5167                 /* First call, allocate structure to hold IRQ map */
5168                 pimap = kvmppc_alloc_pimap();
5169                 if (pimap == NULL) {
5170                         mutex_unlock(&kvm->lock);
5171                         return -ENOMEM;
5172                 }
5173                 kvm->arch.pimap = pimap;
5174         }
5175
5176         /*
5177          * For now, we only support interrupts for which the EOI operation
5178          * is an OPAL call followed by a write to XIRR, since that's
5179          * what our real-mode EOI code does, or a XIVE interrupt
5180          */
5181         chip = irq_data_get_irq_chip(&desc->irq_data);
5182         if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5183                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5184                         host_irq, guest_gsi);
5185                 mutex_unlock(&kvm->lock);
5186                 return -ENOENT;
5187         }
5188
5189         /*
5190          * See if we already have an entry for this guest IRQ number.
5191          * If it's mapped to a hardware IRQ number, that's an error,
5192          * otherwise re-use this entry.
5193          */
5194         for (i = 0; i < pimap->n_mapped; i++) {
5195                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5196                         if (pimap->mapped[i].r_hwirq) {
5197                                 mutex_unlock(&kvm->lock);
5198                                 return -EINVAL;
5199                         }
5200                         break;
5201                 }
5202         }
5203
5204         if (i == KVMPPC_PIRQ_MAPPED) {
5205                 mutex_unlock(&kvm->lock);
5206                 return -EAGAIN;         /* table is full */
5207         }
5208
5209         irq_map = &pimap->mapped[i];
5210
5211         irq_map->v_hwirq = guest_gsi;
5212         irq_map->desc = desc;
5213
5214         /*
5215          * Order the above two stores before the next to serialize with
5216          * the KVM real mode handler.
5217          */
5218         smp_wmb();
5219         irq_map->r_hwirq = desc->irq_data.hwirq;
5220
5221         if (i == pimap->n_mapped)
5222                 pimap->n_mapped++;
5223
5224         if (xics_on_xive())
5225                 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5226         else
5227                 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5228         if (rc)
5229                 irq_map->r_hwirq = 0;
5230
5231         mutex_unlock(&kvm->lock);
5232
5233         return 0;
5234 }
5235
5236 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5237 {
5238         struct irq_desc *desc;
5239         struct kvmppc_passthru_irqmap *pimap;
5240         int i, rc = 0;
5241
5242         if (!kvm_irq_bypass)
5243                 return 0;
5244
5245         desc = irq_to_desc(host_irq);
5246         if (!desc)
5247                 return -EIO;
5248
5249         mutex_lock(&kvm->lock);
5250         if (!kvm->arch.pimap)
5251                 goto unlock;
5252
5253         pimap = kvm->arch.pimap;
5254
5255         for (i = 0; i < pimap->n_mapped; i++) {
5256                 if (guest_gsi == pimap->mapped[i].v_hwirq)
5257                         break;
5258         }
5259
5260         if (i == pimap->n_mapped) {
5261                 mutex_unlock(&kvm->lock);
5262                 return -ENODEV;
5263         }
5264
5265         if (xics_on_xive())
5266                 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5267         else
5268                 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5269
5270         /* invalidate the entry (what do do on error from the above ?) */
5271         pimap->mapped[i].r_hwirq = 0;
5272
5273         /*
5274          * We don't free this structure even when the count goes to
5275          * zero. The structure is freed when we destroy the VM.
5276          */
5277  unlock:
5278         mutex_unlock(&kvm->lock);
5279         return rc;
5280 }
5281
5282 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5283                                              struct irq_bypass_producer *prod)
5284 {
5285         int ret = 0;
5286         struct kvm_kernel_irqfd *irqfd =
5287                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5288
5289         irqfd->producer = prod;
5290
5291         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5292         if (ret)
5293                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5294                         prod->irq, irqfd->gsi, ret);
5295
5296         return ret;
5297 }
5298
5299 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5300                                               struct irq_bypass_producer *prod)
5301 {
5302         int ret;
5303         struct kvm_kernel_irqfd *irqfd =
5304                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5305
5306         irqfd->producer = NULL;
5307
5308         /*
5309          * When producer of consumer is unregistered, we change back to
5310          * default external interrupt handling mode - KVM real mode
5311          * will switch back to host.
