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 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1639                 bool preserve_top32)
1640 {
1641         struct kvm *kvm = vcpu->kvm;
1642         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1643         u64 mask;
1644
1645         spin_lock(&vc->lock);
1646         /*
1647          * If ILE (interrupt little-endian) has changed, update the
1648          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1649          */
1650         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1651                 struct kvm_vcpu *vcpu;
1652                 int i;
1653
1654                 kvm_for_each_vcpu(i, vcpu, kvm) {
1655                         if (vcpu->arch.vcore != vc)
1656                                 continue;
1657                         if (new_lpcr & LPCR_ILE)
1658                                 vcpu->arch.intr_msr |= MSR_LE;
1659                         else
1660                                 vcpu->arch.intr_msr &= ~MSR_LE;
1661                 }
1662         }
1663
1664         /*
1665          * Userspace can only modify DPFD (default prefetch depth),
1666          * ILE (interrupt little-endian) and TC (translation control).
1667          * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1668          */
1669         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1670         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1671                 mask |= LPCR_AIL;
1672         /*
1673          * On POWER9, allow userspace to enable large decrementer for the
1674          * guest, whether or not the host has it enabled.
1675          */
1676         if (cpu_has_feature(CPU_FTR_ARCH_300))
1677                 mask |= LPCR_LD;
1678
1679         /* Broken 32-bit version of LPCR must not clear top bits */
1680         if (preserve_top32)
1681                 mask &= 0xFFFFFFFF;
1682         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1683         spin_unlock(&vc->lock);
1684 }
1685
1686 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1687                                  union kvmppc_one_reg *val)
1688 {
1689         int r = 0;
1690         long int i;
1691
1692         switch (id) {
1693         case KVM_REG_PPC_DEBUG_INST:
1694                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1695                 break;
1696         case KVM_REG_PPC_HIOR:
1697                 *val = get_reg_val(id, 0);
1698                 break;
1699         case KVM_REG_PPC_DABR:
1700                 *val = get_reg_val(id, vcpu->arch.dabr);
1701                 break;
1702         case KVM_REG_PPC_DABRX:
1703                 *val = get_reg_val(id, vcpu->arch.dabrx);
1704                 break;
1705         case KVM_REG_PPC_DSCR:
1706                 *val = get_reg_val(id, vcpu->arch.dscr);
1707                 break;
1708         case KVM_REG_PPC_PURR:
1709                 *val = get_reg_val(id, vcpu->arch.purr);
1710                 break;
1711         case KVM_REG_PPC_SPURR:
1712                 *val = get_reg_val(id, vcpu->arch.spurr);
1713                 break;
1714         case KVM_REG_PPC_AMR:
1715                 *val = get_reg_val(id, vcpu->arch.amr);
1716                 break;
1717         case KVM_REG_PPC_UAMOR:
1718                 *val = get_reg_val(id, vcpu->arch.uamor);
1719                 break;
1720         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
1721                 i = id - KVM_REG_PPC_MMCR0;
1722                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1723                 break;
1724         case KVM_REG_PPC_MMCR2:
1725                 *val = get_reg_val(id, vcpu->arch.mmcr[2]);
1726                 break;
1727         case KVM_REG_PPC_MMCRA:
1728                 *val = get_reg_val(id, vcpu->arch.mmcra);
1729                 break;
1730         case KVM_REG_PPC_MMCRS:
1731                 *val = get_reg_val(id, vcpu->arch.mmcrs);
1732                 break;
1733         case KVM_REG_PPC_MMCR3:
1734                 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
1735                 break;
1736         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1737                 i = id - KVM_REG_PPC_PMC1;
1738                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1739                 break;
1740         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1741                 i = id - KVM_REG_PPC_SPMC1;
1742                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1743                 break;
1744         case KVM_REG_PPC_SIAR:
1745                 *val = get_reg_val(id, vcpu->arch.siar);
1746                 break;
1747         case KVM_REG_PPC_SDAR:
1748                 *val = get_reg_val(id, vcpu->arch.sdar);
1749                 break;
1750         case KVM_REG_PPC_SIER:
1751                 *val = get_reg_val(id, vcpu->arch.sier[0]);
1752                 break;
1753         case KVM_REG_PPC_SIER2:
1754                 *val = get_reg_val(id, vcpu->arch.sier[1]);
1755                 break;
1756         case KVM_REG_PPC_SIER3:
1757                 *val = get_reg_val(id, vcpu->arch.sier[2]);
1758                 break;
1759         case KVM_REG_PPC_IAMR:
1760                 *val = get_reg_val(id, vcpu->arch.iamr);
1761                 break;
1762         case KVM_REG_PPC_PSPB:
1763                 *val = get_reg_val(id, vcpu->arch.pspb);
1764                 break;
1765         case KVM_REG_PPC_DPDES:
1766                 /*
1767                  * On POWER9, where we are emulating msgsndp etc.,
1768                  * we return 1 bit for each vcpu, which can come from
1769                  * either vcore->dpdes or doorbell_request.
1770                  * On POWER8, doorbell_request is 0.
1771                  */
1772                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes |
1773                                    vcpu->arch.doorbell_request);
1774                 break;
1775         case KVM_REG_PPC_VTB:
1776                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1777                 break;
1778         case KVM_REG_PPC_DAWR:
1779                 *val = get_reg_val(id, vcpu->arch.dawr0);
1780                 break;
1781         case KVM_REG_PPC_DAWRX:
1782                 *val = get_reg_val(id, vcpu->arch.dawrx0);
1783                 break;
1784         case KVM_REG_PPC_DAWR1:
1785                 *val = get_reg_val(id, vcpu->arch.dawr1);
1786                 break;
1787         case KVM_REG_PPC_DAWRX1:
1788                 *val = get_reg_val(id, vcpu->arch.dawrx1);
1789                 break;
1790         case KVM_REG_PPC_CIABR:
1791                 *val = get_reg_val(id, vcpu->arch.ciabr);
1792                 break;
1793         case KVM_REG_PPC_CSIGR:
1794                 *val = get_reg_val(id, vcpu->arch.csigr);
1795                 break;
1796         case KVM_REG_PPC_TACR:
1797                 *val = get_reg_val(id, vcpu->arch.tacr);
1798                 break;
1799         case KVM_REG_PPC_TCSCR:
1800                 *val = get_reg_val(id, vcpu->arch.tcscr);
1801                 break;
1802         case KVM_REG_PPC_PID:
1803                 *val = get_reg_val(id, vcpu->arch.pid);
1804                 break;
1805         case KVM_REG_PPC_ACOP:
1806                 *val = get_reg_val(id, vcpu->arch.acop);
1807                 break;
1808         case KVM_REG_PPC_WORT:
1809                 *val = get_reg_val(id, vcpu->arch.wort);
1810                 break;
1811         case KVM_REG_PPC_TIDR:
1812                 *val = get_reg_val(id, vcpu->arch.tid);
1813                 break;
1814         case KVM_REG_PPC_PSSCR:
1815                 *val = get_reg_val(id, vcpu->arch.psscr);
1816                 break;
1817         case KVM_REG_PPC_VPA_ADDR:
1818                 spin_lock(&vcpu->arch.vpa_update_lock);
1819                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1820                 spin_unlock(&vcpu->arch.vpa_update_lock);
1821                 break;
1822         case KVM_REG_PPC_VPA_SLB:
1823                 spin_lock(&vcpu->arch.vpa_update_lock);
1824                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1825                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1826                 spin_unlock(&vcpu->arch.vpa_update_lock);
1827                 break;
1828         case KVM_REG_PPC_VPA_DTL:
1829                 spin_lock(&vcpu->arch.vpa_update_lock);
1830                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1831                 val->vpaval.length = vcpu->arch.dtl.len;
1832                 spin_unlock(&vcpu->arch.vpa_update_lock);
1833                 break;
1834         case KVM_REG_PPC_TB_OFFSET:
1835                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1836                 break;
1837         case KVM_REG_PPC_LPCR:
1838         case KVM_REG_PPC_LPCR_64:
1839                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1840                 break;
1841         case KVM_REG_PPC_PPR:
1842                 *val = get_reg_val(id, vcpu->arch.ppr);
1843                 break;
1844 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1845         case KVM_REG_PPC_TFHAR:
1846                 *val = get_reg_val(id, vcpu->arch.tfhar);
1847                 break;
1848         case KVM_REG_PPC_TFIAR:
1849                 *val = get_reg_val(id, vcpu->arch.tfiar);
1850                 break;
1851         case KVM_REG_PPC_TEXASR:
1852                 *val = get_reg_val(id, vcpu->arch.texasr);
1853                 break;
1854         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1855                 i = id - KVM_REG_PPC_TM_GPR0;
1856                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1857                 break;
1858         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1859         {
1860                 int j;
1861                 i = id - KVM_REG_PPC_TM_VSR0;
1862                 if (i < 32)
1863                         for (j = 0; j < TS_FPRWIDTH; j++)
1864                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1865                 else {
1866                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1867                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1868                         else
1869                                 r = -ENXIO;
1870                 }
1871                 break;
1872         }
1873         case KVM_REG_PPC_TM_CR:
1874                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1875                 break;
1876         case KVM_REG_PPC_TM_XER:
1877                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1878                 break;
1879         case KVM_REG_PPC_TM_LR:
1880                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1881                 break;
1882         case KVM_REG_PPC_TM_CTR:
1883                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1884                 break;
1885         case KVM_REG_PPC_TM_FPSCR:
1886                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1887                 break;
1888         case KVM_REG_PPC_TM_AMR:
1889                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1890                 break;
1891         case KVM_REG_PPC_TM_PPR:
1892                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1893                 break;
1894         case KVM_REG_PPC_TM_VRSAVE:
1895                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1896                 break;
1897         case KVM_REG_PPC_TM_VSCR:
1898                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1899                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1900                 else
1901                         r = -ENXIO;
1902                 break;
1903         case KVM_REG_PPC_TM_DSCR:
1904                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1905                 break;
1906         case KVM_REG_PPC_TM_TAR:
1907                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1908                 break;
1909 #endif
1910         case KVM_REG_PPC_ARCH_COMPAT:
1911                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1912                 break;
1913         case KVM_REG_PPC_DEC_EXPIRY:
1914                 *val = get_reg_val(id, vcpu->arch.dec_expires +
1915                                    vcpu->arch.vcore->tb_offset);
1916                 break;
1917         case KVM_REG_PPC_ONLINE:
1918                 *val = get_reg_val(id, vcpu->arch.online);
1919                 break;
1920         case KVM_REG_PPC_PTCR:
1921                 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1922                 break;
1923         default:
1924                 r = -EINVAL;
1925                 break;
1926         }
1927
1928         return r;
1929 }
1930
1931 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1932                                  union kvmppc_one_reg *val)
1933 {
1934         int r = 0;
1935         long int i;
1936         unsigned long addr, len;
1937
1938         switch (id) {
1939         case KVM_REG_PPC_HIOR:
1940                 /* Only allow this to be set to zero */
1941                 if (set_reg_val(id, *val))
1942                         r = -EINVAL;
1943                 break;
1944         case KVM_REG_PPC_DABR:
1945                 vcpu->arch.dabr = set_reg_val(id, *val);
1946                 break;
1947         case KVM_REG_PPC_DABRX:
1948                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1949                 break;
1950         case KVM_REG_PPC_DSCR:
1951                 vcpu->arch.dscr = set_reg_val(id, *val);
1952                 break;
1953         case KVM_REG_PPC_PURR:
1954                 vcpu->arch.purr = set_reg_val(id, *val);
1955                 break;
1956         case KVM_REG_PPC_SPURR:
1957                 vcpu->arch.spurr = set_reg_val(id, *val);
1958                 break;
1959         case KVM_REG_PPC_AMR:
1960                 vcpu->arch.amr = set_reg_val(id, *val);
1961                 break;
1962         case KVM_REG_PPC_UAMOR:
1963                 vcpu->arch.uamor = set_reg_val(id, *val);
1964                 break;
1965         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
1966                 i = id - KVM_REG_PPC_MMCR0;
1967                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1968                 break;
1969         case KVM_REG_PPC_MMCR2:
1970                 vcpu->arch.mmcr[2] = set_reg_val(id, *val);
1971                 break;
1972         case KVM_REG_PPC_MMCRA:
1973                 vcpu->arch.mmcra = set_reg_val(id, *val);
1974                 break;
1975         case KVM_REG_PPC_MMCRS:
1976                 vcpu->arch.mmcrs = set_reg_val(id, *val);
1977                 break;
1978         case KVM_REG_PPC_MMCR3:
1979                 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
1980                 break;
1981         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1982                 i = id - KVM_REG_PPC_PMC1;
1983                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1984                 break;
1985         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1986                 i = id - KVM_REG_PPC_SPMC1;
1987                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1988                 break;
1989         case KVM_REG_PPC_SIAR:
1990                 vcpu->arch.siar = set_reg_val(id, *val);
1991                 break;
1992         case KVM_REG_PPC_SDAR:
1993                 vcpu->arch.sdar = set_reg_val(id, *val);
1994                 break;
1995         case KVM_REG_PPC_SIER:
1996                 vcpu->arch.sier[0] = set_reg_val(id, *val);
1997                 break;
1998         case KVM_REG_PPC_SIER2:
1999                 vcpu->arch.sier[1] = set_reg_val(id, *val);
2000                 break;
2001         case KVM_REG_PPC_SIER3:
2002                 vcpu->arch.sier[2] = set_reg_val(id, *val);
2003                 break;
2004         case KVM_REG_PPC_IAMR:
2005                 vcpu->arch.iamr = set_reg_val(id, *val);
2006                 break;
2007         case KVM_REG_PPC_PSPB:
2008                 vcpu->arch.pspb = set_reg_val(id, *val);
2009                 break;
2010         case KVM_REG_PPC_DPDES:
2011                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
2012                 break;
2013         case KVM_REG_PPC_VTB:
2014                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
2015                 break;
2016         case KVM_REG_PPC_DAWR:
2017                 vcpu->arch.dawr0 = set_reg_val(id, *val);
2018                 break;
2019         case KVM_REG_PPC_DAWRX:
2020                 vcpu->arch.dawrx0 = set_reg_val(id, *val) & ~DAWRX_HYP;
2021                 break;
2022         case KVM_REG_PPC_DAWR1:
2023                 vcpu->arch.dawr1 = set_reg_val(id, *val);
2024                 break;
2025         case KVM_REG_PPC_DAWRX1:
2026                 vcpu->arch.dawrx1 = set_reg_val(id, *val) & ~DAWRX_HYP;
2027                 break;
2028         case KVM_REG_PPC_CIABR:
2029                 vcpu->arch.ciabr = set_reg_val(id, *val);
2030                 /* Don't allow setting breakpoints in hypervisor code */
2031                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
2032                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
2033                 break;
2034         case KVM_REG_PPC_CSIGR:
2035                 vcpu->arch.csigr = set_reg_val(id, *val);
2036                 break;
2037         case KVM_REG_PPC_TACR:
2038                 vcpu->arch.tacr = set_reg_val(id, *val);
2039                 break;
2040         case KVM_REG_PPC_TCSCR:
2041                 vcpu->arch.tcscr = set_reg_val(id, *val);
2042                 break;
2043         case KVM_REG_PPC_PID:
2044                 vcpu->arch.pid = set_reg_val(id, *val);
2045                 break;
2046         case KVM_REG_PPC_ACOP:
2047                 vcpu->arch.acop = set_reg_val(id, *val);
2048                 break;
2049         case KVM_REG_PPC_WORT:
2050                 vcpu->arch.wort = set_reg_val(id, *val);
2051                 break;
2052         case KVM_REG_PPC_TIDR:
2053                 vcpu->arch.tid = set_reg_val(id, *val);
2054                 break;
2055         case KVM_REG_PPC_PSSCR:
2056                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2057                 break;
2058         case KVM_REG_PPC_VPA_ADDR:
2059                 addr = set_reg_val(id, *val);
2060                 r = -EINVAL;
2061                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2062                               vcpu->arch.dtl.next_gpa))
2063                         break;
2064                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2065                 break;
2066         case KVM_REG_PPC_VPA_SLB:
2067                 addr = val->vpaval.addr;
2068                 len = val->vpaval.length;
2069                 r = -EINVAL;
2070                 if (addr && !vcpu->arch.vpa.next_gpa)
2071                         break;
2072                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2073                 break;
2074         case KVM_REG_PPC_VPA_DTL:
2075                 addr = val->vpaval.addr;
2076                 len = val->vpaval.length;
2077                 r = -EINVAL;
2078                 if (addr && (len < sizeof(struct dtl_entry) ||
2079                              !vcpu->arch.vpa.next_gpa))
2080                         break;
2081                 len -= len % sizeof(struct dtl_entry);
2082                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2083                 break;
2084         case KVM_REG_PPC_TB_OFFSET:
2085                 /* round up to multiple of 2^24 */
2086                 vcpu->arch.vcore->tb_offset =
2087                         ALIGN(set_reg_val(id, *val), 1UL << 24);
2088                 break;
2089         case KVM_REG_PPC_LPCR:
2090                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2091                 break;
2092         case KVM_REG_PPC_LPCR_64:
2093                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2094                 break;
2095         case KVM_REG_PPC_PPR:
2096                 vcpu->arch.ppr = set_reg_val(id, *val);
2097                 break;
2098 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2099         case KVM_REG_PPC_TFHAR:
2100                 vcpu->arch.tfhar = set_reg_val(id, *val);
2101                 break;
2102         case KVM_REG_PPC_TFIAR:
2103                 vcpu->arch.tfiar = set_reg_val(id, *val);
2104                 break;
2105         case KVM_REG_PPC_TEXASR:
2106                 vcpu->arch.texasr = set_reg_val(id, *val);
2107                 break;
2108         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2109                 i = id - KVM_REG_PPC_TM_GPR0;
2110                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2111                 break;
2112         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2113         {
2114                 int j;
2115                 i = id - KVM_REG_PPC_TM_VSR0;
2116                 if (i < 32)
2117                         for (j = 0; j < TS_FPRWIDTH; j++)
2118                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2119                 else
2120                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
2121                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2122                         else
2123                                 r = -ENXIO;
2124                 break;
2125         }
2126         case KVM_REG_PPC_TM_CR:
2127                 vcpu->arch.cr_tm = set_reg_val(id, *val);
2128                 break;
2129         case KVM_REG_PPC_TM_XER:
2130                 vcpu->arch.xer_tm = set_reg_val(id, *val);
2131                 break;
2132         case KVM_REG_PPC_TM_LR:
2133                 vcpu->arch.lr_tm = set_reg_val(id, *val);
2134                 break;
2135         case KVM_REG_PPC_TM_CTR:
2136                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2137                 break;
2138         case KVM_REG_PPC_TM_FPSCR:
2139                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2140                 break;
2141         case KVM_REG_PPC_TM_AMR:
2142                 vcpu->arch.amr_tm = set_reg_val(id, *val);
2143                 break;
2144         case KVM_REG_PPC_TM_PPR:
2145                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2146                 break;
2147         case KVM_REG_PPC_TM_VRSAVE:
2148                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2149                 break;
2150         case KVM_REG_PPC_TM_VSCR:
2151                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2152                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2153                 else
2154                         r = - ENXIO;
2155                 break;
2156         case KVM_REG_PPC_TM_DSCR:
2157                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2158                 break;
2159         case KVM_REG_PPC_TM_TAR:
2160                 vcpu->arch.tar_tm = set_reg_val(id, *val);
2161                 break;
2162 #endif
2163         case KVM_REG_PPC_ARCH_COMPAT:
2164                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2165                 break;
2166         case KVM_REG_PPC_DEC_EXPIRY:
2167                 vcpu->arch.dec_expires = set_reg_val(id, *val) -
2168                         vcpu->arch.vcore->tb_offset;
2169                 break;
2170         case KVM_REG_PPC_ONLINE:
2171                 i = set_reg_val(id, *val);
2172                 if (i && !vcpu->arch.online)
2173                         atomic_inc(&vcpu->arch.vcore->online_count);
2174                 else if (!i && vcpu->arch.online)
2175                         atomic_dec(&vcpu->arch.vcore->online_count);
2176                 vcpu->arch.online = i;
2177                 break;
2178         case KVM_REG_PPC_PTCR:
2179                 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2180                 break;
2181         default:
2182                 r = -EINVAL;
2183                 break;
2184         }
2185
2186         return r;
2187 }
2188
2189 /*
2190  * On POWER9, threads are independent and can be in different partitions.
