arch/powerpc/kvm/book3s_hv.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
   4  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
   5  *
   6  * Authors:
   7  *    Paul Mackerras <paulus@au1.ibm.com>
   8  *    Alexander Graf <agraf@suse.de>
   9  *    Kevin Wolf <mail@kevin-wolf.de>
  10  *
  11  * Description: KVM functions specific to running on Book 3S
  12  * processors in hypervisor mode (specifically POWER7 and later).
  13  *
  14  * This file is derived from arch/powerpc/kvm/book3s.c,
  15  * by Alexander Graf <agraf@suse.de>.
  16  */
  17
  18 #include <linux/kvm_host.h>
  19 #include <linux/kernel.h>
  20 #include <linux/err.h>
  21 #include <linux/slab.h>
  22 #include <linux/preempt.h>
  23 #include <linux/sched/signal.h>
  24 #include <linux/sched/stat.h>
  25 #include <linux/delay.h>
  26 #include <linux/export.h>
  27 #include <linux/fs.h>
  28 #include <linux/anon_inodes.h>
  29 #include <linux/cpu.h>
  30 #include <linux/cpumask.h>
  31 #include <linux/spinlock.h>
  32 #include <linux/page-flags.h>
  33 #include <linux/srcu.h>
  34 #include <linux/miscdevice.h>
  35 #include <linux/debugfs.h>
  36 #include <linux/gfp.h>
  37 #include <linux/vmalloc.h>
  38 #include <linux/highmem.h>
  39 #include <linux/hugetlb.h>
  40 #include <linux/kvm_irqfd.h>
  41 #include <linux/irqbypass.h>
  42 #include <linux/module.h>
  43 #include <linux/compiler.h>
  44 #include <linux/of.h>
  45
  46 #include <asm/ftrace.h>
  47 #include <asm/reg.h>
  48 #include <asm/ppc-opcode.h>
  49 #include <asm/asm-prototypes.h>
  50 #include <asm/archrandom.h>
  51 #include <asm/debug.h>
  52 #include <asm/disassemble.h>
  53 #include <asm/cputable.h>
  54 #include <asm/cacheflush.h>
  55 #include <linux/uaccess.h>
  56 #include <asm/interrupt.h>
  57 #include <asm/io.h>
  58 #include <asm/kvm_ppc.h>
  59 #include <asm/kvm_book3s.h>
  60 #include <asm/mmu_context.h>
  61 #include <asm/lppaca.h>
  62 #include <asm/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 __read_mostly;
 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.dawr0  = value1;
 787                 vcpu->arch.dawrx0 = value2;
 788                 return H_SUCCESS;
 789         case H_SET_MODE_RESOURCE_SET_DAWR1:
 790                 if (!kvmppc_power8_compatible(vcpu))
 791                         return H_P2;
 792                 if (!ppc_breakpoint_available())
 793                         return H_P2;
 794                 if (!cpu_has_feature(CPU_FTR_DAWR1))
 795                         return H_P2;
 796                 if (!vcpu->kvm->arch.dawr1_enabled)
 797                         return H_FUNCTION;
 798                 if (mflags)
 799                         return H_UNSUPPORTED_FLAG_START;
 800                 if (value2 & DABRX_HYP)
 801                         return H_P4;
 802                 vcpu->arch.dawr1  = value1;
 803                 vcpu->arch.dawrx1 = value2;
 804                 return H_SUCCESS;
 805         case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
 806                 /* KVM does not support mflags=2 (AIL=2) */
 807                 if (mflags != 0 && mflags != 3)
 808                         return H_UNSUPPORTED_FLAG_START;
 809                 return H_TOO_HARD;
 810         default:
 811                 return H_TOO_HARD;
 812         }
 813 }
 814
 815 /* Copy guest memory in place - must reside within a single memslot */
 816 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
 817                                   unsigned long len)
 818 {
 819         struct kvm_memory_slot *to_memslot = NULL;
 820         struct kvm_memory_slot *from_memslot = NULL;
 821         unsigned long to_addr, from_addr;
 822         int r;
 823
 824         /* Get HPA for from address */
 825         from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
 826         if (!from_memslot)
 827                 return -EFAULT;
 828         if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
 829                              << PAGE_SHIFT))
 830                 return -EINVAL;
 831         from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
 832         if (kvm_is_error_hva(from_addr))
 833                 return -EFAULT;
 834         from_addr |= (from & (PAGE_SIZE - 1));
 835
 836         /* Get HPA for to address */
 837         to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
 838         if (!to_memslot)
 839                 return -EFAULT;
 840         if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
 841                            << PAGE_SHIFT))
 842                 return -EINVAL;
 843         to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
 844         if (kvm_is_error_hva(to_addr))
 845                 return -EFAULT;
 846         to_addr |= (to & (PAGE_SIZE - 1));
 847
 848         /* Perform copy */
 849         r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
 850                              len);
 851         if (r)
 852                 return -EFAULT;
 853         mark_page_dirty(kvm, to >> PAGE_SHIFT);
 854         return 0;
 855 }
 856
 857 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
 858                                unsigned long dest, unsigned long src)
 859 {
 860         u64 pg_sz = SZ_4K;              /* 4K page size */
 861         u64 pg_mask = SZ_4K - 1;
 862         int ret;
 863
 864         /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
 865         if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
 866                       H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
 867                 return H_PARAMETER;
 868
 869         /* dest (and src if copy_page flag set) must be page aligned */
 870         if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
 871                 return H_PARAMETER;
 872
 873         /* zero and/or copy the page as determined by the flags */
 874         if (flags & H_COPY_PAGE) {
 875                 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
 876                 if (ret < 0)
 877                         return H_PARAMETER;
 878         } else if (flags & H_ZERO_PAGE) {
 879                 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
 880                 if (ret < 0)
 881                         return H_PARAMETER;
 882         }
 883
 884         /* We can ignore the remaining flags */
 885
 886         return H_SUCCESS;
 887 }
 888
 889 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 890 {
 891         struct kvmppc_vcore *vcore = target->arch.vcore;
 892
 893         /*
 894          * We expect to have been called by the real mode handler
 895          * (kvmppc_rm_h_confer()) which would have directly returned
 896          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 897          * have useful work to do and should not confer) so we don't
 898          * recheck that here.
 899          */
 900
 901         spin_lock(&vcore->lock);
 902         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 903             vcore->vcore_state != VCORE_INACTIVE &&
 904             vcore->runner)
 905                 target = vcore->runner;
 906         spin_unlock(&vcore->lock);
 907
 908         return kvm_vcpu_yield_to(target);
 909 }
 910
 911 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 912 {
 913         int yield_count = 0;
 914         struct lppaca *lppaca;
 915
 916         spin_lock(&vcpu->arch.vpa_update_lock);
 917         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 918         if (lppaca)
 919                 yield_count = be32_to_cpu(lppaca->yield_count);
 920         spin_unlock(&vcpu->arch.vpa_update_lock);
 921         return yield_count;
 922 }
 923
 924 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 925 {
 926         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 927         unsigned long target, ret = H_SUCCESS;
 928         int yield_count;
 929         struct kvm_vcpu *tvcpu;
 930         int idx, rc;
 931
 932         if (req <= MAX_HCALL_OPCODE &&
 933             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 934                 return RESUME_HOST;
 935
 936         switch (req) {
 937         case H_CEDE:
 938                 break;
 939         case H_PROD:
 940                 target = kvmppc_get_gpr(vcpu, 4);
 941                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 942                 if (!tvcpu) {
 943                         ret = H_PARAMETER;
 944                         break;
 945                 }
 946                 tvcpu->arch.prodded = 1;
 947                 smp_mb();
 948                 if (tvcpu->arch.ceded)
 949                         kvmppc_fast_vcpu_kick_hv(tvcpu);
 950                 break;
 951         case H_CONFER:
 952                 target = kvmppc_get_gpr(vcpu, 4);
 953                 if (target == -1)
 954                         break;
 955                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 956                 if (!tvcpu) {
 957                         ret = H_PARAMETER;
 958                         break;
 959                 }
 960                 yield_count = kvmppc_get_gpr(vcpu, 5);
 961                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 962                         break;
 963                 kvm_arch_vcpu_yield_to(tvcpu);
 964                 break;
 965         case H_REGISTER_VPA:
 966                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 967                                         kvmppc_get_gpr(vcpu, 5),
 968                                         kvmppc_get_gpr(vcpu, 6));
 969                 break;
 970         case H_RTAS:
 971                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 972                         return RESUME_HOST;
 973
 974                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 975                 rc = kvmppc_rtas_hcall(vcpu);
 976                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 977
 978                 if (rc == -ENOENT)
 979                         return RESUME_HOST;
 980                 else if (rc == 0)
 981                         break;
 982
 983                 /* Send the error out to userspace via KVM_RUN */
 984                 return rc;
 985         case H_LOGICAL_CI_LOAD:
 986                 ret = kvmppc_h_logical_ci_load(vcpu);
 987                 if (ret == H_TOO_HARD)
 988                         return RESUME_HOST;
 989                 break;
 990         case H_LOGICAL_CI_STORE:
 991                 ret = kvmppc_h_logical_ci_store(vcpu);
 992                 if (ret == H_TOO_HARD)
 993                         return RESUME_HOST;
 994                 break;
 995         case H_SET_MODE:
 996                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 997                                         kvmppc_get_gpr(vcpu, 5),
 998                                         kvmppc_get_gpr(vcpu, 6),
 999                                         kvmppc_get_gpr(vcpu, 7));
1000                 if (ret == H_TOO_HARD)
1001                         return RESUME_HOST;
1002                 break;
1003         case H_XIRR:
1004         case H_CPPR:
1005         case H_EOI:
1006         case H_IPI:
1007         case H_IPOLL:
1008         case H_XIRR_X:
1009                 if (kvmppc_xics_enabled(vcpu)) {
1010                         if (xics_on_xive()) {
1011                                 ret = H_NOT_AVAILABLE;
1012                                 return RESUME_GUEST;
1013                         }
1014                         ret = kvmppc_xics_hcall(vcpu, req);
1015                         break;
1016                 }
1017                 return RESUME_HOST;
1018         case H_SET_DABR:
1019                 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1020                 break;
1021         case H_SET_XDABR:
1022                 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1023                                                 kvmppc_get_gpr(vcpu, 5));
1024                 break;
1025 #ifdef CONFIG_SPAPR_TCE_IOMMU
1026         case H_GET_TCE:
1027                 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1028                                                 kvmppc_get_gpr(vcpu, 5));
1029                 if (ret == H_TOO_HARD)
1030                         return RESUME_HOST;
1031                 break;
1032         case H_PUT_TCE:
1033                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1034                                                 kvmppc_get_gpr(vcpu, 5),
1035                                                 kvmppc_get_gpr(vcpu, 6));
1036                 if (ret == H_TOO_HARD)
1037                         return RESUME_HOST;
1038                 break;
1039         case H_PUT_TCE_INDIRECT:
1040                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1041                                                 kvmppc_get_gpr(vcpu, 5),
1042                                                 kvmppc_get_gpr(vcpu, 6),
1043                                                 kvmppc_get_gpr(vcpu, 7));
1044                 if (ret == H_TOO_HARD)
1045                         return RESUME_HOST;
1046                 break;
1047         case H_STUFF_TCE:
1048                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1049                                                 kvmppc_get_gpr(vcpu, 5),
1050                                                 kvmppc_get_gpr(vcpu, 6),
1051                                                 kvmppc_get_gpr(vcpu, 7));
1052                 if (ret == H_TOO_HARD)
1053                         return RESUME_HOST;
1054                 break;
1055 #endif
1056         case H_RANDOM:
1057                 if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1058                         ret = H_HARDWARE;
1059                 break;
1060
1061         case H_SET_PARTITION_TABLE:
1062                 ret = H_FUNCTION;
1063                 if (nesting_enabled(vcpu->kvm))
1064                         ret = kvmhv_set_partition_table(vcpu);
1065                 break;
1066         case H_ENTER_NESTED:
1067                 ret = H_FUNCTION;
1068                 if (!nesting_enabled(vcpu->kvm))
1069                         break;
1070                 ret = kvmhv_enter_nested_guest(vcpu);
1071                 if (ret == H_INTERRUPT) {
1072                         kvmppc_set_gpr(vcpu, 3, 0);
1073                         vcpu->arch.hcall_needed = 0;
1074                         return -EINTR;
1075                 } else if (ret == H_TOO_HARD) {
1076                         kvmppc_set_gpr(vcpu, 3, 0);
1077                         vcpu->arch.hcall_needed = 0;
1078                         return RESUME_HOST;
1079                 }
1080                 break;
1081         case H_TLB_INVALIDATE:
1082                 ret = H_FUNCTION;
1083                 if (nesting_enabled(vcpu->kvm))
1084                         ret = kvmhv_do_nested_tlbie(vcpu);
1085                 break;
1086         case H_COPY_TOFROM_GUEST:
1087                 ret = H_FUNCTION;
1088                 if (nesting_enabled(vcpu->kvm))
1089                         ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1090                 break;
1091         case H_PAGE_INIT:
1092                 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1093                                          kvmppc_get_gpr(vcpu, 5),
1094                                          kvmppc_get_gpr(vcpu, 6));
1095                 break;
1096         case H_SVM_PAGE_IN:
1097                 ret = H_UNSUPPORTED;
1098                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1099                         ret = kvmppc_h_svm_page_in(vcpu->kvm,
1100                                                    kvmppc_get_gpr(vcpu, 4),
1101                                                    kvmppc_get_gpr(vcpu, 5),
1102                                                    kvmppc_get_gpr(vcpu, 6));
1103                 break;
1104         case H_SVM_PAGE_OUT:
1105                 ret = H_UNSUPPORTED;
1106                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1107                         ret = kvmppc_h_svm_page_out(vcpu->kvm,
1108                                                     kvmppc_get_gpr(vcpu, 4),
1109                                                     kvmppc_get_gpr(vcpu, 5),
1110                                                     kvmppc_get_gpr(vcpu, 6));
1111                 break;
1112         case H_SVM_INIT_START:
1113                 ret = H_UNSUPPORTED;
1114                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1115                         ret = kvmppc_h_svm_init_start(vcpu->kvm);
1116                 break;
1117         case H_SVM_INIT_DONE:
1118                 ret = H_UNSUPPORTED;
1119                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1120                         ret = kvmppc_h_svm_init_done(vcpu->kvm);
1121                 break;
1122         case H_SVM_INIT_ABORT:
1123                 /*
1124                  * Even if that call is made by the Ultravisor, the SSR1 value
1125                  * is the guest context one, with the secure bit clear as it has
1126                  * not yet been secured. So we can't check it here.
1127                  * Instead the kvm->arch.secure_guest flag is checked inside
1128                  * kvmppc_h_svm_init_abort().
1129                  */
1130                 ret = kvmppc_h_svm_init_abort(vcpu->kvm);
1131                 break;
1132
1133         default:
1134                 return RESUME_HOST;
1135         }
1136         kvmppc_set_gpr(vcpu, 3, ret);
1137         vcpu->arch.hcall_needed = 0;
1138         return RESUME_GUEST;
1139 }
1140
1141 /*
1142  * Handle H_CEDE in the nested virtualization case where we haven't
1143  * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1144  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1145  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1146  */
1147 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1148 {
1149         vcpu->arch.shregs.msr |= MSR_EE;
1150         vcpu->arch.ceded = 1;
1151         smp_mb();
1152         if (vcpu->arch.prodded) {
1153                 vcpu->arch.prodded = 0;
1154                 smp_mb();
1155                 vcpu->arch.ceded = 0;
1156         }
1157 }
1158
1159 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1160 {
1161         switch (cmd) {
1162         case H_CEDE:
1163         case H_PROD:
1164         case H_CONFER:
1165         case H_REGISTER_VPA:
1166         case H_SET_MODE:
1167         case H_LOGICAL_CI_LOAD:
1168         case H_LOGICAL_CI_STORE:
1169 #ifdef CONFIG_KVM_XICS
1170         case H_XIRR:
1171         case H_CPPR:
1172         case H_EOI:
1173         case H_IPI:
1174         case H_IPOLL:
1175         case H_XIRR_X:
1176 #endif
1177         case H_PAGE_INIT:
1178                 return 1;
1179         }
1180
1181         /* See if it's in the real-mode table */
1182         return kvmppc_hcall_impl_hv_realmode(cmd);
1183 }
1184
1185 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1186 {
1187         u32 last_inst;
1188
1189         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1190                                         EMULATE_DONE) {
1191                 /*
1192                  * Fetch failed, so return to guest and
1193                  * try executing it again.
1194                  */
1195                 return RESUME_GUEST;
1196         }
1197
1198         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1199                 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1200                 vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1201                 return RESUME_HOST;
1202         } else {
1203                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1204                 return RESUME_GUEST;
1205         }
1206 }
1207
1208 static void do_nothing(void *x)
1209 {
1210 }
1211
1212 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1213 {
1214         int thr, cpu, pcpu, nthreads;
1215         struct kvm_vcpu *v;
1216         unsigned long dpdes;
1217
1218         nthreads = vcpu->kvm->arch.emul_smt_mode;
1219         dpdes = 0;
1220         cpu = vcpu->vcpu_id & ~(nthreads - 1);
1221         for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1222                 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1223                 if (!v)
1224                         continue;
1225                 /*
1226                  * If the vcpu is currently running on a physical cpu thread,
1227                  * interrupt it in order to pull it out of the guest briefly,
1228                  * which will update its vcore->dpdes value.
1229                  */
1230                 pcpu = READ_ONCE(v->cpu);
1231                 if (pcpu >= 0)
1232                         smp_call_function_single(pcpu, do_nothing, NULL, 1);
1233                 if (kvmppc_doorbell_pending(v))
1234                         dpdes |= 1 << thr;
1235         }
1236         return dpdes;
1237 }
1238
1239 /*
1240  * On POWER9, emulate doorbell-related instructions in order to
1241  * give the guest the illusion of running on a multi-threaded core.
1242  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1243  * and mfspr DPDES.
1244  */
1245 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1246 {
1247         u32 inst, rb, thr;
1248         unsigned long arg;
1249         struct kvm *kvm = vcpu->kvm;
1250         struct kvm_vcpu *tvcpu;
1251
1252         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1253                 return RESUME_GUEST;
1254         if (get_op(inst) != 31)
1255                 return EMULATE_FAIL;
1256         rb = get_rb(inst);
1257         thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1258         switch (get_xop(inst)) {
1259         case OP_31_XOP_MSGSNDP:
1260                 arg = kvmppc_get_gpr(vcpu, rb);
1261                 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1262                         break;
1263                 arg &= 0x7f;
1264                 if (arg >= kvm->arch.emul_smt_mode)
1265                         break;
1266                 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1267                 if (!tvcpu)
1268                         break;
1269                 if (!tvcpu->arch.doorbell_request) {
1270                         tvcpu->arch.doorbell_request = 1;
1271                         kvmppc_fast_vcpu_kick_hv(tvcpu);
1272                 }
1273                 break;
1274         case OP_31_XOP_MSGCLRP:
1275                 arg = kvmppc_get_gpr(vcpu, rb);
1276                 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1277                         break;
1278                 vcpu->arch.vcore->dpdes = 0;
1279                 vcpu->arch.doorbell_request = 0;
1280                 break;
1281         case OP_31_XOP_MFSPR:
1282                 switch (get_sprn(inst)) {
1283                 case SPRN_TIR:
1284                         arg = thr;
1285                         break;
1286                 case SPRN_DPDES:
1287                         arg = kvmppc_read_dpdes(vcpu);
1288                         break;
1289                 default:
1290                         return EMULATE_FAIL;
1291                 }
1292                 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1293                 break;
1294         default:
1295                 return EMULATE_FAIL;
1296         }
1297         kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1298         return RESUME_GUEST;
1299 }
1300
1301 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1302                                  struct task_struct *tsk)
1303 {
1304         struct kvm_run *run = vcpu->run;
1305         int r = RESUME_HOST;
1306
1307         vcpu->stat.sum_exits++;
1308
1309         /*
1310          * This can happen if an interrupt occurs in the last stages
1311          * of guest entry or the first stages of guest exit (i.e. after
1312          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1313          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1314          * That can happen due to a bug, or due to a machine check
1315          * occurring at just the wrong time.
1316          */
1317         if (vcpu->arch.shregs.msr & MSR_HV) {
1318                 printk(KERN_EMERG "KVM trap in HV mode!\n");
1319                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1320                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1321                         vcpu->arch.shregs.msr);
1322                 kvmppc_dump_regs(vcpu);
1323                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1324                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1325                 return RESUME_HOST;
1326         }
1327         run->exit_reason = KVM_EXIT_UNKNOWN;
1328         run->ready_for_interrupt_injection = 1;
1329         switch (vcpu->arch.trap) {
1330         /* We're good on these - the host merely wanted to get our attention */
1331         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1332                 vcpu->stat.dec_exits++;
1333                 r = RESUME_GUEST;
1334                 break;
1335         case BOOK3S_INTERRUPT_EXTERNAL:
1336         case BOOK3S_INTERRUPT_H_DOORBELL:
1337         case BOOK3S_INTERRUPT_H_VIRT:
1338                 vcpu->stat.ext_intr_exits++;
1339                 r = RESUME_GUEST;
1340                 break;
1341         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1342         case BOOK3S_INTERRUPT_HMI:
1343         case BOOK3S_INTERRUPT_PERFMON:
1344         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1345                 r = RESUME_GUEST;
1346                 break;
1347         case BOOK3S_INTERRUPT_MACHINE_CHECK: {
1348                 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1349                                               DEFAULT_RATELIMIT_BURST);
1350                 /*
1351                  * Print the MCE event to host console. Ratelimit so the guest
1352                  * can't flood the host log.