5312          */
5313         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5314         if (ret)
5315                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5316                         prod->irq, irqfd->gsi, ret);
5317 }
5318 #endif
5319
5320 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5321                                  unsigned int ioctl, unsigned long arg)
5322 {
5323         struct kvm *kvm __maybe_unused = filp->private_data;
5324         void __user *argp = (void __user *)arg;
5325         long r;
5326
5327         switch (ioctl) {
5328
5329         case KVM_PPC_ALLOCATE_HTAB: {
5330                 u32 htab_order;
5331
5332                 /* If we're a nested hypervisor, we currently only support radix */
5333                 if (kvmhv_on_pseries()) {
5334                         r = -EOPNOTSUPP;
5335                         break;
5336                 }
5337
5338                 r = -EFAULT;
5339                 if (get_user(htab_order, (u32 __user *)argp))
5340                         break;
5341                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5342                 if (r)
5343                         break;
5344                 r = 0;
5345                 break;
5346         }
5347
5348         case KVM_PPC_GET_HTAB_FD: {
5349                 struct kvm_get_htab_fd ghf;
5350
5351                 r = -EFAULT;
5352                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5353                         break;
5354                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5355                 break;
5356         }
5357
5358         case KVM_PPC_RESIZE_HPT_PREPARE: {
5359                 struct kvm_ppc_resize_hpt rhpt;
5360
5361                 r = -EFAULT;
5362                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5363                         break;
5364
5365                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5366                 break;
5367         }
5368
5369         case KVM_PPC_RESIZE_HPT_COMMIT: {
5370                 struct kvm_ppc_resize_hpt rhpt;
5371
5372                 r = -EFAULT;
5373                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5374                         break;
5375
5376                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5377                 break;
5378         }
5379
5380         default:
5381                 r = -ENOTTY;
5382         }
5383
5384         return r;
5385 }
5386
5387 /*
5388  * List of hcall numbers to enable by default.
5389  * For compatibility with old userspace, we enable by default
5390  * all hcalls that were implemented before the hcall-enabling
5391  * facility was added.  Note this list should not include H_RTAS.
5392  */
5393 static unsigned int default_hcall_list[] = {
5394         H_REMOVE,
5395         H_ENTER,
5396         H_READ,
5397         H_PROTECT,
5398         H_BULK_REMOVE,
5399         H_GET_TCE,
5400         H_PUT_TCE,
5401         H_SET_DABR,
5402         H_SET_XDABR,
5403         H_CEDE,
5404         H_PROD,
5405         H_CONFER,
5406         H_REGISTER_VPA,
5407 #ifdef CONFIG_KVM_XICS
5408         H_EOI,
5409         H_CPPR,
5410         H_IPI,
5411         H_IPOLL,
5412         H_XIRR,
5413         H_XIRR_X,
5414 #endif
5415         0
5416 };
5417
5418 static void init_default_hcalls(void)
5419 {
5420         int i;
5421         unsigned int hcall;
5422
5423         for (i = 0; default_hcall_list[i]; ++i) {
5424                 hcall = default_hcall_list[i];
5425                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5426                 __set_bit(hcall / 4, default_enabled_hcalls);
5427         }
5428 }
5429
5430 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5431 {
5432         unsigned long lpcr;
5433         int radix;
5434         int err;
5435
5436         /* If not on a POWER9, reject it */
5437         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5438                 return -ENODEV;
5439
5440         /* If any unknown flags set, reject it */
5441         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5442                 return -EINVAL;
5443
5444         /* GR (guest radix) bit in process_table field must match */
5445         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5446         if (!!(cfg->process_table & PATB_GR) != radix)
