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                 /*
 807                  * KVM does not support mflags=2 (AIL=2) and AIL=1 is reserved.
 808                  * Keep this in synch with kvmppc_filter_guest_lpcr_hv.
 809                  */
 810                 if (mflags != 0 && mflags != 3)
 811                         return H_UNSUPPORTED_FLAG_START;
 812                 return H_TOO_HARD;
 813         default:
 814                 return H_TOO_HARD;
 815         }
 816 }
 817
 818 /* Copy guest memory in place - must reside within a single memslot */
 819 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
 820                                   unsigned long len)
 821 {
 822         struct kvm_memory_slot *to_memslot = NULL;
 823         struct kvm_memory_slot *from_memslot = NULL;
 824         unsigned long to_addr, from_addr;
 825         int r;
 826
 827         /* Get HPA for from address */
 828         from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
 829         if (!from_memslot)
 830                 return -EFAULT;
 831         if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
 832                              << PAGE_SHIFT))
 833                 return -EINVAL;
 834         from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
 835         if (kvm_is_error_hva(from_addr))
 836                 return -EFAULT;
 837         from_addr |= (from & (PAGE_SIZE - 1));
 838
 839         /* Get HPA for to address */
 840         to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
 841         if (!to_memslot)
 842                 return -EFAULT;
 843         if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
 844                            << PAGE_SHIFT))
 845                 return -EINVAL;
 846         to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
 847         if (kvm_is_error_hva(to_addr))
 848                 return -EFAULT;
 849         to_addr |= (to & (PAGE_SIZE - 1));
 850
 851         /* Perform copy */
 852         r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
 853                              len);
 854         if (r)
 855                 return -EFAULT;
 856         mark_page_dirty(kvm, to >> PAGE_SHIFT);
 857         return 0;
 858 }
 859
 860 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
 861                                unsigned long dest, unsigned long src)
 862 {
 863         u64 pg_sz = SZ_4K;              /* 4K page size */
 864         u64 pg_mask = SZ_4K - 1;
 865         int ret;
 866
 867         /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
 868         if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
 869                       H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
 870                 return H_PARAMETER;
 871
 872         /* dest (and src if copy_page flag set) must be page aligned */
 873         if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
 874                 return H_PARAMETER;
 875
 876         /* zero and/or copy the page as determined by the flags */
 877         if (flags & H_COPY_PAGE) {
 878                 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
 879                 if (ret < 0)
 880                         return H_PARAMETER;
 881         } else if (flags & H_ZERO_PAGE) {
 882                 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
 883                 if (ret < 0)
 884                         return H_PARAMETER;
 885         }
 886
 887         /* We can ignore the remaining flags */
 888
 889         return H_SUCCESS;
 890 }
 891
 892 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 893 {
 894         struct kvmppc_vcore *vcore = target->arch.vcore;
 895
 896         /*
 897          * We expect to have been called by the real mode handler
 898          * (kvmppc_rm_h_confer()) which would have directly returned
 899          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 900          * have useful work to do and should not confer) so we don't
 901          * recheck that here.
 902          */
 903
 904         spin_lock(&vcore->lock);
 905         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 906             vcore->vcore_state != VCORE_INACTIVE &&
 907             vcore->runner)
 908                 target = vcore->runner;
 909         spin_unlock(&vcore->lock);
 910
 911         return kvm_vcpu_yield_to(target);
 912 }
 913
 914 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 915 {
 916         int yield_count = 0;
 917         struct lppaca *lppaca;
 918
 919         spin_lock(&vcpu->arch.vpa_update_lock);
 920         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 921         if (lppaca)
 922                 yield_count = be32_to_cpu(lppaca->yield_count);
 923         spin_unlock(&vcpu->arch.vpa_update_lock);
 924         return yield_count;
 925 }
 926
 927 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 928 {
 929         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 930         unsigned long target, ret = H_SUCCESS;
 931         int yield_count;
 932         struct kvm_vcpu *tvcpu;
 933         int idx, rc;
 934
 935         if (req <= MAX_HCALL_OPCODE &&
 936             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 937                 return RESUME_HOST;
 938
 939         switch (req) {
 940         case H_CEDE:
 941                 break;
 942         case H_PROD:
 943                 target = kvmppc_get_gpr(vcpu, 4);
 944                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 945                 if (!tvcpu) {
 946                         ret = H_PARAMETER;
 947                         break;
 948                 }
 949                 tvcpu->arch.prodded = 1;
 950                 smp_mb();
 951                 if (tvcpu->arch.ceded)
 952                         kvmppc_fast_vcpu_kick_hv(tvcpu);
 953                 break;
 954         case H_CONFER:
 955                 target = kvmppc_get_gpr(vcpu, 4);
 956                 if (target == -1)
 957                         break;
 958                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 959                 if (!tvcpu) {
 960                         ret = H_PARAMETER;
 961                         break;
 962                 }
 963                 yield_count = kvmppc_get_gpr(vcpu, 5);
 964                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 965                         break;
 966                 kvm_arch_vcpu_yield_to(tvcpu);
 967                 break;
 968         case H_REGISTER_VPA:
 969                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 970                                         kvmppc_get_gpr(vcpu, 5),
 971                                         kvmppc_get_gpr(vcpu, 6));
 972                 break;
 973         case H_RTAS:
 974                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 975                         return RESUME_HOST;
 976
 977                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 978                 rc = kvmppc_rtas_hcall(vcpu);
 979                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 980
 981                 if (rc == -ENOENT)
 982                         return RESUME_HOST;
 983                 else if (rc == 0)
 984                         break;
 985
 986                 /* Send the error out to userspace via KVM_RUN */
 987                 return rc;
 988         case H_LOGICAL_CI_LOAD:
 989                 ret = kvmppc_h_logical_ci_load(vcpu);
 990                 if (ret == H_TOO_HARD)
 991                         return RESUME_HOST;
 992                 break;
 993         case H_LOGICAL_CI_STORE:
 994                 ret = kvmppc_h_logical_ci_store(vcpu);
 995                 if (ret == H_TOO_HARD)
 996                         return RESUME_HOST;
 997                 break;
 998         case H_SET_MODE:
 999                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
1000                                         kvmppc_get_gpr(vcpu, 5),
1001                                         kvmppc_get_gpr(vcpu, 6),
1002                                         kvmppc_get_gpr(vcpu, 7));
1003                 if (ret == H_TOO_HARD)
1004                         return RESUME_HOST;
1005                 break;
1006         case H_XIRR:
1007         case H_CPPR:
1008         case H_EOI:
1009         case H_IPI:
1010         case H_IPOLL:
1011         case H_XIRR_X:
1012                 if (kvmppc_xics_enabled(vcpu)) {
1013                         if (xics_on_xive()) {
1014                                 ret = H_NOT_AVAILABLE;
1015                                 return RESUME_GUEST;
1016                         }
1017                         ret = kvmppc_xics_hcall(vcpu, req);
1018                         break;
1019                 }
1020                 return RESUME_HOST;
1021         case H_SET_DABR:
1022                 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1023                 break;
1024         case H_SET_XDABR:
1025                 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1026                                                 kvmppc_get_gpr(vcpu, 5));
1027                 break;
1028 #ifdef CONFIG_SPAPR_TCE_IOMMU
1029         case H_GET_TCE:
1030                 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1031                                                 kvmppc_get_gpr(vcpu, 5));
1032                 if (ret == H_TOO_HARD)
1033                         return RESUME_HOST;
1034                 break;
1035         case H_PUT_TCE:
1036                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1037                                                 kvmppc_get_gpr(vcpu, 5),
1038                                                 kvmppc_get_gpr(vcpu, 6));
1039                 if (ret == H_TOO_HARD)
1040                         return RESUME_HOST;
1041                 break;
1042         case H_PUT_TCE_INDIRECT:
1043                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1044                                                 kvmppc_get_gpr(vcpu, 5),
1045                                                 kvmppc_get_gpr(vcpu, 6),
1046                                                 kvmppc_get_gpr(vcpu, 7));
1047                 if (ret == H_TOO_HARD)
1048                         return RESUME_HOST;
1049                 break;
1050         case H_STUFF_TCE:
1051                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1052                                                 kvmppc_get_gpr(vcpu, 5),
1053                                                 kvmppc_get_gpr(vcpu, 6),
1054                                                 kvmppc_get_gpr(vcpu, 7));
1055                 if (ret == H_TOO_HARD)
1056                         return RESUME_HOST;
1057                 break;
1058 #endif
1059         case H_RANDOM:
1060                 if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1061                         ret = H_HARDWARE;
1062                 break;
1063
1064         case H_SET_PARTITION_TABLE:
1065                 ret = H_FUNCTION;
1066                 if (nesting_enabled(vcpu->kvm))
1067                         ret = kvmhv_set_partition_table(vcpu);
1068                 break;
1069         case H_ENTER_NESTED:
1070                 ret = H_FUNCTION;
1071                 if (!nesting_enabled(vcpu->kvm))
1072                         break;
1073                 ret = kvmhv_enter_nested_guest(vcpu);
1074                 if (ret == H_INTERRUPT) {
1075                         kvmppc_set_gpr(vcpu, 3, 0);
1076                         vcpu->arch.hcall_needed = 0;
1077                         return -EINTR;
1078                 } else if (ret == H_TOO_HARD) {
1079                         kvmppc_set_gpr(vcpu, 3, 0);
1080                         vcpu->arch.hcall_needed = 0;
1081                         return RESUME_HOST;
1082                 }
1083                 break;
1084         case H_TLB_INVALIDATE:
1085                 ret = H_FUNCTION;
1086                 if (nesting_enabled(vcpu->kvm))
1087                         ret = kvmhv_do_nested_tlbie(vcpu);
1088                 break;
1089         case H_COPY_TOFROM_GUEST:
1090                 ret = H_FUNCTION;
1091                 if (nesting_enabled(vcpu->kvm))
1092                         ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1093                 break;
1094         case H_PAGE_INIT:
1095                 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1096                                          kvmppc_get_gpr(vcpu, 5),
1097                                          kvmppc_get_gpr(vcpu, 6));
1098                 break;
1099         case H_SVM_PAGE_IN:
1100                 ret = H_UNSUPPORTED;
1101                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1102                         ret = kvmppc_h_svm_page_in(vcpu->kvm,
1103                                                    kvmppc_get_gpr(vcpu, 4),
1104                                                    kvmppc_get_gpr(vcpu, 5),
1105                                                    kvmppc_get_gpr(vcpu, 6));
1106                 break;
1107         case H_SVM_PAGE_OUT:
1108                 ret = H_UNSUPPORTED;
1109                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1110                         ret = kvmppc_h_svm_page_out(vcpu->kvm,
1111                                                     kvmppc_get_gpr(vcpu, 4),
1112                                                     kvmppc_get_gpr(vcpu, 5),
1113                                                     kvmppc_get_gpr(vcpu, 6));
1114                 break;
1115         case H_SVM_INIT_START:
1116                 ret = H_UNSUPPORTED;
1117                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1118                         ret = kvmppc_h_svm_init_start(vcpu->kvm);
1119                 break;
1120         case H_SVM_INIT_DONE:
1121                 ret = H_UNSUPPORTED;
1122                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1123                         ret = kvmppc_h_svm_init_done(vcpu->kvm);
1124                 break;
1125         case H_SVM_INIT_ABORT:
1126                 /*
1127                  * Even if that call is made by the Ultravisor, the SSR1 value
1128                  * is the guest context one, with the secure bit clear as it has
1129                  * not yet been secured. So we can't check it here.
1130                  * Instead the kvm->arch.secure_guest flag is checked inside
1131                  * kvmppc_h_svm_init_abort().
1132                  */
1133                 ret = kvmppc_h_svm_init_abort(vcpu->kvm);
1134                 break;
1135
1136         default:
1137                 return RESUME_HOST;
1138         }
1139         kvmppc_set_gpr(vcpu, 3, ret);
1140         vcpu->arch.hcall_needed = 0;
1141         return RESUME_GUEST;
1142 }
1143
1144 /*
1145  * Handle H_CEDE in the nested virtualization case where we haven't
1146  * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1147  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1148  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1149  */
1150 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1151 {
1152         vcpu->arch.shregs.msr |= MSR_EE;
1153         vcpu->arch.ceded = 1;
1154         smp_mb();
1155         if (vcpu->arch.prodded) {
1156                 vcpu->arch.prodded = 0;
1157                 smp_mb();
1158                 vcpu->arch.ceded = 0;
1159         }
1160 }
1161
1162 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1163 {
1164         switch (cmd) {
1165         case H_CEDE:
1166         case H_PROD:
1167         case H_CONFER:
1168         case H_REGISTER_VPA:
1169         case H_SET_MODE:
1170         case H_LOGICAL_CI_LOAD:
1171         case H_LOGICAL_CI_STORE:
1172 #ifdef CONFIG_KVM_XICS
1173         case H_XIRR:
1174         case H_CPPR:
1175         case H_EOI:
1176         case H_IPI:
1177         case H_IPOLL:
1178         case H_XIRR_X:
1179 #endif
1180         case H_PAGE_INIT:
1181                 return 1;
1182         }
1183
1184         /* See if it's in the real-mode table */
1185         return kvmppc_hcall_impl_hv_realmode(cmd);
1186 }
1187
1188 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1189 {
1190         u32 last_inst;
1191
1192         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1193                                         EMULATE_DONE) {
1194                 /*
1195                  * Fetch failed, so return to guest and
1196                  * try executing it again.
1197                  */
1198                 return RESUME_GUEST;
1199         }
1200
1201         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1202                 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1203                 vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1204                 return RESUME_HOST;
1205         } else {
1206                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1207                 return RESUME_GUEST;
1208         }
1209 }
1210
1211 static void do_nothing(void *x)
1212 {
1213 }
1214
1215 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1216 {
1217         int thr, cpu, pcpu, nthreads;
1218         struct kvm_vcpu *v;
1219         unsigned long dpdes;
1220
1221         nthreads = vcpu->kvm->arch.emul_smt_mode;
1222         dpdes = 0;
1223         cpu = vcpu->vcpu_id & ~(nthreads - 1);
1224         for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1225                 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1226                 if (!v)
1227                         continue;
1228                 /*
1229                  * If the vcpu is currently running on a physical cpu thread,
1230                  * interrupt it in order to pull it out of the guest briefly,
1231                  * which will update its vcore->dpdes value.
1232                  */
1233                 pcpu = READ_ONCE(v->cpu);
1234                 if (pcpu >= 0)
1235                         smp_call_function_single(pcpu, do_nothing, NULL, 1);
1236                 if (kvmppc_doorbell_pending(v))
1237                         dpdes |= 1 << thr;
1238         }
1239         return dpdes;
1240 }
1241
1242 /*
1243  * On POWER9, emulate doorbell-related instructions in order to
1244  * give the guest the illusion of running on a multi-threaded core.
1245  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1246  * and mfspr DPDES.
1247  */
1248 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1249 {
1250         u32 inst, rb, thr;
1251         unsigned long arg;
1252         struct kvm *kvm = vcpu->kvm;
1253         struct kvm_vcpu *tvcpu;
1254
1255         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1256                 return RESUME_GUEST;
1257         if (get_op(inst) != 31)
1258                 return EMULATE_FAIL;
1259         rb = get_rb(inst);
1260         thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1261         switch (get_xop(inst)) {
1262         case OP_31_XOP_MSGSNDP:
1263                 arg = kvmppc_get_gpr(vcpu, rb);
1264                 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1265                         break;
1266                 arg &= 0x7f;
1267                 if (arg >= kvm->arch.emul_smt_mode)
1268                         break;
1269                 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1270                 if (!tvcpu)
1271                         break;
1272                 if (!tvcpu->arch.doorbell_request) {
1273                         tvcpu->arch.doorbell_request = 1;
1274                         kvmppc_fast_vcpu_kick_hv(tvcpu);
1275                 }
1276                 break;
1277         case OP_31_XOP_MSGCLRP:
1278                 arg = kvmppc_get_gpr(vcpu, rb);
1279                 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1280                         break;
1281                 vcpu->arch.vcore->dpdes = 0;
1282                 vcpu->arch.doorbell_request = 0;
1283                 break;
1284         case OP_31_XOP_MFSPR:
1285                 switch (get_sprn(inst)) {
1286                 case SPRN_TIR:
1287                         arg = thr;
1288                         break;
1289                 case SPRN_DPDES:
1290                         arg = kvmppc_read_dpdes(vcpu);
1291                         break;
1292                 default:
1293                         return EMULATE_FAIL;
1294                 }
1295                 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1296                 break;
1297         default:
1298                 return EMULATE_FAIL;
1299         }
1300         kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1301         return RESUME_GUEST;
1302 }
1303
1304 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1305                                  struct task_struct *tsk)
1306 {
1307         struct kvm_run *run = vcpu->run;
1308         int r = RESUME_HOST;
1309
1310         vcpu->stat.sum_exits++;
1311
1312         /*
1313          * This can happen if an interrupt occurs in the last stages
1314          * of guest entry or the first stages of guest exit (i.e. after
1315          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1316          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1317          * That can happen due to a bug, or due to a machine check
1318          * occurring at just the wrong time.
1319          */
1320         if (vcpu->arch.shregs.msr & MSR_HV) {
1321                 printk(KERN_EMERG "KVM trap in HV mode!\n");
1322                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1323                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1324                         vcpu->arch.shregs.msr);
1325                 kvmppc_dump_regs(vcpu);
1326                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1327                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1328                 return RESUME_HOST;
1329         }
1330         run->exit_reason = KVM_EXIT_UNKNOWN;
1331         run->ready_for_interrupt_injection = 1;
1332         switch (vcpu->arch.trap) {
1333         /* We're good on these - the host merely wanted to get our attention */
1334         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1335                 vcpu->stat.dec_exits++;
1336                 r = RESUME_GUEST;
1337                 break;
1338         case BOOK3S_INTERRUPT_EXTERNAL:
1339         case BOOK3S_INTERRUPT_H_DOORBELL:
1340         case BOOK3S_INTERRUPT_H_VIRT:
1341                 vcpu->stat.ext_intr_exits++;
1342                 r = RESUME_GUEST;
1343                 break;
1344         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1345         case BOOK3S_INTERRUPT_HMI:
1346         case BOOK3S_INTERRUPT_PERFMON:
1347         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1348                 r = RESUME_GUEST;
1349                 break;
1350         case BOOK3S_INTERRUPT_MACHINE_CHECK: {
1351                 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1352                                               DEFAULT_RATELIMIT_BURST);
1353                 /*
1354                  * Print the MCE event to host console. Ratelimit so the guest
1355                  * can't flood the host log.