1353                  */
1354                 if (__ratelimit(&rs))
1355                         machine_check_print_event_info(&vcpu->arch.mce_evt,false, true);
1356
1357                 /*
1358                  * If the guest can do FWNMI, exit to userspace so it can
1359                  * deliver a FWNMI to the guest.
1360                  * Otherwise we synthesize a machine check for the guest
1361                  * so that it knows that the machine check occurred.
1362                  */
1363                 if (!vcpu->kvm->arch.fwnmi_enabled) {
1364                         ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1365                         kvmppc_core_queue_machine_check(vcpu, flags);
1366                         r = RESUME_GUEST;
1367                         break;
1368                 }
1369
1370                 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1371                 run->exit_reason = KVM_EXIT_NMI;
1372                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1373                 /* Clear out the old NMI status from run->flags */
1374                 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1375                 /* Now set the NMI status */
1376                 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1377                         run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1378                 else
1379                         run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1380
1381                 r = RESUME_HOST;
1382                 break;
1383         }
1384         case BOOK3S_INTERRUPT_PROGRAM:
1385         {
1386                 ulong flags;
1387                 /*
1388                  * Normally program interrupts are delivered directly
1389                  * to the guest by the hardware, but we can get here
1390                  * as a result of a hypervisor emulation interrupt
1391                  * (e40) getting turned into a 700 by BML RTAS.
1392                  */
1393                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1394                 kvmppc_core_queue_program(vcpu, flags);
1395                 r = RESUME_GUEST;
1396                 break;
1397         }
1398         case BOOK3S_INTERRUPT_SYSCALL:
1399         {
1400                 /* hcall - punt to userspace */
1401                 int i;
1402
1403                 /* hypercall with MSR_PR has already been handled in rmode,
1404                  * and never reaches here.
1405                  */
1406
1407                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1408                 for (i = 0; i < 9; ++i)
1409                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1410                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1411                 vcpu->arch.hcall_needed = 1;
1412                 r = RESUME_HOST;
1413                 break;
1414         }
1415         /*
1416          * We get these next two if the guest accesses a page which it thinks
1417          * it has mapped but which is not actually present, either because
1418          * it is for an emulated I/O device or because the corresonding
1419          * host page has been paged out.  Any other HDSI/HISI interrupts
1420          * have been handled already.
1421          */
1422         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1423                 r = RESUME_PAGE_FAULT;
1424                 break;
1425         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1426                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1427                 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1428                         DSISR_SRR1_MATCH_64S;
1429                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1430                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1431                 r = RESUME_PAGE_FAULT;
1432                 break;
1433         /*
1434          * This occurs if the guest executes an illegal instruction.
1435          * If the guest debug is disabled, generate a program interrupt
1436          * to the guest. If guest debug is enabled, we need to check
1437          * whether the instruction is a software breakpoint instruction.
1438          * Accordingly return to Guest or Host.
1439          */
1440         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1441                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1442                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1443                                 swab32(vcpu->arch.emul_inst) :
1444                                 vcpu->arch.emul_inst;
1445                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1446                         r = kvmppc_emulate_debug_inst(vcpu);
1447                 } else {
1448                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1449                         r = RESUME_GUEST;
1450                 }
1451                 break;
1452         /*
1453          * This occurs if the guest (kernel or userspace), does something that
1454          * is prohibited by HFSCR.
1455          * On POWER9, this could be a doorbell instruction that we need
1456          * to emulate.
1457          * Otherwise, we just generate a program interrupt to the guest.
1458          */
1459         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1460                 r = EMULATE_FAIL;
1461                 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1462                     cpu_has_feature(CPU_FTR_ARCH_300))
1463                         r = kvmppc_emulate_doorbell_instr(vcpu);
1464                 if (r == EMULATE_FAIL) {
1465                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1466                         r = RESUME_GUEST;
1467                 }
1468                 break;
1469
1470 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1471         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1472                 /*
1473                  * This occurs for various TM-related instructions that
1474                  * we need to emulate on POWER9 DD2.2.  We have already
1475                  * handled the cases where the guest was in real-suspend
1476                  * mode and was transitioning to transactional state.
1477                  */
1478                 r = kvmhv_p9_tm_emulation(vcpu);
1479                 break;
1480 #endif
1481
1482         case BOOK3S_INTERRUPT_HV_RM_HARD:
1483                 r = RESUME_PASSTHROUGH;
1484                 break;
1485         default:
1486                 kvmppc_dump_regs(vcpu);
1487                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1488                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1489                         vcpu->arch.shregs.msr);
1490                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1491                 r = RESUME_HOST;
1492                 break;
1493         }
1494
1495         return r;
1496 }
1497
1498 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1499 {
1500         int r;
1501         int srcu_idx;
1502
1503         vcpu->stat.sum_exits++;
1504
1505         /*
1506          * This can happen if an interrupt occurs in the last stages
1507          * of guest entry or the first stages of guest exit (i.e. after
1508          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1509          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1510          * That can happen due to a bug, or due to a machine check
1511          * occurring at just the wrong time.
1512          */
1513         if (vcpu->arch.shregs.msr & MSR_HV) {
1514                 pr_emerg("KVM trap in HV mode while nested!\n");
1515                 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1516                          vcpu->arch.trap, kvmppc_get_pc(vcpu),
1517                          vcpu->arch.shregs.msr);
1518                 kvmppc_dump_regs(vcpu);
1519                 return RESUME_HOST;
1520         }
1521         switch (vcpu->arch.trap) {
1522         /* We're good on these - the host merely wanted to get our attention */
1523         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1524                 vcpu->stat.dec_exits++;
1525                 r = RESUME_GUEST;
1526                 break;
1527         case BOOK3S_INTERRUPT_EXTERNAL:
1528                 vcpu->stat.ext_intr_exits++;
1529                 r = RESUME_HOST;
1530                 break;
1531         case BOOK3S_INTERRUPT_H_DOORBELL:
1532         case BOOK3S_INTERRUPT_H_VIRT:
1533                 vcpu->stat.ext_intr_exits++;
1534                 r = RESUME_GUEST;
1535                 break;
1536         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1537         case BOOK3S_INTERRUPT_HMI:
1538         case BOOK3S_INTERRUPT_PERFMON:
1539         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1540                 r = RESUME_GUEST;
1541                 break;
1542         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1543         {
1544                 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1545                                               DEFAULT_RATELIMIT_BURST);
1546                 /* Pass the machine check to the L1 guest */
1547                 r = RESUME_HOST;
1548                 /* Print the MCE event to host console. */
1549                 if (__ratelimit(&rs))
1550                         machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1551                 break;
1552         }
1553         /*
1554          * We get these next two if the guest accesses a page which it thinks
1555          * it has mapped but which is not actually present, either because
1556          * it is for an emulated I/O device or because the corresonding
1557          * host page has been paged out.
1558          */
1559         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1560                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1561                 r = kvmhv_nested_page_fault(vcpu);
1562                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1563                 break;
1564         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1565                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1566                 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1567                                          DSISR_SRR1_MATCH_64S;
1568                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1569                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1570                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1571                 r = kvmhv_nested_page_fault(vcpu);
1572                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1573                 break;
1574
1575 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1576         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1577                 /*
1578                  * This occurs for various TM-related instructions that
1579                  * we need to emulate on POWER9 DD2.2.  We have already
1580                  * handled the cases where the guest was in real-suspend
1581                  * mode and was transitioning to transactional state.
1582                  */
1583                 r = kvmhv_p9_tm_emulation(vcpu);
1584                 break;
1585 #endif
1586
1587         case BOOK3S_INTERRUPT_HV_RM_HARD:
1588                 vcpu->arch.trap = 0;
1589                 r = RESUME_GUEST;
1590                 if (!xics_on_xive())
1591                         kvmppc_xics_rm_complete(vcpu, 0);
1592                 break;
1593         default:
1594                 r = RESUME_HOST;
1595                 break;
1596         }
1597
1598         return r;
1599 }
1600
1601 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1602                                             struct kvm_sregs *sregs)
1603 {
1604         int i;
1605
1606         memset(sregs, 0, sizeof(struct kvm_sregs));
1607         sregs->pvr = vcpu->arch.pvr;
1608         for (i = 0; i < vcpu->arch.slb_max; i++) {
1609                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1610                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1611         }
1612
1613         return 0;
1614 }
1615
1616 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1617                                             struct kvm_sregs *sregs)
1618 {
1619         int i, j;
1620
1621         /* Only accept the same PVR as the host's, since we can't spoof it */
1622         if (sregs->pvr != vcpu->arch.pvr)
1623                 return -EINVAL;
1624
1625         j = 0;
1626         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1627                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1628                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1629                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1630                         ++j;
1631                 }
1632         }
1633         vcpu->arch.slb_max = j;
1634
1635         return 0;
1636 }
1637
1638 /*
1639  * Enforce limits on guest LPCR values based on hardware availability,
1640  * guest configuration, and possibly hypervisor support and security
1641  * concerns.
1642  */
1643 unsigned long kvmppc_filter_lpcr_hv(struct kvm *kvm, unsigned long lpcr)
1644 {
1645         /* On POWER8 and above, userspace can modify AIL */
1646         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
1647                 lpcr &= ~LPCR_AIL;
1648
1649         /*
1650          * On POWER9, allow userspace to enable large decrementer for the
1651          * guest, whether or not the host has it enabled.
1652          */
1653         if (!cpu_has_feature(CPU_FTR_ARCH_300))
1654                 lpcr &= ~LPCR_LD;
1655
1656         return lpcr;
1657 }
1658
1659 static void verify_lpcr(struct kvm *kvm, unsigned long lpcr)
1660 {
1661         if (lpcr != kvmppc_filter_lpcr_hv(kvm, lpcr)) {
1662                 WARN_ONCE(1, "lpcr 0x%lx differs from filtered 0x%lx\n",
1663                           lpcr, kvmppc_filter_lpcr_hv(kvm, lpcr));
1664         }
1665 }
1666
1667 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1668                 bool preserve_top32)
1669 {
1670         struct kvm *kvm = vcpu->kvm;
1671         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1672         u64 mask;
1673
1674         spin_lock(&vc->lock);
1675
1676         /*
1677          * Userspace can only modify
1678          * DPFD (default prefetch depth), ILE (interrupt little-endian),
1679          * TC (translation control), AIL (alternate interrupt location),
1680          * LD (large decrementer).
1681          * These are subject to restrictions from kvmppc_filter_lcpr_hv().
1682          */
1683         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD;
1684
1685         /* Broken 32-bit version of LPCR must not clear top bits */
1686         if (preserve_top32)
1687                 mask &= 0xFFFFFFFF;
1688
1689         new_lpcr = kvmppc_filter_lpcr_hv(kvm,
1690                         (vc->lpcr & ~mask) | (new_lpcr & mask));
1691
1692         /*
1693          * If ILE (interrupt little-endian) has changed, update the
1694          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1695          */
1696         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1697                 struct kvm_vcpu *vcpu;
1698                 int i;
1699
1700                 kvm_for_each_vcpu(i, vcpu, kvm) {
1701                         if (vcpu->arch.vcore != vc)
1702                                 continue;
1703                         if (new_lpcr & LPCR_ILE)
1704                                 vcpu->arch.intr_msr |= MSR_LE;
1705                         else
1706                                 vcpu->arch.intr_msr &= ~MSR_LE;
1707                 }
1708         }
1709
1710         vc->lpcr = new_lpcr;
1711
1712         spin_unlock(&vc->lock);
1713 }
1714
1715 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1716                                  union kvmppc_one_reg *val)
1717 {
1718         int r = 0;
1719         long int i;
1720
1721         switch (id) {
1722         case KVM_REG_PPC_DEBUG_INST:
1723                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1724                 break;
1725         case KVM_REG_PPC_HIOR:
1726                 *val = get_reg_val(id, 0);
1727                 break;
1728         case KVM_REG_PPC_DABR:
1729                 *val = get_reg_val(id, vcpu->arch.dabr);
1730                 break;
1731         case KVM_REG_PPC_DABRX:
1732                 *val = get_reg_val(id, vcpu->arch.dabrx);
1733                 break;
1734         case KVM_REG_PPC_DSCR:
1735                 *val = get_reg_val(id, vcpu->arch.dscr);
1736                 break;
1737         case KVM_REG_PPC_PURR:
1738                 *val = get_reg_val(id, vcpu->arch.purr);
1739                 break;
1740         case KVM_REG_PPC_SPURR:
1741                 *val = get_reg_val(id, vcpu->arch.spurr);
1742                 break;
1743         case KVM_REG_PPC_AMR:
1744                 *val = get_reg_val(id, vcpu->arch.amr);
1745                 break;
1746         case KVM_REG_PPC_UAMOR:
1747                 *val = get_reg_val(id, vcpu->arch.uamor);
1748                 break;
1749         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
1750                 i = id - KVM_REG_PPC_MMCR0;
1751                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1752                 break;
1753         case KVM_REG_PPC_MMCR2:
1754                 *val = get_reg_val(id, vcpu->arch.mmcr[2]);
1755                 break;
1756         case KVM_REG_PPC_MMCRA:
1757                 *val = get_reg_val(id, vcpu->arch.mmcra);
1758                 break;
1759         case KVM_REG_PPC_MMCRS:
1760                 *val = get_reg_val(id, vcpu->arch.mmcrs);
1761                 break;
1762         case KVM_REG_PPC_MMCR3:
1763                 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
1764                 break;
1765         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1766                 i = id - KVM_REG_PPC_PMC1;
1767                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1768                 break;
1769         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1770                 i = id - KVM_REG_PPC_SPMC1;
1771                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1772                 break;
1773         case KVM_REG_PPC_SIAR:
1774                 *val = get_reg_val(id, vcpu->arch.siar);
1775                 break;
1776         case KVM_REG_PPC_SDAR:
1777                 *val = get_reg_val(id, vcpu->arch.sdar);
1778                 break;
1779         case KVM_REG_PPC_SIER:
1780                 *val = get_reg_val(id, vcpu->arch.sier[0]);
1781                 break;
1782         case KVM_REG_PPC_SIER2:
1783                 *val = get_reg_val(id, vcpu->arch.sier[1]);
1784                 break;
1785         case KVM_REG_PPC_SIER3:
1786                 *val = get_reg_val(id, vcpu->arch.sier[2]);
1787                 break;
1788         case KVM_REG_PPC_IAMR:
1789                 *val = get_reg_val(id, vcpu->arch.iamr);
1790                 break;
1791         case KVM_REG_PPC_PSPB:
1792                 *val = get_reg_val(id, vcpu->arch.pspb);
1793                 break;
1794         case KVM_REG_PPC_DPDES:
1795                 /*
1796                  * On POWER9, where we are emulating msgsndp etc.,
1797                  * we return 1 bit for each vcpu, which can come from
1798                  * either vcore->dpdes or doorbell_request.
1799                  * On POWER8, doorbell_request is 0.