5447                 return -EINVAL;
5448
5449         /* Process table size field must be reasonable, i.e. <= 24 */
5450         if ((cfg->process_table & PRTS_MASK) > 24)
5451                 return -EINVAL;
5452
5453         /* We can change a guest to/from radix now, if the host is radix */
5454         if (radix && !radix_enabled())
5455                 return -EINVAL;
5456
5457         /* If we're a nested hypervisor, we currently only support radix */
5458         if (kvmhv_on_pseries() && !radix)
5459                 return -EINVAL;
5460
5461         mutex_lock(&kvm->arch.mmu_setup_lock);
5462         if (radix != kvm_is_radix(kvm)) {
5463                 if (kvm->arch.mmu_ready) {
5464                         kvm->arch.mmu_ready = 0;
5465                         /* order mmu_ready vs. vcpus_running */
5466                         smp_mb();
5467                         if (atomic_read(&kvm->arch.vcpus_running)) {
5468                                 kvm->arch.mmu_ready = 1;
5469                                 err = -EBUSY;
5470                                 goto out_unlock;
5471                         }
5472                 }
5473                 if (radix)
5474                         err = kvmppc_switch_mmu_to_radix(kvm);
5475                 else
5476                         err = kvmppc_switch_mmu_to_hpt(kvm);
5477                 if (err)
5478                         goto out_unlock;
5479         }
5480
5481         kvm->arch.process_table = cfg->process_table;
5482         kvmppc_setup_partition_table(kvm);
5483
5484         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5485         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5486         err = 0;
5487
5488  out_unlock:
5489         mutex_unlock(&kvm->arch.mmu_setup_lock);
5490         return err;
5491 }
5492
5493 static int kvmhv_enable_nested(struct kvm *kvm)
5494 {
5495         if (!nested)
5496                 return -EPERM;
5497         if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5498                 return -ENODEV;
5499
5500         /* kvm == NULL means the caller is testing if the capability exists */
5501         if (kvm)
5502                 kvm->arch.nested_enable = true;
5503         return 0;
5504 }
5505
5506 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5507                                  int size)
5508 {
5509         int rc = -EINVAL;
5510
5511         if (kvmhv_vcpu_is_radix(vcpu)) {
5512                 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5513
5514                 if (rc > 0)
5515                         rc = -EINVAL;
5516         }
5517
5518         /* For now quadrants are the only way to access nested guest memory */
5519         if (rc && vcpu->arch.nested)
5520                 rc = -EAGAIN;
5521
5522         return rc;
5523 }
5524
5525 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5526                                 int size)
5527 {
5528         int rc = -EINVAL;
5529
5530         if (kvmhv_vcpu_is_radix(vcpu)) {
5531                 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5532
5533                 if (rc > 0)
5534                         rc = -EINVAL;
5535         }
5536
5537         /* For now quadrants are the only way to access nested guest memory */
5538         if (rc && vcpu->arch.nested)
5539                 rc = -EAGAIN;
5540
5541         return rc;
5542 }
5543
5544 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5545 {
5546         unpin_vpa(kvm, vpa);
5547         vpa->gpa = 0;
5548         vpa->pinned_addr = NULL;
5549         vpa->dirty = false;
5550         vpa->update_pending = 0;
5551 }
5552
5553 /*
5554  * Enable a guest to become a secure VM, or test whether
5555  * that could be enabled.
5556  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
5557  * tested (kvm == NULL) or enabled (kvm != NULL).
5558  */
5559 static int kvmhv_enable_svm(struct kvm *kvm)
5560 {
5561         if (!kvmppc_uvmem_available())
5562                 return -EINVAL;
5563         if (kvm)
5564                 kvm->arch.svm_enabled = 1;
5565         return 0;
5566 }
5567
5568 /*
5569  *  IOCTL handler to turn off secure mode of guest
5570  *
5571  * - Release all device pages
5572  * - Issue ucall to terminate the guest on the UV side
5573  * - Unpin the VPA pages.