1356                  */
1357                 if (__ratelimit(&rs))
1358                         machine_check_print_event_info(&vcpu->arch.mce_evt,false, true);
1359
1360                 /*
1361                  * If the guest can do FWNMI, exit to userspace so it can
1362                  * deliver a FWNMI to the guest.
1363                  * Otherwise we synthesize a machine check for the guest
1364                  * so that it knows that the machine check occurred.
1365                  */
1366                 if (!vcpu->kvm->arch.fwnmi_enabled) {
1367                         ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1368                         kvmppc_core_queue_machine_check(vcpu, flags);
1369                         r = RESUME_GUEST;
1370                         break;
1371                 }
1372
1373                 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1374                 run->exit_reason = KVM_EXIT_NMI;
1375                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1376                 /* Clear out the old NMI status from run->flags */
1377                 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1378                 /* Now set the NMI status */
1379                 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1380                         run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1381                 else
1382                         run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1383
1384                 r = RESUME_HOST;
1385                 break;
1386         }
1387         case BOOK3S_INTERRUPT_PROGRAM:
1388         {
1389                 ulong flags;
1390                 /*
1391                  * Normally program interrupts are delivered directly
1392                  * to the guest by the hardware, but we can get here
1393                  * as a result of a hypervisor emulation interrupt
1394                  * (e40) getting turned into a 700 by BML RTAS.
1395                  */
1396                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1397                 kvmppc_core_queue_program(vcpu, flags);
1398                 r = RESUME_GUEST;
1399                 break;
1400         }
1401         case BOOK3S_INTERRUPT_SYSCALL:
1402         {
1403                 /* hcall - punt to userspace */
1404                 int i;
1405
1406                 /* hypercall with MSR_PR has already been handled in rmode,
1407                  * and never reaches here.
1408                  */
1409
1410                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1411                 for (i = 0; i < 9; ++i)
1412                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1413                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1414                 vcpu->arch.hcall_needed = 1;
1415                 r = RESUME_HOST;
1416                 break;
1417         }
1418         /*
1419          * We get these next two if the guest accesses a page which it thinks
1420          * it has mapped but which is not actually present, either because
1421          * it is for an emulated I/O device or because the corresonding
1422          * host page has been paged out.  Any other HDSI/HISI interrupts
1423          * have been handled already.
1424          */
1425         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1426                 r = RESUME_PAGE_FAULT;
1427                 break;
1428         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1429                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1430                 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1431                         DSISR_SRR1_MATCH_64S;
1432                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1433                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1434                 r = RESUME_PAGE_FAULT;
1435                 break;
1436         /*
1437          * This occurs if the guest executes an illegal instruction.
1438          * If the guest debug is disabled, generate a program interrupt
1439          * to the guest. If guest debug is enabled, we need to check
1440          * whether the instruction is a software breakpoint instruction.
1441          * Accordingly return to Guest or Host.
1442          */
1443         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1444                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1445                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1446                                 swab32(vcpu->arch.emul_inst) :
1447                                 vcpu->arch.emul_inst;
1448                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1449                         r = kvmppc_emulate_debug_inst(vcpu);
1450                 } else {
1451                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1452                         r = RESUME_GUEST;
1453                 }
1454                 break;
1455         /*
1456          * This occurs if the guest (kernel or userspace), does something that
1457          * is prohibited by HFSCR.
1458          * On POWER9, this could be a doorbell instruction that we need
1459          * to emulate.
1460          * Otherwise, we just generate a program interrupt to the guest.
1461          */
1462         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1463                 r = EMULATE_FAIL;
1464                 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1465                     cpu_has_feature(CPU_FTR_ARCH_300))
1466                         r = kvmppc_emulate_doorbell_instr(vcpu);
1467                 if (r == EMULATE_FAIL) {
1468                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1469                         r = RESUME_GUEST;
1470                 }
1471                 break;
1472
1473 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1474         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1475                 /*
1476                  * This occurs for various TM-related instructions that
1477                  * we need to emulate on POWER9 DD2.2.  We have already
1478                  * handled the cases where the guest was in real-suspend
1479                  * mode and was transitioning to transactional state.
1480                  */
1481                 r = kvmhv_p9_tm_emulation(vcpu);
1482                 break;
1483 #endif
1484
1485         case BOOK3S_INTERRUPT_HV_RM_HARD:
1486                 r = RESUME_PASSTHROUGH;
1487                 break;
1488         default:
1489                 kvmppc_dump_regs(vcpu);
1490                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1491                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1492                         vcpu->arch.shregs.msr);
1493                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1494                 r = RESUME_HOST;
1495                 break;
1496         }
1497
1498         return r;
1499 }
1500
1501 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1502 {
1503         int r;
1504         int srcu_idx;
1505
1506         vcpu->stat.sum_exits++;
1507
1508         /*
1509          * This can happen if an interrupt occurs in the last stages
1510          * of guest entry or the first stages of guest exit (i.e. after
1511          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1512          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1513          * That can happen due to a bug, or due to a machine check
1514          * occurring at just the wrong time.
1515          */
1516         if (vcpu->arch.shregs.msr & MSR_HV) {
1517                 pr_emerg("KVM trap in HV mode while nested!\n");
1518                 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1519                          vcpu->arch.trap, kvmppc_get_pc(vcpu),
1520                          vcpu->arch.shregs.msr);
1521                 kvmppc_dump_regs(vcpu);
1522                 return RESUME_HOST;
1523         }
1524         switch (vcpu->arch.trap) {
1525         /* We're good on these - the host merely wanted to get our attention */
1526         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1527                 vcpu->stat.dec_exits++;
1528                 r = RESUME_GUEST;
1529                 break;
1530         case BOOK3S_INTERRUPT_EXTERNAL:
1531                 vcpu->stat.ext_intr_exits++;
1532                 r = RESUME_HOST;
1533                 break;
1534         case BOOK3S_INTERRUPT_H_DOORBELL:
1535         case BOOK3S_INTERRUPT_H_VIRT:
1536                 vcpu->stat.ext_intr_exits++;
1537                 r = RESUME_GUEST;
1538                 break;
1539         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1540         case BOOK3S_INTERRUPT_HMI:
1541         case BOOK3S_INTERRUPT_PERFMON:
1542         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1543                 r = RESUME_GUEST;
1544                 break;
1545         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1546         {
1547                 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1548                                               DEFAULT_RATELIMIT_BURST);
1549                 /* Pass the machine check to the L1 guest */
1550                 r = RESUME_HOST;
1551                 /* Print the MCE event to host console. */
1552                 if (__ratelimit(&rs))
1553                         machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1554                 break;
1555         }
1556         /*
1557          * We get these next two if the guest accesses a page which it thinks
1558          * it has mapped but which is not actually present, either because
1559          * it is for an emulated I/O device or because the corresonding
1560          * host page has been paged out.
1561          */
1562         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1563                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1564                 r = kvmhv_nested_page_fault(vcpu);
1565                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1566                 break;
1567         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1568                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1569                 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1570                                          DSISR_SRR1_MATCH_64S;
1571                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1572                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1573                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1574                 r = kvmhv_nested_page_fault(vcpu);
1575                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1576                 break;
1577
1578 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1579         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1580                 /*
1581                  * This occurs for various TM-related instructions that
1582                  * we need to emulate on POWER9 DD2.2.  We have already
1583                  * handled the cases where the guest was in real-suspend
1584                  * mode and was transitioning to transactional state.
1585                  */
1586                 r = kvmhv_p9_tm_emulation(vcpu);
1587                 break;
1588 #endif
1589
1590         case BOOK3S_INTERRUPT_HV_RM_HARD:
1591                 vcpu->arch.trap = 0;
1592                 r = RESUME_GUEST;
1593                 if (!xics_on_xive())
1594                         kvmppc_xics_rm_complete(vcpu, 0);
1595                 break;
1596         default:
1597                 r = RESUME_HOST;
1598                 break;
1599         }
1600
1601         return r;
1602 }
1603
1604 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1605                                             struct kvm_sregs *sregs)
1606 {
1607         int i;
1608
1609         memset(sregs, 0, sizeof(struct kvm_sregs));
1610         sregs->pvr = vcpu->arch.pvr;
1611         for (i = 0; i < vcpu->arch.slb_max; i++) {
1612                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1613                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1614         }
1615
1616         return 0;
1617 }
1618
1619 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1620                                             struct kvm_sregs *sregs)
1621 {
1622         int i, j;
1623
1624         /* Only accept the same PVR as the host's, since we can't spoof it */
1625         if (sregs->pvr != vcpu->arch.pvr)
1626                 return -EINVAL;
1627
1628         j = 0;
1629         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1630                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1631                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1632                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1633                         ++j;
1634                 }
1635         }
1636         vcpu->arch.slb_max = j;
1637
1638         return 0;
1639 }
1640
1641 /*
1642  * Enforce limits on guest LPCR values based on hardware availability,
1643  * guest configuration, and possibly hypervisor support and security
1644  * concerns.
1645  */
1646 unsigned long kvmppc_filter_lpcr_hv(struct kvm *kvm, unsigned long lpcr)
1647 {
1648         /* LPCR_TC only applies to HPT guests */
1649         if (kvm_is_radix(kvm))
1650                 lpcr &= ~LPCR_TC;
1651
1652         /* On POWER8 and above, userspace can modify AIL */
1653         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
1654                 lpcr &= ~LPCR_AIL;
1655         if ((lpcr & LPCR_AIL) != LPCR_AIL_3)
1656                 lpcr &= ~LPCR_AIL; /* LPCR[AIL]=1/2 is disallowed */
1657
1658         /*
1659          * On POWER9, allow userspace to enable large decrementer for the
1660          * guest, whether or not the host has it enabled.
1661          */
1662         if (!cpu_has_feature(CPU_FTR_ARCH_300))
1663                 lpcr &= ~LPCR_LD;
1664
1665         return lpcr;
1666 }
1667
1668 static void verify_lpcr(struct kvm *kvm, unsigned long lpcr)
1669 {
1670         if (lpcr != kvmppc_filter_lpcr_hv(kvm, lpcr)) {
1671                 WARN_ONCE(1, "lpcr 0x%lx differs from filtered 0x%lx\n",
1672                           lpcr, kvmppc_filter_lpcr_hv(kvm, lpcr));
1673         }
1674 }
1675
1676 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1677                 bool preserve_top32)
1678 {
1679         struct kvm *kvm = vcpu->kvm;
1680         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1681         u64 mask;
1682
1683         spin_lock(&vc->lock);
1684
1685         /*
1686          * Userspace can only modify
1687          * DPFD (default prefetch depth), ILE (interrupt little-endian),
1688          * TC (translation control), AIL (alternate interrupt location),
1689          * LD (large decrementer).
1690          * These are subject to restrictions from kvmppc_filter_lcpr_hv().
1691          */
1692         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD;
1693
1694         /* Broken 32-bit version of LPCR must not clear top bits */
1695         if (preserve_top32)
1696                 mask &= 0xFFFFFFFF;
1697
1698         new_lpcr = kvmppc_filter_lpcr_hv(kvm,
1699                         (vc->lpcr & ~mask) | (new_lpcr & mask));
1700
1701         /*
1702          * If ILE (interrupt little-endian) has changed, update the
1703          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1704          */
1705         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1706                 struct kvm_vcpu *vcpu;
1707                 int i;
1708
1709                 kvm_for_each_vcpu(i, vcpu, kvm) {
1710                         if (vcpu->arch.vcore != vc)
1711                                 continue;
1712                         if (new_lpcr & LPCR_ILE)
1713                                 vcpu->arch.intr_msr |= MSR_LE;
1714                         else
1715                                 vcpu->arch.intr_msr &= ~MSR_LE;
1716                 }
1717         }
1718
1719         vc->lpcr = new_lpcr;
1720
1721         spin_unlock(&vc->lock);
1722 }
1723
1724 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1725                                  union kvmppc_one_reg *val)
1726 {
1727         int r = 0;
1728         long int i;
1729
1730         switch (id) {
1731         case KVM_REG_PPC_DEBUG_INST:
1732                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1733                 break;
1734         case KVM_REG_PPC_HIOR:
1735                 *val = get_reg_val(id, 0);
1736                 break;
1737         case KVM_REG_PPC_DABR:
1738                 *val = get_reg_val(id, vcpu->arch.dabr);
1739                 break;
1740         case KVM_REG_PPC_DABRX:
1741                 *val = get_reg_val(id, vcpu->arch.dabrx);
1742                 break;
1743         case KVM_REG_PPC_DSCR:
1744                 *val = get_reg_val(id, vcpu->arch.dscr);
1745                 break;
1746         case KVM_REG_PPC_PURR:
1747                 *val = get_reg_val(id, vcpu->arch.purr);
1748                 break;
1749         case KVM_REG_PPC_SPURR:
1750                 *val = get_reg_val(id, vcpu->arch.spurr);
1751                 break;
1752         case KVM_REG_PPC_AMR:
1753                 *val = get_reg_val(id, vcpu->arch.amr);
1754                 break;
1755         case KVM_REG_PPC_UAMOR:
1756                 *val = get_reg_val(id, vcpu->arch.uamor);
1757                 break;
1758         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
1759                 i = id - KVM_REG_PPC_MMCR0;
1760                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1761                 break;
1762         case KVM_REG_PPC_MMCR2:
1763                 *val = get_reg_val(id, vcpu->arch.mmcr[2]);
1764                 break;
1765         case KVM_REG_PPC_MMCRA:
1766                 *val = get_reg_val(id, vcpu->arch.mmcra);
1767                 break;
1768         case KVM_REG_PPC_MMCRS:
1769                 *val = get_reg_val(id, vcpu->arch.mmcrs);
1770                 break;
1771         case KVM_REG_PPC_MMCR3:
1772                 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
1773                 break;
1774         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1775                 i = id - KVM_REG_PPC_PMC1;
1776                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1777                 break;
1778         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1779                 i = id - KVM_REG_PPC_SPMC1;
1780                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1781                 break;
1782         case KVM_REG_PPC_SIAR:
1783                 *val = get_reg_val(id, vcpu->arch.siar);
1784                 break;
1785         case KVM_REG_PPC_SDAR:
1786                 *val = get_reg_val(id, vcpu->arch.sdar);
1787                 break;
1788         case KVM_REG_PPC_SIER:
1789                 *val = get_reg_val(id, vcpu->arch.sier[0]);
1790                 break;
1791         case KVM_REG_PPC_SIER2:
1792                 *val = get_reg_val(id, vcpu->arch.sier[1]);
1793                 break;
1794         case KVM_REG_PPC_SIER3:
1795                 *val = get_reg_val(id, vcpu->arch.sier[2]);
1796                 break;
1797         case KVM_REG_PPC_IAMR:
1798                 *val = get_reg_val(id, vcpu->arch.iamr);
1799                 break;
1800         case KVM_REG_PPC_PSPB:
1801                 *val = get_reg_val(id, vcpu->arch.pspb);
1802                 break;
1803         case KVM_REG_PPC_DPDES:
1804                 /*
1805                  * On POWER9, where we are emulating msgsndp etc.,
1806                  * we return 1 bit for each vcpu, which can come from
1807                  * either vcore->dpdes or doorbell_request.
1808                  * On POWER8, doorbell_request is 0.