1800                  */
1801                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes |
1802                                    vcpu->arch.doorbell_request);
1803                 break;
1804         case KVM_REG_PPC_VTB:
1805                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1806                 break;
1807         case KVM_REG_PPC_DAWR:
1808                 *val = get_reg_val(id, vcpu->arch.dawr0);
1809                 break;
1810         case KVM_REG_PPC_DAWRX:
1811                 *val = get_reg_val(id, vcpu->arch.dawrx0);
1812                 break;
1813         case KVM_REG_PPC_DAWR1:
1814                 *val = get_reg_val(id, vcpu->arch.dawr1);
1815                 break;
1816         case KVM_REG_PPC_DAWRX1:
1817                 *val = get_reg_val(id, vcpu->arch.dawrx1);
1818                 break;
1819         case KVM_REG_PPC_CIABR:
1820                 *val = get_reg_val(id, vcpu->arch.ciabr);
1821                 break;
1822         case KVM_REG_PPC_CSIGR:
1823                 *val = get_reg_val(id, vcpu->arch.csigr);
1824                 break;
1825         case KVM_REG_PPC_TACR:
1826                 *val = get_reg_val(id, vcpu->arch.tacr);
1827                 break;
1828         case KVM_REG_PPC_TCSCR:
1829                 *val = get_reg_val(id, vcpu->arch.tcscr);
1830                 break;
1831         case KVM_REG_PPC_PID:
1832                 *val = get_reg_val(id, vcpu->arch.pid);
1833                 break;
1834         case KVM_REG_PPC_ACOP:
1835                 *val = get_reg_val(id, vcpu->arch.acop);
1836                 break;
1837         case KVM_REG_PPC_WORT:
1838                 *val = get_reg_val(id, vcpu->arch.wort);
1839                 break;
1840         case KVM_REG_PPC_TIDR:
1841                 *val = get_reg_val(id, vcpu->arch.tid);
1842                 break;
1843         case KVM_REG_PPC_PSSCR:
1844                 *val = get_reg_val(id, vcpu->arch.psscr);
1845                 break;
1846         case KVM_REG_PPC_VPA_ADDR:
1847                 spin_lock(&vcpu->arch.vpa_update_lock);
1848                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1849                 spin_unlock(&vcpu->arch.vpa_update_lock);
1850                 break;
1851         case KVM_REG_PPC_VPA_SLB:
1852                 spin_lock(&vcpu->arch.vpa_update_lock);
1853                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1854                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1855                 spin_unlock(&vcpu->arch.vpa_update_lock);
1856                 break;
1857         case KVM_REG_PPC_VPA_DTL:
1858                 spin_lock(&vcpu->arch.vpa_update_lock);
1859                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1860                 val->vpaval.length = vcpu->arch.dtl.len;
1861                 spin_unlock(&vcpu->arch.vpa_update_lock);
1862                 break;
1863         case KVM_REG_PPC_TB_OFFSET:
1864                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1865                 break;
1866         case KVM_REG_PPC_LPCR:
1867         case KVM_REG_PPC_LPCR_64:
1868                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1869                 break;
1870         case KVM_REG_PPC_PPR:
1871                 *val = get_reg_val(id, vcpu->arch.ppr);
1872                 break;
1873 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1874         case KVM_REG_PPC_TFHAR:
1875                 *val = get_reg_val(id, vcpu->arch.tfhar);
1876                 break;
1877         case KVM_REG_PPC_TFIAR:
1878                 *val = get_reg_val(id, vcpu->arch.tfiar);
1879                 break;
1880         case KVM_REG_PPC_TEXASR:
1881                 *val = get_reg_val(id, vcpu->arch.texasr);
1882                 break;
1883         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1884                 i = id - KVM_REG_PPC_TM_GPR0;
1885                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1886                 break;
1887         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1888         {
1889                 int j;
1890                 i = id - KVM_REG_PPC_TM_VSR0;
1891                 if (i < 32)
1892                         for (j = 0; j < TS_FPRWIDTH; j++)
1893                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1894                 else {
1895                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1896                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1897                         else
1898                                 r = -ENXIO;
1899                 }
1900                 break;
1901         }
1902         case KVM_REG_PPC_TM_CR:
1903                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1904                 break;
1905         case KVM_REG_PPC_TM_XER:
1906                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1907                 break;
1908         case KVM_REG_PPC_TM_LR:
1909                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1910                 break;
1911         case KVM_REG_PPC_TM_CTR:
1912                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1913                 break;
1914         case KVM_REG_PPC_TM_FPSCR:
1915                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1916                 break;
1917         case KVM_REG_PPC_TM_AMR:
1918                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1919                 break;
1920         case KVM_REG_PPC_TM_PPR:
1921                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1922                 break;
1923         case KVM_REG_PPC_TM_VRSAVE:
1924                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1925                 break;
1926         case KVM_REG_PPC_TM_VSCR:
1927                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1928                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1929                 else
1930                         r = -ENXIO;
1931                 break;
1932         case KVM_REG_PPC_TM_DSCR:
1933                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1934                 break;
1935         case KVM_REG_PPC_TM_TAR:
1936                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1937                 break;
1938 #endif
1939         case KVM_REG_PPC_ARCH_COMPAT:
1940                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1941                 break;
1942         case KVM_REG_PPC_DEC_EXPIRY:
1943                 *val = get_reg_val(id, vcpu->arch.dec_expires +
1944                                    vcpu->arch.vcore->tb_offset);
1945                 break;
1946         case KVM_REG_PPC_ONLINE:
1947                 *val = get_reg_val(id, vcpu->arch.online);
1948                 break;
1949         case KVM_REG_PPC_PTCR:
1950                 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1951                 break;
1952         default:
1953                 r = -EINVAL;
1954                 break;
1955         }
1956
1957         return r;
1958 }
1959
1960 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1961                                  union kvmppc_one_reg *val)
1962 {
1963         int r = 0;
1964         long int i;
1965         unsigned long addr, len;
1966
1967         switch (id) {
1968         case KVM_REG_PPC_HIOR:
1969                 /* Only allow this to be set to zero */
1970                 if (set_reg_val(id, *val))
1971                         r = -EINVAL;
1972                 break;
1973         case KVM_REG_PPC_DABR:
1974                 vcpu->arch.dabr = set_reg_val(id, *val);
1975                 break;
1976         case KVM_REG_PPC_DABRX:
1977                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1978                 break;
1979         case KVM_REG_PPC_DSCR:
1980                 vcpu->arch.dscr = set_reg_val(id, *val);
1981                 break;
1982         case KVM_REG_PPC_PURR:
1983                 vcpu->arch.purr = set_reg_val(id, *val);
1984                 break;
1985         case KVM_REG_PPC_SPURR:
1986                 vcpu->arch.spurr = set_reg_val(id, *val);
1987                 break;
1988         case KVM_REG_PPC_AMR:
1989                 vcpu->arch.amr = set_reg_val(id, *val);
1990                 break;
1991         case KVM_REG_PPC_UAMOR:
1992                 vcpu->arch.uamor = set_reg_val(id, *val);
1993                 break;
1994         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
1995                 i = id - KVM_REG_PPC_MMCR0;
1996                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1997                 break;
1998         case KVM_REG_PPC_MMCR2:
1999                 vcpu->arch.mmcr[2] = set_reg_val(id, *val);
2000                 break;
2001         case KVM_REG_PPC_MMCRA:
2002                 vcpu->arch.mmcra = set_reg_val(id, *val);
2003                 break;
2004         case KVM_REG_PPC_MMCRS:
2005                 vcpu->arch.mmcrs = set_reg_val(id, *val);
2006                 break;
2007         case KVM_REG_PPC_MMCR3:
2008                 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
2009                 break;
2010         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2011                 i = id - KVM_REG_PPC_PMC1;
2012                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
2013                 break;
2014         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2015                 i = id - KVM_REG_PPC_SPMC1;
2016                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
2017                 break;
2018         case KVM_REG_PPC_SIAR:
2019                 vcpu->arch.siar = set_reg_val(id, *val);
2020                 break;
2021         case KVM_REG_PPC_SDAR:
2022                 vcpu->arch.sdar = set_reg_val(id, *val);
2023                 break;
2024         case KVM_REG_PPC_SIER:
2025                 vcpu->arch.sier[0] = set_reg_val(id, *val);
2026                 break;
2027         case KVM_REG_PPC_SIER2:
2028                 vcpu->arch.sier[1] = set_reg_val(id, *val);
2029                 break;
2030         case KVM_REG_PPC_SIER3:
2031                 vcpu->arch.sier[2] = set_reg_val(id, *val);
2032                 break;
2033         case KVM_REG_PPC_IAMR:
2034                 vcpu->arch.iamr = set_reg_val(id, *val);
2035                 break;
2036         case KVM_REG_PPC_PSPB:
2037                 vcpu->arch.pspb = set_reg_val(id, *val);
2038                 break;
2039         case KVM_REG_PPC_DPDES:
2040                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
2041                 break;
2042         case KVM_REG_PPC_VTB:
2043                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
2044                 break;
2045         case KVM_REG_PPC_DAWR:
2046                 vcpu->arch.dawr0 = set_reg_val(id, *val);
2047                 break;
2048         case KVM_REG_PPC_DAWRX:
2049                 vcpu->arch.dawrx0 = set_reg_val(id, *val) & ~DAWRX_HYP;
2050                 break;
2051         case KVM_REG_PPC_DAWR1:
2052                 vcpu->arch.dawr1 = set_reg_val(id, *val);
2053                 break;
2054         case KVM_REG_PPC_DAWRX1:
2055                 vcpu->arch.dawrx1 = set_reg_val(id, *val) & ~DAWRX_HYP;
2056                 break;
2057         case KVM_REG_PPC_CIABR:
2058                 vcpu->arch.ciabr = set_reg_val(id, *val);
2059                 /* Don't allow setting breakpoints in hypervisor code */
2060                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
2061                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
2062                 break;
2063         case KVM_REG_PPC_CSIGR:
2064                 vcpu->arch.csigr = set_reg_val(id, *val);
2065                 break;
2066         case KVM_REG_PPC_TACR:
2067                 vcpu->arch.tacr = set_reg_val(id, *val);
2068                 break;
2069         case KVM_REG_PPC_TCSCR:
2070                 vcpu->arch.tcscr = set_reg_val(id, *val);
2071                 break;
2072         case KVM_REG_PPC_PID:
2073                 vcpu->arch.pid = set_reg_val(id, *val);
2074                 break;
2075         case KVM_REG_PPC_ACOP:
2076                 vcpu->arch.acop = set_reg_val(id, *val);
2077                 break;
2078         case KVM_REG_PPC_WORT:
2079                 vcpu->arch.wort = set_reg_val(id, *val);
2080                 break;
2081         case KVM_REG_PPC_TIDR:
2082                 vcpu->arch.tid = set_reg_val(id, *val);
2083                 break;
2084         case KVM_REG_PPC_PSSCR:
2085                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2086                 break;
2087         case KVM_REG_PPC_VPA_ADDR:
2088                 addr = set_reg_val(id, *val);
2089                 r = -EINVAL;
2090                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2091                               vcpu->arch.dtl.next_gpa))
2092                         break;
2093                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2094                 break;
2095         case KVM_REG_PPC_VPA_SLB:
2096                 addr = val->vpaval.addr;
2097                 len = val->vpaval.length;
2098                 r = -EINVAL;
2099                 if (addr && !vcpu->arch.vpa.next_gpa)
2100                         break;
2101                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2102                 break;
2103         case KVM_REG_PPC_VPA_DTL:
2104                 addr = val->vpaval.addr;
2105                 len = val->vpaval.length;
2106                 r = -EINVAL;
2107                 if (addr && (len < sizeof(struct dtl_entry) ||
2108                              !vcpu->arch.vpa.next_gpa))
2109                         break;
2110                 len -= len % sizeof(struct dtl_entry);
2111                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2112                 break;
2113         case KVM_REG_PPC_TB_OFFSET:
2114                 /* round up to multiple of 2^24 */
2115                 vcpu->arch.vcore->tb_offset =
2116                         ALIGN(set_reg_val(id, *val), 1UL << 24);
2117                 break;
2118         case KVM_REG_PPC_LPCR:
2119                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2120                 break;
2121         case KVM_REG_PPC_LPCR_64:
2122                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2123                 break;
2124         case KVM_REG_PPC_PPR:
2125                 vcpu->arch.ppr = set_reg_val(id, *val);
2126                 break;
2127 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2128         case KVM_REG_PPC_TFHAR:
2129                 vcpu->arch.tfhar = set_reg_val(id, *val);
2130                 break;
2131         case KVM_REG_PPC_TFIAR:
2132                 vcpu->arch.tfiar = set_reg_val(id, *val);
2133                 break;
2134         case KVM_REG_PPC_TEXASR:
2135                 vcpu->arch.texasr = set_reg_val(id, *val);
2136                 break;
2137         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2138                 i = id - KVM_REG_PPC_TM_GPR0;
2139                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2140                 break;
2141         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2142         {
2143                 int j;
2144                 i = id - KVM_REG_PPC_TM_VSR0;
2145                 if (i < 32)
2146                         for (j = 0; j < TS_FPRWIDTH; j++)
2147                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2148                 else
2149                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
2150                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2151                         else
2152                                 r = -ENXIO;
2153                 break;
2154         }
2155         case KVM_REG_PPC_TM_CR:
2156                 vcpu->arch.cr_tm = set_reg_val(id, *val);
2157                 break;
2158         case KVM_REG_PPC_TM_XER:
2159                 vcpu->arch.xer_tm = set_reg_val(id, *val);
2160                 break;
2161         case KVM_REG_PPC_TM_LR:
2162                 vcpu->arch.lr_tm = set_reg_val(id, *val);
2163                 break;
2164         case KVM_REG_PPC_TM_CTR:
2165                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2166                 break;
2167         case KVM_REG_PPC_TM_FPSCR:
2168                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2169                 break;
2170         case KVM_REG_PPC_TM_AMR:
2171                 vcpu->arch.amr_tm = set_reg_val(id, *val);
2172                 break;
2173         case KVM_REG_PPC_TM_PPR:
2174                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2175                 break;
2176         case KVM_REG_PPC_TM_VRSAVE:
2177                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2178                 break;
2179         case KVM_REG_PPC_TM_VSCR:
2180                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2181                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2182                 else
2183                         r = - ENXIO;
2184                 break;
2185         case KVM_REG_PPC_TM_DSCR:
2186                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2187                 break;
2188         case KVM_REG_PPC_TM_TAR:
2189                 vcpu->arch.tar_tm = set_reg_val(id, *val);
2190                 break;
2191 #endif
2192         case KVM_REG_PPC_ARCH_COMPAT:
2193                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2194                 break;
2195         case KVM_REG_PPC_DEC_EXPIRY:
2196                 vcpu->arch.dec_expires = set_reg_val(id, *val) -
2197                         vcpu->arch.vcore->tb_offset;
2198                 break;
2199         case KVM_REG_PPC_ONLINE:
2200                 i = set_reg_val(id, *val);
2201                 if (i && !vcpu->arch.online)
2202                         atomic_inc(&vcpu->arch.vcore->online_count);
2203                 else if (!i && vcpu->arch.online)
2204                         atomic_dec(&vcpu->arch.vcore->online_count);
2205                 vcpu->arch.online = i;
2206                 break;
2207         case KVM_REG_PPC_PTCR:
2208                 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2209                 break;
2210         default:
2211                 r = -EINVAL;
2212                 break;
2213         }
2214
2215         return r;
2216 }
2217
2218 /*
2219  * On POWER9, threads are independent and can be in different partitions.
2220  * Therefore we consider each thread to be a subcore.
2221  * There is a restriction that all threads have to be in the same
2222  * MMU mode (radix or HPT), unfortunately, but since we only support
2223  * HPT guests on a HPT host so far, that isn't an impediment yet.
2224  */
2225 static int threads_per_vcore(struct kvm *kvm)
2226 {
2227         if (kvm->arch.threads_indep)
2228                 return 1;
2229         return threads_per_subcore;
2230 }
2231
2232 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2233 {
2234         struct kvmppc_vcore *vcore;
2235
2236         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2237
2238         if (vcore == NULL)
2239                 return NULL;
2240
2241         spin_lock_init(&vcore->lock);
2242         spin_lock_init(&vcore->stoltb_lock);
2243         rcuwait_init(&vcore->wait);
2244         vcore->preempt_tb = TB_NIL;
2245         vcore->lpcr = kvm->arch.lpcr;
2246         vcore->first_vcpuid = id;
2247         vcore->kvm = kvm;
2248         INIT_LIST_HEAD(&vcore->preempt_list);
2249
2250         return vcore;
2251 }
2252
2253 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2254 static struct debugfs_timings_element {
2255         const char *name;
2256         size_t offset;
2257 } timings[] = {
2258         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
2259         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
2260         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
2261         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
2262         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
2263 };
2264
2265 #define N_TIMINGS       (ARRAY_SIZE(timings))
2266
2267 struct debugfs_timings_state {
2268         struct kvm_vcpu *vcpu;
2269         unsigned int    buflen;
2270         char            buf[N_TIMINGS * 100];
2271 };
2272
2273 static int debugfs_timings_open(struct inode *inode, struct file *file)
2274 {
2275         struct kvm_vcpu *vcpu = inode->i_private;
2276         struct debugfs_timings_state *p;
2277
2278         p = kzalloc(sizeof(*p), GFP_KERNEL);
2279         if (!p)
2280                 return -ENOMEM;
2281
2282         kvm_get_kvm(vcpu->kvm);
2283         p->vcpu = vcpu;
2284         file->private_data = p;
2285
2286         return nonseekable_open(inode, file);
2287 }
2288
2289 static int debugfs_timings_release(struct inode *inode, struct file *file)
2290 {
2291         struct debugfs_timings_state *p = file->private_data;
2292
2293         kvm_put_kvm(p->vcpu->kvm);
2294         kfree(p);
2295         return 0;
2296 }
2297
2298 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2299                                     size_t len, loff_t *ppos)
2300 {
2301         struct debugfs_timings_state *p = file->private_data;
2302         struct kvm_vcpu *vcpu = p->vcpu;
2303         char *s, *buf_end;
2304         struct kvmhv_tb_accumulator tb;
2305         u64 count;
2306         loff_t pos;
2307         ssize_t n;
2308         int i, loops;
2309         bool ok;
2310
2311         if (!p->buflen) {
2312                 s = p->buf;
2313                 buf_end = s + sizeof(p->buf);
2314                 for (i = 0; i < N_TIMINGS; ++i) {
2315                         struct kvmhv_tb_accumulator *acc;
2316
2317                         acc = (struct kvmhv_tb_accumulator *)
2318                                 ((unsigned long)vcpu + timings[i].offset);
2319                         ok = false;
2320                         for (loops = 0; loops < 1000; ++loops) {
2321                                 count = acc->seqcount;
2322                                 if (!(count & 1)) {
2323                                         smp_rmb();
2324                                         tb = *acc;
2325                                         smp_rmb();
2326                                         if (count == acc->seqcount) {
2327                                                 ok = true;
2328                                                 break;
2329                                         }
2330                                 }
2331                                 udelay(1);
2332                         }
2333                         if (!ok)
2334                                 snprintf(s, buf_end - s, "%s: stuck\n",
2335                                         timings[i].name);
2336                         else
2337                                 snprintf(s, buf_end - s,
2338                                         "%s: %llu %llu %llu %llu\n",
2339                                         timings[i].name, count / 2,
2340                                         tb_to_ns(tb.tb_total),
2341                                         tb_to_ns(tb.tb_min),
2342                                         tb_to_ns(tb.tb_max));
2343                         s += strlen(s);
2344                 }
2345                 p->buflen = s - p->buf;
2346         }
2347
2348         pos = *ppos;
2349         if (pos >= p->buflen)
2350                 return 0;
2351         if (len > p->buflen - pos)
2352                 len = p->buflen - pos;
2353         n = copy_to_user(buf, p->buf + pos, len);
2354         if (n) {
2355                 if (n == len)
2356                         return -EFAULT;
2357                 len -= n;
2358         }
2359         *ppos = pos + len;
2360         return len;
2361 }
2362
2363 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2364                                      size_t len, loff_t *ppos)
2365 {
2366         return -EACCES;
2367 }
2368
2369 static const struct file_operations debugfs_timings_ops = {
2370         .owner   = THIS_MODULE,
2371         .open    = debugfs_timings_open,
2372         .release = debugfs_timings_release,
2373         .read    = debugfs_timings_read,
2374         .write   = debugfs_timings_write,
2375         .llseek  = generic_file_llseek,
2376 };
2377
2378 /* Create a debugfs directory for the vcpu */
2379 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2380 {
2381         char buf[16];
2382         struct kvm *kvm = vcpu->kvm;
2383
2384         snprintf(buf, sizeof(buf), "vcpu%u", id);
2385         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2386         debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir, vcpu,
2387                             &debugfs_timings_ops);
2388 }
2389
2390 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2391 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2392 {
2393 }
2394 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2395
2396 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2397 {
2398         int err;
2399         int core;
2400         struct kvmppc_vcore *vcore;
2401         struct kvm *kvm;
2402         unsigned int id;
2403
2404         kvm = vcpu->kvm;
2405         id = vcpu->vcpu_id;
2406
2407         vcpu->arch.shared = &vcpu->arch.shregs;
2408 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2409         /*
2410          * The shared struct is never shared on HV,
2411          * so we can always use host endianness
2412          */
2413 #ifdef __BIG_ENDIAN__
2414         vcpu->arch.shared_big_endian = true;
2415 #else
2416         vcpu->arch.shared_big_endian = false;
2417 #endif
2418 #endif
2419         vcpu->arch.mmcr[0] = MMCR0_FC;
2420         vcpu->arch.ctrl = CTRL_RUNLATCH;
2421         /* default to host PVR, since we can't spoof it */
2422         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2423         spin_lock_init(&vcpu->arch.vpa_update_lock);
2424         spin_lock_init(&vcpu->arch.tbacct_lock);
2425         vcpu->arch.busy_preempt = TB_NIL;
2426         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2427
2428         /*
2429          * Set the default HFSCR for the guest from the host value.
2430          * This value is only used on POWER9.
2431          * On POWER9, we want to virtualize the doorbell facility, so we
2432          * don't set the HFSCR_MSGP bit, and that causes those instructions
2433          * to trap and then we emulate them.
2434          */
2435         vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2436                 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP | HFSCR_PREFIX;
2437         if (cpu_has_feature(CPU_FTR_HVMODE)) {
2438                 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2439                 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2440                         vcpu->arch.hfscr |= HFSCR_TM;
2441         }
2442         if (cpu_has_feature(CPU_FTR_TM_COMP))
2443                 vcpu->arch.hfscr |= HFSCR_TM;
2444
2445         kvmppc_mmu_book3s_hv_init(vcpu);
2446
2447         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2448
2449         init_waitqueue_head(&vcpu->arch.cpu_run);
2450
2451         mutex_lock(&kvm->lock);
2452         vcore = NULL;
2453         err = -EINVAL;
2454         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2455                 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2456                         pr_devel("KVM: VCPU ID too high\n");
2457                         core = KVM_MAX_VCORES;
2458                 } else {
2459                         BUG_ON(kvm->arch.smt_mode != 1);
2460                         core = kvmppc_pack_vcpu_id(kvm, id);
2461                 }
2462         } else {
2463                 core = id / kvm->arch.smt_mode;
2464         }
2465         if (core < KVM_MAX_VCORES) {
2466                 vcore = kvm->arch.vcores[core];
2467                 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2468                         pr_devel("KVM: collision on id %u", id);
2469                         vcore = NULL;
2470                 } else if (!vcore) {
2471                         /*
2472                          * Take mmu_setup_lock for mutual exclusion
2473                          * with kvmppc_update_lpcr().
2474                          */
2475                         err = -ENOMEM;
2476                         vcore = kvmppc_vcore_create(kvm,
2477                                         id & ~(kvm->arch.smt_mode - 1));
2478                         mutex_lock(&kvm->arch.mmu_setup_lock);
2479                         kvm->arch.vcores[core] = vcore;
2480                         kvm->arch.online_vcores++;
2481                         mutex_unlock(&kvm->arch.mmu_setup_lock);
2482                 }
2483         }
2484         mutex_unlock(&kvm->lock);
2485
2486         if (!vcore)
2487                 return err;
2488
2489         spin_lock(&vcore->lock);
2490         ++vcore->num_threads;
2491         spin_unlock(&vcore->lock);
2492         vcpu->arch.vcore = vcore;
2493         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2494         vcpu->arch.thread_cpu = -1;
2495         vcpu->arch.prev_cpu = -1;
2496
2497         vcpu->arch.cpu_type = KVM_CPU_3S_64;
2498         kvmppc_sanity_check(vcpu);
2499
2500         debugfs_vcpu_init(vcpu, id);
2501
2502         return 0;
2503 }
2504
2505 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2506                               unsigned long flags)
2507 {
2508         int err;
2509         int esmt = 0;
2510
2511         if (flags)
2512                 return -EINVAL;
2513         if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2514                 return -EINVAL;
2515         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2516                 /*
2517                  * On POWER8 (or POWER7), the threading mode is "strict",
2518                  * so we pack smt_mode vcpus per vcore.
2519                  */
2520                 if (smt_mode > threads_per_subcore)
2521                         return -EINVAL;
2522         } else {
2523                 /*
2524                  * On POWER9, the threading mode is "loose",
2525                  * so each vcpu gets its own vcore.