2191  * Therefore we consider each thread to be a subcore.
2192  * There is a restriction that all threads have to be in the same
2193  * MMU mode (radix or HPT), unfortunately, but since we only support
2194  * HPT guests on a HPT host so far, that isn't an impediment yet.
2195  */
2196 static int threads_per_vcore(struct kvm *kvm)
2197 {
2198         if (kvm->arch.threads_indep)
2199                 return 1;
2200         return threads_per_subcore;
2201 }
2202
2203 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2204 {
2205         struct kvmppc_vcore *vcore;
2206
2207         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2208
2209         if (vcore == NULL)
2210                 return NULL;
2211
2212         spin_lock_init(&vcore->lock);
2213         spin_lock_init(&vcore->stoltb_lock);
2214         rcuwait_init(&vcore->wait);
2215         vcore->preempt_tb = TB_NIL;
2216         vcore->lpcr = kvm->arch.lpcr;
2217         vcore->first_vcpuid = id;
2218         vcore->kvm = kvm;
2219         INIT_LIST_HEAD(&vcore->preempt_list);
2220
2221         return vcore;
2222 }
2223
2224 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2225 static struct debugfs_timings_element {
2226         const char *name;
2227         size_t offset;
2228 } timings[] = {
2229         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
2230         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
2231         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
2232         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
2233         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
2234 };
2235
2236 #define N_TIMINGS       (ARRAY_SIZE(timings))
2237
2238 struct debugfs_timings_state {
2239         struct kvm_vcpu *vcpu;
2240         unsigned int    buflen;
2241         char            buf[N_TIMINGS * 100];
2242 };
2243
2244 static int debugfs_timings_open(struct inode *inode, struct file *file)
2245 {
2246         struct kvm_vcpu *vcpu = inode->i_private;
2247         struct debugfs_timings_state *p;
2248
2249         p = kzalloc(sizeof(*p), GFP_KERNEL);
2250         if (!p)
2251                 return -ENOMEM;
2252
2253         kvm_get_kvm(vcpu->kvm);
2254         p->vcpu = vcpu;
2255         file->private_data = p;
2256
2257         return nonseekable_open(inode, file);
2258 }
2259
2260 static int debugfs_timings_release(struct inode *inode, struct file *file)
2261 {
2262         struct debugfs_timings_state *p = file->private_data;
2263
2264         kvm_put_kvm(p->vcpu->kvm);
2265         kfree(p);
2266         return 0;
2267 }
2268
2269 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2270                                     size_t len, loff_t *ppos)
2271 {
2272         struct debugfs_timings_state *p = file->private_data;
2273         struct kvm_vcpu *vcpu = p->vcpu;
2274         char *s, *buf_end;
2275         struct kvmhv_tb_accumulator tb;
2276         u64 count;
2277         loff_t pos;
2278         ssize_t n;
2279         int i, loops;
2280         bool ok;
2281
2282         if (!p->buflen) {
2283                 s = p->buf;
2284                 buf_end = s + sizeof(p->buf);
2285                 for (i = 0; i < N_TIMINGS; ++i) {
2286                         struct kvmhv_tb_accumulator *acc;
2287
2288                         acc = (struct kvmhv_tb_accumulator *)
2289                                 ((unsigned long)vcpu + timings[i].offset);
2290                         ok = false;
2291                         for (loops = 0; loops < 1000; ++loops) {
2292                                 count = acc->seqcount;
2293                                 if (!(count & 1)) {
2294                                         smp_rmb();
2295                                         tb = *acc;
2296                                         smp_rmb();
2297                                         if (count == acc->seqcount) {
2298                                                 ok = true;
2299                                                 break;
2300                                         }
2301                                 }
2302                                 udelay(1);
2303                         }
2304                         if (!ok)
2305                                 snprintf(s, buf_end - s, "%s: stuck\n",
2306                                         timings[i].name);
2307                         else
2308                                 snprintf(s, buf_end - s,
2309                                         "%s: %llu %llu %llu %llu\n",
2310                                         timings[i].name, count / 2,
2311                                         tb_to_ns(tb.tb_total),
2312                                         tb_to_ns(tb.tb_min),
2313                                         tb_to_ns(tb.tb_max));
2314                         s += strlen(s);
2315                 }
2316                 p->buflen = s - p->buf;
2317         }
2318
2319         pos = *ppos;
2320         if (pos >= p->buflen)
2321                 return 0;
2322         if (len > p->buflen - pos)
2323                 len = p->buflen - pos;
2324         n = copy_to_user(buf, p->buf + pos, len);
2325         if (n) {
2326                 if (n == len)
2327                         return -EFAULT;
2328                 len -= n;
2329         }
2330         *ppos = pos + len;
2331         return len;
2332 }
2333
2334 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2335                                      size_t len, loff_t *ppos)
2336 {
2337         return -EACCES;
2338 }
2339
2340 static const struct file_operations debugfs_timings_ops = {
2341         .owner   = THIS_MODULE,
2342         .open    = debugfs_timings_open,
2343         .release = debugfs_timings_release,
2344         .read    = debugfs_timings_read,
2345         .write   = debugfs_timings_write,
2346         .llseek  = generic_file_llseek,
2347 };
2348
2349 /* Create a debugfs directory for the vcpu */
2350 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2351 {
2352         char buf[16];
2353         struct kvm *kvm = vcpu->kvm;
2354
2355         snprintf(buf, sizeof(buf), "vcpu%u", id);
2356         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2357         debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir, vcpu,
2358                             &debugfs_timings_ops);
2359 }
2360
2361 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2362 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2363 {
2364 }
2365 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2366
2367 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2368 {
2369         int err;
2370         int core;
2371         struct kvmppc_vcore *vcore;
2372         struct kvm *kvm;
2373         unsigned int id;
2374
2375         kvm = vcpu->kvm;
2376         id = vcpu->vcpu_id;
2377
2378         vcpu->arch.shared = &vcpu->arch.shregs;
2379 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2380         /*
2381          * The shared struct is never shared on HV,
2382          * so we can always use host endianness
2383          */
2384 #ifdef __BIG_ENDIAN__
2385         vcpu->arch.shared_big_endian = true;
2386 #else
2387         vcpu->arch.shared_big_endian = false;
2388 #endif
2389 #endif
2390         vcpu->arch.mmcr[0] = MMCR0_FC;
2391         vcpu->arch.ctrl = CTRL_RUNLATCH;
2392         /* default to host PVR, since we can't spoof it */
2393         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2394         spin_lock_init(&vcpu->arch.vpa_update_lock);
2395         spin_lock_init(&vcpu->arch.tbacct_lock);
2396         vcpu->arch.busy_preempt = TB_NIL;
2397         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2398
2399         /*
2400          * Set the default HFSCR for the guest from the host value.
2401          * This value is only used on POWER9.
2402          * On POWER9, we want to virtualize the doorbell facility, so we
2403          * don't set the HFSCR_MSGP bit, and that causes those instructions
2404          * to trap and then we emulate them.
2405          */
2406         vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2407                 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP | HFSCR_PREFIX;
2408         if (cpu_has_feature(CPU_FTR_HVMODE)) {
2409                 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2410                 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2411                         vcpu->arch.hfscr |= HFSCR_TM;
2412         }
2413         if (cpu_has_feature(CPU_FTR_TM_COMP))
2414                 vcpu->arch.hfscr |= HFSCR_TM;
2415
2416         kvmppc_mmu_book3s_hv_init(vcpu);
2417
2418         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2419
2420         init_waitqueue_head(&vcpu->arch.cpu_run);
2421
2422         mutex_lock(&kvm->lock);
2423         vcore = NULL;
2424         err = -EINVAL;
2425         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2426                 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2427                         pr_devel("KVM: VCPU ID too high\n");
2428                         core = KVM_MAX_VCORES;
2429                 } else {
2430                         BUG_ON(kvm->arch.smt_mode != 1);
2431                         core = kvmppc_pack_vcpu_id(kvm, id);
2432                 }
2433         } else {
2434                 core = id / kvm->arch.smt_mode;
2435         }
2436         if (core < KVM_MAX_VCORES) {
2437                 vcore = kvm->arch.vcores[core];
2438                 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2439                         pr_devel("KVM: collision on id %u", id);
2440                         vcore = NULL;
2441                 } else if (!vcore) {
2442                         /*
2443                          * Take mmu_setup_lock for mutual exclusion
2444                          * with kvmppc_update_lpcr().
2445                          */
2446                         err = -ENOMEM;
2447                         vcore = kvmppc_vcore_create(kvm,
2448                                         id & ~(kvm->arch.smt_mode - 1));
2449                         mutex_lock(&kvm->arch.mmu_setup_lock);
2450                         kvm->arch.vcores[core] = vcore;
2451                         kvm->arch.online_vcores++;
2452                         mutex_unlock(&kvm->arch.mmu_setup_lock);
2453                 }
2454         }
2455         mutex_unlock(&kvm->lock);
2456
2457         if (!vcore)
2458                 return err;
2459
2460         spin_lock(&vcore->lock);
2461         ++vcore->num_threads;
2462         spin_unlock(&vcore->lock);
2463         vcpu->arch.vcore = vcore;
2464         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2465         vcpu->arch.thread_cpu = -1;
2466         vcpu->arch.prev_cpu = -1;
2467
2468         vcpu->arch.cpu_type = KVM_CPU_3S_64;
2469         kvmppc_sanity_check(vcpu);
2470
2471         debugfs_vcpu_init(vcpu, id);
2472
2473         return 0;
2474 }
2475
2476 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2477                               unsigned long flags)
2478 {
2479         int err;
2480         int esmt = 0;
2481
2482         if (flags)
2483                 return -EINVAL;
2484         if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2485                 return -EINVAL;
2486         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2487                 /*
2488                  * On POWER8 (or POWER7), the threading mode is "strict",
2489                  * so we pack smt_mode vcpus per vcore.
2490                  */
2491                 if (smt_mode > threads_per_subcore)
2492                         return -EINVAL;
2493         } else {
2494                 /*
2495                  * On POWER9, the threading mode is "loose",
2496                  * so each vcpu gets its own vcore.
2497                  */
2498                 esmt = smt_mode;
2499                 smt_mode = 1;
2500         }
2501         mutex_lock(&kvm->lock);
2502         err = -EBUSY;
2503         if (!kvm->arch.online_vcores) {
2504                 kvm->arch.smt_mode = smt_mode;
2505                 kvm->arch.emul_smt_mode = esmt;
2506                 err = 0;
2507         }
2508         mutex_unlock(&kvm->lock);
2509
2510         return err;
2511 }
2512
2513 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2514 {
2515         if (vpa->pinned_addr)
2516                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2517                                         vpa->dirty);
2518 }
2519
2520 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2521 {
2522         spin_lock(&vcpu->arch.vpa_update_lock);
2523         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2524         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2525         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2526         spin_unlock(&vcpu->arch.vpa_update_lock);
2527 }
2528
2529 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2530 {
2531         /* Indicate we want to get back into the guest */
2532         return 1;
2533 }
2534
2535 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2536 {
2537         unsigned long dec_nsec, now;
2538
2539         now = get_tb();
2540         if (now > vcpu->arch.dec_expires) {
2541                 /* decrementer has already gone negative */
2542                 kvmppc_core_queue_dec(vcpu);
2543                 kvmppc_core_prepare_to_enter(vcpu);
2544                 return;
2545         }
2546         dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2547         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2548         vcpu->arch.timer_running = 1;
2549 }
2550
2551 extern int __kvmppc_vcore_entry(void);
2552
2553 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2554                                    struct kvm_vcpu *vcpu)
2555 {
2556         u64 now;
2557
2558         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2559                 return;
2560         spin_lock_irq(&vcpu->arch.tbacct_lock);
2561         now = mftb();
2562         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2563                 vcpu->arch.stolen_logged;
2564         vcpu->arch.busy_preempt = now;
2565         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2566         spin_unlock_irq(&vcpu->arch.tbacct_lock);
2567         --vc->n_runnable;
2568         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2569 }
2570
2571 static int kvmppc_grab_hwthread(int cpu)
2572 {
2573         struct paca_struct *tpaca;
2574         long timeout = 10000;
2575
2576         tpaca = paca_ptrs[cpu];
2577
2578         /* Ensure the thread won't go into the kernel if it wakes */
2579         tpaca->kvm_hstate.kvm_vcpu = NULL;
2580         tpaca->kvm_hstate.kvm_vcore = NULL;
2581         tpaca->kvm_hstate.napping = 0;
2582         smp_wmb();
2583         tpaca->kvm_hstate.hwthread_req = 1;
2584
2585         /*
2586          * If the thread is already executing in the kernel (e.g. handling
2587          * a stray interrupt), wait for it to get back to nap mode.
2588          * The smp_mb() is to ensure that our setting of hwthread_req
2589          * is visible before we look at hwthread_state, so if this
2590          * races with the code at system_reset_pSeries and the thread
2591          * misses our setting of hwthread_req, we are sure to see its
2592          * setting of hwthread_state, and vice versa.
2593          */
2594         smp_mb();
2595         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2596                 if (--timeout <= 0) {
2597                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
2598                         return -EBUSY;
2599                 }
2600                 udelay(1);
2601         }
2602         return 0;
2603 }
2604
2605 static void kvmppc_release_hwthread(int cpu)
2606 {
2607         struct paca_struct *tpaca;
2608
2609         tpaca = paca_ptrs[cpu];
2610         tpaca->kvm_hstate.hwthread_req = 0;
2611         tpaca->kvm_hstate.kvm_vcpu = NULL;
2612         tpaca->kvm_hstate.kvm_vcore = NULL;
2613         tpaca->kvm_hstate.kvm_split_mode = NULL;
2614 }
2615
2616 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2617 {
2618         struct kvm_nested_guest *nested = vcpu->arch.nested;
2619         cpumask_t *cpu_in_guest;
2620         int i;
2621
2622         cpu = cpu_first_thread_sibling(cpu);
2623         if (nested) {
2624                 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2625                 cpu_in_guest = &nested->cpu_in_guest;
2626         } else {
2627                 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2628                 cpu_in_guest = &kvm->arch.cpu_in_guest;
2629         }
2630         /*
2631          * Make sure setting of bit in need_tlb_flush precedes
2632          * testing of cpu_in_guest bits.  The matching barrier on
2633          * the other side is the first smp_mb() in kvmppc_run_core().