1809                  */
1810                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes |
1811                                    vcpu->arch.doorbell_request);
1812                 break;
1813         case KVM_REG_PPC_VTB:
1814                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1815                 break;
1816         case KVM_REG_PPC_DAWR:
1817                 *val = get_reg_val(id, vcpu->arch.dawr0);
1818                 break;
1819         case KVM_REG_PPC_DAWRX:
1820                 *val = get_reg_val(id, vcpu->arch.dawrx0);
1821                 break;
1822         case KVM_REG_PPC_DAWR1:
1823                 *val = get_reg_val(id, vcpu->arch.dawr1);
1824                 break;
1825         case KVM_REG_PPC_DAWRX1:
1826                 *val = get_reg_val(id, vcpu->arch.dawrx1);
1827                 break;
1828         case KVM_REG_PPC_CIABR:
1829                 *val = get_reg_val(id, vcpu->arch.ciabr);
1830                 break;
1831         case KVM_REG_PPC_CSIGR:
1832                 *val = get_reg_val(id, vcpu->arch.csigr);
1833                 break;
1834         case KVM_REG_PPC_TACR:
1835                 *val = get_reg_val(id, vcpu->arch.tacr);
1836                 break;
1837         case KVM_REG_PPC_TCSCR:
1838                 *val = get_reg_val(id, vcpu->arch.tcscr);
1839                 break;
1840         case KVM_REG_PPC_PID:
1841                 *val = get_reg_val(id, vcpu->arch.pid);
1842                 break;
1843         case KVM_REG_PPC_ACOP:
1844                 *val = get_reg_val(id, vcpu->arch.acop);
1845                 break;
1846         case KVM_REG_PPC_WORT:
1847                 *val = get_reg_val(id, vcpu->arch.wort);
1848                 break;
1849         case KVM_REG_PPC_TIDR:
1850                 *val = get_reg_val(id, vcpu->arch.tid);
1851                 break;
1852         case KVM_REG_PPC_PSSCR:
1853                 *val = get_reg_val(id, vcpu->arch.psscr);
1854                 break;
1855         case KVM_REG_PPC_VPA_ADDR:
1856                 spin_lock(&vcpu->arch.vpa_update_lock);
1857                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1858                 spin_unlock(&vcpu->arch.vpa_update_lock);
1859                 break;
1860         case KVM_REG_PPC_VPA_SLB:
1861                 spin_lock(&vcpu->arch.vpa_update_lock);
1862                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1863                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1864                 spin_unlock(&vcpu->arch.vpa_update_lock);
1865                 break;
1866         case KVM_REG_PPC_VPA_DTL:
1867                 spin_lock(&vcpu->arch.vpa_update_lock);
1868                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1869                 val->vpaval.length = vcpu->arch.dtl.len;
1870                 spin_unlock(&vcpu->arch.vpa_update_lock);
1871                 break;
1872         case KVM_REG_PPC_TB_OFFSET:
1873                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1874                 break;
1875         case KVM_REG_PPC_LPCR:
1876         case KVM_REG_PPC_LPCR_64:
1877                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1878                 break;
1879         case KVM_REG_PPC_PPR:
1880                 *val = get_reg_val(id, vcpu->arch.ppr);
1881                 break;
1882 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1883         case KVM_REG_PPC_TFHAR:
1884                 *val = get_reg_val(id, vcpu->arch.tfhar);
1885                 break;
1886         case KVM_REG_PPC_TFIAR:
1887                 *val = get_reg_val(id, vcpu->arch.tfiar);
1888                 break;
1889         case KVM_REG_PPC_TEXASR:
1890                 *val = get_reg_val(id, vcpu->arch.texasr);
1891                 break;
1892         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1893                 i = id - KVM_REG_PPC_TM_GPR0;
1894                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1895                 break;
1896         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1897         {
1898                 int j;
1899                 i = id - KVM_REG_PPC_TM_VSR0;
1900                 if (i < 32)
1901                         for (j = 0; j < TS_FPRWIDTH; j++)
1902                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1903                 else {
1904                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1905                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1906                         else
1907                                 r = -ENXIO;
1908                 }
1909                 break;
1910         }
1911         case KVM_REG_PPC_TM_CR:
1912                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1913                 break;
1914         case KVM_REG_PPC_TM_XER:
1915                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1916                 break;
1917         case KVM_REG_PPC_TM_LR:
1918                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1919                 break;
1920         case KVM_REG_PPC_TM_CTR:
1921                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1922                 break;
1923         case KVM_REG_PPC_TM_FPSCR:
1924                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1925                 break;
1926         case KVM_REG_PPC_TM_AMR:
1927                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1928                 break;
1929         case KVM_REG_PPC_TM_PPR:
1930                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1931                 break;
1932         case KVM_REG_PPC_TM_VRSAVE:
1933                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1934                 break;
1935         case KVM_REG_PPC_TM_VSCR:
1936                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1937                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1938                 else
1939                         r = -ENXIO;
1940                 break;
1941         case KVM_REG_PPC_TM_DSCR:
1942                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1943                 break;
1944         case KVM_REG_PPC_TM_TAR:
1945                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1946                 break;
1947 #endif
1948         case KVM_REG_PPC_ARCH_COMPAT:
1949                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1950                 break;
1951         case KVM_REG_PPC_DEC_EXPIRY:
1952                 *val = get_reg_val(id, vcpu->arch.dec_expires +
1953                                    vcpu->arch.vcore->tb_offset);
1954                 break;
1955         case KVM_REG_PPC_ONLINE:
1956                 *val = get_reg_val(id, vcpu->arch.online);
1957                 break;
1958         case KVM_REG_PPC_PTCR:
1959                 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1960                 break;
1961         default:
1962                 r = -EINVAL;
1963                 break;
1964         }
1965
1966         return r;
1967 }
1968
1969 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1970                                  union kvmppc_one_reg *val)
1971 {
1972         int r = 0;
1973         long int i;
1974         unsigned long addr, len;
1975
1976         switch (id) {
1977         case KVM_REG_PPC_HIOR:
1978                 /* Only allow this to be set to zero */
1979                 if (set_reg_val(id, *val))
1980                         r = -EINVAL;
1981                 break;
1982         case KVM_REG_PPC_DABR:
1983                 vcpu->arch.dabr = set_reg_val(id, *val);
1984                 break;
1985         case KVM_REG_PPC_DABRX:
1986                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1987                 break;
1988         case KVM_REG_PPC_DSCR:
1989                 vcpu->arch.dscr = set_reg_val(id, *val);
1990                 break;
1991         case KVM_REG_PPC_PURR:
1992                 vcpu->arch.purr = set_reg_val(id, *val);
1993                 break;
1994         case KVM_REG_PPC_SPURR:
1995                 vcpu->arch.spurr = set_reg_val(id, *val);
1996                 break;
1997         case KVM_REG_PPC_AMR:
1998                 vcpu->arch.amr = set_reg_val(id, *val);
1999                 break;
2000         case KVM_REG_PPC_UAMOR:
2001                 vcpu->arch.uamor = set_reg_val(id, *val);
2002                 break;
2003         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2004                 i = id - KVM_REG_PPC_MMCR0;
2005                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
2006                 break;
2007         case KVM_REG_PPC_MMCR2:
2008                 vcpu->arch.mmcr[2] = set_reg_val(id, *val);
2009                 break;
2010         case KVM_REG_PPC_MMCRA:
2011                 vcpu->arch.mmcra = set_reg_val(id, *val);
2012                 break;
2013         case KVM_REG_PPC_MMCRS:
2014                 vcpu->arch.mmcrs = set_reg_val(id, *val);
2015                 break;
2016         case KVM_REG_PPC_MMCR3:
2017                 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
2018                 break;
2019         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2020                 i = id - KVM_REG_PPC_PMC1;
2021                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
2022                 break;
2023         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2024                 i = id - KVM_REG_PPC_SPMC1;
2025                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
2026                 break;
2027         case KVM_REG_PPC_SIAR:
2028                 vcpu->arch.siar = set_reg_val(id, *val);
2029                 break;
2030         case KVM_REG_PPC_SDAR:
2031                 vcpu->arch.sdar = set_reg_val(id, *val);
2032                 break;
2033         case KVM_REG_PPC_SIER:
2034                 vcpu->arch.sier[0] = set_reg_val(id, *val);
2035                 break;
2036         case KVM_REG_PPC_SIER2:
2037                 vcpu->arch.sier[1] = set_reg_val(id, *val);
2038                 break;
2039         case KVM_REG_PPC_SIER3:
2040                 vcpu->arch.sier[2] = set_reg_val(id, *val);
2041                 break;
2042         case KVM_REG_PPC_IAMR:
2043                 vcpu->arch.iamr = set_reg_val(id, *val);
2044                 break;
2045         case KVM_REG_PPC_PSPB:
2046                 vcpu->arch.pspb = set_reg_val(id, *val);
2047                 break;
2048         case KVM_REG_PPC_DPDES:
2049                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
2050                 break;
2051         case KVM_REG_PPC_VTB:
2052                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
2053                 break;
2054         case KVM_REG_PPC_DAWR:
2055                 vcpu->arch.dawr0 = set_reg_val(id, *val);
2056                 break;
2057         case KVM_REG_PPC_DAWRX:
2058                 vcpu->arch.dawrx0 = set_reg_val(id, *val) & ~DAWRX_HYP;
2059                 break;
2060         case KVM_REG_PPC_DAWR1:
2061                 vcpu->arch.dawr1 = set_reg_val(id, *val);
2062                 break;
2063         case KVM_REG_PPC_DAWRX1:
2064                 vcpu->arch.dawrx1 = set_reg_val(id, *val) & ~DAWRX_HYP;
2065                 break;
2066         case KVM_REG_PPC_CIABR:
2067                 vcpu->arch.ciabr = set_reg_val(id, *val);
2068                 /* Don't allow setting breakpoints in hypervisor code */
2069                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
2070                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
2071                 break;
2072         case KVM_REG_PPC_CSIGR:
2073                 vcpu->arch.csigr = set_reg_val(id, *val);
2074                 break;
2075         case KVM_REG_PPC_TACR:
2076                 vcpu->arch.tacr = set_reg_val(id, *val);
2077                 break;
2078         case KVM_REG_PPC_TCSCR:
2079                 vcpu->arch.tcscr = set_reg_val(id, *val);
2080                 break;
2081         case KVM_REG_PPC_PID:
2082                 vcpu->arch.pid = set_reg_val(id, *val);
2083                 break;
2084         case KVM_REG_PPC_ACOP:
2085                 vcpu->arch.acop = set_reg_val(id, *val);
2086                 break;
2087         case KVM_REG_PPC_WORT:
2088                 vcpu->arch.wort = set_reg_val(id, *val);
2089                 break;
2090         case KVM_REG_PPC_TIDR:
2091                 vcpu->arch.tid = set_reg_val(id, *val);
2092                 break;
2093         case KVM_REG_PPC_PSSCR:
2094                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2095                 break;
2096         case KVM_REG_PPC_VPA_ADDR:
2097                 addr = set_reg_val(id, *val);
2098                 r = -EINVAL;
2099                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2100                               vcpu->arch.dtl.next_gpa))
2101                         break;
2102                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2103                 break;
2104         case KVM_REG_PPC_VPA_SLB:
2105                 addr = val->vpaval.addr;
2106                 len = val->vpaval.length;
2107                 r = -EINVAL;
2108                 if (addr && !vcpu->arch.vpa.next_gpa)
2109                         break;
2110                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2111                 break;
2112         case KVM_REG_PPC_VPA_DTL:
2113                 addr = val->vpaval.addr;
2114                 len = val->vpaval.length;
2115                 r = -EINVAL;
2116                 if (addr && (len < sizeof(struct dtl_entry) ||
2117                              !vcpu->arch.vpa.next_gpa))
2118                         break;
2119                 len -= len % sizeof(struct dtl_entry);
2120                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2121                 break;
2122         case KVM_REG_PPC_TB_OFFSET:
2123                 /* round up to multiple of 2^24 */
2124                 vcpu->arch.vcore->tb_offset =
2125                         ALIGN(set_reg_val(id, *val), 1UL << 24);
2126                 break;
2127         case KVM_REG_PPC_LPCR:
2128                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2129                 break;
2130         case KVM_REG_PPC_LPCR_64:
2131                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2132                 break;
2133         case KVM_REG_PPC_PPR:
2134                 vcpu->arch.ppr = set_reg_val(id, *val);
2135                 break;
2136 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2137         case KVM_REG_PPC_TFHAR:
2138                 vcpu->arch.tfhar = set_reg_val(id, *val);
2139                 break;
2140         case KVM_REG_PPC_TFIAR:
2141                 vcpu->arch.tfiar = set_reg_val(id, *val);
2142                 break;
2143         case KVM_REG_PPC_TEXASR:
2144                 vcpu->arch.texasr = set_reg_val(id, *val);
2145                 break;
2146         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2147                 i = id - KVM_REG_PPC_TM_GPR0;
2148                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2149                 break;
2150         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2151         {
2152                 int j;
2153                 i = id - KVM_REG_PPC_TM_VSR0;
2154                 if (i < 32)
2155                         for (j = 0; j < TS_FPRWIDTH; j++)
2156                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2157                 else
2158                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
2159                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2160                         else
2161                                 r = -ENXIO;
2162                 break;
2163         }
2164         case KVM_REG_PPC_TM_CR:
2165                 vcpu->arch.cr_tm = set_reg_val(id, *val);
2166                 break;
2167         case KVM_REG_PPC_TM_XER:
2168                 vcpu->arch.xer_tm = set_reg_val(id, *val);
2169                 break;
2170         case KVM_REG_PPC_TM_LR:
2171                 vcpu->arch.lr_tm = set_reg_val(id, *val);
2172                 break;
2173         case KVM_REG_PPC_TM_CTR:
2174                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2175                 break;
2176         case KVM_REG_PPC_TM_FPSCR:
2177                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2178                 break;
2179         case KVM_REG_PPC_TM_AMR:
2180                 vcpu->arch.amr_tm = set_reg_val(id, *val);
2181                 break;
2182         case KVM_REG_PPC_TM_PPR:
2183                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2184                 break;
2185         case KVM_REG_PPC_TM_VRSAVE:
2186                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2187                 break;
2188         case KVM_REG_PPC_TM_VSCR:
2189                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2190                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2191                 else
2192                         r = - ENXIO;
2193                 break;
2194         case KVM_REG_PPC_TM_DSCR:
2195                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2196                 break;
2197         case KVM_REG_PPC_TM_TAR:
2198                 vcpu->arch.tar_tm = set_reg_val(id, *val);
2199                 break;
2200 #endif
2201         case KVM_REG_PPC_ARCH_COMPAT:
2202                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2203                 break;
2204         case KVM_REG_PPC_DEC_EXPIRY:
2205                 vcpu->arch.dec_expires = set_reg_val(id, *val) -
2206                         vcpu->arch.vcore->tb_offset;
2207                 break;
2208         case KVM_REG_PPC_ONLINE:
2209                 i = set_reg_val(id, *val);
2210                 if (i && !vcpu->arch.online)
2211                         atomic_inc(&vcpu->arch.vcore->online_count);
2212                 else if (!i && vcpu->arch.online)
2213                         atomic_dec(&vcpu->arch.vcore->online_count);
2214                 vcpu->arch.online = i;
2215                 break;
2216         case KVM_REG_PPC_PTCR:
2217                 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2218                 break;
2219         default:
2220                 r = -EINVAL;
2221                 break;
2222         }
2223
2224         return r;
2225 }
2226
2227 /*
2228  * On POWER9, threads are independent and can be in different partitions.
2229  * Therefore we consider each thread to be a subcore.
2230  * There is a restriction that all threads have to be in the same
2231  * MMU mode (radix or HPT), unfortunately, but since we only support
2232  * HPT guests on a HPT host so far, that isn't an impediment yet.
2233  */
2234 static int threads_per_vcore(struct kvm *kvm)
2235 {
2236         if (kvm->arch.threads_indep)
2237                 return 1;
2238         return threads_per_subcore;
2239 }
2240
2241 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2242 {
2243         struct kvmppc_vcore *vcore;
2244
2245         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2246
2247         if (vcore == NULL)
2248                 return NULL;
2249
2250         spin_lock_init(&vcore->lock);
2251         spin_lock_init(&vcore->stoltb_lock);
2252         rcuwait_init(&vcore->wait);
2253         vcore->preempt_tb = TB_NIL;
2254         vcore->lpcr = kvm->arch.lpcr;
2255         vcore->first_vcpuid = id;
2256         vcore->kvm = kvm;
2257         INIT_LIST_HEAD(&vcore->preempt_list);
2258
2259         return vcore;
2260 }
2261
2262 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2263 static struct debugfs_timings_element {
2264         const char *name;
2265         size_t offset;
2266 } timings[] = {
2267         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
2268         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
2269         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
2270         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
2271         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
2272 };
2273
2274 #define N_TIMINGS       (ARRAY_SIZE(timings))
2275
2276 struct debugfs_timings_state {
2277         struct kvm_vcpu *vcpu;
2278         unsigned int    buflen;
2279         char            buf[N_TIMINGS * 100];
2280 };
2281
2282 static int debugfs_timings_open(struct inode *inode, struct file *file)
2283 {
2284         struct kvm_vcpu *vcpu = inode->i_private;
2285         struct debugfs_timings_state *p;
2286
2287         p = kzalloc(sizeof(*p), GFP_KERNEL);
2288         if (!p)
2289                 return -ENOMEM;
2290
2291         kvm_get_kvm(vcpu->kvm);
2292         p->vcpu = vcpu;
2293         file->private_data = p;
2294
2295         return nonseekable_open(inode, file);
2296 }
2297
2298 static int debugfs_timings_release(struct inode *inode, struct file *file)
2299 {
2300         struct debugfs_timings_state *p = file->private_data;
2301
2302         kvm_put_kvm(p->vcpu->kvm);
2303         kfree(p);
2304         return 0;
2305 }
2306
2307 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2308                                     size_t len, loff_t *ppos)
2309 {
2310         struct debugfs_timings_state *p = file->private_data;
2311         struct kvm_vcpu *vcpu = p->vcpu;
2312         char *s, *buf_end;
2313         struct kvmhv_tb_accumulator tb;
2314         u64 count;
2315         loff_t pos;
2316         ssize_t n;
2317         int i, loops;
2318         bool ok;
2319
2320         if (!p->buflen) {
2321                 s = p->buf;
2322                 buf_end = s + sizeof(p->buf);
2323                 for (i = 0; i < N_TIMINGS; ++i) {
2324                         struct kvmhv_tb_accumulator *acc;
2325
2326                         acc = (struct kvmhv_tb_accumulator *)
2327                                 ((unsigned long)vcpu + timings[i].offset);
2328                         ok = false;
2329                         for (loops = 0; loops < 1000; ++loops) {
2330                                 count = acc->seqcount;
2331                                 if (!(count & 1)) {
2332                                         smp_rmb();
2333                                         tb = *acc;
2334                                         smp_rmb();
2335                                         if (count == acc->seqcount) {
2336                                                 ok = true;
2337                                                 break;
2338                                         }
2339                                 }
2340                                 udelay(1);
2341                         }
2342                         if (!ok)
2343                                 snprintf(s, buf_end - s, "%s: stuck\n",
2344                                         timings[i].name);
2345                         else
2346                                 snprintf(s, buf_end - s,
2347                                         "%s: %llu %llu %llu %llu\n",
2348                                         timings[i].name, count / 2,
2349                                         tb_to_ns(tb.tb_total),
2350                                         tb_to_ns(tb.tb_min),
2351                                         tb_to_ns(tb.tb_max));
2352                         s += strlen(s);
2353                 }
2354                 p->buflen = s - p->buf;
2355         }
2356
2357         pos = *ppos;
2358         if (pos >= p->buflen)
2359                 return 0;
2360         if (len > p->buflen - pos)
2361                 len = p->buflen - pos;
2362         n = copy_to_user(buf, p->buf + pos, len);
2363         if (n) {
2364                 if (n == len)
2365                         return -EFAULT;
2366                 len -= n;
2367         }
2368         *ppos = pos + len;
2369         return len;
2370 }
2371
2372 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2373                                      size_t len, loff_t *ppos)
2374 {
2375         return -EACCES;
2376 }
2377
2378 static const struct file_operations debugfs_timings_ops = {
2379         .owner   = THIS_MODULE,
2380         .open    = debugfs_timings_open,
2381         .release = debugfs_timings_release,
2382         .read    = debugfs_timings_read,
2383         .write   = debugfs_timings_write,
2384         .llseek  = generic_file_llseek,
2385 };
2386
2387 /* Create a debugfs directory for the vcpu */
2388 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2389 {
2390         char buf[16];
2391         struct kvm *kvm = vcpu->kvm;
2392
2393         snprintf(buf, sizeof(buf), "vcpu%u", id);
2394         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2395         debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir, vcpu,
2396                             &debugfs_timings_ops);
2397 }
2398
2399 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2400 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2401 {
2402 }
2403 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2404
2405 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2406 {
2407         int err;
2408         int core;
2409         struct kvmppc_vcore *vcore;
2410         struct kvm *kvm;
2411         unsigned int id;
2412
2413         kvm = vcpu->kvm;
2414         id = vcpu->vcpu_id;
2415
2416         vcpu->arch.shared = &vcpu->arch.shregs;
2417 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2418         /*
2419          * The shared struct is never shared on HV,
2420          * so we can always use host endianness
2421          */
2422 #ifdef __BIG_ENDIAN__
2423         vcpu->arch.shared_big_endian = true;
2424 #else
2425         vcpu->arch.shared_big_endian = false;
2426 #endif
2427 #endif
2428         vcpu->arch.mmcr[0] = MMCR0_FC;
2429         vcpu->arch.ctrl = CTRL_RUNLATCH;
2430         /* default to host PVR, since we can't spoof it */
2431         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2432         spin_lock_init(&vcpu->arch.vpa_update_lock);
2433         spin_lock_init(&vcpu->arch.tbacct_lock);
2434         vcpu->arch.busy_preempt = TB_NIL;
2435         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2436
2437         /*
2438          * Set the default HFSCR for the guest from the host value.