5574  * - Reinit the partition scoped page tables
5575  */
5576 static int kvmhv_svm_off(struct kvm *kvm)
5577 {
5578         struct kvm_vcpu *vcpu;
5579         int mmu_was_ready;
5580         int srcu_idx;
5581         int ret = 0;
5582         int i;
5583
5584         if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5585                 return ret;
5586
5587         mutex_lock(&kvm->arch.mmu_setup_lock);
5588         mmu_was_ready = kvm->arch.mmu_ready;
5589         if (kvm->arch.mmu_ready) {
5590                 kvm->arch.mmu_ready = 0;
5591                 /* order mmu_ready vs. vcpus_running */
5592                 smp_mb();
5593                 if (atomic_read(&kvm->arch.vcpus_running)) {
5594                         kvm->arch.mmu_ready = 1;
5595                         ret = -EBUSY;
5596                         goto out;
5597                 }
5598         }
5599
5600         srcu_idx = srcu_read_lock(&kvm->srcu);
5601         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5602                 struct kvm_memory_slot *memslot;
5603                 struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5604
5605                 if (!slots)
5606                         continue;
5607
5608                 kvm_for_each_memslot(memslot, slots) {
5609                         kvmppc_uvmem_drop_pages(memslot, kvm, true);
5610                         uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5611                 }
5612         }
5613         srcu_read_unlock(&kvm->srcu, srcu_idx);
5614
5615         ret = uv_svm_terminate(kvm->arch.lpid);
5616         if (ret != U_SUCCESS) {
5617                 ret = -EINVAL;
5618                 goto out;
5619         }
5620
5621         /*
5622          * When secure guest is reset, all the guest pages are sent
5623          * to UV via UV_PAGE_IN before the non-boot vcpus get a
5624          * chance to run and unpin their VPA pages. Unpinning of all
5625          * VPA pages is done here explicitly so that VPA pages
5626          * can be migrated to the secure side.
5627          *
5628          * This is required to for the secure SMP guest to reboot
5629          * correctly.
5630          */
5631         kvm_for_each_vcpu(i, vcpu, kvm) {
5632                 spin_lock(&vcpu->arch.vpa_update_lock);
5633                 unpin_vpa_reset(kvm, &vcpu->arch.dtl);
5634                 unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
5635                 unpin_vpa_reset(kvm, &vcpu->arch.vpa);
5636                 spin_unlock(&vcpu->arch.vpa_update_lock);
5637         }
5638
5639         kvmppc_setup_partition_table(kvm);
5640         kvm->arch.secure_guest = 0;
5641         kvm->arch.mmu_ready = mmu_was_ready;
5642 out:
5643         mutex_unlock(&kvm->arch.mmu_setup_lock);
5644         return ret;
5645 }
5646
5647 static struct kvmppc_ops kvm_ops_hv = {
5648         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5649         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5650         .get_one_reg = kvmppc_get_one_reg_hv,
5651         .set_one_reg = kvmppc_set_one_reg_hv,
5652         .vcpu_load   = kvmppc_core_vcpu_load_hv,
5653         .vcpu_put    = kvmppc_core_vcpu_put_hv,
5654         .inject_interrupt = kvmppc_inject_interrupt_hv,
5655         .set_msr     = kvmppc_set_msr_hv,
5656         .vcpu_run    = kvmppc_vcpu_run_hv,
5657         .vcpu_create = kvmppc_core_vcpu_create_hv,
5658         .vcpu_free   = kvmppc_core_vcpu_free_hv,
5659         .check_requests = kvmppc_core_check_requests_hv,
5660         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5661         .flush_memslot  = kvmppc_core_flush_memslot_hv,
5662         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5663         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5664         .unmap_hva_range = kvm_unmap_hva_range_hv,
5665         .age_hva  = kvm_age_hva_hv,
5666         .test_age_hva = kvm_test_age_hva_hv,
5667         .set_spte_hva = kvm_set_spte_hva_hv,
5668         .free_memslot = kvmppc_core_free_memslot_hv,
5669         .init_vm =  kvmppc_core_init_vm_hv,
5670         .destroy_vm = kvmppc_core_destroy_vm_hv,
5671         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5672         .emulate_op = kvmppc_core_emulate_op_hv,
5673         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5674         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5675         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5676         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5677         .hcall_implemented = kvmppc_hcall_impl_hv,
5678 #ifdef CONFIG_KVM_XICS
5679         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5680         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5681 #endif
5682         .configure_mmu = kvmhv_configure_mmu,