2634          */
2635         smp_mb();
2636         for (i = 0; i < threads_per_core; ++i)
2637                 if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2638                         smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2639 }
2640
2641 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2642 {
2643         struct kvm_nested_guest *nested = vcpu->arch.nested;
2644         struct kvm *kvm = vcpu->kvm;
2645         int prev_cpu;
2646
2647         if (!cpu_has_feature(CPU_FTR_HVMODE))
2648                 return;
2649
2650         if (nested)
2651                 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2652         else
2653                 prev_cpu = vcpu->arch.prev_cpu;
2654
2655         /*
2656          * With radix, the guest can do TLB invalidations itself,
2657          * and it could choose to use the local form (tlbiel) if
2658          * it is invalidating a translation that has only ever been
2659          * used on one vcpu.  However, that doesn't mean it has
2660          * only ever been used on one physical cpu, since vcpus
2661          * can move around between pcpus.  To cope with this, when
2662          * a vcpu moves from one pcpu to another, we need to tell
2663          * any vcpus running on the same core as this vcpu previously
2664          * ran to flush the TLB.  The TLB is shared between threads,
2665          * so we use a single bit in .need_tlb_flush for all 4 threads.
2666          */
2667         if (prev_cpu != pcpu) {
2668                 if (prev_cpu >= 0 &&
2669                     cpu_first_thread_sibling(prev_cpu) !=
2670                     cpu_first_thread_sibling(pcpu))
2671                         radix_flush_cpu(kvm, prev_cpu, vcpu);
2672                 if (nested)
2673                         nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2674                 else
2675                         vcpu->arch.prev_cpu = pcpu;
2676         }
2677 }
2678
2679 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2680 {
2681         int cpu;
2682         struct paca_struct *tpaca;
2683         struct kvm *kvm = vc->kvm;
2684
2685         cpu = vc->pcpu;
2686         if (vcpu) {
2687                 if (vcpu->arch.timer_running) {
2688                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2689                         vcpu->arch.timer_running = 0;
2690                 }
2691                 cpu += vcpu->arch.ptid;
2692                 vcpu->cpu = vc->pcpu;
2693                 vcpu->arch.thread_cpu = cpu;
2694                 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2695         }
2696         tpaca = paca_ptrs[cpu];
2697         tpaca->kvm_hstate.kvm_vcpu = vcpu;
2698         tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2699         tpaca->kvm_hstate.fake_suspend = 0;
2700         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2701         smp_wmb();
2702         tpaca->kvm_hstate.kvm_vcore = vc;
2703         if (cpu != smp_processor_id())
2704                 kvmppc_ipi_thread(cpu);
2705 }
2706
2707 static void kvmppc_wait_for_nap(int n_threads)
2708 {
2709         int cpu = smp_processor_id();
2710         int i, loops;
2711
2712         if (n_threads <= 1)
2713                 return;
2714         for (loops = 0; loops < 1000000; ++loops) {
2715                 /*
2716                  * Check if all threads are finished.
2717                  * We set the vcore pointer when starting a thread
2718                  * and the thread clears it when finished, so we look
2719                  * for any threads that still have a non-NULL vcore ptr.
2720                  */
2721                 for (i = 1; i < n_threads; ++i)
2722                         if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2723                                 break;
2724                 if (i == n_threads) {
2725                         HMT_medium();
2726                         return;
2727                 }
2728                 HMT_low();
2729         }
2730         HMT_medium();
2731         for (i = 1; i < n_threads; ++i)
2732                 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2733                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2734 }
2735
2736 /*
2737  * Check that we are on thread 0 and that any other threads in
2738  * this core are off-line.  Then grab the threads so they can't
2739  * enter the kernel.
2740  */
2741 static int on_primary_thread(void)
2742 {
2743         int cpu = smp_processor_id();
2744         int thr;
2745
2746         /* Are we on a primary subcore? */
2747         if (cpu_thread_in_subcore(cpu))
2748                 return 0;
2749
2750         thr = 0;
2751         while (++thr < threads_per_subcore)
2752                 if (cpu_online(cpu + thr))
2753                         return 0;
2754
2755         /* Grab all hw threads so they can't go into the kernel */
2756         for (thr = 1; thr < threads_per_subcore; ++thr) {
2757                 if (kvmppc_grab_hwthread(cpu + thr)) {
2758                         /* Couldn't grab one; let the others go */
2759                         do {
2760                                 kvmppc_release_hwthread(cpu + thr);
2761                         } while (--thr > 0);
2762                         return 0;
2763                 }
2764         }
2765         return 1;
2766 }
2767
2768 /*
2769  * A list of virtual cores for each physical CPU.
2770  * These are vcores that could run but their runner VCPU tasks are
2771  * (or may be) preempted.
2772  */
2773 struct preempted_vcore_list {
2774         struct list_head        list;
2775         spinlock_t              lock;
2776 };
2777
2778 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2779
2780 static void init_vcore_lists(void)
2781 {
2782         int cpu;
2783
2784         for_each_possible_cpu(cpu) {
2785                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2786                 spin_lock_init(&lp->lock);
2787                 INIT_LIST_HEAD(&lp->list);
2788         }
2789 }
2790
2791 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2792 {
2793         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2794
2795         vc->vcore_state = VCORE_PREEMPT;
2796         vc->pcpu = smp_processor_id();
2797         if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2798                 spin_lock(&lp->lock);
2799                 list_add_tail(&vc->preempt_list, &lp->list);
2800                 spin_unlock(&lp->lock);
2801         }
2802
2803         /* Start accumulating stolen time */
2804         kvmppc_core_start_stolen(vc);
2805 }
2806
2807 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2808 {
2809         struct preempted_vcore_list *lp;
2810
2811         kvmppc_core_end_stolen(vc);
2812         if (!list_empty(&vc->preempt_list)) {
2813                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2814                 spin_lock(&lp->lock);
2815                 list_del_init(&vc->preempt_list);
2816                 spin_unlock(&lp->lock);
2817         }
2818         vc->vcore_state = VCORE_INACTIVE;
2819 }
2820
2821 /*
2822  * This stores information about the virtual cores currently
2823  * assigned to a physical core.
2824  */
2825 struct core_info {
2826         int             n_subcores;
2827         int             max_subcore_threads;
2828         int             total_threads;
2829         int             subcore_threads[MAX_SUBCORES];
2830         struct kvmppc_vcore *vc[MAX_SUBCORES];
2831 };
2832
2833 /*
2834  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2835  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2836  */
2837 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2838
2839 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2840 {
2841         memset(cip, 0, sizeof(*cip));
2842         cip->n_subcores = 1;
2843         cip->max_subcore_threads = vc->num_threads;
2844         cip->total_threads = vc->num_threads;
2845         cip->subcore_threads[0] = vc->num_threads;
2846         cip->vc[0] = vc;
2847 }
2848
2849 static bool subcore_config_ok(int n_subcores, int n_threads)
2850 {
2851         /*
2852          * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2853          * split-core mode, with one thread per subcore.
2854          */
2855         if (cpu_has_feature(CPU_FTR_ARCH_300))
2856                 return n_subcores <= 4 && n_threads == 1;
2857
2858         /* On POWER8, can only dynamically split if unsplit to begin with */
2859         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2860                 return false;
2861         if (n_subcores > MAX_SUBCORES)
2862                 return false;
2863         if (n_subcores > 1) {
2864                 if (!(dynamic_mt_modes & 2))
2865                         n_subcores = 4;
2866                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2867                         return false;
2868         }
2869
2870         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2871 }
2872
2873 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2874 {
2875         vc->entry_exit_map = 0;
2876         vc->in_guest = 0;
2877         vc->napping_threads = 0;
2878         vc->conferring_threads = 0;
2879         vc->tb_offset_applied = 0;
2880 }
2881
2882 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2883 {
2884         int n_threads = vc->num_threads;
2885         int sub;
2886
2887         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2888                 return false;
2889
2890         /* In one_vm_per_core mode, require all vcores to be from the same vm */
2891         if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2892                 return false;
2893
2894         if (n_threads < cip->max_subcore_threads)
2895                 n_threads = cip->max_subcore_threads;
2896         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2897                 return false;
2898         cip->max_subcore_threads = n_threads;
2899
2900         sub = cip->n_subcores;
2901         ++cip->n_subcores;
2902         cip->total_threads += vc->num_threads;
2903         cip->subcore_threads[sub] = vc->num_threads;
2904         cip->vc[sub] = vc;
2905         init_vcore_to_run(vc);
2906         list_del_init(&vc->preempt_list);
2907
2908         return true;
2909 }
2910
2911 /*
2912  * Work out whether it is possible to piggyback the execution of
2913  * vcore *pvc onto the execution of the other vcores described in *cip.
2914  */
2915 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2916                           int target_threads)
2917 {
2918         if (cip->total_threads + pvc->num_threads > target_threads)
2919                 return false;
2920
2921         return can_dynamic_split(pvc, cip);
2922 }
2923
2924 static void prepare_threads(struct kvmppc_vcore *vc)
2925 {
2926         int i;
2927         struct kvm_vcpu *vcpu;
2928
2929         for_each_runnable_thread(i, vcpu, vc) {
2930                 if (signal_pending(vcpu->arch.run_task))
2931                         vcpu->arch.ret = -EINTR;
2932                 else if (no_mixing_hpt_and_radix &&
2933                          kvm_is_radix(vc->kvm) != radix_enabled())
2934                         vcpu->arch.ret = -EINVAL;
2935                 else if (vcpu->arch.vpa.update_pending ||
2936                          vcpu->arch.slb_shadow.update_pending ||
2937                          vcpu->arch.dtl.update_pending)
2938                         vcpu->arch.ret = RESUME_GUEST;
2939                 else
2940                         continue;
2941                 kvmppc_remove_runnable(vc, vcpu);
2942                 wake_up(&vcpu->arch.cpu_run);
2943         }
2944 }
2945
2946 static void collect_piggybacks(struct core_info *cip, int target_threads)
2947 {
2948         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2949         struct kvmppc_vcore *pvc, *vcnext;
2950
2951         spin_lock(&lp->lock);
2952         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2953                 if (!spin_trylock(&pvc->lock))
2954                         continue;
2955                 prepare_threads(pvc);
2956                 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
2957                         list_del_init(&pvc->preempt_list);
2958                         if (pvc->runner == NULL) {
2959                                 pvc->vcore_state = VCORE_INACTIVE;
2960                                 kvmppc_core_end_stolen(pvc);
2961                         }
2962                         spin_unlock(&pvc->lock);
2963                         continue;
2964                 }
2965                 if (!can_piggyback(pvc, cip, target_threads)) {
2966                         spin_unlock(&pvc->lock);
2967                         continue;
2968                 }
2969                 kvmppc_core_end_stolen(pvc);
2970                 pvc->vcore_state = VCORE_PIGGYBACK;
2971                 if (cip->total_threads >= target_threads)
2972                         break;
2973         }
2974         spin_unlock(&lp->lock);
2975 }
2976
2977 static bool recheck_signals_and_mmu(struct core_info *cip)
2978 {
2979         int sub, i;
2980         struct kvm_vcpu *vcpu;
2981         struct kvmppc_vcore *vc;
2982
2983         for (sub = 0; sub < cip->n_subcores; ++sub) {
2984                 vc = cip->vc[sub];
2985                 if (!vc->kvm->arch.mmu_ready)
2986                         return true;
2987                 for_each_runnable_thread(i, vcpu, vc)
2988                         if (signal_pending(vcpu->arch.run_task))
2989                                 return true;
2990         }
2991         return false;
2992 }
2993
2994 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2995 {
2996         int still_running = 0, i;
2997         u64 now;
2998         long ret;
2999         struct kvm_vcpu *vcpu;
3000
3001         spin_lock(&vc->lock);
3002         now = get_tb();
3003         for_each_runnable_thread(i, vcpu, vc) {
3004                 /*
3005                  * It's safe to unlock the vcore in the loop here, because
3006                  * for_each_runnable_thread() is safe against removal of
3007                  * the vcpu, and the vcore state is VCORE_EXITING here,
3008                  * so any vcpus becoming runnable will have their arch.trap
3009                  * set to zero and can't actually run in the guest.
3010                  */
3011                 spin_unlock(&vc->lock);
3012                 /* cancel pending dec exception if dec is positive */
3013                 if (now < vcpu->arch.dec_expires &&
3014                     kvmppc_core_pending_dec(vcpu))
3015                         kvmppc_core_dequeue_dec(vcpu);
3016
3017                 trace_kvm_guest_exit(vcpu);
3018
3019                 ret = RESUME_GUEST;
3020                 if (vcpu->arch.trap)
3021                         ret = kvmppc_handle_exit_hv(vcpu,
3022                                                     vcpu->arch.run_task);
3023
3024                 vcpu->arch.ret = ret;
3025                 vcpu->arch.trap = 0;
3026
3027                 spin_lock(&vc->lock);
3028                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
3029                         if (vcpu->arch.pending_exceptions)
3030                                 kvmppc_core_prepare_to_enter(vcpu);
3031                         if (vcpu->arch.ceded)
3032                                 kvmppc_set_timer(vcpu);
3033                         else
3034                                 ++still_running;
3035                 } else {
3036                         kvmppc_remove_runnable(vc, vcpu);
3037                         wake_up(&vcpu->arch.cpu_run);
3038                 }
3039         }
3040         if (!is_master) {
3041                 if (still_running > 0) {
3042                         kvmppc_vcore_preempt(vc);
3043                 } else if (vc->runner) {
3044                         vc->vcore_state = VCORE_PREEMPT;
3045                         kvmppc_core_start_stolen(vc);
3046                 } else {
3047                         vc->vcore_state = VCORE_INACTIVE;
3048                 }
3049                 if (vc->n_runnable > 0 && vc->runner == NULL) {
3050                         /* make sure there's a candidate runner awake */
3051                         i = -1;
3052                         vcpu = next_runnable_thread(vc, &i);
3053                         wake_up(&vcpu->arch.cpu_run);
3054                 }
3055         }
3056         spin_unlock(&vc->lock);
3057 }
3058
3059 /*
3060  * Clear core from the list of active host cores as we are about to
3061  * enter the guest. Only do this if it is the primary thread of the
3062  * core (not if a subcore) that is entering the guest.
3063  */
3064 static inline int kvmppc_clear_host_core(unsigned int cpu)
3065 {
3066         int core;
3067
3068         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3069                 return 0;
3070         /*
3071          * Memory barrier can be omitted here as we will do a smp_wmb()
3072          * later in kvmppc_start_thread and we need ensure that state is
3073          * visible to other CPUs only after we enter guest.
3074          */
3075         core = cpu >> threads_shift;
3076         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3077         return 0;
3078 }
3079
3080 /*
3081  * Advertise this core as an active host core since we exited the guest
3082  * Only need to do this if it is the primary thread of the core that is
3083  * exiting.
3084  */
3085 static inline int kvmppc_set_host_core(unsigned int cpu)
3086 {
3087         int core;
3088
3089         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3090                 return 0;
3091
3092         /*
3093          * Memory barrier can be omitted here because we do a spin_unlock
3094          * immediately after this which provides the memory barrier.
3095          */
3096         core = cpu >> threads_shift;
3097         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3098         return 0;
3099 }
3100
3101 static void set_irq_happened(int trap)
3102 {
3103         switch (trap) {
3104         case BOOK3S_INTERRUPT_EXTERNAL:
3105                 local_paca->irq_happened |= PACA_IRQ_EE;
3106                 break;
3107         case BOOK3S_INTERRUPT_H_DOORBELL:
3108                 local_paca->irq_happened |= PACA_IRQ_DBELL;
3109                 break;
3110         case BOOK3S_INTERRUPT_HMI:
3111                 local_paca->irq_happened |= PACA_IRQ_HMI;
3112                 break;
3113         case BOOK3S_INTERRUPT_SYSTEM_RESET:
3114                 replay_system_reset();
3115                 break;
3116         }
3117 }
3118
3119 /*
3120  * Run a set of guest threads on a physical core.
3121  * Called with vc->lock held.
3122  */
3123 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3124 {
3125         struct kvm_vcpu *vcpu;
3126         int i;
3127         int srcu_idx;
3128         struct core_info core_info;
3129         struct kvmppc_vcore *pvc;
3130         struct kvm_split_mode split_info, *sip;
3131         int split, subcore_size, active;
3132         int sub;
3133         bool thr0_done;
3134         unsigned long cmd_bit, stat_bit;
3135         int pcpu, thr;
3136         int target_threads;
3137         int controlled_threads;
3138         int trap;
3139         bool is_power8;
3140
3141         /*
3142          * Remove from the list any threads that have a signal pending
3143          * or need a VPA update done
3144          */
3145         prepare_threads(vc);
3146
3147         /* if the runner is no longer runnable, let the caller pick a new one */
3148         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3149                 return;
3150
3151         /*
3152          * Initialize *vc.
3153          */
3154         init_vcore_to_run(vc);
3155         vc->preempt_tb = TB_NIL;
3156
3157         /*
3158          * Number of threads that we will be controlling: the same as
3159          * the number of threads per subcore, except on POWER9,
3160          * where it's 1 because the threads are (mostly) independent.
3161          */
3162         controlled_threads = threads_per_vcore(vc->kvm);
3163
3164         /*
3165          * Make sure we are running on primary threads, and that secondary
3166          * threads are offline.  Also check if the number of threads in this
3167          * guest are greater than the current system threads per guest.
3168          * On POWER9, we need to be not in independent-threads mode if
3169          * this is a HPT guest on a radix host machine where the
3170          * CPU threads may not be in different MMU modes.