2526                  */
2527                 esmt = smt_mode;
2528                 smt_mode = 1;
2529         }
2530         mutex_lock(&kvm->lock);
2531         err = -EBUSY;
2532         if (!kvm->arch.online_vcores) {
2533                 kvm->arch.smt_mode = smt_mode;
2534                 kvm->arch.emul_smt_mode = esmt;
2535                 err = 0;
2536         }
2537         mutex_unlock(&kvm->lock);
2538
2539         return err;
2540 }
2541
2542 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2543 {
2544         if (vpa->pinned_addr)
2545                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2546                                         vpa->dirty);
2547 }
2548
2549 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2550 {
2551         spin_lock(&vcpu->arch.vpa_update_lock);
2552         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2553         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2554         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2555         spin_unlock(&vcpu->arch.vpa_update_lock);
2556 }
2557
2558 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2559 {
2560         /* Indicate we want to get back into the guest */
2561         return 1;
2562 }
2563
2564 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2565 {
2566         unsigned long dec_nsec, now;
2567
2568         now = get_tb();
2569         if (now > vcpu->arch.dec_expires) {
2570                 /* decrementer has already gone negative */
2571                 kvmppc_core_queue_dec(vcpu);
2572                 kvmppc_core_prepare_to_enter(vcpu);
2573                 return;
2574         }
2575         dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2576         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2577         vcpu->arch.timer_running = 1;
2578 }
2579
2580 extern int __kvmppc_vcore_entry(void);
2581
2582 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2583                                    struct kvm_vcpu *vcpu)
2584 {
2585         u64 now;
2586
2587         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2588                 return;
2589         spin_lock_irq(&vcpu->arch.tbacct_lock);
2590         now = mftb();
2591         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2592                 vcpu->arch.stolen_logged;
2593         vcpu->arch.busy_preempt = now;
2594         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2595         spin_unlock_irq(&vcpu->arch.tbacct_lock);
2596         --vc->n_runnable;
2597         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2598 }
2599
2600 static int kvmppc_grab_hwthread(int cpu)
2601 {
2602         struct paca_struct *tpaca;
2603         long timeout = 10000;
2604
2605         tpaca = paca_ptrs[cpu];
2606
2607         /* Ensure the thread won't go into the kernel if it wakes */
2608         tpaca->kvm_hstate.kvm_vcpu = NULL;
2609         tpaca->kvm_hstate.kvm_vcore = NULL;
2610         tpaca->kvm_hstate.napping = 0;
2611         smp_wmb();
2612         tpaca->kvm_hstate.hwthread_req = 1;
2613
2614         /*
2615          * If the thread is already executing in the kernel (e.g. handling
2616          * a stray interrupt), wait for it to get back to nap mode.
2617          * The smp_mb() is to ensure that our setting of hwthread_req
2618          * is visible before we look at hwthread_state, so if this
2619          * races with the code at system_reset_pSeries and the thread
2620          * misses our setting of hwthread_req, we are sure to see its
2621          * setting of hwthread_state, and vice versa.
2622          */
2623         smp_mb();
2624         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2625                 if (--timeout <= 0) {
2626                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
2627                         return -EBUSY;
2628                 }
2629                 udelay(1);
2630         }
2631         return 0;
2632 }
2633
2634 static void kvmppc_release_hwthread(int cpu)
2635 {
2636         struct paca_struct *tpaca;
2637
2638         tpaca = paca_ptrs[cpu];
2639         tpaca->kvm_hstate.hwthread_req = 0;
2640         tpaca->kvm_hstate.kvm_vcpu = NULL;
2641         tpaca->kvm_hstate.kvm_vcore = NULL;
2642         tpaca->kvm_hstate.kvm_split_mode = NULL;
2643 }
2644
2645 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2646 {
2647         struct kvm_nested_guest *nested = vcpu->arch.nested;
2648         cpumask_t *cpu_in_guest;
2649         int i;
2650
2651         cpu = cpu_first_thread_sibling(cpu);
2652         if (nested) {
2653                 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2654                 cpu_in_guest = &nested->cpu_in_guest;
2655         } else {
2656                 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2657                 cpu_in_guest = &kvm->arch.cpu_in_guest;
2658         }
2659         /*
2660          * Make sure setting of bit in need_tlb_flush precedes
2661          * testing of cpu_in_guest bits.  The matching barrier on
2662          * the other side is the first smp_mb() in kvmppc_run_core().
2663          */
2664         smp_mb();
2665         for (i = 0; i < threads_per_core; ++i)
2666                 if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2667                         smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2668 }
2669
2670 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2671 {
2672         struct kvm_nested_guest *nested = vcpu->arch.nested;
2673         struct kvm *kvm = vcpu->kvm;
2674         int prev_cpu;
2675
2676         if (!cpu_has_feature(CPU_FTR_HVMODE))
2677                 return;
2678
2679         if (nested)
2680                 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2681         else
2682                 prev_cpu = vcpu->arch.prev_cpu;
2683
2684         /*
2685          * With radix, the guest can do TLB invalidations itself,
2686          * and it could choose to use the local form (tlbiel) if
2687          * it is invalidating a translation that has only ever been
2688          * used on one vcpu.  However, that doesn't mean it has
2689          * only ever been used on one physical cpu, since vcpus
2690          * can move around between pcpus.  To cope with this, when
2691          * a vcpu moves from one pcpu to another, we need to tell
2692          * any vcpus running on the same core as this vcpu previously
2693          * ran to flush the TLB.  The TLB is shared between threads,
2694          * so we use a single bit in .need_tlb_flush for all 4 threads.
2695          */
2696         if (prev_cpu != pcpu) {
2697                 if (prev_cpu >= 0 &&
2698                     cpu_first_thread_sibling(prev_cpu) !=
2699                     cpu_first_thread_sibling(pcpu))
2700                         radix_flush_cpu(kvm, prev_cpu, vcpu);
2701                 if (nested)
2702                         nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2703                 else
2704                         vcpu->arch.prev_cpu = pcpu;
2705         }
2706 }
2707
2708 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2709 {
2710         int cpu;
2711         struct paca_struct *tpaca;
2712         struct kvm *kvm = vc->kvm;
2713
2714         cpu = vc->pcpu;
2715         if (vcpu) {
2716                 if (vcpu->arch.timer_running) {
2717                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2718                         vcpu->arch.timer_running = 0;
2719                 }
2720                 cpu += vcpu->arch.ptid;
2721                 vcpu->cpu = vc->pcpu;
2722                 vcpu->arch.thread_cpu = cpu;
2723                 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2724         }
2725         tpaca = paca_ptrs[cpu];
2726         tpaca->kvm_hstate.kvm_vcpu = vcpu;
2727         tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2728         tpaca->kvm_hstate.fake_suspend = 0;
2729         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2730         smp_wmb();
2731         tpaca->kvm_hstate.kvm_vcore = vc;
2732         if (cpu != smp_processor_id())
2733                 kvmppc_ipi_thread(cpu);
2734 }
2735
2736 static void kvmppc_wait_for_nap(int n_threads)
2737 {
2738         int cpu = smp_processor_id();
2739         int i, loops;
2740
2741         if (n_threads <= 1)
2742                 return;
2743         for (loops = 0; loops < 1000000; ++loops) {
2744                 /*
2745                  * Check if all threads are finished.
2746                  * We set the vcore pointer when starting a thread
2747                  * and the thread clears it when finished, so we look
2748                  * for any threads that still have a non-NULL vcore ptr.
2749                  */
2750                 for (i = 1; i < n_threads; ++i)
2751                         if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2752                                 break;
2753                 if (i == n_threads) {
2754                         HMT_medium();
2755                         return;
2756                 }
2757                 HMT_low();
2758         }
2759         HMT_medium();
2760         for (i = 1; i < n_threads; ++i)
2761                 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2762                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2763 }
2764
2765 /*
2766  * Check that we are on thread 0 and that any other threads in
2767  * this core are off-line.  Then grab the threads so they can't
2768  * enter the kernel.
2769  */
2770 static int on_primary_thread(void)
2771 {
2772         int cpu = smp_processor_id();
2773         int thr;
2774
2775         /* Are we on a primary subcore? */
2776         if (cpu_thread_in_subcore(cpu))
2777                 return 0;
2778
2779         thr = 0;
2780         while (++thr < threads_per_subcore)
2781                 if (cpu_online(cpu + thr))
2782                         return 0;
2783
2784         /* Grab all hw threads so they can't go into the kernel */
2785         for (thr = 1; thr < threads_per_subcore; ++thr) {
2786                 if (kvmppc_grab_hwthread(cpu + thr)) {
2787                         /* Couldn't grab one; let the others go */
2788                         do {
2789                                 kvmppc_release_hwthread(cpu + thr);
2790                         } while (--thr > 0);
2791                         return 0;
2792                 }
2793         }
2794         return 1;
2795 }
2796
2797 /*
2798  * A list of virtual cores for each physical CPU.
2799  * These are vcores that could run but their runner VCPU tasks are
2800  * (or may be) preempted.
2801  */
2802 struct preempted_vcore_list {
2803         struct list_head        list;
2804         spinlock_t              lock;
2805 };
2806
2807 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2808
2809 static void init_vcore_lists(void)
2810 {
2811         int cpu;
2812
2813         for_each_possible_cpu(cpu) {
2814                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2815                 spin_lock_init(&lp->lock);
2816                 INIT_LIST_HEAD(&lp->list);
2817         }
2818 }
2819
2820 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2821 {
2822         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2823
2824         vc->vcore_state = VCORE_PREEMPT;
2825         vc->pcpu = smp_processor_id();
2826         if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2827                 spin_lock(&lp->lock);
2828                 list_add_tail(&vc->preempt_list, &lp->list);
2829                 spin_unlock(&lp->lock);
2830         }
2831
2832         /* Start accumulating stolen time */
2833         kvmppc_core_start_stolen(vc);
2834 }
2835
2836 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2837 {
2838         struct preempted_vcore_list *lp;
2839
2840         kvmppc_core_end_stolen(vc);
2841         if (!list_empty(&vc->preempt_list)) {
2842                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2843                 spin_lock(&lp->lock);
2844                 list_del_init(&vc->preempt_list);
2845                 spin_unlock(&lp->lock);
2846         }
2847         vc->vcore_state = VCORE_INACTIVE;
2848 }
2849
2850 /*
2851  * This stores information about the virtual cores currently
2852  * assigned to a physical core.
2853  */
2854 struct core_info {
2855         int             n_subcores;
2856         int             max_subcore_threads;
2857         int             total_threads;
2858         int             subcore_threads[MAX_SUBCORES];
2859         struct kvmppc_vcore *vc[MAX_SUBCORES];
2860 };
2861
2862 /*
2863  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2864  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2865  */
2866 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2867
2868 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2869 {
2870         memset(cip, 0, sizeof(*cip));
2871         cip->n_subcores = 1;
2872         cip->max_subcore_threads = vc->num_threads;
2873         cip->total_threads = vc->num_threads;
2874         cip->subcore_threads[0] = vc->num_threads;
2875         cip->vc[0] = vc;
2876 }
2877
2878 static bool subcore_config_ok(int n_subcores, int n_threads)
2879 {
2880         /*
2881          * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2882          * split-core mode, with one thread per subcore.
2883          */
2884         if (cpu_has_feature(CPU_FTR_ARCH_300))
2885                 return n_subcores <= 4 && n_threads == 1;
2886
2887         /* On POWER8, can only dynamically split if unsplit to begin with */
2888         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2889                 return false;
2890         if (n_subcores > MAX_SUBCORES)
2891                 return false;
2892         if (n_subcores > 1) {
2893                 if (!(dynamic_mt_modes & 2))
2894                         n_subcores = 4;
2895                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2896                         return false;
2897         }
2898
2899         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2900 }
2901
2902 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2903 {
2904         vc->entry_exit_map = 0;
2905         vc->in_guest = 0;
2906         vc->napping_threads = 0;
2907         vc->conferring_threads = 0;
2908         vc->tb_offset_applied = 0;
2909 }
2910
2911 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2912 {
2913         int n_threads = vc->num_threads;
2914         int sub;
2915
2916         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2917                 return false;
2918
2919         /* In one_vm_per_core mode, require all vcores to be from the same vm */
2920         if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2921                 return false;
2922
2923         if (n_threads < cip->max_subcore_threads)
2924                 n_threads = cip->max_subcore_threads;
2925         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2926                 return false;
2927         cip->max_subcore_threads = n_threads;
2928
2929         sub = cip->n_subcores;
2930         ++cip->n_subcores;
2931         cip->total_threads += vc->num_threads;
2932         cip->subcore_threads[sub] = vc->num_threads;
2933         cip->vc[sub] = vc;
2934         init_vcore_to_run(vc);
2935         list_del_init(&vc->preempt_list);
2936
2937         return true;
2938 }
2939
2940 /*
2941  * Work out whether it is possible to piggyback the execution of
2942  * vcore *pvc onto the execution of the other vcores described in *cip.
2943  */
2944 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2945                           int target_threads)
2946 {
2947         if (cip->total_threads + pvc->num_threads > target_threads)
2948                 return false;
2949
2950         return can_dynamic_split(pvc, cip);
2951 }
2952
2953 static void prepare_threads(struct kvmppc_vcore *vc)
2954 {
2955         int i;
2956         struct kvm_vcpu *vcpu;
2957
2958         for_each_runnable_thread(i, vcpu, vc) {
2959                 if (signal_pending(vcpu->arch.run_task))
2960                         vcpu->arch.ret = -EINTR;
2961                 else if (no_mixing_hpt_and_radix &&
2962                          kvm_is_radix(vc->kvm) != radix_enabled())
2963                         vcpu->arch.ret = -EINVAL;
2964                 else if (vcpu->arch.vpa.update_pending ||
2965                          vcpu->arch.slb_shadow.update_pending ||
2966                          vcpu->arch.dtl.update_pending)
2967                         vcpu->arch.ret = RESUME_GUEST;
2968                 else
2969                         continue;
2970                 kvmppc_remove_runnable(vc, vcpu);
2971                 wake_up(&vcpu->arch.cpu_run);
2972         }
2973 }
2974
2975 static void collect_piggybacks(struct core_info *cip, int target_threads)
2976 {
2977         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2978         struct kvmppc_vcore *pvc, *vcnext;
2979
2980         spin_lock(&lp->lock);
2981         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2982                 if (!spin_trylock(&pvc->lock))
2983                         continue;
2984                 prepare_threads(pvc);
2985                 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
2986                         list_del_init(&pvc->preempt_list);
2987                         if (pvc->runner == NULL) {
2988                                 pvc->vcore_state = VCORE_INACTIVE;
2989                                 kvmppc_core_end_stolen(pvc);
2990                         }
2991                         spin_unlock(&pvc->lock);
2992                         continue;
2993                 }
2994                 if (!can_piggyback(pvc, cip, target_threads)) {
2995                         spin_unlock(&pvc->lock);
2996                         continue;
2997                 }
2998                 kvmppc_core_end_stolen(pvc);
2999                 pvc->vcore_state = VCORE_PIGGYBACK;
3000                 if (cip->total_threads >= target_threads)
3001                         break;
3002         }
3003         spin_unlock(&lp->lock);
3004 }
3005
3006 static bool recheck_signals_and_mmu(struct core_info *cip)
3007 {
3008         int sub, i;
3009         struct kvm_vcpu *vcpu;
3010         struct kvmppc_vcore *vc;
3011
3012         for (sub = 0; sub < cip->n_subcores; ++sub) {
3013                 vc = cip->vc[sub];
3014                 if (!vc->kvm->arch.mmu_ready)
3015                         return true;
3016                 for_each_runnable_thread(i, vcpu, vc)
3017                         if (signal_pending(vcpu->arch.run_task))
3018                                 return true;
3019         }
3020         return false;
3021 }
3022
3023 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
3024 {
3025         int still_running = 0, i;
3026         u64 now;
3027         long ret;
3028         struct kvm_vcpu *vcpu;
3029
3030         spin_lock(&vc->lock);
3031         now = get_tb();
3032         for_each_runnable_thread(i, vcpu, vc) {
3033                 /*
3034                  * It's safe to unlock the vcore in the loop here, because
3035                  * for_each_runnable_thread() is safe against removal of
3036                  * the vcpu, and the vcore state is VCORE_EXITING here,
3037                  * so any vcpus becoming runnable will have their arch.trap
3038                  * set to zero and can't actually run in the guest.
3039                  */
3040                 spin_unlock(&vc->lock);
3041                 /* cancel pending dec exception if dec is positive */
3042                 if (now < vcpu->arch.dec_expires &&
3043                     kvmppc_core_pending_dec(vcpu))
3044                         kvmppc_core_dequeue_dec(vcpu);
3045
3046                 trace_kvm_guest_exit(vcpu);
3047
3048                 ret = RESUME_GUEST;
3049                 if (vcpu->arch.trap)
3050                         ret = kvmppc_handle_exit_hv(vcpu,
3051                                                     vcpu->arch.run_task);
3052
3053                 vcpu->arch.ret = ret;
3054                 vcpu->arch.trap = 0;
3055
3056                 spin_lock(&vc->lock);
3057                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
3058                         if (vcpu->arch.pending_exceptions)
3059                                 kvmppc_core_prepare_to_enter(vcpu);
3060                         if (vcpu->arch.ceded)
3061                                 kvmppc_set_timer(vcpu);
3062                         else
3063                                 ++still_running;
3064                 } else {
3065                         kvmppc_remove_runnable(vc, vcpu);
3066                         wake_up(&vcpu->arch.cpu_run);
3067                 }
3068         }
3069         if (!is_master) {
3070                 if (still_running > 0) {
3071                         kvmppc_vcore_preempt(vc);
3072                 } else if (vc->runner) {
3073                         vc->vcore_state = VCORE_PREEMPT;
3074                         kvmppc_core_start_stolen(vc);
3075                 } else {
3076                         vc->vcore_state = VCORE_INACTIVE;
3077                 }
3078                 if (vc->n_runnable > 0 && vc->runner == NULL) {
3079                         /* make sure there's a candidate runner awake */
3080                         i = -1;
3081                         vcpu = next_runnable_thread(vc, &i);
3082                         wake_up(&vcpu->arch.cpu_run);
3083                 }
3084         }
3085         spin_unlock(&vc->lock);
3086 }
3087
3088 /*
3089  * Clear core from the list of active host cores as we are about to
3090  * enter the guest. Only do this if it is the primary thread of the
3091  * core (not if a subcore) that is entering the guest.
3092  */
3093 static inline int kvmppc_clear_host_core(unsigned int cpu)
3094 {
3095         int core;
3096
3097         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3098                 return 0;
3099         /*
3100          * Memory barrier can be omitted here as we will do a smp_wmb()
3101          * later in kvmppc_start_thread and we need ensure that state is
3102          * visible to other CPUs only after we enter guest.
3103          */
3104         core = cpu >> threads_shift;
3105         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3106         return 0;
3107 }
3108
3109 /*
3110  * Advertise this core as an active host core since we exited the guest
3111  * Only need to do this if it is the primary thread of the core that is
3112  * exiting.
3113  */
3114 static inline int kvmppc_set_host_core(unsigned int cpu)
3115 {
3116         int core;
3117
3118         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3119                 return 0;
3120
3121         /*
3122          * Memory barrier can be omitted here because we do a spin_unlock
3123          * immediately after this which provides the memory barrier.
3124          */
3125         core = cpu >> threads_shift;
3126         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3127         return 0;
3128 }
3129
3130 static void set_irq_happened(int trap)
3131 {
3132         switch (trap) {
3133         case BOOK3S_INTERRUPT_EXTERNAL:
3134                 local_paca->irq_happened |= PACA_IRQ_EE;
3135                 break;
3136         case BOOK3S_INTERRUPT_H_DOORBELL:
3137                 local_paca->irq_happened |= PACA_IRQ_DBELL;
3138                 break;
3139         case BOOK3S_INTERRUPT_HMI:
3140                 local_paca->irq_happened |= PACA_IRQ_HMI;
3141                 break;
3142         case BOOK3S_INTERRUPT_SYSTEM_RESET:
3143                 replay_system_reset();
3144                 break;
3145         }
3146 }
3147
3148 /*
3149  * Run a set of guest threads on a physical core.
3150  * Called with vc->lock held.
3151  */
3152 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3153 {
3154         struct kvm_vcpu *vcpu;
3155         int i;
3156         int srcu_idx;
3157         struct core_info core_info;
3158         struct kvmppc_vcore *pvc;
3159         struct kvm_split_mode split_info, *sip;
3160         int split, subcore_size, active;
3161         int sub;
3162         bool thr0_done;
3163         unsigned long cmd_bit, stat_bit;
3164         int pcpu, thr;
3165         int target_threads;
3166         int controlled_threads;
3167         int trap;
3168         bool is_power8;
3169
3170         /*
3171          * Remove from the list any threads that have a signal pending
3172          * or need a VPA update done
3173          */
3174         prepare_threads(vc);
3175
3176         /* if the runner is no longer runnable, let the caller pick a new one */
3177         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3178                 return;
3179
3180         /*
3181          * Initialize *vc.
3182          */
3183         init_vcore_to_run(vc);
3184         vc->preempt_tb = TB_NIL;
3185
3186         /*
3187          * Number of threads that we will be controlling: the same as
3188          * the number of threads per subcore, except on POWER9,
3189          * where it's 1 because the threads are (mostly) independent.
3190          */
3191         controlled_threads = threads_per_vcore(vc->kvm);
3192
3193         /*
3194          * Make sure we are running on primary threads, and that secondary
3195          * threads are offline.  Also check if the number of threads in this
3196          * guest are greater than the current system threads per guest.
3197          * On POWER9, we need to be not in independent-threads mode if
3198          * this is a HPT guest on a radix host machine where the
3199          * CPU threads may not be in different MMU modes.
3200          */
3201         if ((controlled_threads > 1) &&
3202             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
3203                 for_each_runnable_thread(i, vcpu, vc) {
3204                         vcpu->arch.ret = -EBUSY;
3205                         kvmppc_remove_runnable(vc, vcpu);
3206                         wake_up(&vcpu->arch.cpu_run);
3207                 }
3208                 goto out;
3209         }
3210
3211         /*
3212          * See if we could run any other vcores on the physical core
3213          * along with this one.