2439          * This value is only used on POWER9.
2440          * On POWER9, we want to virtualize the doorbell facility, so we
2441          * don't set the HFSCR_MSGP bit, and that causes those instructions
2442          * to trap and then we emulate them.
2443          */
2444         vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2445                 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP | HFSCR_PREFIX;
2446         if (cpu_has_feature(CPU_FTR_HVMODE)) {
2447                 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2448                 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2449                         vcpu->arch.hfscr |= HFSCR_TM;
2450         }
2451         if (cpu_has_feature(CPU_FTR_TM_COMP))
2452                 vcpu->arch.hfscr |= HFSCR_TM;
2453
2454         kvmppc_mmu_book3s_hv_init(vcpu);
2455
2456         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2457
2458         init_waitqueue_head(&vcpu->arch.cpu_run);
2459
2460         mutex_lock(&kvm->lock);
2461         vcore = NULL;
2462         err = -EINVAL;
2463         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2464                 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2465                         pr_devel("KVM: VCPU ID too high\n");
2466                         core = KVM_MAX_VCORES;
2467                 } else {
2468                         BUG_ON(kvm->arch.smt_mode != 1);
2469                         core = kvmppc_pack_vcpu_id(kvm, id);
2470                 }
2471         } else {
2472                 core = id / kvm->arch.smt_mode;
2473         }
2474         if (core < KVM_MAX_VCORES) {
2475                 vcore = kvm->arch.vcores[core];
2476                 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2477                         pr_devel("KVM: collision on id %u", id);
2478                         vcore = NULL;
2479                 } else if (!vcore) {
2480                         /*
2481                          * Take mmu_setup_lock for mutual exclusion
2482                          * with kvmppc_update_lpcr().
2483                          */
2484                         err = -ENOMEM;
2485                         vcore = kvmppc_vcore_create(kvm,
2486                                         id & ~(kvm->arch.smt_mode - 1));
2487                         mutex_lock(&kvm->arch.mmu_setup_lock);
2488                         kvm->arch.vcores[core] = vcore;
2489                         kvm->arch.online_vcores++;
2490                         mutex_unlock(&kvm->arch.mmu_setup_lock);
2491                 }
2492         }
2493         mutex_unlock(&kvm->lock);
2494
2495         if (!vcore)
2496                 return err;
2497
2498         spin_lock(&vcore->lock);
2499         ++vcore->num_threads;
2500         spin_unlock(&vcore->lock);
2501         vcpu->arch.vcore = vcore;
2502         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2503         vcpu->arch.thread_cpu = -1;
2504         vcpu->arch.prev_cpu = -1;
2505
2506         vcpu->arch.cpu_type = KVM_CPU_3S_64;
2507         kvmppc_sanity_check(vcpu);
2508
2509         debugfs_vcpu_init(vcpu, id);
2510
2511         return 0;
2512 }
2513
2514 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2515                               unsigned long flags)
2516 {
2517         int err;
2518         int esmt = 0;
2519
2520         if (flags)
2521                 return -EINVAL;
2522         if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2523                 return -EINVAL;
2524         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2525                 /*
2526                  * On POWER8 (or POWER7), the threading mode is "strict",
2527                  * so we pack smt_mode vcpus per vcore.
2528                  */
2529                 if (smt_mode > threads_per_subcore)
2530                         return -EINVAL;
2531         } else {
2532                 /*
2533                  * On POWER9, the threading mode is "loose",
2534                  * so each vcpu gets its own vcore.
2535                  */
2536                 esmt = smt_mode;
2537                 smt_mode = 1;
2538         }
2539         mutex_lock(&kvm->lock);
2540         err = -EBUSY;
2541         if (!kvm->arch.online_vcores) {
2542                 kvm->arch.smt_mode = smt_mode;
2543                 kvm->arch.emul_smt_mode = esmt;
2544                 err = 0;
2545         }
2546         mutex_unlock(&kvm->lock);
2547
2548         return err;
2549 }
2550
2551 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2552 {
2553         if (vpa->pinned_addr)
2554                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2555                                         vpa->dirty);
2556 }
2557
2558 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2559 {
2560         spin_lock(&vcpu->arch.vpa_update_lock);
2561         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2562         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2563         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2564         spin_unlock(&vcpu->arch.vpa_update_lock);
2565 }
2566
2567 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2568 {
2569         /* Indicate we want to get back into the guest */
2570         return 1;
2571 }
2572
2573 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2574 {
2575         unsigned long dec_nsec, now;
2576
2577         now = get_tb();
2578         if (now > vcpu->arch.dec_expires) {
2579                 /* decrementer has already gone negative */
2580                 kvmppc_core_queue_dec(vcpu);
2581                 kvmppc_core_prepare_to_enter(vcpu);
2582                 return;
2583         }
2584         dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2585         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2586         vcpu->arch.timer_running = 1;
2587 }
2588
2589 extern int __kvmppc_vcore_entry(void);
2590
2591 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2592                                    struct kvm_vcpu *vcpu)
2593 {
2594         u64 now;
2595
2596         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2597                 return;
2598         spin_lock_irq(&vcpu->arch.tbacct_lock);
2599         now = mftb();
2600         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2601                 vcpu->arch.stolen_logged;
2602         vcpu->arch.busy_preempt = now;
2603         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2604         spin_unlock_irq(&vcpu->arch.tbacct_lock);
2605         --vc->n_runnable;
2606         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2607 }
2608
2609 static int kvmppc_grab_hwthread(int cpu)
2610 {
2611         struct paca_struct *tpaca;
2612         long timeout = 10000;
2613
2614         tpaca = paca_ptrs[cpu];
2615
2616         /* Ensure the thread won't go into the kernel if it wakes */
2617         tpaca->kvm_hstate.kvm_vcpu = NULL;
2618         tpaca->kvm_hstate.kvm_vcore = NULL;
2619         tpaca->kvm_hstate.napping = 0;
2620         smp_wmb();
2621         tpaca->kvm_hstate.hwthread_req = 1;
2622
2623         /*
2624          * If the thread is already executing in the kernel (e.g. handling
2625          * a stray interrupt), wait for it to get back to nap mode.
2626          * The smp_mb() is to ensure that our setting of hwthread_req
2627          * is visible before we look at hwthread_state, so if this
2628          * races with the code at system_reset_pSeries and the thread
2629          * misses our setting of hwthread_req, we are sure to see its
2630          * setting of hwthread_state, and vice versa.
2631          */
2632         smp_mb();
2633         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2634                 if (--timeout <= 0) {
2635                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
2636                         return -EBUSY;
2637                 }
2638                 udelay(1);
2639         }
2640         return 0;
2641 }
2642
2643 static void kvmppc_release_hwthread(int cpu)
2644 {
2645         struct paca_struct *tpaca;
2646
2647         tpaca = paca_ptrs[cpu];
2648         tpaca->kvm_hstate.hwthread_req = 0;
2649         tpaca->kvm_hstate.kvm_vcpu = NULL;
2650         tpaca->kvm_hstate.kvm_vcore = NULL;
2651         tpaca->kvm_hstate.kvm_split_mode = NULL;
2652 }
2653
2654 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2655 {
2656         struct kvm_nested_guest *nested = vcpu->arch.nested;
2657         cpumask_t *cpu_in_guest;
2658         int i;
2659
2660         cpu = cpu_first_thread_sibling(cpu);
2661         if (nested) {
2662                 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2663                 cpu_in_guest = &nested->cpu_in_guest;
2664         } else {
2665                 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2666                 cpu_in_guest = &kvm->arch.cpu_in_guest;
2667         }
2668         /*
2669          * Make sure setting of bit in need_tlb_flush precedes
2670          * testing of cpu_in_guest bits.  The matching barrier on
2671          * the other side is the first smp_mb() in kvmppc_run_core().
2672          */
2673         smp_mb();
2674         for (i = 0; i < threads_per_core; ++i)
2675                 if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2676                         smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2677 }
2678
2679 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2680 {
2681         struct kvm_nested_guest *nested = vcpu->arch.nested;
2682         struct kvm *kvm = vcpu->kvm;
2683         int prev_cpu;
2684
2685         if (!cpu_has_feature(CPU_FTR_HVMODE))
2686                 return;
2687
2688         if (nested)
2689                 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2690         else
2691                 prev_cpu = vcpu->arch.prev_cpu;
2692
2693         /*
2694          * With radix, the guest can do TLB invalidations itself,
2695          * and it could choose to use the local form (tlbiel) if
2696          * it is invalidating a translation that has only ever been
2697          * used on one vcpu.  However, that doesn't mean it has
2698          * only ever been used on one physical cpu, since vcpus
2699          * can move around between pcpus.  To cope with this, when
2700          * a vcpu moves from one pcpu to another, we need to tell
2701          * any vcpus running on the same core as this vcpu previously
2702          * ran to flush the TLB.  The TLB is shared between threads,
2703          * so we use a single bit in .need_tlb_flush for all 4 threads.
2704          */
2705         if (prev_cpu != pcpu) {
2706                 if (prev_cpu >= 0 &&
2707                     cpu_first_thread_sibling(prev_cpu) !=
2708                     cpu_first_thread_sibling(pcpu))
2709                         radix_flush_cpu(kvm, prev_cpu, vcpu);
2710                 if (nested)
2711                         nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2712                 else
2713                         vcpu->arch.prev_cpu = pcpu;
2714         }
2715 }
2716
2717 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2718 {
2719         int cpu;
2720         struct paca_struct *tpaca;
2721         struct kvm *kvm = vc->kvm;
2722
2723         cpu = vc->pcpu;
2724         if (vcpu) {
2725                 if (vcpu->arch.timer_running) {
2726                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2727                         vcpu->arch.timer_running = 0;
2728                 }
2729                 cpu += vcpu->arch.ptid;
2730                 vcpu->cpu = vc->pcpu;
2731                 vcpu->arch.thread_cpu = cpu;
2732                 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2733         }
2734         tpaca = paca_ptrs[cpu];
2735         tpaca->kvm_hstate.kvm_vcpu = vcpu;
2736         tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2737         tpaca->kvm_hstate.fake_suspend = 0;
2738         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2739         smp_wmb();
2740         tpaca->kvm_hstate.kvm_vcore = vc;
2741         if (cpu != smp_processor_id())
2742                 kvmppc_ipi_thread(cpu);
2743 }
2744
2745 static void kvmppc_wait_for_nap(int n_threads)
2746 {
2747         int cpu = smp_processor_id();
2748         int i, loops;
2749
2750         if (n_threads <= 1)
2751                 return;
2752         for (loops = 0; loops < 1000000; ++loops) {
2753                 /*
2754                  * Check if all threads are finished.
2755                  * We set the vcore pointer when starting a thread
2756                  * and the thread clears it when finished, so we look
2757                  * for any threads that still have a non-NULL vcore ptr.
2758                  */
2759                 for (i = 1; i < n_threads; ++i)
2760                         if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2761                                 break;
2762                 if (i == n_threads) {
2763                         HMT_medium();
2764                         return;
2765                 }
2766                 HMT_low();
2767         }
2768         HMT_medium();
2769         for (i = 1; i < n_threads; ++i)
2770                 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2771                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2772 }
2773
2774 /*
2775  * Check that we are on thread 0 and that any other threads in
2776  * this core are off-line.  Then grab the threads so they can't
2777  * enter the kernel.
2778  */
2779 static int on_primary_thread(void)
2780 {
2781         int cpu = smp_processor_id();
2782         int thr;
2783
2784         /* Are we on a primary subcore? */
2785         if (cpu_thread_in_subcore(cpu))
2786                 return 0;
2787
2788         thr = 0;
2789         while (++thr < threads_per_subcore)
2790                 if (cpu_online(cpu + thr))
2791                         return 0;
2792
2793         /* Grab all hw threads so they can't go into the kernel */
2794         for (thr = 1; thr < threads_per_subcore; ++thr) {
2795                 if (kvmppc_grab_hwthread(cpu + thr)) {
2796                         /* Couldn't grab one; let the others go */
2797                         do {
2798                                 kvmppc_release_hwthread(cpu + thr);
2799                         } while (--thr > 0);
2800                         return 0;
2801                 }
2802         }
2803         return 1;
2804 }
2805
2806 /*
2807  * A list of virtual cores for each physical CPU.
2808  * These are vcores that could run but their runner VCPU tasks are
2809  * (or may be) preempted.
2810  */
2811 struct preempted_vcore_list {
2812         struct list_head        list;
2813         spinlock_t              lock;
2814 };
2815
2816 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2817
2818 static void init_vcore_lists(void)
2819 {
2820         int cpu;
2821
2822         for_each_possible_cpu(cpu) {
2823                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2824                 spin_lock_init(&lp->lock);
2825                 INIT_LIST_HEAD(&lp->list);
2826         }
2827 }
2828
2829 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2830 {
2831         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2832
2833         vc->vcore_state = VCORE_PREEMPT;
2834         vc->pcpu = smp_processor_id();
2835         if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2836                 spin_lock(&lp->lock);
2837                 list_add_tail(&vc->preempt_list, &lp->list);
2838                 spin_unlock(&lp->lock);
2839         }
2840
2841         /* Start accumulating stolen time */
2842         kvmppc_core_start_stolen(vc);
2843 }
2844
2845 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2846 {
2847         struct preempted_vcore_list *lp;
2848
2849         kvmppc_core_end_stolen(vc);
2850         if (!list_empty(&vc->preempt_list)) {
2851                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2852                 spin_lock(&lp->lock);
2853                 list_del_init(&vc->preempt_list);
2854                 spin_unlock(&lp->lock);
2855         }
2856         vc->vcore_state = VCORE_INACTIVE;
2857 }
2858
2859 /*
2860  * This stores information about the virtual cores currently
2861  * assigned to a physical core.
2862  */
2863 struct core_info {
2864         int             n_subcores;
2865         int             max_subcore_threads;
2866         int             total_threads;
2867         int             subcore_threads[MAX_SUBCORES];
2868         struct kvmppc_vcore *vc[MAX_SUBCORES];
2869 };
2870
2871 /*
2872  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2873  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2874  */
2875 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2876
2877 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2878 {
2879         memset(cip, 0, sizeof(*cip));
2880         cip->n_subcores = 1;
2881         cip->max_subcore_threads = vc->num_threads;
2882         cip->total_threads = vc->num_threads;
2883         cip->subcore_threads[0] = vc->num_threads;
2884         cip->vc[0] = vc;
2885 }
2886
2887 static bool subcore_config_ok(int n_subcores, int n_threads)
2888 {
2889         /*
2890          * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2891          * split-core mode, with one thread per subcore.
2892          */
2893         if (cpu_has_feature(CPU_FTR_ARCH_300))
2894                 return n_subcores <= 4 && n_threads == 1;
2895
2896         /* On POWER8, can only dynamically split if unsplit to begin with */
2897         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2898                 return false;
2899         if (n_subcores > MAX_SUBCORES)
2900                 return false;
2901         if (n_subcores > 1) {
2902                 if (!(dynamic_mt_modes & 2))
2903                         n_subcores = 4;
2904                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2905                         return false;
2906         }
2907
2908         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2909 }
2910
2911 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2912 {
2913         vc->entry_exit_map = 0;
2914         vc->in_guest = 0;
2915         vc->napping_threads = 0;
2916         vc->conferring_threads = 0;
2917         vc->tb_offset_applied = 0;
2918 }
2919
2920 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2921 {
2922         int n_threads = vc->num_threads;
2923         int sub;
2924
2925         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2926                 return false;
2927
2928         /* In one_vm_per_core mode, require all vcores to be from the same vm */
2929         if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2930                 return false;
2931
2932         if (n_threads < cip->max_subcore_threads)
2933                 n_threads = cip->max_subcore_threads;
2934         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2935                 return false;
2936         cip->max_subcore_threads = n_threads;
2937
2938         sub = cip->n_subcores;
2939         ++cip->n_subcores;
2940         cip->total_threads += vc->num_threads;
2941         cip->subcore_threads[sub] = vc->num_threads;
2942         cip->vc[sub] = vc;
2943         init_vcore_to_run(vc);
2944         list_del_init(&vc->preempt_list);
2945
2946         return true;
2947 }
2948
2949 /*
2950  * Work out whether it is possible to piggyback the execution of
2951  * vcore *pvc onto the execution of the other vcores described in *cip.