3171          */
3172         if ((controlled_threads > 1) &&
3173             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
3174                 for_each_runnable_thread(i, vcpu, vc) {
3175                         vcpu->arch.ret = -EBUSY;
3176                         kvmppc_remove_runnable(vc, vcpu);
3177                         wake_up(&vcpu->arch.cpu_run);
3178                 }
3179                 goto out;
3180         }
3181
3182         /*
3183          * See if we could run any other vcores on the physical core
3184          * along with this one.
3185          */
3186         init_core_info(&core_info, vc);
3187         pcpu = smp_processor_id();
3188         target_threads = controlled_threads;
3189         if (target_smt_mode && target_smt_mode < target_threads)
3190                 target_threads = target_smt_mode;
3191         if (vc->num_threads < target_threads)
3192                 collect_piggybacks(&core_info, target_threads);
3193
3194         /*
3195          * On radix, arrange for TLB flushing if necessary.
3196          * This has to be done before disabling interrupts since
3197          * it uses smp_call_function().
3198          */
3199         pcpu = smp_processor_id();
3200         if (kvm_is_radix(vc->kvm)) {
3201                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3202                         for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3203                                 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3204         }
3205
3206         /*
3207          * Hard-disable interrupts, and check resched flag and signals.
3208          * If we need to reschedule or deliver a signal, clean up
3209          * and return without going into the guest(s).
3210          * If the mmu_ready flag has been cleared, don't go into the
3211          * guest because that means a HPT resize operation is in progress.
3212          */
3213         local_irq_disable();
3214         hard_irq_disable();
3215         if (lazy_irq_pending() || need_resched() ||
3216             recheck_signals_and_mmu(&core_info)) {
3217                 local_irq_enable();
3218                 vc->vcore_state = VCORE_INACTIVE;
3219                 /* Unlock all except the primary vcore */
3220                 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3221                         pvc = core_info.vc[sub];
3222                         /* Put back on to the preempted vcores list */
3223                         kvmppc_vcore_preempt(pvc);
3224                         spin_unlock(&pvc->lock);
3225                 }
3226                 for (i = 0; i < controlled_threads; ++i)
3227                         kvmppc_release_hwthread(pcpu + i);
3228                 return;
3229         }
3230
3231         kvmppc_clear_host_core(pcpu);
3232
3233         /* Decide on micro-threading (split-core) mode */
3234         subcore_size = threads_per_subcore;
3235         cmd_bit = stat_bit = 0;
3236         split = core_info.n_subcores;
3237         sip = NULL;
3238         is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3239                 && !cpu_has_feature(CPU_FTR_ARCH_300);
3240
3241         if (split > 1) {
3242                 sip = &split_info;
3243                 memset(&split_info, 0, sizeof(split_info));
3244                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3245                         split_info.vc[sub] = core_info.vc[sub];
3246
3247                 if (is_power8) {
3248                         if (split == 2 && (dynamic_mt_modes & 2)) {
3249                                 cmd_bit = HID0_POWER8_1TO2LPAR;
3250                                 stat_bit = HID0_POWER8_2LPARMODE;
3251                         } else {
3252                                 split = 4;
3253                                 cmd_bit = HID0_POWER8_1TO4LPAR;
3254                                 stat_bit = HID0_POWER8_4LPARMODE;
3255                         }
3256                         subcore_size = MAX_SMT_THREADS / split;
3257                         split_info.rpr = mfspr(SPRN_RPR);
3258                         split_info.pmmar = mfspr(SPRN_PMMAR);
3259                         split_info.ldbar = mfspr(SPRN_LDBAR);
3260                         split_info.subcore_size = subcore_size;
3261                 } else {
3262                         split_info.subcore_size = 1;
3263                 }
3264
3265                 /* order writes to split_info before kvm_split_mode pointer */
3266                 smp_wmb();
3267         }
3268
3269         for (thr = 0; thr < controlled_threads; ++thr) {
3270                 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3271
3272                 paca->kvm_hstate.napping = 0;
3273                 paca->kvm_hstate.kvm_split_mode = sip;
3274         }
3275
3276         /* Initiate micro-threading (split-core) on POWER8 if required */
3277         if (cmd_bit) {
3278                 unsigned long hid0 = mfspr(SPRN_HID0);
3279
3280                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3281                 mb();
3282                 mtspr(SPRN_HID0, hid0);
3283                 isync();
3284                 for (;;) {
3285                         hid0 = mfspr(SPRN_HID0);
3286                         if (hid0 & stat_bit)
3287                                 break;
3288                         cpu_relax();
3289                 }
3290         }
3291
3292         /*
3293          * On POWER8, set RWMR register.
3294          * Since it only affects PURR and SPURR, it doesn't affect
3295          * the host, so we don't save/restore the host value.
3296          */
3297         if (is_power8) {
3298                 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3299                 int n_online = atomic_read(&vc->online_count);
3300
3301                 /*
3302                  * Use the 8-thread value if we're doing split-core
3303                  * or if the vcore's online count looks bogus.
3304                  */
3305                 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3306                     n_online >= 1 && n_online <= MAX_SMT_THREADS)
3307                         rwmr_val = p8_rwmr_values[n_online];
3308                 mtspr(SPRN_RWMR, rwmr_val);
3309         }
3310
3311         /* Start all the threads */
3312         active = 0;
3313         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3314                 thr = is_power8 ? subcore_thread_map[sub] : sub;
3315                 thr0_done = false;
3316                 active |= 1 << thr;
3317                 pvc = core_info.vc[sub];
3318                 pvc->pcpu = pcpu + thr;
3319                 for_each_runnable_thread(i, vcpu, pvc) {
3320                         kvmppc_start_thread(vcpu, pvc);
3321                         kvmppc_create_dtl_entry(vcpu, pvc);
3322                         trace_kvm_guest_enter(vcpu);
3323                         if (!vcpu->arch.ptid)
3324                                 thr0_done = true;
3325                         active |= 1 << (thr + vcpu->arch.ptid);
3326                 }
3327                 /*
3328                  * We need to start the first thread of each subcore
3329                  * even if it doesn't have a vcpu.
3330                  */
3331                 if (!thr0_done)
3332                         kvmppc_start_thread(NULL, pvc);
3333         }
3334
3335         /*
3336          * Ensure that split_info.do_nap is set after setting
3337          * the vcore pointer in the PACA of the secondaries.
3338          */
3339         smp_mb();
3340
3341         /*
3342          * When doing micro-threading, poke the inactive threads as well.
3343          * This gets them to the nap instruction after kvm_do_nap,
3344          * which reduces the time taken to unsplit later.
3345          */
3346         if (cmd_bit) {
3347                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
3348                 for (thr = 1; thr < threads_per_subcore; ++thr)
3349                         if (!(active & (1 << thr)))
3350                                 kvmppc_ipi_thread(pcpu + thr);
3351         }
3352
3353         vc->vcore_state = VCORE_RUNNING;
3354         preempt_disable();
3355
3356         trace_kvmppc_run_core(vc, 0);
3357
3358         for (sub = 0; sub < core_info.n_subcores; ++sub)
3359                 spin_unlock(&core_info.vc[sub]->lock);
3360
3361         guest_enter_irqoff();
3362
3363         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3364
3365         this_cpu_disable_ftrace();
3366
3367         /*
3368          * Interrupts will be enabled once we get into the guest,
3369          * so tell lockdep that we're about to enable interrupts.
3370          */
3371         trace_hardirqs_on();
3372
3373         trap = __kvmppc_vcore_entry();
3374
3375         trace_hardirqs_off();
3376
3377         this_cpu_enable_ftrace();
3378
3379         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3380
3381         set_irq_happened(trap);
3382
3383         spin_lock(&vc->lock);
3384         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3385         vc->vcore_state = VCORE_EXITING;
3386
3387         /* wait for secondary threads to finish writing their state to memory */
3388         kvmppc_wait_for_nap(controlled_threads);
3389
3390         /* Return to whole-core mode if we split the core earlier */
3391         if (cmd_bit) {
3392                 unsigned long hid0 = mfspr(SPRN_HID0);
3393                 unsigned long loops = 0;
3394
3395                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3396                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3397                 mb();
3398                 mtspr(SPRN_HID0, hid0);
3399                 isync();
3400                 for (;;) {
3401                         hid0 = mfspr(SPRN_HID0);
3402                         if (!(hid0 & stat_bit))
3403                                 break;
3404                         cpu_relax();
3405                         ++loops;
3406                 }
3407                 split_info.do_nap = 0;
3408         }
3409
3410         kvmppc_set_host_core(pcpu);
3411
3412         guest_exit_irqoff();
3413
3414         local_irq_enable();
3415
3416         /* Let secondaries go back to the offline loop */
3417         for (i = 0; i < controlled_threads; ++i) {
3418                 kvmppc_release_hwthread(pcpu + i);
3419                 if (sip && sip->napped[i])
3420                         kvmppc_ipi_thread(pcpu + i);
3421                 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3422         }
3423
3424         spin_unlock(&vc->lock);
3425
3426         /* make sure updates to secondary vcpu structs are visible now */
3427         smp_mb();
3428
3429         preempt_enable();
3430
3431         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3432                 pvc = core_info.vc[sub];
3433                 post_guest_process(pvc, pvc == vc);
3434         }
3435
3436         spin_lock(&vc->lock);
3437
3438  out:
3439         vc->vcore_state = VCORE_INACTIVE;
3440         trace_kvmppc_run_core(vc, 1);
3441 }
3442
3443 /*
3444  * Load up hypervisor-mode registers on P9.
3445  */
3446 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3447                                      unsigned long lpcr)
3448 {
3449         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3450         s64 hdec;
3451         u64 tb, purr, spurr;
3452         int trap;
3453         unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3454         unsigned long host_ciabr = mfspr(SPRN_CIABR);
3455         unsigned long host_dawr0 = mfspr(SPRN_DAWR0);
3456         unsigned long host_dawrx0 = mfspr(SPRN_DAWRX0);
3457         unsigned long host_psscr = mfspr(SPRN_PSSCR);
3458         unsigned long host_pidr = mfspr(SPRN_PID);
3459         unsigned long host_dawr1 = 0;
3460         unsigned long host_dawrx1 = 0;
3461
3462         if (cpu_has_feature(CPU_FTR_DAWR1)) {
3463                 host_dawr1 = mfspr(SPRN_DAWR1);
3464                 host_dawrx1 = mfspr(SPRN_DAWRX1);
3465         }
3466
3467         /*
3468          * P8 and P9 suppress the HDEC exception when LPCR[HDICE] = 0,
3469          * so set HDICE before writing HDEC.
3470          */
3471         mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr | LPCR_HDICE);
3472         isync();
3473
3474         hdec = time_limit - mftb();
3475         if (hdec < 0) {
3476                 mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3477                 isync();
3478                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3479         }
3480         mtspr(SPRN_HDEC, hdec);
3481
3482         if (vc->tb_offset) {
3483                 u64 new_tb = mftb() + vc->tb_offset;
3484                 mtspr(SPRN_TBU40, new_tb);
3485                 tb = mftb();
3486                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3487                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3488                 vc->tb_offset_applied = vc->tb_offset;
3489         }
3490
3491         if (vc->pcr)
3492                 mtspr(SPRN_PCR, vc->pcr | PCR_MASK);
3493         mtspr(SPRN_DPDES, vc->dpdes);
3494         mtspr(SPRN_VTB, vc->vtb);
3495
3496         local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3497         local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3498         mtspr(SPRN_PURR, vcpu->arch.purr);
3499         mtspr(SPRN_SPURR, vcpu->arch.spurr);
3500
3501         if (dawr_enabled()) {
3502                 mtspr(SPRN_DAWR0, vcpu->arch.dawr0);
3503                 mtspr(SPRN_DAWRX0, vcpu->arch.dawrx0);
3504                 if (cpu_has_feature(CPU_FTR_DAWR1)) {
3505                         mtspr(SPRN_DAWR1, vcpu->arch.dawr1);
3506                         mtspr(SPRN_DAWRX1, vcpu->arch.dawrx1);
3507                 }
3508         }
3509         mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3510         mtspr(SPRN_IC, vcpu->arch.ic);
3511         mtspr(SPRN_PID, vcpu->arch.pid);
3512
3513         mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3514               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3515
3516         mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3517
3518         mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3519         mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3520         mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3521         mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3522
3523         mtspr(SPRN_AMOR, ~0UL);
3524
3525         mtspr(SPRN_LPCR, lpcr);
3526         isync();
3527
3528         kvmppc_xive_push_vcpu(vcpu);
3529
3530         mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3531         mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3532
3533         trap = __kvmhv_vcpu_entry_p9(vcpu);
3534
3535         /* Advance host PURR/SPURR by the amount used by guest */
3536         purr = mfspr(SPRN_PURR);
3537         spurr = mfspr(SPRN_SPURR);
3538         mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3539               purr - vcpu->arch.purr);
3540         mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3541               spurr - vcpu->arch.spurr);
3542         vcpu->arch.purr = purr;
3543         vcpu->arch.spurr = spurr;
3544
3545         vcpu->arch.ic = mfspr(SPRN_IC);
3546         vcpu->arch.pid = mfspr(SPRN_PID);
3547         vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3548
3549         vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3550         vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3551         vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3552         vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3553
3554         /* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3555         mtspr(SPRN_PSSCR, host_psscr |
3556               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3557         mtspr(SPRN_HFSCR, host_hfscr);
3558         mtspr(SPRN_CIABR, host_ciabr);
3559         mtspr(SPRN_DAWR0, host_dawr0);
3560         mtspr(SPRN_DAWRX0, host_dawrx0);
3561         if (cpu_has_feature(CPU_FTR_DAWR1)) {
3562                 mtspr(SPRN_DAWR1, host_dawr1);
3563                 mtspr(SPRN_DAWRX1, host_dawrx1);
3564         }
3565         mtspr(SPRN_PID, host_pidr);
3566
3567         /*
3568          * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3569          * case we interrupted the guest between a tlbie and a ptesync.
3570          */
3571         asm volatile("eieio; tlbsync; ptesync");
3572
3573         /*
3574          * cp_abort is required if the processor supports local copy-paste
3575          * to clear the copy buffer that was under control of the guest.
3576          */
3577         if (cpu_has_feature(CPU_FTR_ARCH_31))
3578                 asm volatile(PPC_CP_ABORT);
3579
3580         mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid);    /* restore host LPID */
3581         isync();
3582
3583         vc->dpdes = mfspr(SPRN_DPDES);
3584         vc->vtb = mfspr(SPRN_VTB);
3585         mtspr(SPRN_DPDES, 0);
3586         if (vc->pcr)
3587                 mtspr(SPRN_PCR, PCR_MASK);
3588
3589         if (vc->tb_offset_applied) {
3590                 u64 new_tb = mftb() - vc->tb_offset_applied;
3591                 mtspr(SPRN_TBU40, new_tb);
3592                 tb = mftb();
3593                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3594                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3595                 vc->tb_offset_applied = 0;
3596         }
3597
3598         mtspr(SPRN_HDEC, 0x7fffffff);
3599         mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3600
3601         return trap;
3602 }
3603
3604 /*
3605  * Virtual-mode guest entry for POWER9 and later when the host and
3606  * guest are both using the radix MMU.  The LPIDR has already been set.
3607  */
3608 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3609                          unsigned long lpcr)
3610 {
3611         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3612         unsigned long host_dscr = mfspr(SPRN_DSCR);
3613         unsigned long host_tidr = mfspr(SPRN_TIDR);
3614         unsigned long host_iamr = mfspr(SPRN_IAMR);
3615         unsigned long host_amr = mfspr(SPRN_AMR);
3616         unsigned long host_fscr = mfspr(SPRN_FSCR);
3617         s64 dec;
3618         u64 tb;
3619         int trap, save_pmu;
3620
3621         dec = mfspr(SPRN_DEC);
3622         tb = mftb();
3623         if (dec < 0)
3624                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3625         local_paca->kvm_hstate.dec_expires = dec + tb;
3626         if (local_paca->kvm_hstate.dec_expires < time_limit)
3627                 time_limit = local_paca->kvm_hstate.dec_expires;
3628
3629         vcpu->arch.ceded = 0;
3630
3631         kvmhv_save_host_pmu();          /* saves it to PACA kvm_hstate */
3632
3633         kvmppc_subcore_enter_guest();
3634
3635         vc->entry_exit_map = 1;
3636         vc->in_guest = 1;
3637
3638         if (vcpu->arch.vpa.pinned_addr) {
3639                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3640                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3641                 lp->yield_count = cpu_to_be32(yield_count);
3642                 vcpu->arch.vpa.dirty = 1;
3643         }
3644
3645         if (cpu_has_feature(CPU_FTR_TM) ||
3646             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3647                 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3648
3649         kvmhv_load_guest_pmu(vcpu);
3650
3651         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3652         load_fp_state(&vcpu->arch.fp);
3653 #ifdef CONFIG_ALTIVEC
3654         load_vr_state(&vcpu->arch.vr);
3655 #endif
3656         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3657
3658         mtspr(SPRN_DSCR, vcpu->arch.dscr);
3659         mtspr(SPRN_IAMR, vcpu->arch.iamr);
3660         mtspr(SPRN_PSPB, vcpu->arch.pspb);
3661         mtspr(SPRN_FSCR, vcpu->arch.fscr);
3662         mtspr(SPRN_TAR, vcpu->arch.tar);
3663         mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3664         mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3665         mtspr(SPRN_BESCR, vcpu->arch.bescr);
3666         mtspr(SPRN_WORT, vcpu->arch.wort);
3667         mtspr(SPRN_TIDR, vcpu->arch.tid);
3668         mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3669         mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3670         mtspr(SPRN_AMR, vcpu->arch.amr);
3671         mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3672
3673         if (!(vcpu->arch.ctrl & 1))
3674                 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3675
3676         mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3677
3678         if (kvmhv_on_pseries()) {
3679                 /*
3680                  * We need to save and restore the guest visible part of the
3681                  * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3682                  * doesn't do this for us. Note only required if pseries since
3683                  * this is done in kvmhv_load_hv_regs_and_go() below otherwise.