3214          */
3215         init_core_info(&core_info, vc);
3216         pcpu = smp_processor_id();
3217         target_threads = controlled_threads;
3218         if (target_smt_mode && target_smt_mode < target_threads)
3219                 target_threads = target_smt_mode;
3220         if (vc->num_threads < target_threads)
3221                 collect_piggybacks(&core_info, target_threads);
3222
3223         /*
3224          * On radix, arrange for TLB flushing if necessary.
3225          * This has to be done before disabling interrupts since
3226          * it uses smp_call_function().
3227          */
3228         pcpu = smp_processor_id();
3229         if (kvm_is_radix(vc->kvm)) {
3230                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3231                         for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3232                                 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3233         }
3234
3235         /*
3236          * Hard-disable interrupts, and check resched flag and signals.
3237          * If we need to reschedule or deliver a signal, clean up
3238          * and return without going into the guest(s).
3239          * If the mmu_ready flag has been cleared, don't go into the
3240          * guest because that means a HPT resize operation is in progress.
3241          */
3242         local_irq_disable();
3243         hard_irq_disable();
3244         if (lazy_irq_pending() || need_resched() ||
3245             recheck_signals_and_mmu(&core_info)) {
3246                 local_irq_enable();
3247                 vc->vcore_state = VCORE_INACTIVE;
3248                 /* Unlock all except the primary vcore */
3249                 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3250                         pvc = core_info.vc[sub];
3251                         /* Put back on to the preempted vcores list */
3252                         kvmppc_vcore_preempt(pvc);
3253                         spin_unlock(&pvc->lock);
3254                 }
3255                 for (i = 0; i < controlled_threads; ++i)
3256                         kvmppc_release_hwthread(pcpu + i);
3257                 return;
3258         }
3259
3260         kvmppc_clear_host_core(pcpu);
3261
3262         /* Decide on micro-threading (split-core) mode */
3263         subcore_size = threads_per_subcore;
3264         cmd_bit = stat_bit = 0;
3265         split = core_info.n_subcores;
3266         sip = NULL;
3267         is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3268                 && !cpu_has_feature(CPU_FTR_ARCH_300);
3269
3270         if (split > 1) {
3271                 sip = &split_info;
3272                 memset(&split_info, 0, sizeof(split_info));
3273                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3274                         split_info.vc[sub] = core_info.vc[sub];
3275
3276                 if (is_power8) {
3277                         if (split == 2 && (dynamic_mt_modes & 2)) {
3278                                 cmd_bit = HID0_POWER8_1TO2LPAR;
3279                                 stat_bit = HID0_POWER8_2LPARMODE;
3280                         } else {
3281                                 split = 4;
3282                                 cmd_bit = HID0_POWER8_1TO4LPAR;
3283                                 stat_bit = HID0_POWER8_4LPARMODE;
3284                         }
3285                         subcore_size = MAX_SMT_THREADS / split;
3286                         split_info.rpr = mfspr(SPRN_RPR);
3287                         split_info.pmmar = mfspr(SPRN_PMMAR);
3288                         split_info.ldbar = mfspr(SPRN_LDBAR);
3289                         split_info.subcore_size = subcore_size;
3290                 } else {
3291                         split_info.subcore_size = 1;
3292                 }
3293
3294                 /* order writes to split_info before kvm_split_mode pointer */
3295                 smp_wmb();
3296         }
3297
3298         for (thr = 0; thr < controlled_threads; ++thr) {
3299                 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3300
3301                 paca->kvm_hstate.napping = 0;
3302                 paca->kvm_hstate.kvm_split_mode = sip;
3303         }
3304
3305         /* Initiate micro-threading (split-core) on POWER8 if required */
3306         if (cmd_bit) {
3307                 unsigned long hid0 = mfspr(SPRN_HID0);
3308
3309                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3310                 mb();
3311                 mtspr(SPRN_HID0, hid0);
3312                 isync();
3313                 for (;;) {
3314                         hid0 = mfspr(SPRN_HID0);
3315                         if (hid0 & stat_bit)
3316                                 break;
3317                         cpu_relax();
3318                 }
3319         }
3320
3321         /*
3322          * On POWER8, set RWMR register.
3323          * Since it only affects PURR and SPURR, it doesn't affect
3324          * the host, so we don't save/restore the host value.
3325          */
3326         if (is_power8) {
3327                 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3328                 int n_online = atomic_read(&vc->online_count);
3329
3330                 /*
3331                  * Use the 8-thread value if we're doing split-core
3332                  * or if the vcore's online count looks bogus.
3333                  */
3334                 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3335                     n_online >= 1 && n_online <= MAX_SMT_THREADS)
3336                         rwmr_val = p8_rwmr_values[n_online];
3337                 mtspr(SPRN_RWMR, rwmr_val);
3338         }
3339
3340         /* Start all the threads */
3341         active = 0;
3342         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3343                 thr = is_power8 ? subcore_thread_map[sub] : sub;
3344                 thr0_done = false;
3345                 active |= 1 << thr;
3346                 pvc = core_info.vc[sub];
3347                 pvc->pcpu = pcpu + thr;
3348                 for_each_runnable_thread(i, vcpu, pvc) {
3349                         kvmppc_start_thread(vcpu, pvc);
3350                         kvmppc_create_dtl_entry(vcpu, pvc);
3351                         trace_kvm_guest_enter(vcpu);
3352                         if (!vcpu->arch.ptid)
3353                                 thr0_done = true;
3354                         active |= 1 << (thr + vcpu->arch.ptid);
3355                 }
3356                 /*
3357                  * We need to start the first thread of each subcore
3358                  * even if it doesn't have a vcpu.
3359                  */
3360                 if (!thr0_done)
3361                         kvmppc_start_thread(NULL, pvc);
3362         }
3363
3364         /*
3365          * Ensure that split_info.do_nap is set after setting
3366          * the vcore pointer in the PACA of the secondaries.
3367          */
3368         smp_mb();
3369
3370         /*
3371          * When doing micro-threading, poke the inactive threads as well.
3372          * This gets them to the nap instruction after kvm_do_nap,
3373          * which reduces the time taken to unsplit later.
3374          */
3375         if (cmd_bit) {
3376                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
3377                 for (thr = 1; thr < threads_per_subcore; ++thr)
3378                         if (!(active & (1 << thr)))
3379                                 kvmppc_ipi_thread(pcpu + thr);
3380         }
3381
3382         vc->vcore_state = VCORE_RUNNING;
3383         preempt_disable();
3384
3385         trace_kvmppc_run_core(vc, 0);
3386
3387         for (sub = 0; sub < core_info.n_subcores; ++sub)
3388                 spin_unlock(&core_info.vc[sub]->lock);
3389
3390         guest_enter_irqoff();
3391
3392         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3393
3394         this_cpu_disable_ftrace();
3395
3396         /*
3397          * Interrupts will be enabled once we get into the guest,
3398          * so tell lockdep that we're about to enable interrupts.
3399          */
3400         trace_hardirqs_on();
3401
3402         trap = __kvmppc_vcore_entry();
3403
3404         trace_hardirqs_off();
3405
3406         this_cpu_enable_ftrace();
3407
3408         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3409
3410         set_irq_happened(trap);
3411
3412         spin_lock(&vc->lock);
3413         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3414         vc->vcore_state = VCORE_EXITING;
3415
3416         /* wait for secondary threads to finish writing their state to memory */
3417         kvmppc_wait_for_nap(controlled_threads);
3418
3419         /* Return to whole-core mode if we split the core earlier */
3420         if (cmd_bit) {
3421                 unsigned long hid0 = mfspr(SPRN_HID0);
3422                 unsigned long loops = 0;
3423
3424                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3425                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3426                 mb();
3427                 mtspr(SPRN_HID0, hid0);
3428                 isync();
3429                 for (;;) {
3430                         hid0 = mfspr(SPRN_HID0);
3431                         if (!(hid0 & stat_bit))
3432                                 break;
3433                         cpu_relax();
3434                         ++loops;
3435                 }
3436                 split_info.do_nap = 0;
3437         }
3438
3439         kvmppc_set_host_core(pcpu);
3440
3441         guest_exit_irqoff();
3442
3443         local_irq_enable();
3444
3445         /* Let secondaries go back to the offline loop */
3446         for (i = 0; i < controlled_threads; ++i) {
3447                 kvmppc_release_hwthread(pcpu + i);
3448                 if (sip && sip->napped[i])
3449                         kvmppc_ipi_thread(pcpu + i);
3450                 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3451         }
3452
3453         spin_unlock(&vc->lock);
3454
3455         /* make sure updates to secondary vcpu structs are visible now */
3456         smp_mb();
3457
3458         preempt_enable();
3459
3460         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3461                 pvc = core_info.vc[sub];
3462                 post_guest_process(pvc, pvc == vc);
3463         }
3464
3465         spin_lock(&vc->lock);
3466
3467  out:
3468         vc->vcore_state = VCORE_INACTIVE;
3469         trace_kvmppc_run_core(vc, 1);
3470 }
3471
3472 /*
3473  * Load up hypervisor-mode registers on P9.
3474  */
3475 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3476                                      unsigned long lpcr)
3477 {
3478         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3479         s64 hdec;
3480         u64 tb, purr, spurr;
3481         int trap;
3482         unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3483         unsigned long host_ciabr = mfspr(SPRN_CIABR);
3484         unsigned long host_dawr0 = mfspr(SPRN_DAWR0);
3485         unsigned long host_dawrx0 = mfspr(SPRN_DAWRX0);
3486         unsigned long host_psscr = mfspr(SPRN_PSSCR);
3487         unsigned long host_pidr = mfspr(SPRN_PID);
3488         unsigned long host_dawr1 = 0;
3489         unsigned long host_dawrx1 = 0;
3490
3491         if (cpu_has_feature(CPU_FTR_DAWR1)) {
3492                 host_dawr1 = mfspr(SPRN_DAWR1);
3493                 host_dawrx1 = mfspr(SPRN_DAWRX1);
3494         }
3495
3496         /*
3497          * P8 and P9 suppress the HDEC exception when LPCR[HDICE] = 0,
3498          * so set HDICE before writing HDEC.
3499          */
3500         mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr | LPCR_HDICE);
3501         isync();
3502
3503         hdec = time_limit - mftb();
3504         if (hdec < 0) {
3505                 mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3506                 isync();
3507                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3508         }
3509         mtspr(SPRN_HDEC, hdec);
3510
3511         if (vc->tb_offset) {
3512                 u64 new_tb = mftb() + vc->tb_offset;
3513                 mtspr(SPRN_TBU40, new_tb);
3514                 tb = mftb();
3515                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3516                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3517                 vc->tb_offset_applied = vc->tb_offset;
3518         }
3519
3520         if (vc->pcr)
3521                 mtspr(SPRN_PCR, vc->pcr | PCR_MASK);
3522         mtspr(SPRN_DPDES, vc->dpdes);
3523         mtspr(SPRN_VTB, vc->vtb);
3524
3525         local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3526         local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3527         mtspr(SPRN_PURR, vcpu->arch.purr);
3528         mtspr(SPRN_SPURR, vcpu->arch.spurr);
3529
3530         if (dawr_enabled()) {
3531                 mtspr(SPRN_DAWR0, vcpu->arch.dawr0);
3532                 mtspr(SPRN_DAWRX0, vcpu->arch.dawrx0);
3533                 if (cpu_has_feature(CPU_FTR_DAWR1)) {
3534                         mtspr(SPRN_DAWR1, vcpu->arch.dawr1);
3535                         mtspr(SPRN_DAWRX1, vcpu->arch.dawrx1);
3536                 }
3537         }
3538         mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3539         mtspr(SPRN_IC, vcpu->arch.ic);
3540         mtspr(SPRN_PID, vcpu->arch.pid);
3541
3542         mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3543               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3544
3545         mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3546
3547         mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3548         mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3549         mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3550         mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3551
3552         mtspr(SPRN_AMOR, ~0UL);
3553
3554         mtspr(SPRN_LPCR, lpcr);
3555         isync();
3556
3557         kvmppc_xive_push_vcpu(vcpu);
3558
3559         mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3560         mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3561
3562         trap = __kvmhv_vcpu_entry_p9(vcpu);
3563
3564         /* Advance host PURR/SPURR by the amount used by guest */
3565         purr = mfspr(SPRN_PURR);
3566         spurr = mfspr(SPRN_SPURR);
3567         mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3568               purr - vcpu->arch.purr);
3569         mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3570               spurr - vcpu->arch.spurr);
3571         vcpu->arch.purr = purr;
3572         vcpu->arch.spurr = spurr;
3573
3574         vcpu->arch.ic = mfspr(SPRN_IC);
3575         vcpu->arch.pid = mfspr(SPRN_PID);
3576         vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3577
3578         vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3579         vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3580         vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3581         vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3582
3583         /* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3584         mtspr(SPRN_PSSCR, host_psscr |
3585               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3586         mtspr(SPRN_HFSCR, host_hfscr);
3587         mtspr(SPRN_CIABR, host_ciabr);
3588         mtspr(SPRN_DAWR0, host_dawr0);
3589         mtspr(SPRN_DAWRX0, host_dawrx0);
3590         if (cpu_has_feature(CPU_FTR_DAWR1)) {
3591                 mtspr(SPRN_DAWR1, host_dawr1);
3592                 mtspr(SPRN_DAWRX1, host_dawrx1);
3593         }
3594         mtspr(SPRN_PID, host_pidr);
3595
3596         /*
3597          * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3598          * case we interrupted the guest between a tlbie and a ptesync.
3599          */
3600         asm volatile("eieio; tlbsync; ptesync");
3601
3602         /*
3603          * cp_abort is required if the processor supports local copy-paste
3604          * to clear the copy buffer that was under control of the guest.
3605          */
3606         if (cpu_has_feature(CPU_FTR_ARCH_31))
3607                 asm volatile(PPC_CP_ABORT);
3608
3609         mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid);    /* restore host LPID */
3610         isync();
3611
3612         vc->dpdes = mfspr(SPRN_DPDES);
3613         vc->vtb = mfspr(SPRN_VTB);
3614         mtspr(SPRN_DPDES, 0);
3615         if (vc->pcr)
3616                 mtspr(SPRN_PCR, PCR_MASK);
3617
3618         if (vc->tb_offset_applied) {
3619                 u64 new_tb = mftb() - vc->tb_offset_applied;
3620                 mtspr(SPRN_TBU40, new_tb);
3621                 tb = mftb();
3622                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3623                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3624                 vc->tb_offset_applied = 0;
3625         }
3626
3627         mtspr(SPRN_HDEC, 0x7fffffff);
3628         mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3629
3630         return trap;
3631 }
3632
3633 /*
3634  * Virtual-mode guest entry for POWER9 and later when the host and
3635  * guest are both using the radix MMU.  The LPIDR has already been set.
3636  */
3637 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3638                          unsigned long lpcr)
3639 {
3640         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3641         unsigned long host_dscr = mfspr(SPRN_DSCR);
3642         unsigned long host_tidr = mfspr(SPRN_TIDR);
3643         unsigned long host_iamr = mfspr(SPRN_IAMR);
3644         unsigned long host_amr = mfspr(SPRN_AMR);
3645         unsigned long host_fscr = mfspr(SPRN_FSCR);
3646         s64 dec;
3647         u64 tb;
3648         int trap, save_pmu;
3649
3650         dec = mfspr(SPRN_DEC);
3651         tb = mftb();
3652         if (dec < 0)
3653                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3654         local_paca->kvm_hstate.dec_expires = dec + tb;
3655         if (local_paca->kvm_hstate.dec_expires < time_limit)
3656                 time_limit = local_paca->kvm_hstate.dec_expires;
3657
3658         vcpu->arch.ceded = 0;
3659
3660         kvmhv_save_host_pmu();          /* saves it to PACA kvm_hstate */
3661
3662         kvmppc_subcore_enter_guest();
3663
3664         vc->entry_exit_map = 1;
3665         vc->in_guest = 1;
3666
3667         if (vcpu->arch.vpa.pinned_addr) {
3668                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3669                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3670                 lp->yield_count = cpu_to_be32(yield_count);
3671                 vcpu->arch.vpa.dirty = 1;
3672         }
3673
3674         if (cpu_has_feature(CPU_FTR_TM) ||
3675             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3676                 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3677
3678         kvmhv_load_guest_pmu(vcpu);
3679
3680         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3681         load_fp_state(&vcpu->arch.fp);
3682 #ifdef CONFIG_ALTIVEC
3683         load_vr_state(&vcpu->arch.vr);
3684 #endif
3685         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3686
3687         mtspr(SPRN_DSCR, vcpu->arch.dscr);
3688         mtspr(SPRN_IAMR, vcpu->arch.iamr);
3689         mtspr(SPRN_PSPB, vcpu->arch.pspb);
3690         mtspr(SPRN_FSCR, vcpu->arch.fscr);
3691         mtspr(SPRN_TAR, vcpu->arch.tar);
3692         mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3693         mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3694         mtspr(SPRN_BESCR, vcpu->arch.bescr);
3695         mtspr(SPRN_WORT, vcpu->arch.wort);
3696         mtspr(SPRN_TIDR, vcpu->arch.tid);
3697         mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3698         mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3699         mtspr(SPRN_AMR, vcpu->arch.amr);
3700         mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3701
3702         if (!(vcpu->arch.ctrl & 1))
3703                 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3704
3705         mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3706
3707         if (kvmhv_on_pseries()) {
3708                 /*
3709                  * We need to save and restore the guest visible part of the
3710                  * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3711                  * doesn't do this for us. Note only required if pseries since
3712                  * this is done in kvmhv_load_hv_regs_and_go() below otherwise.
3713                  */
3714                 unsigned long host_psscr;
3715                 /* call our hypervisor to load up HV regs and go */
3716                 struct hv_guest_state hvregs;
3717
3718                 host_psscr = mfspr(SPRN_PSSCR_PR);
3719                 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3720                 kvmhv_save_hv_regs(vcpu, &hvregs);
3721                 hvregs.lpcr = lpcr;
3722                 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3723                 hvregs.version = HV_GUEST_STATE_VERSION;
3724                 if (vcpu->arch.nested) {
3725                         hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3726                         hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3727                 } else {
3728                         hvregs.lpid = vcpu->kvm->arch.lpid;
3729                         hvregs.vcpu_token = vcpu->vcpu_id;
3730                 }
3731                 hvregs.hdec_expiry = time_limit;
3732                 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3733                                           __pa(&vcpu->arch.regs));
3734                 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3735                 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3736                 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3737                 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3738                 vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3739                 mtspr(SPRN_PSSCR_PR, host_psscr);
3740
3741                 /* H_CEDE has to be handled now, not later */
3742                 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3743                     kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3744                         kvmppc_nested_cede(vcpu);
3745                         kvmppc_set_gpr(vcpu, 3, 0);
3746                         trap = 0;
3747                 }
3748         } else {
3749                 trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3750         }
3751
3752         vcpu->arch.slb_max = 0;
3753         dec = mfspr(SPRN_DEC);
3754         if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
3755                 dec = (s32) dec;
3756         tb = mftb();
3757         vcpu->arch.dec_expires = dec + tb;
3758         vcpu->cpu = -1;
3759         vcpu->arch.thread_cpu = -1;
3760         /* Save guest CTRL register, set runlatch to 1 */
3761         vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3762         if (!(vcpu->arch.ctrl & 1))
3763                 mtspr(SPRN_CTRLT, vcpu->arch.ctrl | 1);
3764
3765         vcpu->arch.iamr = mfspr(SPRN_IAMR);
3766         vcpu->arch.pspb = mfspr(SPRN_PSPB);
3767         vcpu->arch.fscr = mfspr(SPRN_FSCR);
3768         vcpu->arch.tar = mfspr(SPRN_TAR);
3769         vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3770         vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3771         vcpu->arch.bescr = mfspr(SPRN_BESCR);
3772         vcpu->arch.wort = mfspr(SPRN_WORT);
3773         vcpu->arch.tid = mfspr(SPRN_TIDR);
3774         vcpu->arch.amr = mfspr(SPRN_AMR);
3775         vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3776         vcpu->arch.dscr = mfspr(SPRN_DSCR);
3777
3778         mtspr(SPRN_PSPB, 0);
3779         mtspr(SPRN_WORT, 0);
3780         mtspr(SPRN_UAMOR, 0);
3781         mtspr(SPRN_DSCR, host_dscr);
3782         mtspr(SPRN_TIDR, host_tidr);
3783         mtspr(SPRN_IAMR, host_iamr);
3784         mtspr(SPRN_PSPB, 0);
3785
3786         if (host_amr != vcpu->arch.amr)
3787                 mtspr(SPRN_AMR, host_amr);
3788
3789         if (host_fscr != vcpu->arch.fscr)
3790                 mtspr(SPRN_FSCR, host_fscr);
3791
3792         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3793         store_fp_state(&vcpu->arch.fp);
3794 #ifdef CONFIG_ALTIVEC
3795         store_vr_state(&vcpu->arch.vr);
3796 #endif
3797         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3798
3799         if (cpu_has_feature(CPU_FTR_TM) ||
3800             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3801                 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3802
3803         save_pmu = 1;
3804         if (vcpu->arch.vpa.pinned_addr) {
3805                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3806                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3807                 lp->yield_count = cpu_to_be32(yield_count);
3808                 vcpu->arch.vpa.dirty = 1;
3809                 save_pmu = lp->pmcregs_in_use;
3810         }
3811         /* Must save pmu if this guest is capable of running nested guests */
3812         save_pmu |= nesting_enabled(vcpu->kvm);
3813
3814         kvmhv_save_guest_pmu(vcpu, save_pmu);
3815
3816         vc->entry_exit_map = 0x101;
3817         vc->in_guest = 0;
3818
3819         mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3820         mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3821
3822         kvmhv_load_host_pmu();
3823
3824         kvmppc_subcore_exit_guest();
3825
3826         return trap;
3827 }
3828
3829 /*
3830  * Wait for some other vcpu thread to execute us, and
3831  * wake us up when we need to handle something in the host.