2952  */
2953 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2954                           int target_threads)
2955 {
2956         if (cip->total_threads + pvc->num_threads > target_threads)
2957                 return false;
2958
2959         return can_dynamic_split(pvc, cip);
2960 }
2961
2962 static void prepare_threads(struct kvmppc_vcore *vc)
2963 {
2964         int i;
2965         struct kvm_vcpu *vcpu;
2966
2967         for_each_runnable_thread(i, vcpu, vc) {
2968                 if (signal_pending(vcpu->arch.run_task))
2969                         vcpu->arch.ret = -EINTR;
2970                 else if (no_mixing_hpt_and_radix &&
2971                          kvm_is_radix(vc->kvm) != radix_enabled())
2972                         vcpu->arch.ret = -EINVAL;
2973                 else if (vcpu->arch.vpa.update_pending ||
2974                          vcpu->arch.slb_shadow.update_pending ||
2975                          vcpu->arch.dtl.update_pending)
2976                         vcpu->arch.ret = RESUME_GUEST;
2977                 else
2978                         continue;
2979                 kvmppc_remove_runnable(vc, vcpu);
2980                 wake_up(&vcpu->arch.cpu_run);
2981         }
2982 }
2983
2984 static void collect_piggybacks(struct core_info *cip, int target_threads)
2985 {
2986         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2987         struct kvmppc_vcore *pvc, *vcnext;
2988
2989         spin_lock(&lp->lock);
2990         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2991                 if (!spin_trylock(&pvc->lock))
2992                         continue;
2993                 prepare_threads(pvc);
2994                 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
2995                         list_del_init(&pvc->preempt_list);
2996                         if (pvc->runner == NULL) {
2997                                 pvc->vcore_state = VCORE_INACTIVE;
2998                                 kvmppc_core_end_stolen(pvc);
2999                         }
3000                         spin_unlock(&pvc->lock);
3001                         continue;
3002                 }
3003                 if (!can_piggyback(pvc, cip, target_threads)) {
3004                         spin_unlock(&pvc->lock);
3005                         continue;
3006                 }
3007                 kvmppc_core_end_stolen(pvc);
3008                 pvc->vcore_state = VCORE_PIGGYBACK;
3009                 if (cip->total_threads >= target_threads)
3010                         break;
3011         }
3012         spin_unlock(&lp->lock);
3013 }
3014
3015 static bool recheck_signals_and_mmu(struct core_info *cip)
3016 {
3017         int sub, i;
3018         struct kvm_vcpu *vcpu;
3019         struct kvmppc_vcore *vc;
3020
3021         for (sub = 0; sub < cip->n_subcores; ++sub) {
3022                 vc = cip->vc[sub];
3023                 if (!vc->kvm->arch.mmu_ready)
3024                         return true;
3025                 for_each_runnable_thread(i, vcpu, vc)
3026                         if (signal_pending(vcpu->arch.run_task))
3027                                 return true;
3028         }
3029         return false;
3030 }
3031
3032 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
3033 {
3034         int still_running = 0, i;
3035         u64 now;
3036         long ret;
3037         struct kvm_vcpu *vcpu;
3038
3039         spin_lock(&vc->lock);
3040         now = get_tb();
3041         for_each_runnable_thread(i, vcpu, vc) {
3042                 /*
3043                  * It's safe to unlock the vcore in the loop here, because
3044                  * for_each_runnable_thread() is safe against removal of
3045                  * the vcpu, and the vcore state is VCORE_EXITING here,
3046                  * so any vcpus becoming runnable will have their arch.trap
3047                  * set to zero and can't actually run in the guest.
3048                  */
3049                 spin_unlock(&vc->lock);
3050                 /* cancel pending dec exception if dec is positive */
3051                 if (now < vcpu->arch.dec_expires &&
3052                     kvmppc_core_pending_dec(vcpu))
3053                         kvmppc_core_dequeue_dec(vcpu);
3054
3055                 trace_kvm_guest_exit(vcpu);
3056
3057                 ret = RESUME_GUEST;
3058                 if (vcpu->arch.trap)
3059                         ret = kvmppc_handle_exit_hv(vcpu,
3060                                                     vcpu->arch.run_task);
3061
3062                 vcpu->arch.ret = ret;
3063                 vcpu->arch.trap = 0;
3064
3065                 spin_lock(&vc->lock);
3066                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
3067                         if (vcpu->arch.pending_exceptions)
3068                                 kvmppc_core_prepare_to_enter(vcpu);
3069                         if (vcpu->arch.ceded)
3070                                 kvmppc_set_timer(vcpu);
3071                         else
3072                                 ++still_running;
3073                 } else {
3074                         kvmppc_remove_runnable(vc, vcpu);
3075                         wake_up(&vcpu->arch.cpu_run);
3076                 }
3077         }
3078         if (!is_master) {
3079                 if (still_running > 0) {
3080                         kvmppc_vcore_preempt(vc);
3081                 } else if (vc->runner) {
3082                         vc->vcore_state = VCORE_PREEMPT;
3083                         kvmppc_core_start_stolen(vc);
3084                 } else {
3085                         vc->vcore_state = VCORE_INACTIVE;
3086                 }
3087                 if (vc->n_runnable > 0 && vc->runner == NULL) {
3088                         /* make sure there's a candidate runner awake */
3089                         i = -1;
3090                         vcpu = next_runnable_thread(vc, &i);
3091                         wake_up(&vcpu->arch.cpu_run);
3092                 }
3093         }
3094         spin_unlock(&vc->lock);
3095 }
3096
3097 /*
3098  * Clear core from the list of active host cores as we are about to
3099  * enter the guest. Only do this if it is the primary thread of the
3100  * core (not if a subcore) that is entering the guest.
3101  */
3102 static inline int kvmppc_clear_host_core(unsigned int cpu)
3103 {
3104         int core;
3105
3106         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3107                 return 0;
3108         /*
3109          * Memory barrier can be omitted here as we will do a smp_wmb()
3110          * later in kvmppc_start_thread and we need ensure that state is
3111          * visible to other CPUs only after we enter guest.
3112          */
3113         core = cpu >> threads_shift;
3114         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3115         return 0;
3116 }
3117
3118 /*
3119  * Advertise this core as an active host core since we exited the guest
3120  * Only need to do this if it is the primary thread of the core that is
3121  * exiting.
3122  */
3123 static inline int kvmppc_set_host_core(unsigned int cpu)
3124 {
3125         int core;
3126
3127         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3128                 return 0;
3129
3130         /*
3131          * Memory barrier can be omitted here because we do a spin_unlock
3132          * immediately after this which provides the memory barrier.
3133          */
3134         core = cpu >> threads_shift;
3135         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3136         return 0;
3137 }
3138
3139 static void set_irq_happened(int trap)
3140 {
3141         switch (trap) {
3142         case BOOK3S_INTERRUPT_EXTERNAL:
3143                 local_paca->irq_happened |= PACA_IRQ_EE;
3144                 break;
3145         case BOOK3S_INTERRUPT_H_DOORBELL:
3146                 local_paca->irq_happened |= PACA_IRQ_DBELL;
3147                 break;
3148         case BOOK3S_INTERRUPT_HMI:
3149                 local_paca->irq_happened |= PACA_IRQ_HMI;
3150                 break;
3151         case BOOK3S_INTERRUPT_SYSTEM_RESET:
3152                 replay_system_reset();
3153                 break;
3154         }
3155 }
3156
3157 /*
3158  * Run a set of guest threads on a physical core.
3159  * Called with vc->lock held.
3160  */
3161 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3162 {
3163         struct kvm_vcpu *vcpu;
3164         int i;
3165         int srcu_idx;
3166         struct core_info core_info;
3167         struct kvmppc_vcore *pvc;
3168         struct kvm_split_mode split_info, *sip;
3169         int split, subcore_size, active;
3170         int sub;
3171         bool thr0_done;
3172         unsigned long cmd_bit, stat_bit;
3173         int pcpu, thr;
3174         int target_threads;
3175         int controlled_threads;
3176         int trap;
3177         bool is_power8;
3178
3179         /*
3180          * Remove from the list any threads that have a signal pending
3181          * or need a VPA update done
3182          */
3183         prepare_threads(vc);
3184
3185         /* if the runner is no longer runnable, let the caller pick a new one */
3186         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3187                 return;
3188
3189         /*
3190          * Initialize *vc.
3191          */
3192         init_vcore_to_run(vc);
3193         vc->preempt_tb = TB_NIL;
3194
3195         /*
3196          * Number of threads that we will be controlling: the same as
3197          * the number of threads per subcore, except on POWER9,
3198          * where it's 1 because the threads are (mostly) independent.
3199          */
3200         controlled_threads = threads_per_vcore(vc->kvm);
3201
3202         /*
3203          * Make sure we are running on primary threads, and that secondary
3204          * threads are offline.  Also check if the number of threads in this
3205          * guest are greater than the current system threads per guest.
3206          * On POWER9, we need to be not in independent-threads mode if
3207          * this is a HPT guest on a radix host machine where the
3208          * CPU threads may not be in different MMU modes.
3209          */
3210         if ((controlled_threads > 1) &&
3211             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
3212                 for_each_runnable_thread(i, vcpu, vc) {
3213                         vcpu->arch.ret = -EBUSY;
3214                         kvmppc_remove_runnable(vc, vcpu);
3215                         wake_up(&vcpu->arch.cpu_run);
3216                 }
3217                 goto out;
3218         }
3219
3220         /*
3221          * See if we could run any other vcores on the physical core
3222          * along with this one.
3223          */
3224         init_core_info(&core_info, vc);
3225         pcpu = smp_processor_id();
3226         target_threads = controlled_threads;
3227         if (target_smt_mode && target_smt_mode < target_threads)
3228                 target_threads = target_smt_mode;
3229         if (vc->num_threads < target_threads)
3230                 collect_piggybacks(&core_info, target_threads);
3231
3232         /*
3233          * On radix, arrange for TLB flushing if necessary.
3234          * This has to be done before disabling interrupts since
3235          * it uses smp_call_function().
3236          */
3237         pcpu = smp_processor_id();
3238         if (kvm_is_radix(vc->kvm)) {
3239                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3240                         for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3241                                 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3242         }
3243
3244         /*
3245          * Hard-disable interrupts, and check resched flag and signals.
3246          * If we need to reschedule or deliver a signal, clean up
3247          * and return without going into the guest(s).
3248          * If the mmu_ready flag has been cleared, don't go into the
3249          * guest because that means a HPT resize operation is in progress.
3250          */
3251         local_irq_disable();
3252         hard_irq_disable();
3253         if (lazy_irq_pending() || need_resched() ||
3254             recheck_signals_and_mmu(&core_info)) {
3255                 local_irq_enable();
3256                 vc->vcore_state = VCORE_INACTIVE;
3257                 /* Unlock all except the primary vcore */
3258                 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3259                         pvc = core_info.vc[sub];
3260                         /* Put back on to the preempted vcores list */
3261                         kvmppc_vcore_preempt(pvc);
3262                         spin_unlock(&pvc->lock);
3263                 }
3264                 for (i = 0; i < controlled_threads; ++i)
3265                         kvmppc_release_hwthread(pcpu + i);
3266                 return;
3267         }
3268
3269         kvmppc_clear_host_core(pcpu);
3270
3271         /* Decide on micro-threading (split-core) mode */
3272         subcore_size = threads_per_subcore;
3273         cmd_bit = stat_bit = 0;
3274         split = core_info.n_subcores;
3275         sip = NULL;
3276         is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3277                 && !cpu_has_feature(CPU_FTR_ARCH_300);
3278
3279         if (split > 1) {
3280                 sip = &split_info;
3281                 memset(&split_info, 0, sizeof(split_info));
3282                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3283                         split_info.vc[sub] = core_info.vc[sub];
3284
3285                 if (is_power8) {
3286                         if (split == 2 && (dynamic_mt_modes & 2)) {
3287                                 cmd_bit = HID0_POWER8_1TO2LPAR;
3288                                 stat_bit = HID0_POWER8_2LPARMODE;
3289                         } else {
3290                                 split = 4;
3291                                 cmd_bit = HID0_POWER8_1TO4LPAR;
3292                                 stat_bit = HID0_POWER8_4LPARMODE;
3293                         }
3294                         subcore_size = MAX_SMT_THREADS / split;
3295                         split_info.rpr = mfspr(SPRN_RPR);
3296                         split_info.pmmar = mfspr(SPRN_PMMAR);
3297                         split_info.ldbar = mfspr(SPRN_LDBAR);
3298                         split_info.subcore_size = subcore_size;
3299                 } else {
3300                         split_info.subcore_size = 1;
3301                 }
3302
3303                 /* order writes to split_info before kvm_split_mode pointer */
3304                 smp_wmb();
3305         }
3306
3307         for (thr = 0; thr < controlled_threads; ++thr) {
3308                 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3309
3310                 paca->kvm_hstate.napping = 0;
3311                 paca->kvm_hstate.kvm_split_mode = sip;
3312         }
3313
3314         /* Initiate micro-threading (split-core) on POWER8 if required */
3315         if (cmd_bit) {
3316                 unsigned long hid0 = mfspr(SPRN_HID0);
3317
3318                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3319                 mb();
3320                 mtspr(SPRN_HID0, hid0);
3321                 isync();
3322                 for (;;) {
3323                         hid0 = mfspr(SPRN_HID0);
3324                         if (hid0 & stat_bit)
3325                                 break;
3326                         cpu_relax();
3327                 }
3328         }
3329
3330         /*
3331          * On POWER8, set RWMR register.
3332          * Since it only affects PURR and SPURR, it doesn't affect
3333          * the host, so we don't save/restore the host value.
3334          */
3335         if (is_power8) {
3336                 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3337                 int n_online = atomic_read(&vc->online_count);
3338
3339                 /*
3340                  * Use the 8-thread value if we're doing split-core
3341                  * or if the vcore's online count looks bogus.
3342                  */
3343                 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3344                     n_online >= 1 && n_online <= MAX_SMT_THREADS)
3345                         rwmr_val = p8_rwmr_values[n_online];
3346                 mtspr(SPRN_RWMR, rwmr_val);
3347         }
3348
3349         /* Start all the threads */
3350         active = 0;
3351         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3352                 thr = is_power8 ? subcore_thread_map[sub] : sub;
3353                 thr0_done = false;
3354                 active |= 1 << thr;
3355                 pvc = core_info.vc[sub];
3356                 pvc->pcpu = pcpu + thr;
3357                 for_each_runnable_thread(i, vcpu, pvc) {
3358                         kvmppc_start_thread(vcpu, pvc);
3359                         kvmppc_create_dtl_entry(vcpu, pvc);
3360                         trace_kvm_guest_enter(vcpu);
3361                         if (!vcpu->arch.ptid)
3362                                 thr0_done = true;
3363                         active |= 1 << (thr + vcpu->arch.ptid);
3364                 }
3365                 /*
3366                  * We need to start the first thread of each subcore
3367                  * even if it doesn't have a vcpu.
3368                  */
3369                 if (!thr0_done)
3370                         kvmppc_start_thread(NULL, pvc);
3371         }
3372
3373         /*
3374          * Ensure that split_info.do_nap is set after setting
3375          * the vcore pointer in the PACA of the secondaries.
3376          */
3377         smp_mb();
3378
3379         /*
3380          * When doing micro-threading, poke the inactive threads as well.
3381          * This gets them to the nap instruction after kvm_do_nap,
3382          * which reduces the time taken to unsplit later.
3383          */
3384         if (cmd_bit) {
3385                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
3386                 for (thr = 1; thr < threads_per_subcore; ++thr)
3387                         if (!(active & (1 << thr)))
3388                                 kvmppc_ipi_thread(pcpu + thr);
3389         }
3390
3391         vc->vcore_state = VCORE_RUNNING;
3392         preempt_disable();
3393
3394         trace_kvmppc_run_core(vc, 0);
3395
3396         for (sub = 0; sub < core_info.n_subcores; ++sub)
3397                 spin_unlock(&core_info.vc[sub]->lock);
3398
3399         guest_enter_irqoff();
3400
3401         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3402
3403         this_cpu_disable_ftrace();
3404
3405         /*
3406          * Interrupts will be enabled once we get into the guest,
3407          * so tell lockdep that we're about to enable interrupts.
3408          */
3409         trace_hardirqs_on();
3410
3411         trap = __kvmppc_vcore_entry();
3412
3413         trace_hardirqs_off();
3414
3415         this_cpu_enable_ftrace();
3416
3417         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3418
3419         set_irq_happened(trap);
3420
3421         spin_lock(&vc->lock);
3422         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3423         vc->vcore_state = VCORE_EXITING;
3424
3425         /* wait for secondary threads to finish writing their state to memory */
3426         kvmppc_wait_for_nap(controlled_threads);
3427
3428         /* Return to whole-core mode if we split the core earlier */
3429         if (cmd_bit) {
3430                 unsigned long hid0 = mfspr(SPRN_HID0);
3431                 unsigned long loops = 0;
3432
3433                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3434                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3435                 mb();
3436                 mtspr(SPRN_HID0, hid0);
3437                 isync();
3438                 for (;;) {
3439                         hid0 = mfspr(SPRN_HID0);
3440                         if (!(hid0 & stat_bit))
3441                                 break;
3442                         cpu_relax();
3443                         ++loops;
3444                 }
3445                 split_info.do_nap = 0;
3446         }
3447
3448         kvmppc_set_host_core(pcpu);
3449
3450         guest_exit_irqoff();
3451
3452         local_irq_enable();
3453
3454         /* Let secondaries go back to the offline loop */
3455         for (i = 0; i < controlled_threads; ++i) {
3456                 kvmppc_release_hwthread(pcpu + i);
3457                 if (sip && sip->napped[i])
3458                         kvmppc_ipi_thread(pcpu + i);
3459                 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3460         }
3461
3462         spin_unlock(&vc->lock);
3463
3464         /* make sure updates to secondary vcpu structs are visible now */
3465         smp_mb();
3466
3467         preempt_enable();
3468
3469         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3470                 pvc = core_info.vc[sub];
3471                 post_guest_process(pvc, pvc == vc);
3472         }
3473
3474         spin_lock(&vc->lock);
3475
3476  out:
3477         vc->vcore_state = VCORE_INACTIVE;
3478         trace_kvmppc_run_core(vc, 1);
3479 }
3480
3481 /*
3482  * Load up hypervisor-mode registers on P9.