3684                  */
3685                 unsigned long host_psscr;
3686                 /* call our hypervisor to load up HV regs and go */
3687                 struct hv_guest_state hvregs;
3688
3689                 host_psscr = mfspr(SPRN_PSSCR_PR);
3690                 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3691                 kvmhv_save_hv_regs(vcpu, &hvregs);
3692                 hvregs.lpcr = lpcr;
3693                 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3694                 hvregs.version = HV_GUEST_STATE_VERSION;
3695                 if (vcpu->arch.nested) {
3696                         hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3697                         hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3698                 } else {
3699                         hvregs.lpid = vcpu->kvm->arch.lpid;
3700                         hvregs.vcpu_token = vcpu->vcpu_id;
3701                 }
3702                 hvregs.hdec_expiry = time_limit;
3703                 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3704                                           __pa(&vcpu->arch.regs));
3705                 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3706                 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3707                 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3708                 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3709                 vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3710                 mtspr(SPRN_PSSCR_PR, host_psscr);
3711
3712                 /* H_CEDE has to be handled now, not later */
3713                 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3714                     kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3715                         kvmppc_nested_cede(vcpu);
3716                         kvmppc_set_gpr(vcpu, 3, 0);
3717                         trap = 0;
3718                 }
3719         } else {
3720                 trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3721         }
3722
3723         vcpu->arch.slb_max = 0;
3724         dec = mfspr(SPRN_DEC);
3725         if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
3726                 dec = (s32) dec;
3727         tb = mftb();
3728         vcpu->arch.dec_expires = dec + tb;
3729         vcpu->cpu = -1;
3730         vcpu->arch.thread_cpu = -1;
3731         vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3732
3733         vcpu->arch.iamr = mfspr(SPRN_IAMR);
3734         vcpu->arch.pspb = mfspr(SPRN_PSPB);
3735         vcpu->arch.fscr = mfspr(SPRN_FSCR);
3736         vcpu->arch.tar = mfspr(SPRN_TAR);
3737         vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3738         vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3739         vcpu->arch.bescr = mfspr(SPRN_BESCR);
3740         vcpu->arch.wort = mfspr(SPRN_WORT);
3741         vcpu->arch.tid = mfspr(SPRN_TIDR);
3742         vcpu->arch.amr = mfspr(SPRN_AMR);
3743         vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3744         vcpu->arch.dscr = mfspr(SPRN_DSCR);
3745
3746         mtspr(SPRN_PSPB, 0);
3747         mtspr(SPRN_WORT, 0);
3748         mtspr(SPRN_UAMOR, 0);
3749         mtspr(SPRN_DSCR, host_dscr);
3750         mtspr(SPRN_TIDR, host_tidr);
3751         mtspr(SPRN_IAMR, host_iamr);
3752         mtspr(SPRN_PSPB, 0);
3753
3754         if (host_amr != vcpu->arch.amr)
3755                 mtspr(SPRN_AMR, host_amr);
3756
3757         if (host_fscr != vcpu->arch.fscr)
3758                 mtspr(SPRN_FSCR, host_fscr);
3759
3760         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3761         store_fp_state(&vcpu->arch.fp);
3762 #ifdef CONFIG_ALTIVEC
3763         store_vr_state(&vcpu->arch.vr);
3764 #endif
3765         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3766
3767         if (cpu_has_feature(CPU_FTR_TM) ||
3768             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3769                 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3770
3771         save_pmu = 1;
3772         if (vcpu->arch.vpa.pinned_addr) {
3773                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3774                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3775                 lp->yield_count = cpu_to_be32(yield_count);
3776                 vcpu->arch.vpa.dirty = 1;
3777                 save_pmu = lp->pmcregs_in_use;
3778         }
3779         /* Must save pmu if this guest is capable of running nested guests */
3780         save_pmu |= nesting_enabled(vcpu->kvm);
3781
3782         kvmhv_save_guest_pmu(vcpu, save_pmu);
3783
3784         vc->entry_exit_map = 0x101;
3785         vc->in_guest = 0;
3786
3787         mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3788         mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3789
3790         kvmhv_load_host_pmu();
3791
3792         kvmppc_subcore_exit_guest();
3793
3794         return trap;
3795 }
3796
3797 /*
3798  * Wait for some other vcpu thread to execute us, and
3799  * wake us up when we need to handle something in the host.
3800  */
3801 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3802                                  struct kvm_vcpu *vcpu, int wait_state)
3803 {
3804         DEFINE_WAIT(wait);
3805
3806         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3807         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3808                 spin_unlock(&vc->lock);
3809                 schedule();
3810                 spin_lock(&vc->lock);
3811         }
3812         finish_wait(&vcpu->arch.cpu_run, &wait);
3813 }
3814
3815 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3816 {
3817         if (!halt_poll_ns_grow)
3818                 return;
3819
3820         vc->halt_poll_ns *= halt_poll_ns_grow;
3821         if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3822                 vc->halt_poll_ns = halt_poll_ns_grow_start;
3823 }
3824
3825 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3826 {
3827         if (halt_poll_ns_shrink == 0)
3828                 vc->halt_poll_ns = 0;
3829         else
3830                 vc->halt_poll_ns /= halt_poll_ns_shrink;
3831 }
3832
3833 #ifdef CONFIG_KVM_XICS
3834 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3835 {
3836         if (!xics_on_xive())
3837                 return false;
3838         return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3839                 vcpu->arch.xive_saved_state.cppr;
3840 }
3841 #else
3842 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3843 {
3844         return false;
3845 }
3846 #endif /* CONFIG_KVM_XICS */
3847
3848 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3849 {
3850         if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3851             kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3852                 return true;
3853
3854         return false;
3855 }
3856
3857 /*
3858  * Check to see if any of the runnable vcpus on the vcore have pending
3859  * exceptions or are no longer ceded
3860  */
3861 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3862 {
3863         struct kvm_vcpu *vcpu;
3864         int i;
3865
3866         for_each_runnable_thread(i, vcpu, vc) {
3867                 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3868                         return 1;
3869         }
3870
3871         return 0;
3872 }
3873
3874 /*
3875  * All the vcpus in this vcore are idle, so wait for a decrementer
3876  * or external interrupt to one of the vcpus.  vc->lock is held.
3877  */
3878 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3879 {
3880         ktime_t cur, start_poll, start_wait;
3881         int do_sleep = 1;
3882         u64 block_ns;
3883
3884         /* Poll for pending exceptions and ceded state */
3885         cur = start_poll = ktime_get();
3886         if (vc->halt_poll_ns) {
3887                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3888                 ++vc->runner->stat.halt_attempted_poll;
3889
3890                 vc->vcore_state = VCORE_POLLING;
3891                 spin_unlock(&vc->lock);
3892
3893                 do {
3894                         if (kvmppc_vcore_check_block(vc)) {
3895                                 do_sleep = 0;
3896                                 break;
3897                         }
3898                         cur = ktime_get();
3899                 } while (single_task_running() && ktime_before(cur, stop));
3900
3901                 spin_lock(&vc->lock);
3902                 vc->vcore_state = VCORE_INACTIVE;
3903
3904                 if (!do_sleep) {
3905                         ++vc->runner->stat.halt_successful_poll;
3906                         goto out;
3907                 }
3908         }
3909
3910         prepare_to_rcuwait(&vc->wait);
3911         set_current_state(TASK_INTERRUPTIBLE);
3912         if (kvmppc_vcore_check_block(vc)) {
3913                 finish_rcuwait(&vc->wait);
3914                 do_sleep = 0;
3915                 /* If we polled, count this as a successful poll */
3916                 if (vc->halt_poll_ns)
3917                         ++vc->runner->stat.halt_successful_poll;
3918                 goto out;
3919         }
3920
3921         start_wait = ktime_get();
3922
3923         vc->vcore_state = VCORE_SLEEPING;
3924         trace_kvmppc_vcore_blocked(vc, 0);
3925         spin_unlock(&vc->lock);
3926         schedule();
3927         finish_rcuwait(&vc->wait);
3928         spin_lock(&vc->lock);
3929         vc->vcore_state = VCORE_INACTIVE;
3930         trace_kvmppc_vcore_blocked(vc, 1);
3931         ++vc->runner->stat.halt_successful_wait;
3932
3933         cur = ktime_get();
3934
3935 out:
3936         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3937
3938         /* Attribute wait time */
3939         if (do_sleep) {
3940                 vc->runner->stat.halt_wait_ns +=
3941                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
3942                 /* Attribute failed poll time */
3943                 if (vc->halt_poll_ns)
3944                         vc->runner->stat.halt_poll_fail_ns +=
3945                                 ktime_to_ns(start_wait) -
3946                                 ktime_to_ns(start_poll);
3947         } else {
3948                 /* Attribute successful poll time */
3949                 if (vc->halt_poll_ns)
3950                         vc->runner->stat.halt_poll_success_ns +=
3951                                 ktime_to_ns(cur) -
3952                                 ktime_to_ns(start_poll);
3953         }
3954
3955         /* Adjust poll time */
3956         if (halt_poll_ns) {
3957                 if (block_ns <= vc->halt_poll_ns)
3958                         ;
3959                 /* We slept and blocked for longer than the max halt time */
3960                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3961                         shrink_halt_poll_ns(vc);
3962                 /* We slept and our poll time is too small */
3963                 else if (vc->halt_poll_ns < halt_poll_ns &&
3964                                 block_ns < halt_poll_ns)
3965                         grow_halt_poll_ns(vc);
3966                 if (vc->halt_poll_ns > halt_poll_ns)
3967                         vc->halt_poll_ns = halt_poll_ns;
3968         } else
3969                 vc->halt_poll_ns = 0;
3970
3971         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3972 }
3973
3974 /*
3975  * This never fails for a radix guest, as none of the operations it does
3976  * for a radix guest can fail or have a way to report failure.
3977  * kvmhv_run_single_vcpu() relies on this fact.
3978  */
3979 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3980 {
3981         int r = 0;
3982         struct kvm *kvm = vcpu->kvm;
3983
3984         mutex_lock(&kvm->arch.mmu_setup_lock);
3985         if (!kvm->arch.mmu_ready) {
3986                 if (!kvm_is_radix(kvm))
3987                         r = kvmppc_hv_setup_htab_rma(vcpu);
3988                 if (!r) {
3989                         if (cpu_has_feature(CPU_FTR_ARCH_300))
3990                                 kvmppc_setup_partition_table(kvm);
3991                         kvm->arch.mmu_ready = 1;
3992                 }
3993         }
3994         mutex_unlock(&kvm->arch.mmu_setup_lock);
3995         return r;
3996 }
3997
3998 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
3999 {
4000         struct kvm_run *run = vcpu->run;
4001         int n_ceded, i, r;
4002         struct kvmppc_vcore *vc;
4003         struct kvm_vcpu *v;
4004
4005         trace_kvmppc_run_vcpu_enter(vcpu);
4006
4007         run->exit_reason = 0;
4008         vcpu->arch.ret = RESUME_GUEST;
4009         vcpu->arch.trap = 0;
4010         kvmppc_update_vpas(vcpu);
4011
4012         /*
4013          * Synchronize with other threads in this virtual core
4014          */
4015         vc = vcpu->arch.vcore;
4016         spin_lock(&vc->lock);
4017         vcpu->arch.ceded = 0;
4018         vcpu->arch.run_task = current;
4019         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4020         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4021         vcpu->arch.busy_preempt = TB_NIL;
4022         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4023         ++vc->n_runnable;
4024
4025         /*
4026          * This happens the first time this is called for a vcpu.
4027          * If the vcore is already running, we may be able to start
4028          * this thread straight away and have it join in.
4029          */
4030         if (!signal_pending(current)) {
4031                 if ((vc->vcore_state == VCORE_PIGGYBACK ||
4032                      vc->vcore_state == VCORE_RUNNING) &&
4033                            !VCORE_IS_EXITING(vc)) {
4034                         kvmppc_create_dtl_entry(vcpu, vc);
4035                         kvmppc_start_thread(vcpu, vc);
4036                         trace_kvm_guest_enter(vcpu);
4037                 } else if (vc->vcore_state == VCORE_SLEEPING) {
4038                         rcuwait_wake_up(&vc->wait);
4039                 }
4040
4041         }
4042
4043         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4044                !signal_pending(current)) {
4045                 /* See if the MMU is ready to go */
4046                 if (!vcpu->kvm->arch.mmu_ready) {
4047                         spin_unlock(&vc->lock);
4048                         r = kvmhv_setup_mmu(vcpu);
4049                         spin_lock(&vc->lock);
4050                         if (r) {
4051                                 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4052                                 run->fail_entry.