5683         .get_rmmu_info = kvmhv_get_rmmu_info,
5684         .set_smt_mode = kvmhv_set_smt_mode,
5685         .enable_nested = kvmhv_enable_nested,
5686         .load_from_eaddr = kvmhv_load_from_eaddr,
5687         .store_to_eaddr = kvmhv_store_to_eaddr,
5688         .enable_svm = kvmhv_enable_svm,
5689         .svm_off = kvmhv_svm_off,
5690 };
5691
5692 static int kvm_init_subcore_bitmap(void)
5693 {
5694         int i, j;
5695         int nr_cores = cpu_nr_cores();
5696         struct sibling_subcore_state *sibling_subcore_state;
5697
5698         for (i = 0; i < nr_cores; i++) {
5699                 int first_cpu = i * threads_per_core;
5700                 int node = cpu_to_node(first_cpu);
5701
5702                 /* Ignore if it is already allocated. */
5703                 if (paca_ptrs[first_cpu]->sibling_subcore_state)
5704                         continue;
5705
5706                 sibling_subcore_state =
5707                         kzalloc_node(sizeof(struct sibling_subcore_state),
5708                                                         GFP_KERNEL, node);
5709                 if (!sibling_subcore_state)
5710                         return -ENOMEM;
5711
5712
5713                 for (j = 0; j < threads_per_core; j++) {
5714                         int cpu = first_cpu + j;
5715
5716                         paca_ptrs[cpu]->sibling_subcore_state =
5717                                                 sibling_subcore_state;
5718                 }
5719         }
5720         return 0;
5721 }
5722
5723 static int kvmppc_radix_possible(void)
5724 {
5725         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5726 }
5727
5728 static int kvmppc_book3s_init_hv(void)
5729 {
5730         int r;
5731
5732         if (!tlbie_capable) {
5733                 pr_err("KVM-HV: Host does not support TLBIE\n");
5734                 return -ENODEV;
5735         }
5736
5737         /*
5738          * FIXME!! Do we need to check on all cpus ?
5739          */
5740         r = kvmppc_core_check_processor_compat_hv();
5741         if (r < 0)
5742                 return -ENODEV;
5743
5744         r = kvmhv_nested_init();
5745         if (r)
5746                 return r;
5747
5748         r = kvm_init_subcore_bitmap();
5749         if (r)
5750                 return r;
5751
5752         /*
5753          * We need a way of accessing the XICS interrupt controller,
5754          * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5755          * indirectly, via OPAL.
5756          */
5757 #ifdef CONFIG_SMP
5758         if (!xics_on_xive() && !kvmhv_on_pseries() &&
5759             !local_paca->kvm_hstate.xics_phys) {
5760                 struct device_node *np;
5761
5762                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5763                 if (!np) {
5764                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5765                         return -ENODEV;
5766                 }
5767                 /* presence of intc confirmed - node can be dropped again */
5768                 of_node_put(np);
5769         }
5770 #endif
5771
5772         kvm_ops_hv.owner = THIS_MODULE;
5773         kvmppc_hv_ops = &kvm_ops_hv;
5774
5775         init_default_hcalls();
5776
5777         init_vcore_lists();
5778
5779         r = kvmppc_mmu_hv_init();
5780         if (r)
5781                 return r;
5782
5783         if (kvmppc_radix_possible()) {
5784                 r = kvmppc_radix_init();
5785                 if (r)
5786                         return r;
5787         }
5788
5789         /*
5790          * POWER9 chips before version 2.02 can't have some threads in
5791          * HPT mode and some in radix mode on the same core.
5792          */
5793         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5794                 unsigned int pvr = mfspr(SPRN_PVR);
5795                 if ((pvr >> 16) == PVR_POWER9 &&
5796                     (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5797                      ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5798                         no_mixing_hpt_and_radix = true;
5799         }
5800
5801         r = kvmppc_uvmem_init();
5802         if (r < 0)
5803                 pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
5804
5805         return r;
5806 }
5807
5808 static void kvmppc_book3s_exit_hv(void)
5809 {
5810         kvmppc_uvmem_free();
5811         kvmppc_free_host_rm_ops();
5812         if (kvmppc_radix_possible())
5813                 kvmppc_radix_exit();
5814         kvmppc_hv_ops = NULL;
5815         kvmhv_nested_exit();
5816 }
5817
5818 module_init(kvmppc_book3s_init_hv);
5819 module_exit(kvmppc_book3s_exit_hv);
5820 MODULE_LICENSE("GPL");
5821 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5822 MODULE_ALIAS("devname:kvm");