3832  */
3833 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3834                                  struct kvm_vcpu *vcpu, int wait_state)
3835 {
3836         DEFINE_WAIT(wait);
3837
3838         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3839         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3840                 spin_unlock(&vc->lock);
3841                 schedule();
3842                 spin_lock(&vc->lock);
3843         }
3844         finish_wait(&vcpu->arch.cpu_run, &wait);
3845 }
3846
3847 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3848 {
3849         if (!halt_poll_ns_grow)
3850                 return;
3851
3852         vc->halt_poll_ns *= halt_poll_ns_grow;
3853         if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3854                 vc->halt_poll_ns = halt_poll_ns_grow_start;
3855 }
3856
3857 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3858 {
3859         if (halt_poll_ns_shrink == 0)
3860                 vc->halt_poll_ns = 0;
3861         else
3862                 vc->halt_poll_ns /= halt_poll_ns_shrink;
3863 }
3864
3865 #ifdef CONFIG_KVM_XICS
3866 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3867 {
3868         if (!xics_on_xive())
3869                 return false;
3870         return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3871                 vcpu->arch.xive_saved_state.cppr;
3872 }
3873 #else
3874 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3875 {
3876         return false;
3877 }
3878 #endif /* CONFIG_KVM_XICS */
3879
3880 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3881 {
3882         if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3883             kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3884                 return true;
3885
3886         return false;
3887 }
3888
3889 /*
3890  * Check to see if any of the runnable vcpus on the vcore have pending
3891  * exceptions or are no longer ceded
3892  */
3893 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3894 {
3895         struct kvm_vcpu *vcpu;
3896         int i;
3897
3898         for_each_runnable_thread(i, vcpu, vc) {
3899                 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3900                         return 1;
3901         }
3902
3903         return 0;
3904 }
3905
3906 /*
3907  * All the vcpus in this vcore are idle, so wait for a decrementer
3908  * or external interrupt to one of the vcpus.  vc->lock is held.
3909  */
3910 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3911 {
3912         ktime_t cur, start_poll, start_wait;
3913         int do_sleep = 1;
3914         u64 block_ns;
3915
3916         /* Poll for pending exceptions and ceded state */
3917         cur = start_poll = ktime_get();
3918         if (vc->halt_poll_ns) {
3919                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3920                 ++vc->runner->stat.halt_attempted_poll;
3921
3922                 vc->vcore_state = VCORE_POLLING;
3923                 spin_unlock(&vc->lock);
3924
3925                 do {
3926                         if (kvmppc_vcore_check_block(vc)) {
3927                                 do_sleep = 0;
3928                                 break;
3929                         }
3930                         cur = ktime_get();
3931                 } while (single_task_running() && ktime_before(cur, stop));
3932
3933                 spin_lock(&vc->lock);
3934                 vc->vcore_state = VCORE_INACTIVE;
3935
3936                 if (!do_sleep) {
3937                         ++vc->runner->stat.halt_successful_poll;
3938                         goto out;
3939                 }
3940         }
3941
3942         prepare_to_rcuwait(&vc->wait);
3943         set_current_state(TASK_INTERRUPTIBLE);
3944         if (kvmppc_vcore_check_block(vc)) {
3945                 finish_rcuwait(&vc->wait);
3946                 do_sleep = 0;
3947                 /* If we polled, count this as a successful poll */
3948                 if (vc->halt_poll_ns)
3949                         ++vc->runner->stat.halt_successful_poll;
3950                 goto out;
3951         }
3952
3953         start_wait = ktime_get();
3954
3955         vc->vcore_state = VCORE_SLEEPING;
3956         trace_kvmppc_vcore_blocked(vc, 0);
3957         spin_unlock(&vc->lock);
3958         schedule();
3959         finish_rcuwait(&vc->wait);
3960         spin_lock(&vc->lock);
3961         vc->vcore_state = VCORE_INACTIVE;
3962         trace_kvmppc_vcore_blocked(vc, 1);
3963         ++vc->runner->stat.halt_successful_wait;
3964
3965         cur = ktime_get();
3966
3967 out:
3968         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3969
3970         /* Attribute wait time */
3971         if (do_sleep) {
3972                 vc->runner->stat.halt_wait_ns +=
3973                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
3974                 /* Attribute failed poll time */
3975                 if (vc->halt_poll_ns)
3976                         vc->runner->stat.halt_poll_fail_ns +=
3977                                 ktime_to_ns(start_wait) -
3978                                 ktime_to_ns(start_poll);
3979         } else {
3980                 /* Attribute successful poll time */
3981                 if (vc->halt_poll_ns)
3982                         vc->runner->stat.halt_poll_success_ns +=
3983                                 ktime_to_ns(cur) -
3984                                 ktime_to_ns(start_poll);
3985         }
3986
3987         /* Adjust poll time */
3988         if (halt_poll_ns) {
3989                 if (block_ns <= vc->halt_poll_ns)
3990                         ;
3991                 /* We slept and blocked for longer than the max halt time */
3992                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3993                         shrink_halt_poll_ns(vc);
3994                 /* We slept and our poll time is too small */
3995                 else if (vc->halt_poll_ns < halt_poll_ns &&
3996                                 block_ns < halt_poll_ns)
3997                         grow_halt_poll_ns(vc);
3998                 if (vc->halt_poll_ns > halt_poll_ns)
3999                         vc->halt_poll_ns = halt_poll_ns;
4000         } else
4001                 vc->halt_poll_ns = 0;
4002
4003         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
4004 }
4005
4006 /*
4007  * This never fails for a radix guest, as none of the operations it does
4008  * for a radix guest can fail or have a way to report failure.
4009  * kvmhv_run_single_vcpu() relies on this fact.
4010  */
4011 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
4012 {
4013         int r = 0;
4014         struct kvm *kvm = vcpu->kvm;
4015
4016         mutex_lock(&kvm->arch.mmu_setup_lock);
4017         if (!kvm->arch.mmu_ready) {
4018                 if (!kvm_is_radix(kvm))
4019                         r = kvmppc_hv_setup_htab_rma(vcpu);
4020                 if (!r) {
4021                         if (cpu_has_feature(CPU_FTR_ARCH_300))
4022                                 kvmppc_setup_partition_table(kvm);
4023                         kvm->arch.mmu_ready = 1;
4024                 }
4025         }
4026         mutex_unlock(&kvm->arch.mmu_setup_lock);
4027         return r;
4028 }
4029
4030 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
4031 {
4032         struct kvm_run *run = vcpu->run;
4033         int n_ceded, i, r;
4034         struct kvmppc_vcore *vc;
4035         struct kvm_vcpu *v;
4036
4037         trace_kvmppc_run_vcpu_enter(vcpu);
4038
4039         run->exit_reason = 0;
4040         vcpu->arch.ret = RESUME_GUEST;
4041         vcpu->arch.trap = 0;
4042         kvmppc_update_vpas(vcpu);
4043
4044         /*
4045          * Synchronize with other threads in this virtual core
4046          */
4047         vc = vcpu->arch.vcore;
4048         spin_lock(&vc->lock);
4049         vcpu->arch.ceded = 0;
4050         vcpu->arch.run_task = current;
4051         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4052         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4053         vcpu->arch.busy_preempt = TB_NIL;
4054         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4055         ++vc->n_runnable;
4056
4057         /*
4058          * This happens the first time this is called for a vcpu.
4059          * If the vcore is already running, we may be able to start
4060          * this thread straight away and have it join in.
4061          */
4062         if (!signal_pending(current)) {
4063                 if ((vc->vcore_state == VCORE_PIGGYBACK ||
4064                      vc->vcore_state == VCORE_RUNNING) &&
4065                            !VCORE_IS_EXITING(vc)) {
4066                         kvmppc_create_dtl_entry(vcpu, vc);
4067                         kvmppc_start_thread(vcpu, vc);
4068                         trace_kvm_guest_enter(vcpu);
4069                 } else if (vc->vcore_state == VCORE_SLEEPING) {
4070                         rcuwait_wake_up(&vc->wait);
4071                 }
4072
4073         }
4074
4075         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4076                !signal_pending(current)) {
4077                 /* See if the MMU is ready to go */
4078                 if (!vcpu->kvm->arch.mmu_ready) {
4079                         spin_unlock(&vc->lock);
4080                         r = kvmhv_setup_mmu(vcpu);
4081                         spin_lock(&vc->lock);
4082                         if (r) {
4083                                 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4084                                 run->fail_entry.
4085                                         hardware_entry_failure_reason = 0;
4086                                 vcpu->arch.ret = r;
4087                                 break;
4088                         }
4089                 }
4090
4091                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4092                         kvmppc_vcore_end_preempt(vc);
4093
4094                 if (vc->vcore_state != VCORE_INACTIVE) {
4095                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4096                         continue;
4097                 }
4098                 for_each_runnable_thread(i, v, vc) {
4099                         kvmppc_core_prepare_to_enter(v);
4100                         if (signal_pending(v->arch.run_task)) {
4101                                 kvmppc_remove_runnable(vc, v);
4102                                 v->stat.signal_exits++;
4103                                 v->run->exit_reason = KVM_EXIT_INTR;
4104                                 v->arch.ret = -EINTR;
4105                                 wake_up(&v->arch.cpu_run);
4106                         }
4107                 }
4108                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4109                         break;
4110                 n_ceded = 0;
4111                 for_each_runnable_thread(i, v, vc) {
4112                         if (!kvmppc_vcpu_woken(v))
4113                                 n_ceded += v->arch.ceded;
4114                         else
4115                                 v->arch.ceded = 0;
4116                 }
4117                 vc->runner = vcpu;
4118                 if (n_ceded == vc->n_runnable) {
4119                         kvmppc_vcore_blocked(vc);
4120                 } else if (need_resched()) {
4121                         kvmppc_vcore_preempt(vc);
4122                         /* Let something else run */
4123                         cond_resched_lock(&vc->lock);
4124                         if (vc->vcore_state == VCORE_PREEMPT)
4125                                 kvmppc_vcore_end_preempt(vc);
4126                 } else {
4127                         kvmppc_run_core(vc);
4128                 }
4129                 vc->runner = NULL;
4130         }
4131
4132         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4133                (vc->vcore_state == VCORE_RUNNING ||
4134                 vc->vcore_state == VCORE_EXITING ||
4135                 vc->vcore_state == VCORE_PIGGYBACK))
4136                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4137
4138         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4139                 kvmppc_vcore_end_preempt(vc);
4140
4141         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4142                 kvmppc_remove_runnable(vc, vcpu);
4143                 vcpu->stat.signal_exits++;
4144                 run->exit_reason = KVM_EXIT_INTR;
4145                 vcpu->arch.ret = -EINTR;
4146         }
4147
4148         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4149                 /* Wake up some vcpu to run the core */
4150                 i = -1;
4151                 v = next_runnable_thread(vc, &i);
4152                 wake_up(&v->arch.cpu_run);
4153         }
4154
4155         trace_kvmppc_run_vcpu_exit(vcpu);
4156         spin_unlock(&vc->lock);
4157         return vcpu->arch.ret;
4158 }
4159
4160 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4161                           unsigned long lpcr)
4162 {
4163         struct kvm_run *run = vcpu->run;
4164         int trap, r, pcpu;
4165         int srcu_idx, lpid;
4166         struct kvmppc_vcore *vc;
4167         struct kvm *kvm = vcpu->kvm;
4168         struct kvm_nested_guest *nested = vcpu->arch.nested;
4169
4170         trace_kvmppc_run_vcpu_enter(vcpu);
4171
4172         run->exit_reason = 0;
4173         vcpu->arch.ret = RESUME_GUEST;
4174         vcpu->arch.trap = 0;
4175
4176         vc = vcpu->arch.vcore;
4177         vcpu->arch.ceded = 0;
4178         vcpu->arch.run_task = current;
4179         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4180         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4181         vcpu->arch.busy_preempt = TB_NIL;
4182         vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4183         vc->runnable_threads[0] = vcpu;
4184         vc->n_runnable = 1;
4185         vc->runner = vcpu;
4186
4187         /* See if the MMU is ready to go */
4188         if (!kvm->arch.mmu_ready)
4189                 kvmhv_setup_mmu(vcpu);
4190
4191         if (need_resched())
4192                 cond_resched();
4193
4194         kvmppc_update_vpas(vcpu);
4195
4196         init_vcore_to_run(vc);
4197         vc->preempt_tb = TB_NIL;
4198
4199         preempt_disable();
4200         pcpu = smp_processor_id();
4201         vc->pcpu = pcpu;
4202         kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4203
4204         local_irq_disable();
4205         hard_irq_disable();
4206         if (signal_pending(current))
4207                 goto sigpend;
4208         if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4209                 goto out;
4210
4211         if (!nested) {
4212                 kvmppc_core_prepare_to_enter(vcpu);
4213                 if (vcpu->arch.doorbell_request) {
4214                         vc->dpdes = 1;
4215                         smp_wmb();
4216                         vcpu->arch.doorbell_request = 0;
4217                 }
4218                 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4219                              &vcpu->arch.pending_exceptions))
4220                         lpcr |= LPCR_MER;
4221         } else if (vcpu->arch.pending_exceptions ||
4222                    vcpu->arch.doorbell_request ||
4223                    xive_interrupt_pending(vcpu)) {
4224                 vcpu->arch.ret = RESUME_HOST;
4225                 goto out;
4226         }
4227
4228         kvmppc_clear_host_core(pcpu);
4229
4230         local_paca->kvm_hstate.napping = 0;
4231         local_paca->kvm_hstate.kvm_split_mode = NULL;
4232         kvmppc_start_thread(vcpu, vc);
4233         kvmppc_create_dtl_entry(vcpu, vc);
4234         trace_kvm_guest_enter(vcpu);
4235
4236         vc->vcore_state = VCORE_RUNNING;
4237         trace_kvmppc_run_core(vc, 0);
4238
4239         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4240                 lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4241                 mtspr(SPRN_LPID, lpid);
4242                 isync();
4243                 kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4244         }
4245
4246         guest_enter_irqoff();
4247
4248         srcu_idx = srcu_read_lock(&kvm->srcu);
4249
4250         this_cpu_disable_ftrace();
4251
4252         /* Tell lockdep that we're about to enable interrupts */
4253         trace_hardirqs_on();
4254
4255         trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4256         vcpu->arch.trap = trap;
4257
4258         trace_hardirqs_off();
4259
4260         this_cpu_enable_ftrace();
4261
4262         srcu_read_unlock(&kvm->srcu, srcu_idx);
4263
4264         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4265                 mtspr(SPRN_LPID, kvm->arch.host_lpid);
4266                 isync();
4267         }
4268
4269         set_irq_happened(trap);
4270
4271         kvmppc_set_host_core(pcpu);
4272
4273         guest_exit_irqoff();
4274
4275         local_irq_enable();
4276
4277         cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4278
4279         preempt_enable();
4280
4281         /*
4282          * cancel pending decrementer exception if DEC is now positive, or if
4283          * entering a nested guest in which case the decrementer is now owned
4284          * by L2 and the L1 decrementer is provided in hdec_expires
4285          */
4286         if (kvmppc_core_pending_dec(vcpu) &&
4287                         ((get_tb() < vcpu->arch.dec_expires) ||
4288                          (trap == BOOK3S_INTERRUPT_SYSCALL &&
4289                           kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4290                 kvmppc_core_dequeue_dec(vcpu);
4291
4292         trace_kvm_guest_exit(vcpu);
4293         r = RESUME_GUEST;
4294         if (trap) {
4295                 if (!nested)
4296                         r = kvmppc_handle_exit_hv(vcpu, current);
4297                 else
4298                         r = kvmppc_handle_nested_exit(vcpu);
4299         }
4300         vcpu->arch.ret = r;
4301
4302         if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4303             !kvmppc_vcpu_woken(vcpu)) {
4304                 kvmppc_set_timer(vcpu);
4305                 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4306                         if (signal_pending(current)) {
4307                                 vcpu->stat.signal_exits++;
4308                                 run->exit_reason = KVM_EXIT_INTR;
4309                                 vcpu->arch.ret = -EINTR;
4310                                 break;
4311                         }
4312                         spin_lock(&vc->lock);
4313                         kvmppc_vcore_blocked(vc);
4314                         spin_unlock(&vc->lock);
4315                 }
4316         }
4317         vcpu->arch.ceded = 0;
4318
4319         vc->vcore_state = VCORE_INACTIVE;
4320         trace_kvmppc_run_core(vc, 1);
4321
4322  done:
4323         kvmppc_remove_runnable(vc, vcpu);
4324         trace_kvmppc_run_vcpu_exit(vcpu);
4325
4326         return vcpu->arch.ret;
4327
4328  sigpend:
4329         vcpu->stat.signal_exits++;
4330         run->exit_reason = KVM_EXIT_INTR;
4331         vcpu->arch.ret = -EINTR;
4332  out:
4333         local_irq_enable();
4334         preempt_enable();
4335         goto done;
4336 }
4337
4338 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
4339 {
4340         struct kvm_run *run = vcpu->run;
4341         int r;
4342         int srcu_idx;
4343         unsigned long ebb_regs[3] = {}; /* shut up GCC */
4344         unsigned long user_tar = 0;
4345         unsigned int user_vrsave;
4346         struct kvm *kvm;
4347
4348         if (!vcpu->arch.sane) {
4349                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4350                 return -EINVAL;
4351         }
4352
4353         /*
4354          * Don't allow entry with a suspended transaction, because
4355          * the guest entry/exit code will lose it.
4356          * If the guest has TM enabled, save away their TM-related SPRs
4357          * (they will get restored by the TM unavailable interrupt).
4358          */
4359 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4360         if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4361             (current->thread.regs->msr & MSR_TM)) {
4362                 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4363                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4364                         run->fail_entry.hardware_entry_failure_reason = 0;
4365                         return -EINVAL;
4366                 }
4367                 /* Enable TM so we can read the TM SPRs */
4368                 mtmsr(mfmsr() | MSR_TM);
4369                 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4370                 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4371                 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4372                 current->thread.regs->msr &= ~MSR_TM;
4373         }
4374 #endif
4375
4376         /*
4377          * Force online to 1 for the sake of old userspace which doesn't
4378          * set it.
4379          */
4380         if (!vcpu->arch.online) {
4381                 atomic_inc(&vcpu->arch.vcore->online_count);
4382                 vcpu->arch.online = 1;
4383         }
4384
4385         kvmppc_core_prepare_to_enter(vcpu);
4386
4387         /* No need to go into the guest when all we'll do is come back out */
4388         if (signal_pending(current)) {
4389                 run->exit_reason = KVM_EXIT_INTR;
4390                 return -EINTR;
4391         }
4392
4393         kvm = vcpu->kvm;
4394         atomic_inc(&kvm->arch.vcpus_running);
4395         /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4396         smp_mb();
4397
4398         flush_all_to_thread(current);
4399
4400         /* Save userspace EBB and other register values */
4401         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4402                 ebb_regs[0] = mfspr(SPRN_EBBHR);
4403                 ebb_regs[1] = mfspr(SPRN_EBBRR);
4404                 ebb_regs[2] = mfspr(SPRN_BESCR);
4405                 user_tar = mfspr(SPRN_TAR);
4406         }
4407         user_vrsave = mfspr(SPRN_VRSAVE);
4408
4409         vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4410         vcpu->arch.pgdir = kvm->mm->pgd;
4411         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4412
4413         do {
4414                 /*
4415                  * The TLB prefetch bug fixup is only in the kvmppc_run_vcpu
4416                  * path, which also handles hash and dependent threads mode.