4053                                         hardware_entry_failure_reason = 0;
4054                                 vcpu->arch.ret = r;
4055                                 break;
4056                         }
4057                 }
4058
4059                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4060                         kvmppc_vcore_end_preempt(vc);
4061
4062                 if (vc->vcore_state != VCORE_INACTIVE) {
4063                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4064                         continue;
4065                 }
4066                 for_each_runnable_thread(i, v, vc) {
4067                         kvmppc_core_prepare_to_enter(v);
4068                         if (signal_pending(v->arch.run_task)) {
4069                                 kvmppc_remove_runnable(vc, v);
4070                                 v->stat.signal_exits++;
4071                                 v->run->exit_reason = KVM_EXIT_INTR;
4072                                 v->arch.ret = -EINTR;
4073                                 wake_up(&v->arch.cpu_run);
4074                         }
4075                 }
4076                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4077                         break;
4078                 n_ceded = 0;
4079                 for_each_runnable_thread(i, v, vc) {
4080                         if (!kvmppc_vcpu_woken(v))
4081                                 n_ceded += v->arch.ceded;
4082                         else
4083                                 v->arch.ceded = 0;
4084                 }
4085                 vc->runner = vcpu;
4086                 if (n_ceded == vc->n_runnable) {
4087                         kvmppc_vcore_blocked(vc);
4088                 } else if (need_resched()) {
4089                         kvmppc_vcore_preempt(vc);
4090                         /* Let something else run */
4091                         cond_resched_lock(&vc->lock);
4092                         if (vc->vcore_state == VCORE_PREEMPT)
4093                                 kvmppc_vcore_end_preempt(vc);
4094                 } else {
4095                         kvmppc_run_core(vc);
4096                 }
4097                 vc->runner = NULL;
4098         }
4099
4100         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4101                (vc->vcore_state == VCORE_RUNNING ||
4102                 vc->vcore_state == VCORE_EXITING ||
4103                 vc->vcore_state == VCORE_PIGGYBACK))
4104                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4105
4106         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4107                 kvmppc_vcore_end_preempt(vc);
4108
4109         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4110                 kvmppc_remove_runnable(vc, vcpu);
4111                 vcpu->stat.signal_exits++;
4112                 run->exit_reason = KVM_EXIT_INTR;
4113                 vcpu->arch.ret = -EINTR;
4114         }
4115
4116         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4117                 /* Wake up some vcpu to run the core */
4118                 i = -1;
4119                 v = next_runnable_thread(vc, &i);
4120                 wake_up(&v->arch.cpu_run);
4121         }
4122
4123         trace_kvmppc_run_vcpu_exit(vcpu);
4124         spin_unlock(&vc->lock);
4125         return vcpu->arch.ret;
4126 }
4127
4128 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4129                           unsigned long lpcr)
4130 {
4131         struct kvm_run *run = vcpu->run;
4132         int trap, r, pcpu;
4133         int srcu_idx, lpid;
4134         struct kvmppc_vcore *vc;
4135         struct kvm *kvm = vcpu->kvm;
4136         struct kvm_nested_guest *nested = vcpu->arch.nested;
4137
4138         trace_kvmppc_run_vcpu_enter(vcpu);
4139
4140         run->exit_reason = 0;
4141         vcpu->arch.ret = RESUME_GUEST;
4142         vcpu->arch.trap = 0;
4143
4144         vc = vcpu->arch.vcore;
4145         vcpu->arch.ceded = 0;
4146         vcpu->arch.run_task = current;
4147         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4148         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4149         vcpu->arch.busy_preempt = TB_NIL;
4150         vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4151         vc->runnable_threads[0] = vcpu;
4152         vc->n_runnable = 1;
4153         vc->runner = vcpu;
4154
4155         /* See if the MMU is ready to go */
4156         if (!kvm->arch.mmu_ready)
4157                 kvmhv_setup_mmu(vcpu);
4158
4159         if (need_resched())
4160                 cond_resched();
4161
4162         kvmppc_update_vpas(vcpu);
4163
4164         init_vcore_to_run(vc);
4165         vc->preempt_tb = TB_NIL;
4166
4167         preempt_disable();
4168         pcpu = smp_processor_id();
4169         vc->pcpu = pcpu;
4170         kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4171
4172         local_irq_disable();
4173         hard_irq_disable();
4174         if (signal_pending(current))
4175                 goto sigpend;
4176         if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4177                 goto out;
4178
4179         if (!nested) {
4180                 kvmppc_core_prepare_to_enter(vcpu);
4181                 if (vcpu->arch.doorbell_request) {
4182                         vc->dpdes = 1;
4183                         smp_wmb();
4184                         vcpu->arch.doorbell_request = 0;
4185                 }
4186                 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4187                              &vcpu->arch.pending_exceptions))
4188                         lpcr |= LPCR_MER;
4189         } else if (vcpu->arch.pending_exceptions ||
4190                    vcpu->arch.doorbell_request ||
4191                    xive_interrupt_pending(vcpu)) {
4192                 vcpu->arch.ret = RESUME_HOST;
4193                 goto out;
4194         }
4195
4196         kvmppc_clear_host_core(pcpu);
4197
4198         local_paca->kvm_hstate.napping = 0;
4199         local_paca->kvm_hstate.kvm_split_mode = NULL;
4200         kvmppc_start_thread(vcpu, vc);
4201         kvmppc_create_dtl_entry(vcpu, vc);
4202         trace_kvm_guest_enter(vcpu);
4203
4204         vc->vcore_state = VCORE_RUNNING;
4205         trace_kvmppc_run_core(vc, 0);
4206
4207         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4208                 lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4209                 mtspr(SPRN_LPID, lpid);
4210                 isync();
4211                 kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4212         }
4213
4214         guest_enter_irqoff();
4215
4216         srcu_idx = srcu_read_lock(&kvm->srcu);
4217
4218         this_cpu_disable_ftrace();
4219
4220         /* Tell lockdep that we're about to enable interrupts */
4221         trace_hardirqs_on();
4222
4223         trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4224         vcpu->arch.trap = trap;
4225
4226         trace_hardirqs_off();
4227
4228         this_cpu_enable_ftrace();
4229
4230         srcu_read_unlock(&kvm->srcu, srcu_idx);
4231
4232         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4233                 mtspr(SPRN_LPID, kvm->arch.host_lpid);
4234                 isync();
4235         }
4236
4237         set_irq_happened(trap);
4238
4239         kvmppc_set_host_core(pcpu);
4240
4241         guest_exit_irqoff();
4242
4243         local_irq_enable();
4244
4245         cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4246
4247         preempt_enable();
4248
4249         /*
4250          * cancel pending decrementer exception if DEC is now positive, or if
4251          * entering a nested guest in which case the decrementer is now owned
4252          * by L2 and the L1 decrementer is provided in hdec_expires
4253          */
4254         if (kvmppc_core_pending_dec(vcpu) &&
4255                         ((get_tb() < vcpu->arch.dec_expires) ||
4256                          (trap == BOOK3S_INTERRUPT_SYSCALL &&
4257                           kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4258                 kvmppc_core_dequeue_dec(vcpu);
4259
4260         trace_kvm_guest_exit(vcpu);
4261         r = RESUME_GUEST;
4262         if (trap) {
4263                 if (!nested)
4264                         r = kvmppc_handle_exit_hv(vcpu, current);
4265                 else
4266                         r = kvmppc_handle_nested_exit(vcpu);
4267         }
4268         vcpu->arch.ret = r;
4269
4270         if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4271             !kvmppc_vcpu_woken(vcpu)) {
4272                 kvmppc_set_timer(vcpu);
4273                 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4274                         if (signal_pending(current)) {
4275                                 vcpu->stat.signal_exits++;
4276                                 run->exit_reason = KVM_EXIT_INTR;
4277                                 vcpu->arch.ret = -EINTR;
4278                                 break;
4279                         }
4280                         spin_lock(&vc->lock);
4281                         kvmppc_vcore_blocked(vc);
4282                         spin_unlock(&vc->lock);
4283                 }
4284         }
4285         vcpu->arch.ceded = 0;
4286
4287         vc->vcore_state = VCORE_INACTIVE;
4288         trace_kvmppc_run_core(vc, 1);
4289
4290  done:
4291         kvmppc_remove_runnable(vc, vcpu);
4292         trace_kvmppc_run_vcpu_exit(vcpu);
4293
4294         return vcpu->arch.ret;
4295
4296  sigpend:
4297         vcpu->stat.signal_exits++;
4298         run->exit_reason = KVM_EXIT_INTR;
4299         vcpu->arch.ret = -EINTR;
4300  out:
4301         local_irq_enable();
4302         preempt_enable();
4303         goto done;
4304 }
4305
4306 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
4307 {
4308         struct kvm_run *run = vcpu->run;
4309         int r;
4310         int srcu_idx;
4311         unsigned long ebb_regs[3] = {}; /* shut up GCC */
4312         unsigned long user_tar = 0;
4313         unsigned int user_vrsave;
4314         struct kvm *kvm;
4315
4316         if (!vcpu->arch.sane) {
4317                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4318                 return -EINVAL;
4319         }
4320
4321         /*
4322          * Don't allow entry with a suspended transaction, because
4323          * the guest entry/exit code will lose it.
4324          * If the guest has TM enabled, save away their TM-related SPRs
4325          * (they will get restored by the TM unavailable interrupt).
4326          */
4327 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4328         if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4329             (current->thread.regs->msr & MSR_TM)) {
4330                 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4331                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4332                         run->fail_entry.hardware_entry_failure_reason = 0;
4333                         return -EINVAL;
4334                 }
4335                 /* Enable TM so we can read the TM SPRs */
4336                 mtmsr(mfmsr() | MSR_TM);
4337                 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4338                 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4339                 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4340                 current->thread.regs->msr &= ~MSR_TM;
4341         }
4342 #endif
4343
4344         /*
4345          * Force online to 1 for the sake of old userspace which doesn't
4346          * set it.
4347          */
4348         if (!vcpu->arch.online) {
4349                 atomic_inc(&vcpu->arch.vcore->online_count);
4350                 vcpu->arch.online = 1;
4351         }
4352
4353         kvmppc_core_prepare_to_enter(vcpu);
4354
4355         /* No need to go into the guest when all we'll do is come back out */
4356         if (signal_pending(current)) {
4357                 run->exit_reason = KVM_EXIT_INTR;
4358                 return -EINTR;
4359         }
4360
4361         kvm = vcpu->kvm;
4362         atomic_inc(&kvm->arch.vcpus_running);
4363         /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4364         smp_mb();
4365
4366         flush_all_to_thread(current);
4367
4368         /* Save userspace EBB and other register values */
4369         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4370                 ebb_regs[0] = mfspr(SPRN_EBBHR);
4371                 ebb_regs[1] = mfspr(SPRN_EBBRR);
4372                 ebb_regs[2] = mfspr(SPRN_BESCR);
4373                 user_tar = mfspr(SPRN_TAR);
4374         }
4375         user_vrsave = mfspr(SPRN_VRSAVE);
4376
4377         vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4378         vcpu->arch.pgdir = kvm->mm->pgd;
4379         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4380
4381         do {
4382                 /*
4383                  * The TLB prefetch bug fixup is only in the kvmppc_run_vcpu
4384                  * path, which also handles hash and dependent threads mode.
4385                  */
4386                 if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4387                     !cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
4388                         r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
4389                                                   vcpu->arch.vcore->lpcr);
4390                 else
4391                         r = kvmppc_run_vcpu(vcpu);
4392
4393                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4394                     !(vcpu->arch.shregs.msr & MSR_PR)) {
4395                         trace_kvm_hcall_enter(vcpu);
4396                         r = kvmppc_pseries_do_hcall(vcpu);
4397                         trace_kvm_hcall_exit(vcpu, r);
4398                         kvmppc_core_prepare_to_enter(vcpu);
4399                 } else if (r == RESUME_PAGE_FAULT) {
4400                         srcu_idx = srcu_read_lock(&kvm->srcu);
4401                         r = kvmppc_book3s_hv_page_fault(vcpu,
4402                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4403                         srcu_read_unlock(&kvm->srcu, srcu_idx);
4404                 } else if (r == RESUME_PASSTHROUGH) {
4405                         if (WARN_ON(xics_on_xive()))
4406                                 r = H_SUCCESS;
4407                         else
4408                                 r = kvmppc_xics_rm_complete(vcpu, 0);
4409                 }
4410         } while (is_kvmppc_resume_guest(r));
4411
4412         /* Restore userspace EBB and other register values */
4413         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4414                 mtspr(SPRN_EBBHR, ebb_regs[0]);
4415                 mtspr(SPRN_EBBRR, ebb_regs[1]);
4416                 mtspr(SPRN_BESCR, ebb_regs[2]);
4417                 mtspr(SPRN_TAR, user_tar);
4418                 mtspr(SPRN_FSCR, current->thread.fscr);
4419         }
4420         mtspr(SPRN_VRSAVE, user_vrsave);
4421
4422         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4423         atomic_dec(&kvm->arch.vcpus_running);
4424         return r;
4425 }
4426
4427 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4428                                      int shift, int sllp)
4429 {
4430         (*sps)->page_shift = shift;
4431         (*sps)->slb_enc = sllp;
4432         (*sps)->enc[0].page_shift = shift;
4433         (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4434         /*
4435          * Add 16MB MPSS support (may get filtered out by userspace)
4436          */
4437         if (shift != 24) {
4438                 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4439                 if (penc != -1) {
4440                         (*sps)->enc[1].page_shift = 24;
4441                         (*sps)->enc[1].pte_enc = penc;
4442                 }
4443         }
4444         (*sps)++;
4445 }
4446
4447 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4448                                          struct kvm_ppc_smmu_info *info)
4449 {
4450         struct kvm_ppc_one_seg_page_size *sps;
4451
4452         /*
4453          * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4454          * POWER7 doesn't support keys for instruction accesses,
4455          * POWER8 and POWER9 do.
4456          */
4457         info->data_keys = 32;
4458         info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4459
4460         /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4461         info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4462         info->slb_size = 32;
4463
4464         /* We only support these sizes for now, and no muti-size segments */
4465         sps = &info->sps[0];
4466         kvmppc_add_seg_page_size(&sps, 12, 0);
4467         kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4468         kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4469
4470         /* If running as a nested hypervisor, we don't support HPT guests */
4471         if (kvmhv_on_pseries())
4472                 info->flags |= KVM_PPC_NO_HASH;
4473
4474         return 0;
4475 }
4476
4477 /*
4478  * Get (and clear) the dirty memory log for a memory slot.
4479  */
4480 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4481                                          struct kvm_dirty_log *log)
4482 {
4483         struct kvm_memslots *slots;
4484         struct kvm_memory_slot *memslot;
4485         int i, r;
4486         unsigned long n;
4487         unsigned long *buf, *p;
4488         struct kvm_vcpu *vcpu;
4489
4490         mutex_lock(&kvm->slots_lock);
4491
4492         r = -EINVAL;
4493         if (log->slot >= KVM_USER_MEM_SLOTS)
4494                 goto out;
4495
4496         slots = kvm_memslots(kvm);
4497         memslot = id_to_memslot(slots, log->slot);
4498         r = -ENOENT;
4499         if (!memslot || !memslot->dirty_bitmap)
4500                 goto out;
4501
4502         /*
4503          * Use second half of bitmap area because both HPT and radix
4504          * accumulate bits in the first half.
4505          */
4506         n = kvm_dirty_bitmap_bytes(memslot);
4507         buf = memslot->dirty_bitmap + n / sizeof(long);
4508         memset(buf, 0, n);
4509
4510         if (kvm_is_radix(kvm))
4511                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4512         else
4513                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4514         if (r)
4515                 goto out;
4516
4517         /*
4518          * We accumulate dirty bits in the first half of the
4519          * memslot's dirty_bitmap area, for when pages are paged
4520          * out or modified by the host directly.  Pick up these
4521          * bits and add them to the map.
4522          */
4523         p = memslot->dirty_bitmap;
4524         for (i = 0; i < n / sizeof(long); ++i)
4525                 buf[i] |= xchg(&p[i], 0);
4526
4527         /* Harvest dirty bits from VPA and DTL updates */
4528         /* Note: we never modify the SLB shadow buffer areas */
4529         kvm_for_each_vcpu(i, vcpu, kvm) {
4530                 spin_lock(&vcpu->arch.vpa_update_lock);
4531                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4532                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4533                 spin_unlock(&vcpu->arch.vpa_update_lock);
4534         }
4535
4536         r = -EFAULT;
4537         if (copy_to_user(log->dirty_bitmap, buf, n))
4538                 goto out;
4539
4540         r = 0;
4541 out:
4542         mutex_unlock(&kvm->slots_lock);
4543         return r;
4544 }
4545
4546 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
4547 {
4548         vfree(slot->arch.rmap);
4549         slot->arch.rmap = NULL;
4550 }
4551
4552 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4553                                         struct kvm_memory_slot *slot,
4554                                         const struct kvm_userspace_memory_region *mem,
4555                                         enum kvm_mr_change change)
4556 {
4557         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4558
4559         if (change == KVM_MR_CREATE) {
4560                 slot->arch.rmap = vzalloc(array_size(npages,
4561                                           sizeof(*slot->arch.rmap)));
4562                 if (!slot->arch.rmap)
4563                         return -ENOMEM;
4564         }
4565
4566         return 0;
4567 }
4568
4569 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4570                                 const struct kvm_userspace_memory_region *mem,
4571                                 const struct kvm_memory_slot *old,
4572                                 const struct kvm_memory_slot *new,
4573                                 enum kvm_mr_change change)
4574 {
4575         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4576
4577         /*
4578          * If we are making a new memslot, it might make
4579          * some address that was previously cached as emulated
4580          * MMIO be no longer emulated MMIO, so invalidate
4581          * all the caches of emulated MMIO translations.
4582          */
4583         if (npages)
4584                 atomic64_inc(&kvm->arch.mmio_update);
4585
4586         /*
4587          * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4588          * have already called kvm_arch_flush_shadow_memslot() to
4589          * flush shadow mappings.  For KVM_MR_CREATE we have no
4590          * previous mappings.  So the only case to handle is
4591          * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4592          * has been changed.
4593          * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4594          * to get rid of any THP PTEs in the partition-scoped page tables
4595          * so we can track dirtiness at the page level; we flush when
4596          * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4597          * using THP PTEs.
4598          */
4599         if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4600             ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4601                 kvmppc_radix_flush_memslot(kvm, old);
4602         /*
4603          * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4604          */
4605         if (!kvm->arch.secure_guest)
4606                 return;
4607
4608         switch (change) {
4609         case KVM_MR_CREATE:
4610                 /*
4611                  * @TODO kvmppc_uvmem_memslot_create() can fail and
4612                  * return error. Fix this.
4613                  */
4614                 kvmppc_uvmem_memslot_create(kvm, new);
4615                 break;
4616         case KVM_MR_DELETE:
4617                 kvmppc_uvmem_memslot_delete(kvm, old);
4618                 break;
4619         default:
4620                 /* TODO: Handle KVM_MR_MOVE */
4621                 break;
4622         }
4623 }
4624
4625 /*
4626  * Update LPCR values in kvm->arch and in vcores.
4627  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4628  * of kvm->arch.lpcr update).
4629  */
4630 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4631 {
4632         long int i;
4633         u32 cores_done = 0;
4634
4635         if ((kvm->arch.lpcr & mask) == lpcr)
4636                 return;
4637
4638         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4639
4640         for (i = 0; i < KVM_MAX_VCORES; ++i) {
4641                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4642                 if (!vc)
4643                         continue;
4644                 spin_lock(&vc->lock);
4645                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4646                 spin_unlock(&vc->lock);
4647                 if (++cores_done >= kvm->arch.online_vcores)
4648                         break;
4649         }
4650 }
4651
4652 void kvmppc_setup_partition_table(struct kvm *kvm)
4653 {
4654         unsigned long dw0, dw1;
4655
4656         if (!kvm_is_radix(kvm)) {
4657                 /* PS field - page size for VRMA */
4658                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4659                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4660                 /* HTABSIZE and HTABORG fields */
4661                 dw0 |= kvm->arch.sdr1;
4662
4663                 /* Second dword as set by userspace */
4664                 dw1 = kvm->arch.process_table;
4665         } else {
4666                 dw0 = PATB_HR | radix__get_tree_size() |
4667                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4668                 dw1 = PATB_GR | kvm->arch.process_table;
4669         }
4670         kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4671 }
4672
4673 /*
4674  * Set up HPT (hashed page table) and RMA (real-mode area).
4675  * Must be called with kvm->arch.mmu_setup_lock held.