3483  */
3484 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3485                                      unsigned long lpcr)
3486 {
3487         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3488         s64 hdec;
3489         u64 tb, purr, spurr;
3490         int trap;
3491         unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3492         unsigned long host_ciabr = mfspr(SPRN_CIABR);
3493         unsigned long host_dawr0 = mfspr(SPRN_DAWR0);
3494         unsigned long host_dawrx0 = mfspr(SPRN_DAWRX0);
3495         unsigned long host_psscr = mfspr(SPRN_PSSCR);
3496         unsigned long host_pidr = mfspr(SPRN_PID);
3497         unsigned long host_dawr1 = 0;
3498         unsigned long host_dawrx1 = 0;
3499
3500         if (cpu_has_feature(CPU_FTR_DAWR1)) {
3501                 host_dawr1 = mfspr(SPRN_DAWR1);
3502                 host_dawrx1 = mfspr(SPRN_DAWRX1);
3503         }
3504
3505         hdec = time_limit - mftb();
3506         if (hdec < 0)
3507                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3508
3509         if (vc->tb_offset) {
3510                 u64 new_tb = mftb() + vc->tb_offset;
3511                 mtspr(SPRN_TBU40, new_tb);
3512                 tb = mftb();
3513                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3514                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3515                 vc->tb_offset_applied = vc->tb_offset;
3516         }
3517
3518         if (vc->pcr)
3519                 mtspr(SPRN_PCR, vc->pcr | PCR_MASK);
3520         mtspr(SPRN_DPDES, vc->dpdes);
3521         mtspr(SPRN_VTB, vc->vtb);
3522
3523         local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3524         local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3525         mtspr(SPRN_PURR, vcpu->arch.purr);
3526         mtspr(SPRN_SPURR, vcpu->arch.spurr);
3527
3528         if (dawr_enabled()) {
3529                 mtspr(SPRN_DAWR0, vcpu->arch.dawr0);
3530                 mtspr(SPRN_DAWRX0, vcpu->arch.dawrx0);
3531                 if (cpu_has_feature(CPU_FTR_DAWR1)) {
3532                         mtspr(SPRN_DAWR1, vcpu->arch.dawr1);
3533                         mtspr(SPRN_DAWRX1, vcpu->arch.dawrx1);
3534                 }
3535         }
3536         mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3537         mtspr(SPRN_IC, vcpu->arch.ic);
3538         mtspr(SPRN_PID, vcpu->arch.pid);
3539
3540         mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3541               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3542
3543         mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3544
3545         mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3546         mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3547         mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3548         mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3549
3550         mtspr(SPRN_AMOR, ~0UL);
3551
3552         mtspr(SPRN_LPCR, lpcr);
3553         isync();
3554
3555         /*
3556          * P9 suppresses the HDEC exception when LPCR[HDICE] = 0,
3557          * so set guest LPCR (with HDICE) before writing HDEC.
3558          */
3559         mtspr(SPRN_HDEC, hdec);
3560
3561         kvmppc_xive_push_vcpu(vcpu);
3562
3563         mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3564         mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3565
3566         trap = __kvmhv_vcpu_entry_p9(vcpu);
3567
3568         kvmppc_xive_pull_vcpu(vcpu);
3569
3570         /* Advance host PURR/SPURR by the amount used by guest */
3571         purr = mfspr(SPRN_PURR);
3572         spurr = mfspr(SPRN_SPURR);
3573         mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3574               purr - vcpu->arch.purr);
3575         mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3576               spurr - vcpu->arch.spurr);
3577         vcpu->arch.purr = purr;
3578         vcpu->arch.spurr = spurr;
3579
3580         vcpu->arch.ic = mfspr(SPRN_IC);
3581         vcpu->arch.pid = mfspr(SPRN_PID);
3582         vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3583
3584         vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3585         vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3586         vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3587         vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3588
3589         /* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3590         mtspr(SPRN_PSSCR, host_psscr |
3591               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3592         mtspr(SPRN_HFSCR, host_hfscr);
3593         mtspr(SPRN_CIABR, host_ciabr);
3594         mtspr(SPRN_DAWR0, host_dawr0);
3595         mtspr(SPRN_DAWRX0, host_dawrx0);
3596         if (cpu_has_feature(CPU_FTR_DAWR1)) {
3597                 mtspr(SPRN_DAWR1, host_dawr1);
3598                 mtspr(SPRN_DAWRX1, host_dawrx1);
3599         }
3600         mtspr(SPRN_PID, host_pidr);
3601
3602         /*
3603          * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3604          * case we interrupted the guest between a tlbie and a ptesync.
3605          */
3606         asm volatile("eieio; tlbsync; ptesync");
3607
3608         /*
3609          * cp_abort is required if the processor supports local copy-paste
3610          * to clear the copy buffer that was under control of the guest.
3611          */
3612         if (cpu_has_feature(CPU_FTR_ARCH_31))
3613                 asm volatile(PPC_CP_ABORT);
3614
3615         mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid);    /* restore host LPID */
3616         isync();
3617
3618         vc->dpdes = mfspr(SPRN_DPDES);
3619         vc->vtb = mfspr(SPRN_VTB);
3620         mtspr(SPRN_DPDES, 0);
3621         if (vc->pcr)
3622                 mtspr(SPRN_PCR, PCR_MASK);
3623
3624         if (vc->tb_offset_applied) {
3625                 u64 new_tb = mftb() - vc->tb_offset_applied;
3626                 mtspr(SPRN_TBU40, new_tb);
3627                 tb = mftb();
3628                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3629                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3630                 vc->tb_offset_applied = 0;
3631         }
3632
3633         mtspr(SPRN_HDEC, 0x7fffffff);
3634         mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3635
3636         return trap;
3637 }
3638
3639 /*
3640  * Virtual-mode guest entry for POWER9 and later when the host and
3641  * guest are both using the radix MMU.  The LPIDR has already been set.
3642  */
3643 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3644                          unsigned long lpcr)
3645 {
3646         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3647         unsigned long host_dscr = mfspr(SPRN_DSCR);
3648         unsigned long host_tidr = mfspr(SPRN_TIDR);
3649         unsigned long host_iamr = mfspr(SPRN_IAMR);
3650         unsigned long host_amr = mfspr(SPRN_AMR);
3651         unsigned long host_fscr = mfspr(SPRN_FSCR);
3652         s64 dec;
3653         u64 tb;
3654         int trap, save_pmu;
3655
3656         dec = mfspr(SPRN_DEC);
3657         tb = mftb();
3658         if (dec < 0)
3659                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3660         local_paca->kvm_hstate.dec_expires = dec + tb;
3661         if (local_paca->kvm_hstate.dec_expires < time_limit)
3662                 time_limit = local_paca->kvm_hstate.dec_expires;
3663
3664         vcpu->arch.ceded = 0;
3665
3666         kvmhv_save_host_pmu();          /* saves it to PACA kvm_hstate */
3667
3668         kvmppc_subcore_enter_guest();
3669
3670         vc->entry_exit_map = 1;
3671         vc->in_guest = 1;
3672
3673         if (vcpu->arch.vpa.pinned_addr) {
3674                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3675                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3676                 lp->yield_count = cpu_to_be32(yield_count);
3677                 vcpu->arch.vpa.dirty = 1;
3678         }
3679
3680         if (cpu_has_feature(CPU_FTR_TM) ||
3681             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3682                 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3683
3684         kvmhv_load_guest_pmu(vcpu);
3685
3686         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3687         load_fp_state(&vcpu->arch.fp);
3688 #ifdef CONFIG_ALTIVEC
3689         load_vr_state(&vcpu->arch.vr);
3690 #endif
3691         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3692
3693         mtspr(SPRN_DSCR, vcpu->arch.dscr);
3694         mtspr(SPRN_IAMR, vcpu->arch.iamr);
3695         mtspr(SPRN_PSPB, vcpu->arch.pspb);
3696         mtspr(SPRN_FSCR, vcpu->arch.fscr);
3697         mtspr(SPRN_TAR, vcpu->arch.tar);
3698         mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3699         mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3700         mtspr(SPRN_BESCR, vcpu->arch.bescr);
3701         mtspr(SPRN_WORT, vcpu->arch.wort);
3702         mtspr(SPRN_TIDR, vcpu->arch.tid);
3703         mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3704         mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3705         mtspr(SPRN_AMR, vcpu->arch.amr);
3706         mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3707
3708         if (!(vcpu->arch.ctrl & 1))
3709                 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3710
3711         mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3712
3713         if (kvmhv_on_pseries()) {
3714                 /*
3715                  * We need to save and restore the guest visible part of the
3716                  * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3717                  * doesn't do this for us. Note only required if pseries since
3718                  * this is done in kvmhv_load_hv_regs_and_go() below otherwise.
3719                  */
3720                 unsigned long host_psscr;
3721                 /* call our hypervisor to load up HV regs and go */
3722                 struct hv_guest_state hvregs;
3723
3724                 host_psscr = mfspr(SPRN_PSSCR_PR);
3725                 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3726                 kvmhv_save_hv_regs(vcpu, &hvregs);
3727                 hvregs.lpcr = lpcr;
3728                 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3729                 hvregs.version = HV_GUEST_STATE_VERSION;
3730                 if (vcpu->arch.nested) {
3731                         hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3732                         hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3733                 } else {
3734                         hvregs.lpid = vcpu->kvm->arch.lpid;
3735                         hvregs.vcpu_token = vcpu->vcpu_id;
3736                 }
3737                 hvregs.hdec_expiry = time_limit;
3738                 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3739                                           __pa(&vcpu->arch.regs));
3740                 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3741                 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3742                 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3743                 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3744                 vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3745                 mtspr(SPRN_PSSCR_PR, host_psscr);
3746
3747                 /* H_CEDE has to be handled now, not later */
3748                 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3749                     kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3750                         kvmppc_nested_cede(vcpu);
3751                         kvmppc_set_gpr(vcpu, 3, 0);
3752                         trap = 0;
3753                 }
3754         } else {
3755                 trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3756         }
3757
3758         vcpu->arch.slb_max = 0;
3759         dec = mfspr(SPRN_DEC);
3760         if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
3761                 dec = (s32) dec;
3762         tb = mftb();
3763         vcpu->arch.dec_expires = dec + tb;
3764         vcpu->cpu = -1;
3765         vcpu->arch.thread_cpu = -1;
3766         /* Save guest CTRL register, set runlatch to 1 */
3767         vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3768         if (!(vcpu->arch.ctrl & 1))
3769                 mtspr(SPRN_CTRLT, vcpu->arch.ctrl | 1);
3770
3771         vcpu->arch.iamr = mfspr(SPRN_IAMR);
3772         vcpu->arch.pspb = mfspr(SPRN_PSPB);
3773         vcpu->arch.fscr = mfspr(SPRN_FSCR);
3774         vcpu->arch.tar = mfspr(SPRN_TAR);
3775         vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3776         vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3777         vcpu->arch.bescr = mfspr(SPRN_BESCR);
3778         vcpu->arch.wort = mfspr(SPRN_WORT);
3779         vcpu->arch.tid = mfspr(SPRN_TIDR);
3780         vcpu->arch.amr = mfspr(SPRN_AMR);
3781         vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3782         vcpu->arch.dscr = mfspr(SPRN_DSCR);
3783
3784         mtspr(SPRN_PSPB, 0);
3785         mtspr(SPRN_WORT, 0);
3786         mtspr(SPRN_UAMOR, 0);
3787         mtspr(SPRN_DSCR, host_dscr);
3788         mtspr(SPRN_TIDR, host_tidr);
3789         mtspr(SPRN_IAMR, host_iamr);
3790
3791         if (host_amr != vcpu->arch.amr)
3792                 mtspr(SPRN_AMR, host_amr);
3793
3794         if (host_fscr != vcpu->arch.fscr)
3795                 mtspr(SPRN_FSCR, host_fscr);
3796
3797         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3798         store_fp_state(&vcpu->arch.fp);
3799 #ifdef CONFIG_ALTIVEC
3800         store_vr_state(&vcpu->arch.vr);
3801 #endif
3802         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3803
3804         if (cpu_has_feature(CPU_FTR_TM) ||
3805             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3806                 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3807
3808         save_pmu = 1;
3809         if (vcpu->arch.vpa.pinned_addr) {
3810                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3811                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3812                 lp->yield_count = cpu_to_be32(yield_count);
3813                 vcpu->arch.vpa.dirty = 1;
3814                 save_pmu = lp->pmcregs_in_use;
3815         }
3816         /* Must save pmu if this guest is capable of running nested guests */
3817         save_pmu |= nesting_enabled(vcpu->kvm);
3818
3819         kvmhv_save_guest_pmu(vcpu, save_pmu);
3820
3821         vc->entry_exit_map = 0x101;
3822         vc->in_guest = 0;
3823
3824         mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3825         mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3826
3827         kvmhv_load_host_pmu();
3828
3829         kvmppc_subcore_exit_guest();
3830
3831         return trap;
3832 }
3833
3834 /*
3835  * Wait for some other vcpu thread to execute us, and
3836  * wake us up when we need to handle something in the host.
3837  */
3838 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3839                                  struct kvm_vcpu *vcpu, int wait_state)
3840 {
3841         DEFINE_WAIT(wait);
3842
3843         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3844         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3845                 spin_unlock(&vc->lock);
3846                 schedule();
3847                 spin_lock(&vc->lock);
3848         }
3849         finish_wait(&vcpu->arch.cpu_run, &wait);
3850 }
3851
3852 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3853 {
3854         if (!halt_poll_ns_grow)
3855                 return;
3856
3857         vc->halt_poll_ns *= halt_poll_ns_grow;
3858         if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3859                 vc->halt_poll_ns = halt_poll_ns_grow_start;
3860 }
3861
3862 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3863 {
3864         if (halt_poll_ns_shrink == 0)
3865                 vc->halt_poll_ns = 0;
3866         else
3867                 vc->halt_poll_ns /= halt_poll_ns_shrink;
3868 }
3869
3870 #ifdef CONFIG_KVM_XICS
3871 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3872 {
3873         if (!xics_on_xive())
3874                 return false;
3875         return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3876                 vcpu->arch.xive_saved_state.cppr;
3877 }
3878 #else
3879 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3880 {
3881         return false;
3882 }
3883 #endif /* CONFIG_KVM_XICS */
3884
3885 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3886 {
3887         if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3888             kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3889                 return true;
3890
3891         return false;
3892 }
3893
3894 /*
3895  * Check to see if any of the runnable vcpus on the vcore have pending
3896  * exceptions or are no longer ceded
3897  */
3898 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3899 {
3900         struct kvm_vcpu *vcpu;
3901         int i;
3902
3903         for_each_runnable_thread(i, vcpu, vc) {
3904                 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3905                         return 1;
3906         }
3907
3908         return 0;
3909 }
3910
3911 /*
3912  * All the vcpus in this vcore are idle, so wait for a decrementer
3913  * or external interrupt to one of the vcpus.  vc->lock is held.
3914  */
3915 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3916 {
3917         ktime_t cur, start_poll, start_wait;
3918         int do_sleep = 1;
3919         u64 block_ns;
3920
3921         /* Poll for pending exceptions and ceded state */
3922         cur = start_poll = ktime_get();
3923         if (vc->halt_poll_ns) {
3924                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3925                 ++vc->runner->stat.halt_attempted_poll;
3926
3927                 vc->vcore_state = VCORE_POLLING;
3928                 spin_unlock(&vc->lock);
3929
3930                 do {
3931                         if (kvmppc_vcore_check_block(vc)) {
3932                                 do_sleep = 0;
3933                                 break;
3934                         }
3935                         cur = ktime_get();
3936                 } while (single_task_running() && ktime_before(cur, stop));
3937
3938                 spin_lock(&vc->lock);
3939                 vc->vcore_state = VCORE_INACTIVE;
3940
3941                 if (!do_sleep) {
3942                         ++vc->runner->stat.halt_successful_poll;
3943                         goto out;
3944                 }
3945         }
3946
3947         prepare_to_rcuwait(&vc->wait);
3948         set_current_state(TASK_INTERRUPTIBLE);
3949         if (kvmppc_vcore_check_block(vc)) {
3950                 finish_rcuwait(&vc->wait);
3951                 do_sleep = 0;
3952                 /* If we polled, count this as a successful poll */
3953                 if (vc->halt_poll_ns)
3954                         ++vc->runner->stat.halt_successful_poll;
3955                 goto out;
3956         }
3957
3958         start_wait = ktime_get();
3959
3960         vc->vcore_state = VCORE_SLEEPING;
3961         trace_kvmppc_vcore_blocked(vc, 0);
3962         spin_unlock(&vc->lock);
3963         schedule();
3964         finish_rcuwait(&vc->wait);
3965         spin_lock(&vc->lock);
3966         vc->vcore_state = VCORE_INACTIVE;
3967         trace_kvmppc_vcore_blocked(vc, 1);
3968         ++vc->runner->stat.halt_successful_wait;
3969
3970         cur = ktime_get();
3971
3972 out:
3973         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3974
3975         /* Attribute wait time */
3976         if (do_sleep) {
3977                 vc->runner->stat.halt_wait_ns +=
3978                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
3979                 /* Attribute failed poll time */
3980                 if (vc->halt_poll_ns)
3981                         vc->runner->stat.halt_poll_fail_ns +=
3982                                 ktime_to_ns(start_wait) -
3983                                 ktime_to_ns(start_poll);
3984         } else {
3985                 /* Attribute successful poll time */
3986                 if (vc->halt_poll_ns)
3987                         vc->runner->stat.halt_poll_success_ns +=
3988                                 ktime_to_ns(cur) -
3989                                 ktime_to_ns(start_poll);
3990         }
3991
3992         /* Adjust poll time */
3993         if (halt_poll_ns) {
3994                 if (block_ns <= vc->halt_poll_ns)
3995                         ;
3996                 /* We slept and blocked for longer than the max halt time */
3997                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3998                         shrink_halt_poll_ns(vc);
3999                 /* We slept and our poll time is too small */
4000                 else if (vc->halt_poll_ns < halt_poll_ns &&
4001                                 block_ns < halt_poll_ns)
4002                         grow_halt_poll_ns(vc);
4003                 if (vc->halt_poll_ns > halt_poll_ns)
4004                         vc->halt_poll_ns = halt_poll_ns;
4005         } else
4006                 vc->halt_poll_ns = 0;
4007
4008         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
4009 }
4010
4011 /*
4012  * This never fails for a radix guest, as none of the operations it does
4013  * for a radix guest can fail or have a way to report failure.
4014  * kvmhv_run_single_vcpu() relies on this fact.