4417                  */
4418                 if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4419                     !cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
4420                         r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
4421                                                   vcpu->arch.vcore->lpcr);
4422                 else
4423                         r = kvmppc_run_vcpu(vcpu);
4424
4425                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4426                     !(vcpu->arch.shregs.msr & MSR_PR)) {
4427                         trace_kvm_hcall_enter(vcpu);
4428                         r = kvmppc_pseries_do_hcall(vcpu);
4429                         trace_kvm_hcall_exit(vcpu, r);
4430                         kvmppc_core_prepare_to_enter(vcpu);
4431                 } else if (r == RESUME_PAGE_FAULT) {
4432                         srcu_idx = srcu_read_lock(&kvm->srcu);
4433                         r = kvmppc_book3s_hv_page_fault(vcpu,
4434                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4435                         srcu_read_unlock(&kvm->srcu, srcu_idx);
4436                 } else if (r == RESUME_PASSTHROUGH) {
4437                         if (WARN_ON(xics_on_xive()))
4438                                 r = H_SUCCESS;
4439                         else
4440                                 r = kvmppc_xics_rm_complete(vcpu, 0);
4441                 }
4442         } while (is_kvmppc_resume_guest(r));
4443
4444         /* Restore userspace EBB and other register values */
4445         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4446                 mtspr(SPRN_EBBHR, ebb_regs[0]);
4447                 mtspr(SPRN_EBBRR, ebb_regs[1]);
4448                 mtspr(SPRN_BESCR, ebb_regs[2]);
4449                 mtspr(SPRN_TAR, user_tar);
4450                 mtspr(SPRN_FSCR, current->thread.fscr);
4451         }
4452         mtspr(SPRN_VRSAVE, user_vrsave);
4453
4454         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4455         atomic_dec(&kvm->arch.vcpus_running);
4456         return r;
4457 }
4458
4459 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4460                                      int shift, int sllp)
4461 {
4462         (*sps)->page_shift = shift;
4463         (*sps)->slb_enc = sllp;
4464         (*sps)->enc[0].page_shift = shift;
4465         (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4466         /*
4467          * Add 16MB MPSS support (may get filtered out by userspace)
4468          */
4469         if (shift != 24) {
4470                 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4471                 if (penc != -1) {
4472                         (*sps)->enc[1].page_shift = 24;
4473                         (*sps)->enc[1].pte_enc = penc;
4474                 }
4475         }
4476         (*sps)++;
4477 }
4478
4479 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4480                                          struct kvm_ppc_smmu_info *info)
4481 {
4482         struct kvm_ppc_one_seg_page_size *sps;
4483
4484         /*
4485          * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4486          * POWER7 doesn't support keys for instruction accesses,
4487          * POWER8 and POWER9 do.
4488          */
4489         info->data_keys = 32;
4490         info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4491
4492         /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4493         info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4494         info->slb_size = 32;
4495
4496         /* We only support these sizes for now, and no muti-size segments */
4497         sps = &info->sps[0];
4498         kvmppc_add_seg_page_size(&sps, 12, 0);
4499         kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4500         kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4501
4502         /* If running as a nested hypervisor, we don't support HPT guests */
4503         if (kvmhv_on_pseries())
4504                 info->flags |= KVM_PPC_NO_HASH;
4505
4506         return 0;
4507 }
4508
4509 /*
4510  * Get (and clear) the dirty memory log for a memory slot.
4511  */
4512 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4513                                          struct kvm_dirty_log *log)
4514 {
4515         struct kvm_memslots *slots;
4516         struct kvm_memory_slot *memslot;
4517         int i, r;
4518         unsigned long n;
4519         unsigned long *buf, *p;
4520         struct kvm_vcpu *vcpu;
4521
4522         mutex_lock(&kvm->slots_lock);
4523
4524         r = -EINVAL;
4525         if (log->slot >= KVM_USER_MEM_SLOTS)
4526                 goto out;
4527
4528         slots = kvm_memslots(kvm);
4529         memslot = id_to_memslot(slots, log->slot);
4530         r = -ENOENT;
4531         if (!memslot || !memslot->dirty_bitmap)
4532                 goto out;
4533
4534         /*
4535          * Use second half of bitmap area because both HPT and radix
4536          * accumulate bits in the first half.
4537          */
4538         n = kvm_dirty_bitmap_bytes(memslot);
4539         buf = memslot->dirty_bitmap + n / sizeof(long);
4540         memset(buf, 0, n);
4541
4542         if (kvm_is_radix(kvm))
4543                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4544         else
4545                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4546         if (r)
4547                 goto out;
4548
4549         /*
4550          * We accumulate dirty bits in the first half of the
4551          * memslot's dirty_bitmap area, for when pages are paged
4552          * out or modified by the host directly.  Pick up these
4553          * bits and add them to the map.
4554          */
4555         p = memslot->dirty_bitmap;
4556         for (i = 0; i < n / sizeof(long); ++i)
4557                 buf[i] |= xchg(&p[i], 0);
4558
4559         /* Harvest dirty bits from VPA and DTL updates */
4560         /* Note: we never modify the SLB shadow buffer areas */
4561         kvm_for_each_vcpu(i, vcpu, kvm) {
4562                 spin_lock(&vcpu->arch.vpa_update_lock);
4563                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4564                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4565                 spin_unlock(&vcpu->arch.vpa_update_lock);
4566         }
4567
4568         r = -EFAULT;
4569         if (copy_to_user(log->dirty_bitmap, buf, n))
4570                 goto out;
4571
4572         r = 0;
4573 out:
4574         mutex_unlock(&kvm->slots_lock);
4575         return r;
4576 }
4577
4578 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
4579 {
4580         vfree(slot->arch.rmap);
4581         slot->arch.rmap = NULL;
4582 }
4583
4584 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4585                                         struct kvm_memory_slot *slot,
4586                                         const struct kvm_userspace_memory_region *mem,
4587                                         enum kvm_mr_change change)
4588 {
4589         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4590
4591         if (change == KVM_MR_CREATE) {
4592                 slot->arch.rmap = vzalloc(array_size(npages,
4593                                           sizeof(*slot->arch.rmap)));
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
4677                 spin_lock(&vc->lock);
4678                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4679                 verify_lpcr(kvm, vc->lpcr);
4680                 spin_unlock(&vc->lock);
4681                 if (++cores_done >= kvm->arch.online_vcores)
4682                         break;
4683         }
4684 }
4685
4686 void kvmppc_setup_partition_table(struct kvm *kvm)
4687 {
4688         unsigned long dw0, dw1;
4689
4690         if (!kvm_is_radix(kvm)) {
4691                 /* PS field - page size for VRMA */
4692                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4693                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4694                 /* HTABSIZE and HTABORG fields */
4695                 dw0 |= kvm->arch.sdr1;
4696
4697                 /* Second dword as set by userspace */
4698                 dw1 = kvm->arch.process_table;
4699         } else {
4700                 dw0 = PATB_HR | radix__get_tree_size() |
4701                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4702                 dw1 = PATB_GR | kvm->arch.process_table;
4703         }
4704         kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4705 }
4706
4707 /*
4708  * Set up HPT (hashed page table) and RMA (real-mode area).
4709  * Must be called with kvm->arch.mmu_setup_lock held.
4710  */
4711 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4712 {
4713         int err = 0;
4714         struct kvm *kvm = vcpu->kvm;
4715         unsigned long hva;
4716         struct kvm_memory_slot *memslot;
4717         struct vm_area_struct *vma;
4718         unsigned long lpcr = 0, senc;
4719         unsigned long psize, porder;
4720         int srcu_idx;
4721
4722         /* Allocate hashed page table (if not done already) and reset it */
4723         if (!kvm->arch.hpt.virt) {
4724                 int order = KVM_DEFAULT_HPT_ORDER;
4725                 struct kvm_hpt_info info;
4726
4727                 err = kvmppc_allocate_hpt(&info, order);
4728                 /* If we get here, it means userspace didn't specify a
4729                  * size explicitly.  So, try successively smaller
4730                  * sizes if the default failed. */
4731                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4732                         err  = kvmppc_allocate_hpt(&info, order);
4733
4734                 if (err < 0) {
4735                         pr_err("KVM: Couldn't alloc HPT\n");
4736                         goto out;
4737                 }
4738
4739                 kvmppc_set_hpt(kvm, &info);
4740         }
4741
4742         /* Look up the memslot for guest physical address 0 */
4743         srcu_idx = srcu_read_lock(&kvm->srcu);
4744         memslot = gfn_to_memslot(kvm, 0);
4745
4746         /* We must have some memory at 0 by now */
4747         err = -EINVAL;
4748         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4749                 goto out_srcu;
4750
4751         /* Look up the VMA for the start of this memory slot */
4752         hva = memslot->userspace_addr;
4753         mmap_read_lock(kvm->mm);
4754         vma = find_vma(kvm->mm, hva);
4755         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4756                 goto up_out;
4757
4758         psize = vma_kernel_pagesize(vma);
4759
4760         mmap_read_unlock(kvm->mm);
4761
4762         /* We can handle 4k, 64k or 16M pages in the VRMA */
4763         if (psize >= 0x1000000)
4764                 psize = 0x1000000;
4765         else if (psize >= 0x10000)
4766                 psize = 0x10000;
4767         else
4768                 psize = 0x1000;
4769         porder = __ilog2(psize);
4770
4771         senc = slb_pgsize_encoding(psize);
4772         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4773                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4774         /* Create HPTEs in the hash page table for the VRMA */
4775         kvmppc_map_vrma(vcpu, memslot, porder);
4776
4777         /* Update VRMASD field in the LPCR */
4778         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4779                 /* the -4 is to account for senc values starting at 0x10 */
4780                 lpcr = senc << (LPCR_VRMASD_SH - 4);
4781                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4782         }
4783
4784         /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4785         smp_wmb();
4786         err = 0;
4787  out_srcu:
4788         srcu_read_unlock(&kvm->srcu, srcu_idx);
4789  out:
4790         return err;
4791
4792  up_out:
4793         mmap_read_unlock(kvm->mm);
4794         goto out_srcu;
4795 }
4796
4797 /*
4798  * Must be called with kvm->arch.mmu_setup_lock held and
4799  * mmu_ready = 0 and no vcpus running.
4800  */
4801 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4802 {
4803         if (nesting_enabled(kvm))
4804                 kvmhv_release_all_nested(kvm);
4805         kvmppc_rmap_reset(kvm);
4806         kvm->arch.process_table = 0;
4807         /* Mutual exclusion with kvm_unmap_hva_range etc. */
4808         spin_lock(&kvm->mmu_lock);
4809         kvm->arch.radix = 0;
4810         spin_unlock(&kvm->mmu_lock);
4811         kvmppc_free_radix(kvm);
4812         kvmppc_update_lpcr(kvm, LPCR_VPM1,
4813                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4814         return 0;
4815 }
4816
4817 /*
4818  * Must be called with kvm->arch.mmu_setup_lock held and
4819  * mmu_ready = 0 and no vcpus running.
4820  */
4821 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4822 {
4823         int err;
4824
4825         err = kvmppc_init_vm_radix(kvm);
4826         if (err)
4827                 return err;
4828         kvmppc_rmap_reset(kvm);
4829         /* Mutual exclusion with kvm_unmap_hva_range etc. */
4830         spin_lock(&kvm->mmu_lock);
4831         kvm->arch.radix = 1;
4832         spin_unlock(&kvm->mmu_lock);
4833         kvmppc_free_hpt(&kvm->arch.hpt);
4834         kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4835                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4836         return 0;
4837 }
4838
4839 #ifdef CONFIG_KVM_XICS
4840 /*
4841  * Allocate a per-core structure for managing state about which cores are
4842  * running in the host versus the guest and for exchanging data between
4843  * real mode KVM and CPU running in the host.
4844  * This is only done for the first VM.
4845  * The allocated structure stays even if all VMs have stopped.
4846  * It is only freed when the kvm-hv module is unloaded.
4847  * It's OK for this routine to fail, we just don't support host
4848  * core operations like redirecting H_IPI wakeups.
4849  */
4850 void kvmppc_alloc_host_rm_ops(void)
4851 {
4852         struct kvmppc_host_rm_ops *ops;
4853         unsigned long l_ops;
4854         int cpu, core;
4855         int size;
4856
4857         /* Not the first time here ? */
4858         if (kvmppc_host_rm_ops_hv != NULL)
4859                 return;
4860
4861         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4862         if (!ops)
4863                 return;
4864
4865         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4866         ops->rm_core = kzalloc(size, GFP_KERNEL);
4867
4868         if (!ops->rm_core) {
4869                 kfree(ops);
4870                 return;
4871         }
4872
4873         cpus_read_lock();
4874
4875         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4876                 if (!cpu_online(cpu))
4877                         continue;
4878
4879                 core = cpu >> threads_shift;
4880                 ops->rm_core[core].rm_state.in_host = 1;
4881         }
4882
4883         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4884
4885         /*
4886          * Make the contents of the kvmppc_host_rm_ops structure visible
4887          * to other CPUs before we assign it to the global variable.
4888          * Do an atomic assignment (no locks used here), but if someone
4889          * beats us to it, just free our copy and return.
4890          */
4891         smp_wmb();
4892         l_ops = (unsigned long) ops;
4893
4894         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4895                 cpus_read_unlock();
4896                 kfree(ops->rm_core);
4897                 kfree(ops);
4898                 return;
4899         }
4900
4901         cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4902                                              "ppc/kvm_book3s:prepare",
4903                                              kvmppc_set_host_core,
4904                                              kvmppc_clear_host_core);
4905         cpus_read_unlock();
4906 }
4907
4908 void kvmppc_free_host_rm_ops(void)
4909 {
4910         if (kvmppc_host_rm_ops_hv) {
4911                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4912                 kfree(kvmppc_host_rm_ops_hv->rm_core);
4913                 kfree(kvmppc_host_rm_ops_hv);
4914                 kvmppc_host_rm_ops_hv = NULL;
4915         }
4916 }
4917 #endif
4918
4919 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4920 {
4921         unsigned long lpcr, lpid;
4922         char buf[32];
4923         int ret;
4924
4925         mutex_init(&kvm->arch.uvmem_lock);
4926         INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
4927         mutex_init(&kvm->arch.mmu_setup_lock);
4928
4929         /* Allocate the guest's logical partition ID */
4930
4931         lpid = kvmppc_alloc_lpid();
4932         if ((long)lpid < 0)
4933                 return -ENOMEM;
4934         kvm->arch.lpid = lpid;
4935
4936         kvmppc_alloc_host_rm_ops();
4937
4938         kvmhv_vm_nested_init(kvm);
4939
4940         /*
4941          * Since we don't flush the TLB when tearing down a VM,
4942          * and this lpid might have previously been used,
4943          * make sure we flush on each core before running the new VM.
4944          * On POWER9, the tlbie in mmu_partition_table_set_entry()
4945          * does this flush for us.
4946          */
4947         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4948                 cpumask_setall(&kvm->arch.need_tlb_flush);
4949
4950         /* Start out with the default set of hcalls enabled */
4951         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4952                sizeof(kvm->arch.enabled_hcalls));
4953
4954         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4955                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4956
4957         /* Init LPCR for virtual RMA mode */
4958         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4959                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
4960                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4961                 lpcr &= LPCR_PECE | LPCR_LPES;
4962         } else {
4963                 lpcr = 0;
4964         }
4965         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4966                 LPCR_VPM0 | LPCR_VPM1;
4967         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4968                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4969         /* On POWER8 turn on online bit to enable PURR/SPURR */
4970         if (cpu_has_feature(CPU_FTR_ARCH_207S))
4971                 lpcr |= LPCR_ONL;
4972         /*
4973          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4974          * Set HVICE bit to enable hypervisor virtualization interrupts.
4975          * Set HEIC to prevent OS interrupts to go to hypervisor (should
4976          * be unnecessary but better safe than sorry in case we re-enable
4977          * EE in HV mode with this LPCR still set)
4978          */
4979         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4980                 lpcr &= ~LPCR_VPM0;
4981                 lpcr |= LPCR_HVICE | LPCR_HEIC;
4982
4983                 /*
4984                  * If xive is enabled, we route 0x500 interrupts directly
4985                  * to the guest.
4986                  */
4987                 if (xics_on_xive())
4988                         lpcr |= LPCR_LPES;
4989         }
4990
4991         /*
4992          * If the host uses radix, the guest starts out as radix.
4993          */
4994         if (radix_enabled()) {
4995                 kvm->arch.radix = 1;
4996                 kvm->arch.mmu_ready = 1;
4997                 lpcr &= ~LPCR_VPM1;
4998                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4999                 ret = kvmppc_init_vm_radix(kvm);
5000                 if (ret) {
5001                         kvmppc_free_lpid(kvm->arch.lpid);
5002                         return ret;
5003                 }
5004                 kvmppc_setup_partition_table(kvm);
5005         }
5006
5007         verify_lpcr(kvm, lpcr);
5008         kvm->arch.lpcr = lpcr;
5009
5010         /* Initialization for future HPT resizes */
5011         kvm->arch.resize_hpt = NULL;
5012
5013         /*
5014          * Work out how many sets the TLB has, for the use of
5015          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
5016          */
5017         if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5018                 /*
5019                  * P10 will flush all the congruence class with a single tlbiel
5020                  */
5021                 kvm->arch.tlb_sets = 1;
5022         } else if (radix_enabled())
5023                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
5024         else if (cpu_has_feature(CPU_FTR_ARCH_300))
5025                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
5026         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
5027                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
5028         else
5029                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
5030
5031         /*
5032          * Track that we now have a HV mode VM active. This blocks secondary
5033          * CPU threads from coming online.
5034          * On POWER9, we only need to do this if the "indep_threads_mode"
5035          * module parameter has been set to N.
5036          */
5037         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5038                 if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
5039                         pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
5040                         kvm->arch.threads_indep = true;
5041                 } else {
5042                         kvm->arch.threads_indep = indep_threads_mode;
5043                 }
5044         }
5045         if (!kvm->arch.threads_indep)
5046                 kvm_hv_vm_activated();
5047
5048         /*
5049          * Initialize smt_mode depending on processor.
5050          * POWER8 and earlier have to use "strict" threading, where
5051          * all vCPUs in a vcore have to run on the same (sub)core,
5052          * whereas on POWER9 the threads can each run a different
5053          * guest.
5054          */
5055         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5056                 kvm->arch.smt_mode = threads_per_subcore;
5057         else
5058                 kvm->arch.smt_mode = 1;
5059         kvm->arch.emul_smt_mode = 1;
5060
5061         /*
5062          * Create a debugfs directory for the VM
5063          */
5064         snprintf(buf, sizeof(buf), "vm%d", current->pid);
5065         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
5066         kvmppc_mmu_debugfs_init(kvm);
5067         if (radix_enabled())
5068                 kvmhv_radix_debugfs_init(kvm);
5069
5070         return 0;
5071 }
5072
5073 static void kvmppc_free_vcores(struct kvm *kvm)
5074 {
5075         long int i;
5076
5077         for (i = 0; i < KVM_MAX_VCORES; ++i)
5078                 kfree(kvm->arch.vcores[i]);
5079         kvm->arch.online_vcores = 0;
5080 }
5081
5082 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5083 {
5084         debugfs_remove_recursive(kvm->arch.debugfs_dir);
5085
5086         if (!kvm->arch.threads_indep)
5087                 kvm_hv_vm_deactivated();
5088
5089         kvmppc_free_vcores(kvm);
5090
5091
5092         if (kvm_is_radix(kvm))
5093                 kvmppc_free_radix(kvm);
5094         else
5095                 kvmppc_free_hpt(&kvm->arch.hpt);
5096
5097         /* Perform global invalidation and return lpid to the pool */
5098         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5099                 if (nesting_enabled(kvm))
5100                         kvmhv_release_all_nested(kvm);
5101                 kvm->arch.process_table = 0;
5102                 if (kvm->arch.secure_guest)
5103                         uv_svm_terminate(kvm->arch.lpid);
5104                 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5105         }
5106
5107         kvmppc_free_lpid(kvm->arch.lpid);
5108
5109         kvmppc_free_pimap(kvm);
5110 }
5111
5112 /* We don't need to emulate any privileged instructions or dcbz */
5113 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5114                                      unsigned int inst, int *advance)
5115 {
5116         return EMULATE_FAIL;
5117 }
5118
5119 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5120                                         ulong spr_val)
5121 {
5122         return EMULATE_FAIL;
5123 }
5124
5125 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5126                                         ulong *spr_val)
5127 {
5128         return EMULATE_FAIL;
5129 }
5130
5131 static int kvmppc_core_check_processor_compat_hv(void)
5132 {
5133         if (cpu_has_feature(CPU_FTR_HVMODE) &&
5134             cpu_has_feature(CPU_FTR_ARCH_206))
5135                 return 0;
5136
5137         /* POWER9 in radix mode is capable of being a nested hypervisor. */
5138         if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5139                 return 0;
5140
5141         return -EIO;
5142 }
5143
5144 #ifdef CONFIG_KVM_XICS
5145
5146 void kvmppc_free_pimap(struct kvm *kvm)
5147 {
5148         kfree(kvm->arch.pimap);
5149 }
5150
5151 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5152 {
5153         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5154 }
5155
5156 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5157 {
5158         struct irq_desc *desc;
5159         struct kvmppc_irq_map *irq_map;
5160         struct kvmppc_passthru_irqmap *pimap;
5161         struct irq_chip *chip;
5162         int i, rc = 0;
5163
5164         if (!kvm_irq_bypass)
5165                 return 1;
5166
5167         desc = irq_to_desc(host_irq);
5168         if (!desc)
5169                 return -EIO;
5170
5171         mutex_lock(&kvm->lock);
5172
5173         pimap = kvm->arch.pimap;
5174         if (pimap == NULL) {
5175                 /* First call, allocate structure to hold IRQ map */
5176                 pimap = kvmppc_alloc_pimap();
5177                 if (pimap == NULL) {
5178                         mutex_unlock(&kvm->lock);
5179                         return -ENOMEM;
5180                 }
5181                 kvm->arch.pimap = pimap;
5182         }
5183
5184         /*
5185          * For now, we only support interrupts for which the EOI operation
5186          * is an OPAL call followed by a write to XIRR, since that's
5187          * what our real-mode EOI code does, or a XIVE interrupt
5188          */
5189         chip = irq_data_get_irq_chip(&desc->irq_data);
5190         if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5191                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5192                         host_irq, guest_gsi);
5193                 mutex_unlock(&kvm->lock);
5194                 return -ENOENT;
5195         }
5196
5197         /*
5198          * See if we already have an entry for this guest IRQ number.