4676  */
4677 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4678 {
4679         int err = 0;
4680         struct kvm *kvm = vcpu->kvm;
4681         unsigned long hva;
4682         struct kvm_memory_slot *memslot;
4683         struct vm_area_struct *vma;
4684         unsigned long lpcr = 0, senc;
4685         unsigned long psize, porder;
4686         int srcu_idx;
4687
4688         /* Allocate hashed page table (if not done already) and reset it */
4689         if (!kvm->arch.hpt.virt) {
4690                 int order = KVM_DEFAULT_HPT_ORDER;
4691                 struct kvm_hpt_info info;
4692
4693                 err = kvmppc_allocate_hpt(&info, order);
4694                 /* If we get here, it means userspace didn't specify a
4695                  * size explicitly.  So, try successively smaller
4696                  * sizes if the default failed. */
4697                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4698                         err  = kvmppc_allocate_hpt(&info, order);
4699
4700                 if (err < 0) {
4701                         pr_err("KVM: Couldn't alloc HPT\n");
4702                         goto out;
4703                 }
4704
4705                 kvmppc_set_hpt(kvm, &info);
4706         }
4707
4708         /* Look up the memslot for guest physical address 0 */
4709         srcu_idx = srcu_read_lock(&kvm->srcu);
4710         memslot = gfn_to_memslot(kvm, 0);
4711
4712         /* We must have some memory at 0 by now */
4713         err = -EINVAL;
4714         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4715                 goto out_srcu;
4716
4717         /* Look up the VMA for the start of this memory slot */
4718         hva = memslot->userspace_addr;
4719         mmap_read_lock(kvm->mm);
4720         vma = find_vma(kvm->mm, hva);
4721         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4722                 goto up_out;
4723
4724         psize = vma_kernel_pagesize(vma);
4725
4726         mmap_read_unlock(kvm->mm);
4727
4728         /* We can handle 4k, 64k or 16M pages in the VRMA */
4729         if (psize >= 0x1000000)
4730                 psize = 0x1000000;
4731         else if (psize >= 0x10000)
4732                 psize = 0x10000;
4733         else
4734                 psize = 0x1000;
4735         porder = __ilog2(psize);
4736
4737         senc = slb_pgsize_encoding(psize);
4738         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4739                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4740         /* Create HPTEs in the hash page table for the VRMA */
4741         kvmppc_map_vrma(vcpu, memslot, porder);
4742
4743         /* Update VRMASD field in the LPCR */
4744         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4745                 /* the -4 is to account for senc values starting at 0x10 */
4746                 lpcr = senc << (LPCR_VRMASD_SH - 4);
4747                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4748         }
4749
4750         /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4751         smp_wmb();
4752         err = 0;
4753  out_srcu:
4754         srcu_read_unlock(&kvm->srcu, srcu_idx);
4755  out:
4756         return err;
4757
4758  up_out:
4759         mmap_read_unlock(kvm->mm);
4760         goto out_srcu;
4761 }
4762
4763 /*
4764  * Must be called with kvm->arch.mmu_setup_lock held and
4765  * mmu_ready = 0 and no vcpus running.
4766  */
4767 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4768 {
4769         if (nesting_enabled(kvm))
4770                 kvmhv_release_all_nested(kvm);
4771         kvmppc_rmap_reset(kvm);
4772         kvm->arch.process_table = 0;
4773         /* Mutual exclusion with kvm_unmap_hva_range etc. */
4774         spin_lock(&kvm->mmu_lock);
4775         kvm->arch.radix = 0;
4776         spin_unlock(&kvm->mmu_lock);
4777         kvmppc_free_radix(kvm);
4778         kvmppc_update_lpcr(kvm, LPCR_VPM1,
4779                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4780         return 0;
4781 }
4782
4783 /*
4784  * Must be called with kvm->arch.mmu_setup_lock held and
4785  * mmu_ready = 0 and no vcpus running.
4786  */
4787 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4788 {
4789         int err;
4790
4791         err = kvmppc_init_vm_radix(kvm);
4792         if (err)
4793                 return err;
4794         kvmppc_rmap_reset(kvm);
4795         /* Mutual exclusion with kvm_unmap_hva_range etc. */
4796         spin_lock(&kvm->mmu_lock);
4797         kvm->arch.radix = 1;
4798         spin_unlock(&kvm->mmu_lock);
4799         kvmppc_free_hpt(&kvm->arch.hpt);
4800         kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4801                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4802         return 0;
4803 }
4804
4805 #ifdef CONFIG_KVM_XICS
4806 /*
4807  * Allocate a per-core structure for managing state about which cores are
4808  * running in the host versus the guest and for exchanging data between
4809  * real mode KVM and CPU running in the host.
4810  * This is only done for the first VM.
4811  * The allocated structure stays even if all VMs have stopped.
4812  * It is only freed when the kvm-hv module is unloaded.
4813  * It's OK for this routine to fail, we just don't support host
4814  * core operations like redirecting H_IPI wakeups.
4815  */
4816 void kvmppc_alloc_host_rm_ops(void)
4817 {
4818         struct kvmppc_host_rm_ops *ops;
4819         unsigned long l_ops;
4820         int cpu, core;
4821         int size;
4822
4823         /* Not the first time here ? */
4824         if (kvmppc_host_rm_ops_hv != NULL)
4825                 return;
4826
4827         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4828         if (!ops)
4829                 return;
4830
4831         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4832         ops->rm_core = kzalloc(size, GFP_KERNEL);
4833
4834         if (!ops->rm_core) {
4835                 kfree(ops);
4836                 return;
4837         }
4838
4839         cpus_read_lock();
4840
4841         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4842                 if (!cpu_online(cpu))
4843                         continue;
4844
4845                 core = cpu >> threads_shift;
4846                 ops->rm_core[core].rm_state.in_host = 1;
4847         }
4848
4849         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4850
4851         /*
4852          * Make the contents of the kvmppc_host_rm_ops structure visible
4853          * to other CPUs before we assign it to the global variable.
4854          * Do an atomic assignment (no locks used here), but if someone
4855          * beats us to it, just free our copy and return.
4856          */
4857         smp_wmb();
4858         l_ops = (unsigned long) ops;
4859
4860         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4861                 cpus_read_unlock();
4862                 kfree(ops->rm_core);
4863                 kfree(ops);
4864                 return;
4865         }
4866
4867         cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4868                                              "ppc/kvm_book3s:prepare",
4869                                              kvmppc_set_host_core,
4870                                              kvmppc_clear_host_core);
4871         cpus_read_unlock();
4872 }
4873
4874 void kvmppc_free_host_rm_ops(void)
4875 {
4876         if (kvmppc_host_rm_ops_hv) {
4877                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4878                 kfree(kvmppc_host_rm_ops_hv->rm_core);
4879                 kfree(kvmppc_host_rm_ops_hv);
4880                 kvmppc_host_rm_ops_hv = NULL;
4881         }
4882 }
4883 #endif
4884
4885 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4886 {
4887         unsigned long lpcr, lpid;
4888         char buf[32];
4889         int ret;
4890
4891         mutex_init(&kvm->arch.uvmem_lock);
4892         INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
4893         mutex_init(&kvm->arch.mmu_setup_lock);
4894
4895         /* Allocate the guest's logical partition ID */
4896
4897         lpid = kvmppc_alloc_lpid();
4898         if ((long)lpid < 0)
4899                 return -ENOMEM;
4900         kvm->arch.lpid = lpid;
4901
4902         kvmppc_alloc_host_rm_ops();
4903
4904         kvmhv_vm_nested_init(kvm);
4905
4906         /*
4907          * Since we don't flush the TLB when tearing down a VM,
4908          * and this lpid might have previously been used,
4909          * make sure we flush on each core before running the new VM.
4910          * On POWER9, the tlbie in mmu_partition_table_set_entry()
4911          * does this flush for us.
4912          */
4913         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4914                 cpumask_setall(&kvm->arch.need_tlb_flush);
4915
4916         /* Start out with the default set of hcalls enabled */
4917         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4918                sizeof(kvm->arch.enabled_hcalls));
4919
4920         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4921                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4922
4923         /* Init LPCR for virtual RMA mode */
4924         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4925                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
4926                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4927                 lpcr &= LPCR_PECE | LPCR_LPES;
4928         } else {
4929                 lpcr = 0;
4930         }
4931         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4932                 LPCR_VPM0 | LPCR_VPM1;
4933         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4934                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4935         /* On POWER8 turn on online bit to enable PURR/SPURR */
4936         if (cpu_has_feature(CPU_FTR_ARCH_207S))
4937                 lpcr |= LPCR_ONL;
4938         /*
4939          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4940          * Set HVICE bit to enable hypervisor virtualization interrupts.
4941          * Set HEIC to prevent OS interrupts to go to hypervisor (should
4942          * be unnecessary but better safe than sorry in case we re-enable
4943          * EE in HV mode with this LPCR still set)
4944          */
4945         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4946                 lpcr &= ~LPCR_VPM0;
4947                 lpcr |= LPCR_HVICE | LPCR_HEIC;
4948
4949                 /*
4950                  * If xive is enabled, we route 0x500 interrupts directly
4951                  * to the guest.
4952                  */
4953                 if (xics_on_xive())
4954                         lpcr |= LPCR_LPES;
4955         }
4956
4957         /*
4958          * If the host uses radix, the guest starts out as radix.
4959          */
4960         if (radix_enabled()) {
4961                 kvm->arch.radix = 1;
4962                 kvm->arch.mmu_ready = 1;
4963                 lpcr &= ~LPCR_VPM1;
4964                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4965                 ret = kvmppc_init_vm_radix(kvm);
4966                 if (ret) {
4967                         kvmppc_free_lpid(kvm->arch.lpid);
4968                         return ret;
4969                 }
4970                 kvmppc_setup_partition_table(kvm);
4971         }
4972
4973         kvm->arch.lpcr = lpcr;
4974
4975         /* Initialization for future HPT resizes */
4976         kvm->arch.resize_hpt = NULL;
4977
4978         /*
4979          * Work out how many sets the TLB has, for the use of
4980          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4981          */
4982         if (cpu_has_feature(CPU_FTR_ARCH_31)) {
4983                 /*
4984                  * P10 will flush all the congruence class with a single tlbiel
4985                  */
4986                 kvm->arch.tlb_sets = 1;
4987         } else if (radix_enabled())
4988                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
4989         else if (cpu_has_feature(CPU_FTR_ARCH_300))
4990                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
4991         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4992                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
4993         else
4994                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
4995
4996         /*
4997          * Track that we now have a HV mode VM active. This blocks secondary
4998          * CPU threads from coming online.
4999          * On POWER9, we only need to do this if the "indep_threads_mode"
5000          * module parameter has been set to N.
5001          */
5002         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5003                 if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
5004                         pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
5005                         kvm->arch.threads_indep = true;
5006                 } else {
5007                         kvm->arch.threads_indep = indep_threads_mode;
5008                 }
5009         }
5010         if (!kvm->arch.threads_indep)
5011                 kvm_hv_vm_activated();
5012
5013         /*
5014          * Initialize smt_mode depending on processor.
5015          * POWER8 and earlier have to use "strict" threading, where
5016          * all vCPUs in a vcore have to run on the same (sub)core,
5017          * whereas on POWER9 the threads can each run a different
5018          * guest.
5019          */
5020         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5021                 kvm->arch.smt_mode = threads_per_subcore;
5022         else
5023                 kvm->arch.smt_mode = 1;
5024         kvm->arch.emul_smt_mode = 1;
5025
5026         /*
5027          * Create a debugfs directory for the VM
5028          */
5029         snprintf(buf, sizeof(buf), "vm%d", current->pid);
5030         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
5031         kvmppc_mmu_debugfs_init(kvm);
5032         if (radix_enabled())
5033                 kvmhv_radix_debugfs_init(kvm);
5034
5035         return 0;
5036 }
5037
5038 static void kvmppc_free_vcores(struct kvm *kvm)
5039 {
5040         long int i;
5041
5042         for (i = 0; i < KVM_MAX_VCORES; ++i)
5043                 kfree(kvm->arch.vcores[i]);
5044         kvm->arch.online_vcores = 0;
5045 }
5046
5047 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5048 {
5049         debugfs_remove_recursive(kvm->arch.debugfs_dir);
5050
5051         if (!kvm->arch.threads_indep)
5052                 kvm_hv_vm_deactivated();
5053
5054         kvmppc_free_vcores(kvm);
5055
5056
5057         if (kvm_is_radix(kvm))
5058                 kvmppc_free_radix(kvm);
5059         else
5060                 kvmppc_free_hpt(&kvm->arch.hpt);
5061
5062         /* Perform global invalidation and return lpid to the pool */
5063         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5064                 if (nesting_enabled(kvm))
5065                         kvmhv_release_all_nested(kvm);
5066                 kvm->arch.process_table = 0;
5067                 if (kvm->arch.secure_guest)
5068                         uv_svm_terminate(kvm->arch.lpid);
5069                 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5070         }
5071
5072         kvmppc_free_lpid(kvm->arch.lpid);
5073
5074         kvmppc_free_pimap(kvm);
5075 }
5076
5077 /* We don't need to emulate any privileged instructions or dcbz */
5078 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5079                                      unsigned int inst, int *advance)
5080 {
5081         return EMULATE_FAIL;
5082 }
5083
5084 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5085                                         ulong spr_val)
5086 {
5087         return EMULATE_FAIL;
5088 }
5089
5090 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5091                                         ulong *spr_val)
5092 {
5093         return EMULATE_FAIL;
5094 }
5095
5096 static int kvmppc_core_check_processor_compat_hv(void)
5097 {
5098         if (cpu_has_feature(CPU_FTR_HVMODE) &&
5099             cpu_has_feature(CPU_FTR_ARCH_206))
5100                 return 0;
5101
5102         /* POWER9 in radix mode is capable of being a nested hypervisor. */
5103         if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5104                 return 0;
5105
5106         return -EIO;
5107 }
5108
5109 #ifdef CONFIG_KVM_XICS
5110
5111 void kvmppc_free_pimap(struct kvm *kvm)
5112 {
5113         kfree(kvm->arch.pimap);
5114 }
5115
5116 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5117 {
5118         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5119 }
5120
5121 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5122 {
5123         struct irq_desc *desc;
5124         struct kvmppc_irq_map *irq_map;
5125         struct kvmppc_passthru_irqmap *pimap;
5126         struct irq_chip *chip;
5127         int i, rc = 0;
5128
5129         if (!kvm_irq_bypass)
5130                 return 1;
5131
5132         desc = irq_to_desc(host_irq);
5133         if (!desc)
5134                 return -EIO;
5135
5136         mutex_lock(&kvm->lock);
5137
5138         pimap = kvm->arch.pimap;
5139         if (pimap == NULL) {
5140                 /* First call, allocate structure to hold IRQ map */
5141                 pimap = kvmppc_alloc_pimap();
5142                 if (pimap == NULL) {
5143                         mutex_unlock(&kvm->lock);
5144                         return -ENOMEM;
5145                 }
5146                 kvm->arch.pimap = pimap;
5147         }
5148
5149         /*
5150          * For now, we only support interrupts for which the EOI operation
5151          * is an OPAL call followed by a write to XIRR, since that's
5152          * what our real-mode EOI code does, or a XIVE interrupt
5153          */
5154         chip = irq_data_get_irq_chip(&desc->irq_data);
5155         if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5156                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5157                         host_irq, guest_gsi);
5158                 mutex_unlock(&kvm->lock);
5159                 return -ENOENT;
5160         }
5161
5162         /*
5163          * See if we already have an entry for this guest IRQ number.
5164          * If it's mapped to a hardware IRQ number, that's an error,
5165          * otherwise re-use this entry.
5166          */
5167         for (i = 0; i < pimap->n_mapped; i++) {
5168                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5169                         if (pimap->mapped[i].r_hwirq) {
5170                                 mutex_unlock(&kvm->lock);
5171                                 return -EINVAL;
5172                         }
5173                         break;
5174                 }
5175         }
5176
5177         if (i == KVMPPC_PIRQ_MAPPED) {
5178                 mutex_unlock(&kvm->lock);
5179                 return -EAGAIN;         /* table is full */
5180         }
5181
5182         irq_map = &pimap->mapped[i];
5183
5184         irq_map->v_hwirq = guest_gsi;
5185         irq_map->desc = desc;
5186
5187         /*
5188          * Order the above two stores before the next to serialize with
5189          * the KVM real mode handler.
5190          */
5191         smp_wmb();
5192         irq_map->r_hwirq = desc->irq_data.hwirq;
5193
5194         if (i == pimap->n_mapped)
5195                 pimap->n_mapped++;
5196
5197         if (xics_on_xive())
5198                 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5199         else
5200                 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5201         if (rc)
5202                 irq_map->r_hwirq = 0;
5203
5204         mutex_unlock(&kvm->lock);
5205
5206         return 0;
5207 }
5208
5209 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5210 {
5211         struct irq_desc *desc;
5212         struct kvmppc_passthru_irqmap *pimap;
5213         int i, rc = 0;
5214
5215         if (!kvm_irq_bypass)
5216                 return 0;
5217
5218         desc = irq_to_desc(host_irq);
5219         if (!desc)
5220                 return -EIO;
5221
5222         mutex_lock(&kvm->lock);
5223         if (!kvm->arch.pimap)
5224                 goto unlock;
5225
5226         pimap = kvm->arch.pimap;
5227
5228         for (i = 0; i < pimap->n_mapped; i++) {
5229                 if (guest_gsi == pimap->mapped[i].v_hwirq)
5230                         break;
5231         }
5232
5233         if (i == pimap->n_mapped) {
5234                 mutex_unlock(&kvm->lock);
5235                 return -ENODEV;
5236         }
5237
5238         if (xics_on_xive())
5239                 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5240         else
5241                 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5242
5243         /* invalidate the entry (what do do on error from the above ?) */
5244         pimap->mapped[i].r_hwirq = 0;
5245
5246         /*
5247          * We don't free this structure even when the count goes to
5248          * zero. The structure is freed when we destroy the VM.