4015  */
4016 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
4017 {
4018         int r = 0;
4019         struct kvm *kvm = vcpu->kvm;
4020
4021         mutex_lock(&kvm->arch.mmu_setup_lock);
4022         if (!kvm->arch.mmu_ready) {
4023                 if (!kvm_is_radix(kvm))
4024                         r = kvmppc_hv_setup_htab_rma(vcpu);
4025                 if (!r) {
4026                         if (cpu_has_feature(CPU_FTR_ARCH_300))
4027                                 kvmppc_setup_partition_table(kvm);
4028                         kvm->arch.mmu_ready = 1;
4029                 }
4030         }
4031         mutex_unlock(&kvm->arch.mmu_setup_lock);
4032         return r;
4033 }
4034
4035 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
4036 {
4037         struct kvm_run *run = vcpu->run;
4038         int n_ceded, i, r;
4039         struct kvmppc_vcore *vc;
4040         struct kvm_vcpu *v;
4041
4042         trace_kvmppc_run_vcpu_enter(vcpu);
4043
4044         run->exit_reason = 0;
4045         vcpu->arch.ret = RESUME_GUEST;
4046         vcpu->arch.trap = 0;
4047         kvmppc_update_vpas(vcpu);
4048
4049         /*
4050          * Synchronize with other threads in this virtual core
4051          */
4052         vc = vcpu->arch.vcore;
4053         spin_lock(&vc->lock);
4054         vcpu->arch.ceded = 0;
4055         vcpu->arch.run_task = current;
4056         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4057         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4058         vcpu->arch.busy_preempt = TB_NIL;
4059         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4060         ++vc->n_runnable;
4061
4062         /*
4063          * This happens the first time this is called for a vcpu.
4064          * If the vcore is already running, we may be able to start
4065          * this thread straight away and have it join in.
4066          */
4067         if (!signal_pending(current)) {
4068                 if ((vc->vcore_state == VCORE_PIGGYBACK ||
4069                      vc->vcore_state == VCORE_RUNNING) &&
4070                            !VCORE_IS_EXITING(vc)) {
4071                         kvmppc_create_dtl_entry(vcpu, vc);
4072                         kvmppc_start_thread(vcpu, vc);
4073                         trace_kvm_guest_enter(vcpu);
4074                 } else if (vc->vcore_state == VCORE_SLEEPING) {
4075                         rcuwait_wake_up(&vc->wait);
4076                 }
4077
4078         }
4079
4080         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4081                !signal_pending(current)) {
4082                 /* See if the MMU is ready to go */
4083                 if (!vcpu->kvm->arch.mmu_ready) {
4084                         spin_unlock(&vc->lock);
4085                         r = kvmhv_setup_mmu(vcpu);
4086                         spin_lock(&vc->lock);
4087                         if (r) {
4088                                 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4089                                 run->fail_entry.
4090                                         hardware_entry_failure_reason = 0;
4091                                 vcpu->arch.ret = r;
4092                                 break;
4093                         }
4094                 }
4095
4096                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4097                         kvmppc_vcore_end_preempt(vc);
4098
4099                 if (vc->vcore_state != VCORE_INACTIVE) {
4100                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4101                         continue;
4102                 }
4103                 for_each_runnable_thread(i, v, vc) {
4104                         kvmppc_core_prepare_to_enter(v);
4105                         if (signal_pending(v->arch.run_task)) {
4106                                 kvmppc_remove_runnable(vc, v);
4107                                 v->stat.signal_exits++;
4108                                 v->run->exit_reason = KVM_EXIT_INTR;
4109                                 v->arch.ret = -EINTR;
4110                                 wake_up(&v->arch.cpu_run);
4111                         }
4112                 }
4113                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4114                         break;
4115                 n_ceded = 0;
4116                 for_each_runnable_thread(i, v, vc) {
4117                         if (!kvmppc_vcpu_woken(v))
4118                                 n_ceded += v->arch.ceded;
4119                         else
4120                                 v->arch.ceded = 0;
4121                 }
4122                 vc->runner = vcpu;
4123                 if (n_ceded == vc->n_runnable) {
4124                         kvmppc_vcore_blocked(vc);
4125                 } else if (need_resched()) {
4126                         kvmppc_vcore_preempt(vc);
4127                         /* Let something else run */
4128                         cond_resched_lock(&vc->lock);
4129                         if (vc->vcore_state == VCORE_PREEMPT)
4130                                 kvmppc_vcore_end_preempt(vc);
4131                 } else {
4132                         kvmppc_run_core(vc);
4133                 }
4134                 vc->runner = NULL;
4135         }
4136
4137         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4138                (vc->vcore_state == VCORE_RUNNING ||
4139                 vc->vcore_state == VCORE_EXITING ||
4140                 vc->vcore_state == VCORE_PIGGYBACK))
4141                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4142
4143         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4144                 kvmppc_vcore_end_preempt(vc);
4145
4146         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4147                 kvmppc_remove_runnable(vc, vcpu);
4148                 vcpu->stat.signal_exits++;
4149                 run->exit_reason = KVM_EXIT_INTR;
4150                 vcpu->arch.ret = -EINTR;
4151         }
4152
4153         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4154                 /* Wake up some vcpu to run the core */
4155                 i = -1;
4156                 v = next_runnable_thread(vc, &i);
4157                 wake_up(&v->arch.cpu_run);
4158         }
4159
4160         trace_kvmppc_run_vcpu_exit(vcpu);
4161         spin_unlock(&vc->lock);
4162         return vcpu->arch.ret;
4163 }
4164
4165 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4166                           unsigned long lpcr)
4167 {
4168         struct kvm_run *run = vcpu->run;
4169         int trap, r, pcpu;
4170         int srcu_idx, lpid;
4171         struct kvmppc_vcore *vc;
4172         struct kvm *kvm = vcpu->kvm;
4173         struct kvm_nested_guest *nested = vcpu->arch.nested;
4174
4175         trace_kvmppc_run_vcpu_enter(vcpu);
4176
4177         run->exit_reason = 0;
4178         vcpu->arch.ret = RESUME_GUEST;
4179         vcpu->arch.trap = 0;
4180
4181         vc = vcpu->arch.vcore;
4182         vcpu->arch.ceded = 0;
4183         vcpu->arch.run_task = current;
4184         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4185         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4186         vcpu->arch.busy_preempt = TB_NIL;
4187         vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4188         vc->runnable_threads[0] = vcpu;
4189         vc->n_runnable = 1;
4190         vc->runner = vcpu;
4191
4192         /* See if the MMU is ready to go */
4193         if (!kvm->arch.mmu_ready)
4194                 kvmhv_setup_mmu(vcpu);
4195
4196         if (need_resched())
4197                 cond_resched();
4198
4199         kvmppc_update_vpas(vcpu);
4200
4201         init_vcore_to_run(vc);
4202         vc->preempt_tb = TB_NIL;
4203
4204         preempt_disable();
4205         pcpu = smp_processor_id();
4206         vc->pcpu = pcpu;
4207         kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4208
4209         local_irq_disable();
4210         hard_irq_disable();
4211         if (signal_pending(current))
4212                 goto sigpend;
4213         if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4214                 goto out;
4215
4216         if (!nested) {
4217                 kvmppc_core_prepare_to_enter(vcpu);
4218                 if (vcpu->arch.doorbell_request) {
4219                         vc->dpdes = 1;
4220                         smp_wmb();
4221                         vcpu->arch.doorbell_request = 0;
4222                 }
4223                 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4224                              &vcpu->arch.pending_exceptions))
4225                         lpcr |= LPCR_MER;
4226         } else if (vcpu->arch.pending_exceptions ||
4227                    vcpu->arch.doorbell_request ||
4228                    xive_interrupt_pending(vcpu)) {
4229                 vcpu->arch.ret = RESUME_HOST;
4230                 goto out;
4231         }
4232
4233         kvmppc_clear_host_core(pcpu);
4234
4235         local_paca->kvm_hstate.napping = 0;
4236         local_paca->kvm_hstate.kvm_split_mode = NULL;
4237         kvmppc_start_thread(vcpu, vc);
4238         kvmppc_create_dtl_entry(vcpu, vc);
4239         trace_kvm_guest_enter(vcpu);
4240
4241         vc->vcore_state = VCORE_RUNNING;
4242         trace_kvmppc_run_core(vc, 0);
4243
4244         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4245                 lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4246                 mtspr(SPRN_LPID, lpid);
4247                 isync();
4248                 kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4249         }
4250
4251         guest_enter_irqoff();
4252
4253         srcu_idx = srcu_read_lock(&kvm->srcu);
4254
4255         this_cpu_disable_ftrace();
4256
4257         /* Tell lockdep that we're about to enable interrupts */
4258         trace_hardirqs_on();
4259
4260         trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4261         vcpu->arch.trap = trap;
4262
4263         trace_hardirqs_off();
4264
4265         this_cpu_enable_ftrace();
4266
4267         srcu_read_unlock(&kvm->srcu, srcu_idx);
4268
4269         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4270                 mtspr(SPRN_LPID, kvm->arch.host_lpid);
4271                 isync();
4272         }
4273
4274         set_irq_happened(trap);
4275
4276         kvmppc_set_host_core(pcpu);
4277
4278         guest_exit_irqoff();
4279
4280         local_irq_enable();
4281
4282         cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4283
4284         preempt_enable();
4285
4286         /*
4287          * cancel pending decrementer exception if DEC is now positive, or if
4288          * entering a nested guest in which case the decrementer is now owned
4289          * by L2 and the L1 decrementer is provided in hdec_expires
4290          */
4291         if (kvmppc_core_pending_dec(vcpu) &&
4292                         ((get_tb() < vcpu->arch.dec_expires) ||
4293                          (trap == BOOK3S_INTERRUPT_SYSCALL &&
4294                           kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4295                 kvmppc_core_dequeue_dec(vcpu);
4296
4297         trace_kvm_guest_exit(vcpu);
4298         r = RESUME_GUEST;
4299         if (trap) {
4300                 if (!nested)
4301                         r = kvmppc_handle_exit_hv(vcpu, current);
4302                 else
4303                         r = kvmppc_handle_nested_exit(vcpu);
4304         }
4305         vcpu->arch.ret = r;
4306
4307         if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4308             !kvmppc_vcpu_woken(vcpu)) {
4309                 kvmppc_set_timer(vcpu);
4310                 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4311                         if (signal_pending(current)) {
4312                                 vcpu->stat.signal_exits++;
4313                                 run->exit_reason = KVM_EXIT_INTR;
4314                                 vcpu->arch.ret = -EINTR;
4315                                 break;
4316                         }
4317                         spin_lock(&vc->lock);
4318                         kvmppc_vcore_blocked(vc);
4319                         spin_unlock(&vc->lock);
4320                 }
4321         }
4322         vcpu->arch.ceded = 0;
4323
4324         vc->vcore_state = VCORE_INACTIVE;
4325         trace_kvmppc_run_core(vc, 1);
4326
4327  done:
4328         kvmppc_remove_runnable(vc, vcpu);
4329         trace_kvmppc_run_vcpu_exit(vcpu);
4330
4331         return vcpu->arch.ret;
4332
4333  sigpend:
4334         vcpu->stat.signal_exits++;
4335         run->exit_reason = KVM_EXIT_INTR;
4336         vcpu->arch.ret = -EINTR;
4337  out:
4338         local_irq_enable();
4339         preempt_enable();
4340         goto done;
4341 }
4342
4343 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
4344 {
4345         struct kvm_run *run = vcpu->run;
4346         int r;
4347         int srcu_idx;
4348         unsigned long ebb_regs[3] = {}; /* shut up GCC */
4349         unsigned long user_tar = 0;
4350         unsigned int user_vrsave;
4351         struct kvm *kvm;
4352
4353         if (!vcpu->arch.sane) {
4354                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4355                 return -EINVAL;
4356         }
4357
4358         /*
4359          * Don't allow entry with a suspended transaction, because
4360          * the guest entry/exit code will lose it.
4361          * If the guest has TM enabled, save away their TM-related SPRs
4362          * (they will get restored by the TM unavailable interrupt).
4363          */
4364 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4365         if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4366             (current->thread.regs->msr & MSR_TM)) {
4367                 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4368                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4369                         run->fail_entry.hardware_entry_failure_reason = 0;
4370                         return -EINVAL;
4371                 }
4372                 /* Enable TM so we can read the TM SPRs */
4373                 mtmsr(mfmsr() | MSR_TM);
4374                 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4375                 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4376                 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4377                 current->thread.regs->msr &= ~MSR_TM;
4378         }
4379 #endif
4380
4381         /*
4382          * Force online to 1 for the sake of old userspace which doesn't
4383          * set it.
4384          */
4385         if (!vcpu->arch.online) {
4386                 atomic_inc(&vcpu->arch.vcore->online_count);
4387                 vcpu->arch.online = 1;
4388         }
4389
4390         kvmppc_core_prepare_to_enter(vcpu);
4391
4392         /* No need to go into the guest when all we'll do is come back out */
4393         if (signal_pending(current)) {
4394                 run->exit_reason = KVM_EXIT_INTR;
4395                 return -EINTR;
4396         }
4397
4398         kvm = vcpu->kvm;
4399         atomic_inc(&kvm->arch.vcpus_running);
4400         /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4401         smp_mb();
4402
4403         flush_all_to_thread(current);
4404
4405         /* Save userspace EBB and other register values */
4406         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4407                 ebb_regs[0] = mfspr(SPRN_EBBHR);
4408                 ebb_regs[1] = mfspr(SPRN_EBBRR);
4409                 ebb_regs[2] = mfspr(SPRN_BESCR);
4410                 user_tar = mfspr(SPRN_TAR);
4411         }
4412         user_vrsave = mfspr(SPRN_VRSAVE);
4413
4414         vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4415         vcpu->arch.pgdir = kvm->mm->pgd;
4416         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4417
4418         do {
4419                 /*
4420                  * The TLB prefetch bug fixup is only in the kvmppc_run_vcpu
4421                  * path, which also handles hash and dependent threads mode.
4422                  */
4423                 if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4424                     !cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
4425                         r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
4426                                                   vcpu->arch.vcore->lpcr);
4427                 else
4428                         r = kvmppc_run_vcpu(vcpu);
4429
4430                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4431                     !(vcpu->arch.shregs.msr & MSR_PR)) {
4432                         trace_kvm_hcall_enter(vcpu);
4433                         r = kvmppc_pseries_do_hcall(vcpu);
4434                         trace_kvm_hcall_exit(vcpu, r);
4435                         kvmppc_core_prepare_to_enter(vcpu);
4436                 } else if (r == RESUME_PAGE_FAULT) {
4437                         srcu_idx = srcu_read_lock(&kvm->srcu);
4438                         r = kvmppc_book3s_hv_page_fault(vcpu,
4439                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4440                         srcu_read_unlock(&kvm->srcu, srcu_idx);
4441                 } else if (r == RESUME_PASSTHROUGH) {
4442                         if (WARN_ON(xics_on_xive()))
4443                                 r = H_SUCCESS;
4444                         else
4445                                 r = kvmppc_xics_rm_complete(vcpu, 0);
4446                 }
4447         } while (is_kvmppc_resume_guest(r));
4448
4449         /* Restore userspace EBB and other register values */
4450         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4451                 mtspr(SPRN_EBBHR, ebb_regs[0]);
4452                 mtspr(SPRN_EBBRR, ebb_regs[1]);
4453                 mtspr(SPRN_BESCR, ebb_regs[2]);
4454                 mtspr(SPRN_TAR, user_tar);
4455         }
4456         mtspr(SPRN_VRSAVE, user_vrsave);
4457
4458         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4459         atomic_dec(&kvm->arch.vcpus_running);
4460         return r;
4461 }
4462
4463 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4464                                      int shift, int sllp)
4465 {
4466         (*sps)->page_shift = shift;
4467         (*sps)->slb_enc = sllp;
4468         (*sps)->enc[0].page_shift = shift;
4469         (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4470         /*
4471          * Add 16MB MPSS support (may get filtered out by userspace)
4472          */
4473         if (shift != 24) {
4474                 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4475                 if (penc != -1) {
4476                         (*sps)->enc[1].page_shift = 24;
4477                         (*sps)->enc[1].pte_enc = penc;
4478                 }
4479         }
4480         (*sps)++;
4481 }
4482
4483 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4484                                          struct kvm_ppc_smmu_info *info)
4485 {
4486         struct kvm_ppc_one_seg_page_size *sps;
4487
4488         /*
4489          * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4490          * POWER7 doesn't support keys for instruction accesses,
4491          * POWER8 and POWER9 do.
4492          */
4493         info->data_keys = 32;
4494         info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4495
4496         /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4497         info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4498         info->slb_size = 32;
4499
4500         /* We only support these sizes for now, and no muti-size segments */
4501         sps = &info->sps[0];
4502         kvmppc_add_seg_page_size(&sps, 12, 0);
4503         kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4504         kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4505
4506         /* If running as a nested hypervisor, we don't support HPT guests */
4507         if (kvmhv_on_pseries())
4508                 info->flags |= KVM_PPC_NO_HASH;
4509
4510         return 0;
4511 }
4512
4513 /*
4514  * Get (and clear) the dirty memory log for a memory slot.
4515  */
4516 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4517                                          struct kvm_dirty_log *log)
4518 {
4519         struct kvm_memslots *slots;
4520         struct kvm_memory_slot *memslot;
4521         int i, r;
4522         unsigned long n;
4523         unsigned long *buf, *p;
4524         struct kvm_vcpu *vcpu;
4525
4526         mutex_lock(&kvm->slots_lock);
4527
4528         r = -EINVAL;
4529         if (log->slot >= KVM_USER_MEM_SLOTS)
4530                 goto out;
4531
4532         slots = kvm_memslots(kvm);
4533         memslot = id_to_memslot(slots, log->slot);
4534         r = -ENOENT;
4535         if (!memslot || !memslot->dirty_bitmap)
4536                 goto out;
4537
4538         /*
4539          * Use second half of bitmap area because both HPT and radix
4540          * accumulate bits in the first half.