5199          * If it's mapped to a hardware IRQ number, that's an error,
5200          * otherwise re-use this entry.
5201          */
5202         for (i = 0; i < pimap->n_mapped; i++) {
5203                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5204                         if (pimap->mapped[i].r_hwirq) {
5205                                 mutex_unlock(&kvm->lock);
5206                                 return -EINVAL;
5207                         }
5208                         break;
5209                 }
5210         }
5211
5212         if (i == KVMPPC_PIRQ_MAPPED) {
5213                 mutex_unlock(&kvm->lock);
5214                 return -EAGAIN;         /* table is full */
5215         }
5216
5217         irq_map = &pimap->mapped[i];
5218
5219         irq_map->v_hwirq = guest_gsi;
5220         irq_map->desc = desc;
5221
5222         /*
5223          * Order the above two stores before the next to serialize with
5224          * the KVM real mode handler.
5225          */
5226         smp_wmb();
5227         irq_map->r_hwirq = desc->irq_data.hwirq;
5228
5229         if (i == pimap->n_mapped)
5230                 pimap->n_mapped++;
5231
5232         if (xics_on_xive())
5233                 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5234         else
5235                 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5236         if (rc)
5237                 irq_map->r_hwirq = 0;
5238
5239         mutex_unlock(&kvm->lock);
5240
5241         return 0;
5242 }
5243
5244 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5245 {
5246         struct irq_desc *desc;
5247         struct kvmppc_passthru_irqmap *pimap;
5248         int i, rc = 0;
5249
5250         if (!kvm_irq_bypass)
5251                 return 0;
5252
5253         desc = irq_to_desc(host_irq);
5254         if (!desc)
5255                 return -EIO;
5256
5257         mutex_lock(&kvm->lock);
5258         if (!kvm->arch.pimap)
5259                 goto unlock;
5260
5261         pimap = kvm->arch.pimap;
5262
5263         for (i = 0; i < pimap->n_mapped; i++) {
5264                 if (guest_gsi == pimap->mapped[i].v_hwirq)
5265                         break;
5266         }
5267
5268         if (i == pimap->n_mapped) {
5269                 mutex_unlock(&kvm->lock);
5270                 return -ENODEV;
5271         }
5272
5273         if (xics_on_xive())
5274                 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5275         else
5276                 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5277
5278         /* invalidate the entry (what do do on error from the above ?) */
5279         pimap->mapped[i].r_hwirq = 0;
5280
5281         /*
5282          * We don't free this structure even when the count goes to
5283          * zero. The structure is freed when we destroy the VM.
5284          */
5285  unlock:
5286         mutex_unlock(&kvm->lock);
5287         return rc;
5288 }
5289
5290 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5291                                              struct irq_bypass_producer *prod)
5292 {
5293         int ret = 0;
5294         struct kvm_kernel_irqfd *irqfd =
5295                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5296
5297         irqfd->producer = prod;
5298
5299         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5300         if (ret)
5301                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5302                         prod->irq, irqfd->gsi, ret);
5303
5304         return ret;
5305 }
5306
5307 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5308                                               struct irq_bypass_producer *prod)
5309 {
5310         int ret;
5311         struct kvm_kernel_irqfd *irqfd =
5312                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5313
5314         irqfd->producer = NULL;
5315
5316         /*
5317          * When producer of consumer is unregistered, we change back to
5318          * default external interrupt handling mode - KVM real mode
5319          * will switch back to host.
5320          */
5321         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5322         if (ret)
5323                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5324                         prod->irq, irqfd->gsi, ret);
5325 }
5326 #endif
5327
5328 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5329                                  unsigned int ioctl, unsigned long arg)
5330 {
5331         struct kvm *kvm __maybe_unused = filp->private_data;
5332         void __user *argp = (void __user *)arg;
5333         long r;
5334
5335         switch (ioctl) {
5336
5337         case KVM_PPC_ALLOCATE_HTAB: {
5338                 u32 htab_order;
5339
5340                 /* If we're a nested hypervisor, we currently only support radix */
5341                 if (kvmhv_on_pseries()) {
5342                         r = -EOPNOTSUPP;
5343                         break;
5344                 }
5345
5346                 r = -EFAULT;
5347                 if (get_user(htab_order, (u32 __user *)argp))
5348                         break;
5349                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5350                 if (r)
5351                         break;
5352                 r = 0;
5353                 break;
5354         }
5355
5356         case KVM_PPC_GET_HTAB_FD: {
5357                 struct kvm_get_htab_fd ghf;
5358
5359                 r = -EFAULT;
5360                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5361                         break;
5362                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5363                 break;
5364         }
5365
5366         case KVM_PPC_RESIZE_HPT_PREPARE: {
5367                 struct kvm_ppc_resize_hpt rhpt;
5368
5369                 r = -EFAULT;
5370                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5371                         break;
5372
5373                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5374                 break;
5375         }
5376
5377         case KVM_PPC_RESIZE_HPT_COMMIT: {
5378                 struct kvm_ppc_resize_hpt rhpt;
5379
5380                 r = -EFAULT;
5381                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5382                         break;
5383
5384                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5385                 break;
5386         }
5387
5388         default:
5389                 r = -ENOTTY;
5390         }
5391
5392         return r;
5393 }
5394
5395 /*
5396  * List of hcall numbers to enable by default.
5397  * For compatibility with old userspace, we enable by default
5398  * all hcalls that were implemented before the hcall-enabling
5399  * facility was added.  Note this list should not include H_RTAS.
5400  */
5401 static unsigned int default_hcall_list[] = {
5402         H_REMOVE,
5403         H_ENTER,
5404         H_READ,
5405         H_PROTECT,
5406         H_BULK_REMOVE,
5407         H_GET_TCE,
5408         H_PUT_TCE,
5409         H_SET_DABR,
5410         H_SET_XDABR,
5411         H_CEDE,
5412         H_PROD,
5413         H_CONFER,
5414         H_REGISTER_VPA,
5415 #ifdef CONFIG_KVM_XICS
5416         H_EOI,
5417         H_CPPR,
5418         H_IPI,
5419         H_IPOLL,
5420         H_XIRR,
5421         H_XIRR_X,
5422 #endif
5423         0
5424 };
5425
5426 static void init_default_hcalls(void)
5427 {
5428         int i;
5429         unsigned int hcall;
5430
5431         for (i = 0; default_hcall_list[i]; ++i) {
5432                 hcall = default_hcall_list[i];
5433                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5434                 __set_bit(hcall / 4, default_enabled_hcalls);
5435         }
5436 }
5437
5438 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5439 {
5440         unsigned long lpcr;
5441         int radix;
5442         int err;
5443
5444         /* If not on a POWER9, reject it */
5445         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5446                 return -ENODEV;
5447
5448         /* If any unknown flags set, reject it */
5449         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5450                 return -EINVAL;
5451
5452         /* GR (guest radix) bit in process_table field must match */
5453         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5454         if (!!(cfg->process_table & PATB_GR) != radix)
5455                 return -EINVAL;
5456
5457         /* Process table size field must be reasonable, i.e. <= 24 */
5458         if ((cfg->process_table & PRTS_MASK) > 24)
5459                 return -EINVAL;
5460
5461         /* We can change a guest to/from radix now, if the host is radix */
5462         if (radix && !radix_enabled())
5463                 return -EINVAL;
5464
5465         /* If we're a nested hypervisor, we currently only support radix */
5466         if (kvmhv_on_pseries() && !radix)
5467                 return -EINVAL;
5468
5469         mutex_lock(&kvm->arch.mmu_setup_lock);
5470         if (radix != kvm_is_radix(kvm)) {
5471                 if (kvm->arch.mmu_ready) {
5472                         kvm->arch.mmu_ready = 0;
5473                         /* order mmu_ready vs. vcpus_running */
5474                         smp_mb();
5475                         if (atomic_read(&kvm->arch.vcpus_running)) {
5476                                 kvm->arch.mmu_ready = 1;
5477                                 err = -EBUSY;
5478                                 goto out_unlock;
5479                         }
5480                 }
5481                 if (radix)
5482                         err = kvmppc_switch_mmu_to_radix(kvm);
5483                 else
5484                         err = kvmppc_switch_mmu_to_hpt(kvm);
5485                 if (err)
5486                         goto out_unlock;
5487         }
5488
5489         kvm->arch.process_table = cfg->process_table;
5490         kvmppc_setup_partition_table(kvm);
5491
5492         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5493         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5494         err = 0;
5495
5496  out_unlock:
5497         mutex_unlock(&kvm->arch.mmu_setup_lock);
5498         return err;
5499 }
5500
5501 static int kvmhv_enable_nested(struct kvm *kvm)
5502 {
5503         if (!nested)
5504                 return -EPERM;
5505         if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5506                 return -ENODEV;
5507
5508         /* kvm == NULL means the caller is testing if the capability exists */
5509         if (kvm)
5510                 kvm->arch.nested_enable = true;
5511         return 0;
5512 }
5513
5514 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5515                                  int size)
5516 {
5517         int rc = -EINVAL;
5518
5519         if (kvmhv_vcpu_is_radix(vcpu)) {
5520                 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5521
5522                 if (rc > 0)
5523                         rc = -EINVAL;
5524         }
5525
5526         /* For now quadrants are the only way to access nested guest memory */
5527         if (rc && vcpu->arch.nested)
5528                 rc = -EAGAIN;
5529
5530         return rc;
5531 }
5532
5533 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5534                                 int size)
5535 {
5536         int rc = -EINVAL;
5537
5538         if (kvmhv_vcpu_is_radix(vcpu)) {
5539                 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5540
5541                 if (rc > 0)
5542                         rc = -EINVAL;
5543         }
5544
5545         /* For now quadrants are the only way to access nested guest memory */
5546         if (rc && vcpu->arch.nested)
5547                 rc = -EAGAIN;
5548
5549         return rc;
5550 }
5551
5552 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5553 {
5554         unpin_vpa(kvm, vpa);
5555         vpa->gpa = 0;
5556         vpa->pinned_addr = NULL;
5557         vpa->dirty = false;
5558         vpa->update_pending = 0;
5559 }
5560
5561 /*
5562  * Enable a guest to become a secure VM, or test whether
5563  * that could be enabled.
5564  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
5565  * tested (kvm == NULL) or enabled (kvm != NULL).
5566  */
5567 static int kvmhv_enable_svm(struct kvm *kvm)
5568 {
5569         if (!kvmppc_uvmem_available())
5570                 return -EINVAL;
5571         if (kvm)
5572                 kvm->arch.svm_enabled = 1;
5573         return 0;
5574 }
5575
5576 /*
5577  *  IOCTL handler to turn off secure mode of guest
5578  *
5579  * - Release all device pages
5580  * - Issue ucall to terminate the guest on the UV side
5581  * - Unpin the VPA pages.
5582  * - Reinit the partition scoped page tables
5583  */
5584 static int kvmhv_svm_off(struct kvm *kvm)
5585 {
5586         struct kvm_vcpu *vcpu;
5587         int mmu_was_ready;
5588         int srcu_idx;
5589         int ret = 0;
5590         int i;
5591
5592         if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5593                 return ret;
5594
5595         mutex_lock(&kvm->arch.mmu_setup_lock);
5596         mmu_was_ready = kvm->arch.mmu_ready;
5597         if (kvm->arch.mmu_ready) {
5598                 kvm->arch.mmu_ready = 0;
5599                 /* order mmu_ready vs. vcpus_running */
5600                 smp_mb();
5601                 if (atomic_read(&kvm->arch.vcpus_running)) {
5602                         kvm->arch.mmu_ready = 1;
5603                         ret = -EBUSY;
5604                         goto out;
5605                 }
5606         }
5607
5608         srcu_idx = srcu_read_lock(&kvm->srcu);
5609         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5610                 struct kvm_memory_slot *memslot;
5611                 struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5612
5613                 if (!slots)
5614                         continue;
5615
5616                 kvm_for_each_memslot(memslot, slots) {
5617                         kvmppc_uvmem_drop_pages(memslot, kvm, true);
5618                         uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5619                 }
5620         }
5621         srcu_read_unlock(&kvm->srcu, srcu_idx);
5622
5623         ret = uv_svm_terminate(kvm->arch.lpid);
5624         if (ret != U_SUCCESS) {
5625                 ret = -EINVAL;
5626                 goto out;
5627         }
5628
5629         /*
5630          * When secure guest is reset, all the guest pages are sent
5631          * to UV via UV_PAGE_IN before the non-boot vcpus get a
5632          * chance to run and unpin their VPA pages. Unpinning of all
5633          * VPA pages is done here explicitly so that VPA pages
5634          * can be migrated to the secure side.
5635          *
5636          * This is required to for the secure SMP guest to reboot
5637          * correctly.
5638          */
5639         kvm_for_each_vcpu(i, vcpu, kvm) {
5640                 spin_lock(&vcpu->arch.vpa_update_lock);
5641                 unpin_vpa_reset(kvm, &vcpu->arch.dtl);
5642                 unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
5643                 unpin_vpa_reset(kvm, &vcpu->arch.vpa);
5644                 spin_unlock(&vcpu->arch.vpa_update_lock);
5645         }
5646
5647         kvmppc_setup_partition_table(kvm);
5648         kvm->arch.secure_guest = 0;
5649         kvm->arch.mmu_ready = mmu_was_ready;
5650 out:
5651         mutex_unlock(&kvm->arch.mmu_setup_lock);
5652         return ret;
5653 }
5654
5655 static int kvmhv_enable_dawr1(struct kvm *kvm)
5656 {
5657         if (!cpu_has_feature(CPU_FTR_DAWR1))
5658                 return -ENODEV;
5659
5660         /* kvm == NULL means the caller is testing if the capability exists */
5661         if (kvm)
5662                 kvm->arch.dawr1_enabled = true;
5663         return 0;
5664 }
5665
5666 static bool kvmppc_hash_v3_possible(void)
5667 {
5668         if (radix_enabled() && no_mixing_hpt_and_radix)
5669                 return false;
5670
5671         return cpu_has_feature(CPU_FTR_ARCH_300) &&
5672                 cpu_has_feature(CPU_FTR_HVMODE);
5673 }
5674
5675 static struct kvmppc_ops kvm_ops_hv = {
5676         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5677         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5678         .get_one_reg = kvmppc_get_one_reg_hv,
5679         .set_one_reg = kvmppc_set_one_reg_hv,
5680         .vcpu_load   = kvmppc_core_vcpu_load_hv,
5681         .vcpu_put    = kvmppc_core_vcpu_put_hv,
5682         .inject_interrupt = kvmppc_inject_interrupt_hv,
5683         .set_msr     = kvmppc_set_msr_hv,
5684         .vcpu_run    = kvmppc_vcpu_run_hv,
5685         .vcpu_create = kvmppc_core_vcpu_create_hv,
5686         .vcpu_free   = kvmppc_core_vcpu_free_hv,
5687         .check_requests = kvmppc_core_check_requests_hv,
5688         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5689         .flush_memslot  = kvmppc_core_flush_memslot_hv,
5690         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5691         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5692         .unmap_hva_range = kvm_unmap_hva_range_hv,
5693         .age_hva  = kvm_age_hva_hv,
5694         .test_age_hva = kvm_test_age_hva_hv,
5695         .set_spte_hva = kvm_set_spte_hva_hv,
5696         .free_memslot = kvmppc_core_free_memslot_hv,
5697         .init_vm =  kvmppc_core_init_vm_hv,
5698         .destroy_vm = kvmppc_core_destroy_vm_hv,
5699         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5700         .emulate_op = kvmppc_core_emulate_op_hv,
5701         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5702         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5703         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5704         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5705         .hcall_implemented = kvmppc_hcall_impl_hv,
5706 #ifdef CONFIG_KVM_XICS
5707         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5708         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5709 #endif
5710         .configure_mmu = kvmhv_configure_mmu,
5711         .get_rmmu_info = kvmhv_get_rmmu_info,
5712         .set_smt_mode = kvmhv_set_smt_mode,
5713         .enable_nested = kvmhv_enable_nested,
5714         .load_from_eaddr = kvmhv_load_from_eaddr,
5715         .store_to_eaddr = kvmhv_store_to_eaddr,
5716         .enable_svm = kvmhv_enable_svm,
5717         .svm_off = kvmhv_svm_off,
5718         .enable_dawr1 = kvmhv_enable_dawr1,
5719         .hash_v3_possible = kvmppc_hash_v3_possible,
5720 };
5721
5722 static int kvm_init_subcore_bitmap(void)
5723 {
5724         int i, j;
5725         int nr_cores = cpu_nr_cores();
5726         struct sibling_subcore_state *sibling_subcore_state;
5727
5728         for (i = 0; i < nr_cores; i++) {
5729                 int first_cpu = i * threads_per_core;
5730                 int node = cpu_to_node(first_cpu);
5731
5732                 /* Ignore if it is already allocated. */
5733                 if (paca_ptrs[first_cpu]->sibling_subcore_state)
5734                         continue;
5735
5736                 sibling_subcore_state =
5737                         kzalloc_node(sizeof(struct sibling_subcore_state),
5738                                                         GFP_KERNEL, node);
5739                 if (!sibling_subcore_state)
5740                         return -ENOMEM;
5741
5742
5743                 for (j = 0; j < threads_per_core; j++) {
5744                         int cpu = first_cpu + j;
5745
5746                         paca_ptrs[cpu]->sibling_subcore_state =
5747                                                 sibling_subcore_state;
5748                 }
5749         }
5750         return 0;
5751 }
5752
5753 static int kvmppc_radix_possible(void)
5754 {
5755         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5756 }
5757
5758 static int kvmppc_book3s_init_hv(void)
5759 {
5760         int r;
5761
5762         if (!tlbie_capable) {
5763                 pr_err("KVM-HV: Host does not support TLBIE\n");
5764                 return -ENODEV;
5765         }
5766
5767         /*
5768          * FIXME!! Do we need to check on all cpus ?
5769          */
5770         r = kvmppc_core_check_processor_compat_hv();
5771         if (r < 0)
5772                 return -ENODEV;
5773
5774         r = kvmhv_nested_init();
5775         if (r)
5776                 return r;
5777
5778         r = kvm_init_subcore_bitmap();
5779         if (r)
5780                 return r;
5781
5782         /*
5783          * We need a way of accessing the XICS interrupt controller,
5784          * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5785          * indirectly, via OPAL.
5786          */
5787 #ifdef CONFIG_SMP
5788         if (!xics_on_xive() && !kvmhv_on_pseries() &&
5789             !local_paca->kvm_hstate.xics_phys) {
5790                 struct device_node *np;
5791
5792                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5793                 if (!np) {
5794                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5795                         return -ENODEV;
5796                 }
5797                 /* presence of intc confirmed - node can be dropped again */
5798                 of_node_put(np);
5799         }
5800 #endif
5801
5802         kvm_ops_hv.owner = THIS_MODULE;
5803         kvmppc_hv_ops = &kvm_ops_hv;
5804
5805         init_default_hcalls();
5806
5807         init_vcore_lists();
5808
5809         r = kvmppc_mmu_hv_init();
5810         if (r)
5811                 return r;
5812
5813         if (kvmppc_radix_possible())
5814                 r = kvmppc_radix_init();
5815
5816         /*
5817          * POWER9 chips before version 2.02 can't have some threads in
5818          * HPT mode and some in radix mode on the same core.
5819          */
5820         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5821                 unsigned int pvr = mfspr(SPRN_PVR);
5822                 if ((pvr >> 16) == PVR_POWER9 &&
5823                     (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5824                      ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5825                         no_mixing_hpt_and_radix = true;
5826         }
5827
5828         r = kvmppc_uvmem_init();
5829         if (r < 0)
5830                 pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
5831
5832         return r;
5833 }
5834
5835 static void kvmppc_book3s_exit_hv(void)
5836 {
5837         kvmppc_uvmem_free();
5838         kvmppc_free_host_rm_ops();
5839         if (kvmppc_radix_possible())
5840                 kvmppc_radix_exit();
5841         kvmppc_hv_ops = NULL;
5842         kvmhv_nested_exit();
5843 }
5844
5845 module_init(kvmppc_book3s_init_hv);
5846 module_exit(kvmppc_book3s_exit_hv);
5847 MODULE_LICENSE("GPL");
5848 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5849 MODULE_ALIAS("devname:kvm");