5249          */
5250  unlock:
5251         mutex_unlock(&kvm->lock);
5252         return rc;
5253 }
5254
5255 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5256                                              struct irq_bypass_producer *prod)
5257 {
5258         int ret = 0;
5259         struct kvm_kernel_irqfd *irqfd =
5260                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5261
5262         irqfd->producer = prod;
5263
5264         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5265         if (ret)
5266                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5267                         prod->irq, irqfd->gsi, ret);
5268
5269         return ret;
5270 }
5271
5272 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5273                                               struct irq_bypass_producer *prod)
5274 {
5275         int ret;
5276         struct kvm_kernel_irqfd *irqfd =
5277                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5278
5279         irqfd->producer = NULL;
5280
5281         /*
5282          * When producer of consumer is unregistered, we change back to
5283          * default external interrupt handling mode - KVM real mode
5284          * will switch back to host.
5285          */
5286         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5287         if (ret)
5288                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5289                         prod->irq, irqfd->gsi, ret);
5290 }
5291 #endif
5292
5293 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5294                                  unsigned int ioctl, unsigned long arg)
5295 {
5296         struct kvm *kvm __maybe_unused = filp->private_data;
5297         void __user *argp = (void __user *)arg;
5298         long r;
5299
5300         switch (ioctl) {
5301
5302         case KVM_PPC_ALLOCATE_HTAB: {
5303                 u32 htab_order;
5304
5305                 /* If we're a nested hypervisor, we currently only support radix */
5306                 if (kvmhv_on_pseries()) {
5307                         r = -EOPNOTSUPP;
5308                         break;
5309                 }
5310
5311                 r = -EFAULT;
5312                 if (get_user(htab_order, (u32 __user *)argp))
5313                         break;
5314                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5315                 if (r)
5316                         break;
5317                 r = 0;
5318                 break;
5319         }
5320
5321         case KVM_PPC_GET_HTAB_FD: {
5322                 struct kvm_get_htab_fd ghf;
5323
5324                 r = -EFAULT;
5325                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5326                         break;
5327                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5328                 break;
5329         }
5330
5331         case KVM_PPC_RESIZE_HPT_PREPARE: {
5332                 struct kvm_ppc_resize_hpt rhpt;
5333
5334                 r = -EFAULT;
5335                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5336                         break;
5337
5338                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5339                 break;
5340         }
5341
5342         case KVM_PPC_RESIZE_HPT_COMMIT: {
5343                 struct kvm_ppc_resize_hpt rhpt;
5344
5345                 r = -EFAULT;
5346                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5347                         break;
5348
5349                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5350                 break;
5351         }
5352
5353         default:
5354                 r = -ENOTTY;
5355         }
5356
5357         return r;
5358 }
5359
5360 /*
5361  * List of hcall numbers to enable by default.
5362  * For compatibility with old userspace, we enable by default
5363  * all hcalls that were implemented before the hcall-enabling
5364  * facility was added.  Note this list should not include H_RTAS.
5365  */
5366 static unsigned int default_hcall_list[] = {
5367         H_REMOVE,
5368         H_ENTER,
5369         H_READ,
5370         H_PROTECT,
5371         H_BULK_REMOVE,
5372         H_GET_TCE,
5373         H_PUT_TCE,
5374         H_SET_DABR,
5375         H_SET_XDABR,
5376         H_CEDE,
5377         H_PROD,
5378         H_CONFER,
5379         H_REGISTER_VPA,
5380 #ifdef CONFIG_KVM_XICS
5381         H_EOI,
5382         H_CPPR,
5383         H_IPI,
5384         H_IPOLL,
5385         H_XIRR,
5386         H_XIRR_X,
5387 #endif
5388         0
5389 };
5390
5391 static void init_default_hcalls(void)
5392 {
5393         int i;
5394         unsigned int hcall;
5395
5396         for (i = 0; default_hcall_list[i]; ++i) {
5397                 hcall = default_hcall_list[i];
5398                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5399                 __set_bit(hcall / 4, default_enabled_hcalls);
5400         }
5401 }
5402
5403 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5404 {
5405         unsigned long lpcr;
5406         int radix;
5407         int err;
5408
5409         /* If not on a POWER9, reject it */
5410         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5411                 return -ENODEV;
5412
5413         /* If any unknown flags set, reject it */
5414         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5415                 return -EINVAL;
5416
5417         /* GR (guest radix) bit in process_table field must match */
5418         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5419         if (!!(cfg->process_table & PATB_GR) != radix)
5420                 return -EINVAL;
5421
5422         /* Process table size field must be reasonable, i.e. <= 24 */
5423         if ((cfg->process_table & PRTS_MASK) > 24)
5424                 return -EINVAL;
5425
5426         /* We can change a guest to/from radix now, if the host is radix */
5427         if (radix && !radix_enabled())
5428                 return -EINVAL;
5429
5430         /* If we're a nested hypervisor, we currently only support radix */
5431         if (kvmhv_on_pseries() && !radix)
5432                 return -EINVAL;
5433
5434         mutex_lock(&kvm->arch.mmu_setup_lock);
5435         if (radix != kvm_is_radix(kvm)) {
5436                 if (kvm->arch.mmu_ready) {
5437                         kvm->arch.mmu_ready = 0;
5438                         /* order mmu_ready vs. vcpus_running */
5439                         smp_mb();
5440                         if (atomic_read(&kvm->arch.vcpus_running)) {
5441                                 kvm->arch.mmu_ready = 1;
5442                                 err = -EBUSY;
5443                                 goto out_unlock;
5444                         }
5445                 }
5446                 if (radix)
5447                         err = kvmppc_switch_mmu_to_radix(kvm);
5448                 else
5449                         err = kvmppc_switch_mmu_to_hpt(kvm);
5450                 if (err)
5451                         goto out_unlock;
5452         }
5453
5454         kvm->arch.process_table = cfg->process_table;
5455         kvmppc_setup_partition_table(kvm);
5456
5457         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5458         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5459         err = 0;
5460
5461  out_unlock:
5462         mutex_unlock(&kvm->arch.mmu_setup_lock);
5463         return err;
5464 }
5465
5466 static int kvmhv_enable_nested(struct kvm *kvm)
5467 {
5468         if (!nested)
5469                 return -EPERM;
5470         if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5471                 return -ENODEV;
5472
5473         /* kvm == NULL means the caller is testing if the capability exists */
5474         if (kvm)
5475                 kvm->arch.nested_enable = true;
5476         return 0;
5477 }
5478
5479 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5480                                  int size)
5481 {
5482         int rc = -EINVAL;
5483
5484         if (kvmhv_vcpu_is_radix(vcpu)) {
5485                 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5486
5487                 if (rc > 0)
5488                         rc = -EINVAL;
5489         }
5490
5491         /* For now quadrants are the only way to access nested guest memory */
5492         if (rc && vcpu->arch.nested)
5493                 rc = -EAGAIN;
5494
5495         return rc;
5496 }
5497
5498 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5499                                 int size)
5500 {
5501         int rc = -EINVAL;
5502
5503         if (kvmhv_vcpu_is_radix(vcpu)) {
5504                 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5505
5506                 if (rc > 0)
5507                         rc = -EINVAL;
5508         }
5509
5510         /* For now quadrants are the only way to access nested guest memory */
5511         if (rc && vcpu->arch.nested)
5512                 rc = -EAGAIN;
5513
5514         return rc;
5515 }
5516
5517 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5518 {
5519         unpin_vpa(kvm, vpa);
5520         vpa->gpa = 0;
5521         vpa->pinned_addr = NULL;
5522         vpa->dirty = false;
5523         vpa->update_pending = 0;
5524 }
5525
5526 /*
5527  * Enable a guest to become a secure VM, or test whether
5528  * that could be enabled.
5529  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
5530  * tested (kvm == NULL) or enabled (kvm != NULL).
5531  */
5532 static int kvmhv_enable_svm(struct kvm *kvm)
5533 {
5534         if (!kvmppc_uvmem_available())
5535                 return -EINVAL;
5536         if (kvm)
5537                 kvm->arch.svm_enabled = 1;
5538         return 0;
5539 }
5540
5541 /*
5542  *  IOCTL handler to turn off secure mode of guest
5543  *
5544  * - Release all device pages
5545  * - Issue ucall to terminate the guest on the UV side
5546  * - Unpin the VPA pages.
5547  * - Reinit the partition scoped page tables
5548  */
5549 static int kvmhv_svm_off(struct kvm *kvm)
5550 {
5551         struct kvm_vcpu *vcpu;
5552         int mmu_was_ready;
5553         int srcu_idx;
5554         int ret = 0;
5555         int i;
5556
5557         if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5558                 return ret;
5559
5560         mutex_lock(&kvm->arch.mmu_setup_lock);
5561         mmu_was_ready = kvm->arch.mmu_ready;
5562         if (kvm->arch.mmu_ready) {
5563                 kvm->arch.mmu_ready = 0;
5564                 /* order mmu_ready vs. vcpus_running */
5565                 smp_mb();
5566                 if (atomic_read(&kvm->arch.vcpus_running)) {
5567                         kvm->arch.mmu_ready = 1;
5568                         ret = -EBUSY;
5569                         goto out;
5570                 }
5571         }
5572
5573         srcu_idx = srcu_read_lock(&kvm->srcu);
5574         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5575                 struct kvm_memory_slot *memslot;
5576                 struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5577
5578                 if (!slots)
5579                         continue;
5580
5581                 kvm_for_each_memslot(memslot, slots) {
5582                         kvmppc_uvmem_drop_pages(memslot, kvm, true);
5583                         uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5584                 }
5585         }
5586         srcu_read_unlock(&kvm->srcu, srcu_idx);
5587
5588         ret = uv_svm_terminate(kvm->arch.lpid);
5589         if (ret != U_SUCCESS) {
5590                 ret = -EINVAL;
5591                 goto out;
5592         }
5593
5594         /*
5595          * When secure guest is reset, all the guest pages are sent
5596          * to UV via UV_PAGE_IN before the non-boot vcpus get a
5597          * chance to run and unpin their VPA pages. Unpinning of all
5598          * VPA pages is done here explicitly so that VPA pages
5599          * can be migrated to the secure side.
5600          *
5601          * This is required to for the secure SMP guest to reboot
5602          * correctly.
5603          */
5604         kvm_for_each_vcpu(i, vcpu, kvm) {
5605                 spin_lock(&vcpu->arch.vpa_update_lock);
5606                 unpin_vpa_reset(kvm, &vcpu->arch.dtl);
5607                 unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
5608                 unpin_vpa_reset(kvm, &vcpu->arch.vpa);
5609                 spin_unlock(&vcpu->arch.vpa_update_lock);
5610         }
5611
5612         kvmppc_setup_partition_table(kvm);
5613         kvm->arch.secure_guest = 0;
5614         kvm->arch.mmu_ready = mmu_was_ready;
5615 out:
5616         mutex_unlock(&kvm->arch.mmu_setup_lock);
5617         return ret;
5618 }
5619
5620 static int kvmhv_enable_dawr1(struct kvm *kvm)
5621 {
5622         if (!cpu_has_feature(CPU_FTR_DAWR1))
5623                 return -ENODEV;
5624
5625         /* kvm == NULL means the caller is testing if the capability exists */
5626         if (kvm)
5627                 kvm->arch.dawr1_enabled = true;
5628         return 0;
5629 }
5630
5631 static bool kvmppc_hash_v3_possible(void)
5632 {
5633         if (radix_enabled() && no_mixing_hpt_and_radix)
5634                 return false;
5635
5636         return cpu_has_feature(CPU_FTR_ARCH_300) &&
5637                 cpu_has_feature(CPU_FTR_HVMODE);
5638 }
5639
5640 static struct kvmppc_ops kvm_ops_hv = {
5641         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5642         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5643         .get_one_reg = kvmppc_get_one_reg_hv,
5644         .set_one_reg = kvmppc_set_one_reg_hv,
5645         .vcpu_load   = kvmppc_core_vcpu_load_hv,
5646         .vcpu_put    = kvmppc_core_vcpu_put_hv,
5647         .inject_interrupt = kvmppc_inject_interrupt_hv,
5648         .set_msr     = kvmppc_set_msr_hv,
5649         .vcpu_run    = kvmppc_vcpu_run_hv,
5650         .vcpu_create = kvmppc_core_vcpu_create_hv,
5651         .vcpu_free   = kvmppc_core_vcpu_free_hv,
5652         .check_requests = kvmppc_core_check_requests_hv,
5653         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5654         .flush_memslot  = kvmppc_core_flush_memslot_hv,
5655         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5656         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5657         .unmap_hva_range = kvm_unmap_hva_range_hv,
5658         .age_hva  = kvm_age_hva_hv,
5659         .test_age_hva = kvm_test_age_hva_hv,
5660         .set_spte_hva = kvm_set_spte_hva_hv,
5661         .free_memslot = kvmppc_core_free_memslot_hv,
5662         .init_vm =  kvmppc_core_init_vm_hv,
5663         .destroy_vm = kvmppc_core_destroy_vm_hv,
5664         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5665         .emulate_op = kvmppc_core_emulate_op_hv,
5666         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5667         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5668         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5669         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5670         .hcall_implemented = kvmppc_hcall_impl_hv,
5671 #ifdef CONFIG_KVM_XICS
5672         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5673         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5674 #endif
5675         .configure_mmu = kvmhv_configure_mmu,
5676         .get_rmmu_info = kvmhv_get_rmmu_info,
5677         .set_smt_mode = kvmhv_set_smt_mode,
5678         .enable_nested = kvmhv_enable_nested,
5679         .load_from_eaddr = kvmhv_load_from_eaddr,
5680         .store_to_eaddr = kvmhv_store_to_eaddr,
5681         .enable_svm = kvmhv_enable_svm,
5682         .svm_off = kvmhv_svm_off,
5683         .enable_dawr1 = kvmhv_enable_dawr1,
5684         .hash_v3_possible = kvmppc_hash_v3_possible,
5685 };
5686
5687 static int kvm_init_subcore_bitmap(void)
5688 {
5689         int i, j;
5690         int nr_cores = cpu_nr_cores();
5691         struct sibling_subcore_state *sibling_subcore_state;
5692
5693         for (i = 0; i < nr_cores; i++) {
5694                 int first_cpu = i * threads_per_core;
5695                 int node = cpu_to_node(first_cpu);
5696
5697                 /* Ignore if it is already allocated. */
5698                 if (paca_ptrs[first_cpu]->sibling_subcore_state)
5699                         continue;
5700
5701                 sibling_subcore_state =
5702                         kzalloc_node(sizeof(struct sibling_subcore_state),
5703                                                         GFP_KERNEL, node);
5704                 if (!sibling_subcore_state)
5705                         return -ENOMEM;
5706
5707
5708                 for (j = 0; j < threads_per_core; j++) {
5709                         int cpu = first_cpu + j;
5710
5711                         paca_ptrs[cpu]->sibling_subcore_state =
5712                                                 sibling_subcore_state;
5713                 }
5714         }
5715         return 0;
5716 }
5717
5718 static int kvmppc_radix_possible(void)
5719 {
5720         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5721 }
5722
5723 static int kvmppc_book3s_init_hv(void)
5724 {
5725         int r;
5726
5727         if (!tlbie_capable) {
5728                 pr_err("KVM-HV: Host does not support TLBIE\n");
5729                 return -ENODEV;
5730         }
5731
5732         /*
5733          * FIXME!! Do we need to check on all cpus ?
5734          */
5735         r = kvmppc_core_check_processor_compat_hv();
5736         if (r < 0)
5737                 return -ENODEV;
5738
5739         r = kvmhv_nested_init();
5740         if (r)
5741                 return r;
5742
5743         r = kvm_init_subcore_bitmap();
5744         if (r)
5745                 return r;
5746
5747         /*
5748          * We need a way of accessing the XICS interrupt controller,
5749          * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5750          * indirectly, via OPAL.
5751          */
5752 #ifdef CONFIG_SMP
5753         if (!xics_on_xive() && !kvmhv_on_pseries() &&
5754             !local_paca->kvm_hstate.xics_phys) {
5755                 struct device_node *np;
5756
5757                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5758                 if (!np) {
5759                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5760                         return -ENODEV;
5761                 }
5762                 /* presence of intc confirmed - node can be dropped again */
5763                 of_node_put(np);
5764         }
5765 #endif
5766
5767         kvm_ops_hv.owner = THIS_MODULE;
5768         kvmppc_hv_ops = &kvm_ops_hv;
5769
5770         init_default_hcalls();
5771
5772         init_vcore_lists();
5773
5774         r = kvmppc_mmu_hv_init();
5775         if (r)
5776                 return r;
5777
5778         if (kvmppc_radix_possible())
5779                 r = kvmppc_radix_init();
5780
5781         /*
5782          * POWER9 chips before version 2.02 can't have some threads in
5783          * HPT mode and some in radix mode on the same core.
5784          */
5785         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5786                 unsigned int pvr = mfspr(SPRN_PVR);
5787                 if ((pvr >> 16) == PVR_POWER9 &&
5788                     (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5789                      ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5790                         no_mixing_hpt_and_radix = true;
5791         }
5792
5793         r = kvmppc_uvmem_init();
5794         if (r < 0)
5795                 pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
5796
5797         return r;
5798 }
5799
5800 static void kvmppc_book3s_exit_hv(void)
5801 {
5802         kvmppc_uvmem_free();
5803         kvmppc_free_host_rm_ops();
5804         if (kvmppc_radix_possible())
5805                 kvmppc_radix_exit();
5806         kvmppc_hv_ops = NULL;
5807         kvmhv_nested_exit();
5808 }
5809
5810 module_init(kvmppc_book3s_init_hv);
5811 module_exit(kvmppc_book3s_exit_hv);
5812 MODULE_LICENSE("GPL");
5813 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5814 MODULE_ALIAS("devname:kvm");