4541          */
4542         n = kvm_dirty_bitmap_bytes(memslot);
4543         buf = memslot->dirty_bitmap + n / sizeof(long);
4544         memset(buf, 0, n);
4545
4546         if (kvm_is_radix(kvm))
4547                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4548         else
4549                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4550         if (r)
4551                 goto out;
4552
4553         /*
4554          * We accumulate dirty bits in the first half of the
4555          * memslot's dirty_bitmap area, for when pages are paged
4556          * out or modified by the host directly.  Pick up these
4557          * bits and add them to the map.
4558          */
4559         p = memslot->dirty_bitmap;
4560         for (i = 0; i < n / sizeof(long); ++i)
4561                 buf[i] |= xchg(&p[i], 0);
4562
4563         /* Harvest dirty bits from VPA and DTL updates */
4564         /* Note: we never modify the SLB shadow buffer areas */
4565         kvm_for_each_vcpu(i, vcpu, kvm) {
4566                 spin_lock(&vcpu->arch.vpa_update_lock);
4567                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4568                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4569                 spin_unlock(&vcpu->arch.vpa_update_lock);
4570         }
4571
4572         r = -EFAULT;
4573         if (copy_to_user(log->dirty_bitmap, buf, n))
4574                 goto out;
4575
4576         r = 0;
4577 out:
4578         mutex_unlock(&kvm->slots_lock);
4579         return r;
4580 }
4581
4582 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
4583 {
4584         vfree(slot->arch.rmap);
4585         slot->arch.rmap = NULL;
4586 }
4587
4588 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4589                                         struct kvm_memory_slot *slot,
4590                                         const struct kvm_userspace_memory_region *mem,
4591                                         enum kvm_mr_change change)
4592 {
4593         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4594
4595         if (change == KVM_MR_CREATE) {
4596                 slot->arch.rmap = vzalloc(array_size(npages,
4597                                           sizeof(*slot->arch.rmap)));
4598                 if (!slot->arch.rmap)
4599                         return -ENOMEM;
4600         }
4601
4602         return 0;
4603 }
4604
4605 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4606                                 const struct kvm_userspace_memory_region *mem,
4607                                 const struct kvm_memory_slot *old,
4608                                 const struct kvm_memory_slot *new,
4609                                 enum kvm_mr_change change)
4610 {
4611         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4612
4613         /*
4614          * If we are making a new memslot, it might make
4615          * some address that was previously cached as emulated
4616          * MMIO be no longer emulated MMIO, so invalidate
4617          * all the caches of emulated MMIO translations.
4618          */
4619         if (npages)
4620                 atomic64_inc(&kvm->arch.mmio_update);
4621
4622         /*
4623          * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4624          * have already called kvm_arch_flush_shadow_memslot() to
4625          * flush shadow mappings.  For KVM_MR_CREATE we have no
4626          * previous mappings.  So the only case to handle is
4627          * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4628          * has been changed.
4629          * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4630          * to get rid of any THP PTEs in the partition-scoped page tables
4631          * so we can track dirtiness at the page level; we flush when
4632          * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4633          * using THP PTEs.
4634          */
4635         if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4636             ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4637                 kvmppc_radix_flush_memslot(kvm, old);
4638         /*
4639          * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4640          */
4641         if (!kvm->arch.secure_guest)
4642                 return;
4643
4644         switch (change) {
4645         case KVM_MR_CREATE:
4646                 /*
4647                  * @TODO kvmppc_uvmem_memslot_create() can fail and
4648                  * return error. Fix this.
4649                  */
4650                 kvmppc_uvmem_memslot_create(kvm, new);
4651                 break;
4652         case KVM_MR_DELETE:
4653                 kvmppc_uvmem_memslot_delete(kvm, old);
4654                 break;
4655         default:
4656                 /* TODO: Handle KVM_MR_MOVE */
4657                 break;
4658         }
4659 }
4660
4661 /*
4662  * Update LPCR values in kvm->arch and in vcores.
4663  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4664  * of kvm->arch.lpcr update).
4665  */
4666 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4667 {
4668         long int i;
4669         u32 cores_done = 0;
4670
4671         if ((kvm->arch.lpcr & mask) == lpcr)
4672                 return;
4673
4674         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4675
4676         for (i = 0; i < KVM_MAX_VCORES; ++i) {
4677                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4678                 if (!vc)
4679                         continue;
4680
4681                 spin_lock(&vc->lock);
4682                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4683                 verify_lpcr(kvm, vc->lpcr);
4684                 spin_unlock(&vc->lock);
4685                 if (++cores_done >= kvm->arch.online_vcores)
4686                         break;
4687         }
4688 }
4689
4690 void kvmppc_setup_partition_table(struct kvm *kvm)
4691 {
4692         unsigned long dw0, dw1;
4693
4694         if (!kvm_is_radix(kvm)) {
4695                 /* PS field - page size for VRMA */
4696                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4697                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4698                 /* HTABSIZE and HTABORG fields */
4699                 dw0 |= kvm->arch.sdr1;
4700
4701                 /* Second dword as set by userspace */
4702                 dw1 = kvm->arch.process_table;
4703         } else {
4704                 dw0 = PATB_HR | radix__get_tree_size() |
4705                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4706                 dw1 = PATB_GR | kvm->arch.process_table;
4707         }
4708         kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4709 }
4710
4711 /*
4712  * Set up HPT (hashed page table) and RMA (real-mode area).
4713  * Must be called with kvm->arch.mmu_setup_lock held.
4714  */
4715 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4716 {
4717         int err = 0;
4718         struct kvm *kvm = vcpu->kvm;
4719         unsigned long hva;
4720         struct kvm_memory_slot *memslot;
4721         struct vm_area_struct *vma;
4722         unsigned long lpcr = 0, senc;
4723         unsigned long psize, porder;
4724         int srcu_idx;
4725
4726         /* Allocate hashed page table (if not done already) and reset it */
4727         if (!kvm->arch.hpt.virt) {
4728                 int order = KVM_DEFAULT_HPT_ORDER;
4729                 struct kvm_hpt_info info;
4730
4731                 err = kvmppc_allocate_hpt(&info, order);
4732                 /* If we get here, it means userspace didn't specify a
4733                  * size explicitly.  So, try successively smaller
4734                  * sizes if the default failed. */
4735                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4736                         err  = kvmppc_allocate_hpt(&info, order);
4737
4738                 if (err < 0) {
4739                         pr_err("KVM: Couldn't alloc HPT\n");
4740                         goto out;
4741                 }
4742
4743                 kvmppc_set_hpt(kvm, &info);
4744         }
4745
4746         /* Look up the memslot for guest physical address 0 */
4747         srcu_idx = srcu_read_lock(&kvm->srcu);
4748         memslot = gfn_to_memslot(kvm, 0);
4749
4750         /* We must have some memory at 0 by now */
4751         err = -EINVAL;
4752         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4753                 goto out_srcu;
4754
4755         /* Look up the VMA for the start of this memory slot */
4756         hva = memslot->userspace_addr;
4757         mmap_read_lock(kvm->mm);
4758         vma = find_vma(kvm->mm, hva);
4759         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4760                 goto up_out;
4761
4762         psize = vma_kernel_pagesize(vma);
4763
4764         mmap_read_unlock(kvm->mm);
4765
4766         /* We can handle 4k, 64k or 16M pages in the VRMA */
4767         if (psize >= 0x1000000)
4768                 psize = 0x1000000;
4769         else if (psize >= 0x10000)
4770                 psize = 0x10000;
4771         else
4772                 psize = 0x1000;
4773         porder = __ilog2(psize);
4774
4775         senc = slb_pgsize_encoding(psize);
4776         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4777                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4778         /* Create HPTEs in the hash page table for the VRMA */
4779         kvmppc_map_vrma(vcpu, memslot, porder);
4780
4781         /* Update VRMASD field in the LPCR */
4782         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4783                 /* the -4 is to account for senc values starting at 0x10 */
4784                 lpcr = senc << (LPCR_VRMASD_SH - 4);
4785                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4786         }
4787
4788         /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4789         smp_wmb();
4790         err = 0;
4791  out_srcu:
4792         srcu_read_unlock(&kvm->srcu, srcu_idx);
4793  out:
4794         return err;
4795
4796  up_out:
4797         mmap_read_unlock(kvm->mm);
4798         goto out_srcu;
4799 }
4800
4801 /*
4802  * Must be called with kvm->arch.mmu_setup_lock held and
4803  * mmu_ready = 0 and no vcpus running.
4804  */
4805 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4806 {
4807         if (nesting_enabled(kvm))
4808                 kvmhv_release_all_nested(kvm);
4809         kvmppc_rmap_reset(kvm);
4810         kvm->arch.process_table = 0;
4811         /* Mutual exclusion with kvm_unmap_gfn_range etc. */
4812         spin_lock(&kvm->mmu_lock);
4813         kvm->arch.radix = 0;
4814         spin_unlock(&kvm->mmu_lock);
4815         kvmppc_free_radix(kvm);
4816         kvmppc_update_lpcr(kvm, LPCR_VPM1,
4817                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4818         return 0;
4819 }
4820
4821 /*
4822  * Must be called with kvm->arch.mmu_setup_lock held and
4823  * mmu_ready = 0 and no vcpus running.
4824  */
4825 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4826 {
4827         int err;
4828
4829         err = kvmppc_init_vm_radix(kvm);
4830         if (err)
4831                 return err;
4832         kvmppc_rmap_reset(kvm);
4833         /* Mutual exclusion with kvm_unmap_gfn_range etc. */
4834         spin_lock(&kvm->mmu_lock);
4835         kvm->arch.radix = 1;
4836         spin_unlock(&kvm->mmu_lock);
4837         kvmppc_free_hpt(&kvm->arch.hpt);
4838         kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4839                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4840         return 0;
4841 }
4842
4843 #ifdef CONFIG_KVM_XICS
4844 /*
4845  * Allocate a per-core structure for managing state about which cores are
4846  * running in the host versus the guest and for exchanging data between
4847  * real mode KVM and CPU running in the host.
4848  * This is only done for the first VM.
4849  * The allocated structure stays even if all VMs have stopped.
4850  * It is only freed when the kvm-hv module is unloaded.
4851  * It's OK for this routine to fail, we just don't support host
4852  * core operations like redirecting H_IPI wakeups.
4853  */
4854 void kvmppc_alloc_host_rm_ops(void)
4855 {
4856         struct kvmppc_host_rm_ops *ops;
4857         unsigned long l_ops;
4858         int cpu, core;
4859         int size;
4860
4861         /* Not the first time here ? */
4862         if (kvmppc_host_rm_ops_hv != NULL)
4863                 return;
4864
4865         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4866         if (!ops)
4867                 return;
4868
4869         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4870         ops->rm_core = kzalloc(size, GFP_KERNEL);
4871
4872         if (!ops->rm_core) {
4873                 kfree(ops);
4874                 return;
4875         }
4876
4877         cpus_read_lock();
4878
4879         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4880                 if (!cpu_online(cpu))
4881                         continue;
4882
4883                 core = cpu >> threads_shift;
4884                 ops->rm_core[core].rm_state.in_host = 1;
4885         }
4886
4887         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4888
4889         /*
4890          * Make the contents of the kvmppc_host_rm_ops structure visible
4891          * to other CPUs before we assign it to the global variable.
4892          * Do an atomic assignment (no locks used here), but if someone
4893          * beats us to it, just free our copy and return.
4894          */
4895         smp_wmb();
4896         l_ops = (unsigned long) ops;
4897
4898         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4899                 cpus_read_unlock();
4900                 kfree(ops->rm_core);
4901                 kfree(ops);
4902                 return;
4903         }
4904
4905         cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4906                                              "ppc/kvm_book3s:prepare",
4907                                              kvmppc_set_host_core,
4908                                              kvmppc_clear_host_core);
4909         cpus_read_unlock();
4910 }
4911
4912 void kvmppc_free_host_rm_ops(void)
4913 {
4914         if (kvmppc_host_rm_ops_hv) {
4915                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4916                 kfree(kvmppc_host_rm_ops_hv->rm_core);
4917                 kfree(kvmppc_host_rm_ops_hv);
4918                 kvmppc_host_rm_ops_hv = NULL;
4919         }
4920 }
4921 #endif
4922
4923 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4924 {
4925         unsigned long lpcr, lpid;
4926         char buf[32];
4927         int ret;
4928
4929         mutex_init(&kvm->arch.uvmem_lock);
4930         INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
4931         mutex_init(&kvm->arch.mmu_setup_lock);
4932
4933         /* Allocate the guest's logical partition ID */
4934
4935         lpid = kvmppc_alloc_lpid();
4936         if ((long)lpid < 0)
4937                 return -ENOMEM;
4938         kvm->arch.lpid = lpid;
4939
4940         kvmppc_alloc_host_rm_ops();
4941
4942         kvmhv_vm_nested_init(kvm);
4943
4944         /*
4945          * Since we don't flush the TLB when tearing down a VM,
4946          * and this lpid might have previously been used,
4947          * make sure we flush on each core before running the new VM.
4948          * On POWER9, the tlbie in mmu_partition_table_set_entry()
4949          * does this flush for us.
4950          */
4951         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4952                 cpumask_setall(&kvm->arch.need_tlb_flush);
4953
4954         /* Start out with the default set of hcalls enabled */
4955         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4956                sizeof(kvm->arch.enabled_hcalls));
4957
4958         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4959                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4960
4961         /* Init LPCR for virtual RMA mode */
4962         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4963                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
4964                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4965                 lpcr &= LPCR_PECE | LPCR_LPES;
4966         } else {
4967                 lpcr = 0;
4968         }
4969         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4970                 LPCR_VPM0 | LPCR_VPM1;
4971         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4972                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4973         /* On POWER8 turn on online bit to enable PURR/SPURR */
4974         if (cpu_has_feature(CPU_FTR_ARCH_207S))
4975                 lpcr |= LPCR_ONL;
4976         /*
4977          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4978          * Set HVICE bit to enable hypervisor virtualization interrupts.
4979          * Set HEIC to prevent OS interrupts to go to hypervisor (should
4980          * be unnecessary but better safe than sorry in case we re-enable
4981          * EE in HV mode with this LPCR still set)
4982          */
4983         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4984                 lpcr &= ~LPCR_VPM0;
4985                 lpcr |= LPCR_HVICE | LPCR_HEIC;
4986
4987                 /*
4988                  * If xive is enabled, we route 0x500 interrupts directly
4989                  * to the guest.
4990                  */
4991                 if (xics_on_xive())
4992                         lpcr |= LPCR_LPES;
4993         }
4994
4995         /*
4996          * If the host uses radix, the guest starts out as radix.
4997          */
4998         if (radix_enabled()) {
4999                 kvm->arch.radix = 1;
5000                 kvm->arch.mmu_ready = 1;
5001                 lpcr &= ~LPCR_VPM1;
5002                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5003                 ret = kvmppc_init_vm_radix(kvm);
5004                 if (ret) {
5005                         kvmppc_free_lpid(kvm->arch.lpid);
5006                         return ret;
5007                 }
5008                 kvmppc_setup_partition_table(kvm);
5009         }
5010
5011         verify_lpcr(kvm, lpcr);
5012         kvm->arch.lpcr = lpcr;
5013
5014         /* Initialization for future HPT resizes */
5015         kvm->arch.resize_hpt = NULL;
5016
5017         /*
5018          * Work out how many sets the TLB has, for the use of
5019          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
5020          */
5021         if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5022                 /*
5023                  * P10 will flush all the congruence class with a single tlbiel
5024                  */
5025                 kvm->arch.tlb_sets = 1;
5026         } else if (radix_enabled())
5027                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
5028         else if (cpu_has_feature(CPU_FTR_ARCH_300))
5029                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
5030         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
5031                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
5032         else
5033                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
5034
5035         /*
5036          * Track that we now have a HV mode VM active. This blocks secondary
5037          * CPU threads from coming online.
5038          * On POWER9, we only need to do this if the "indep_threads_mode"
5039          * module parameter has been set to N.
5040          */
5041         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5042                 if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
5043                         pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
5044                         kvm->arch.threads_indep = true;
5045                 } else {
5046                         kvm->arch.threads_indep = indep_threads_mode;
5047                 }
5048         }
5049         if (!kvm->arch.threads_indep)
5050                 kvm_hv_vm_activated();
5051
5052         /*
5053          * Initialize smt_mode depending on processor.
5054          * POWER8 and earlier have to use "strict" threading, where
5055          * all vCPUs in a vcore have to run on the same (sub)core,
5056          * whereas on POWER9 the threads can each run a different
5057          * guest.
5058          */
5059         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5060                 kvm->arch.smt_mode = threads_per_subcore;
5061         else
5062                 kvm->arch.smt_mode = 1;
5063         kvm->arch.emul_smt_mode = 1;
5064
5065         /*
5066          * Create a debugfs directory for the VM
5067          */
5068         snprintf(buf, sizeof(buf), "vm%d", current->pid);
5069         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
5070         kvmppc_mmu_debugfs_init(kvm);
5071         if (radix_enabled())
5072                 kvmhv_radix_debugfs_init(kvm);
5073
5074         return 0;
5075 }
5076
5077 static void kvmppc_free_vcores(struct kvm *kvm)
5078 {
5079         long int i;
5080
5081         for (i = 0; i < KVM_MAX_VCORES; ++i)
5082                 kfree(kvm->arch.vcores[i]);
5083         kvm->arch.online_vcores = 0;
5084 }
5085
5086 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5087 {
5088         debugfs_remove_recursive(kvm->arch.debugfs_dir);
5089
5090         if (!kvm->arch.threads_indep)
5091                 kvm_hv_vm_deactivated();
5092
5093         kvmppc_free_vcores(kvm);
5094
5095
5096         if (kvm_is_radix(kvm))
5097                 kvmppc_free_radix(kvm);
5098         else
5099                 kvmppc_free_hpt(&kvm->arch.hpt);
5100
5101         /* Perform global invalidation and return lpid to the pool */
5102         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5103                 if (nesting_enabled(kvm))
5104                         kvmhv_release_all_nested(kvm);
5105                 kvm->arch.process_table = 0;
5106                 if (kvm->arch.secure_guest)
5107                         uv_svm_terminate(kvm->arch.lpid);
5108                 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5109         }
5110
5111         kvmppc_free_lpid(kvm->arch.lpid);
5112
5113         kvmppc_free_pimap(kvm);
5114 }
5115
5116 /* We don't need to emulate any privileged instructions or dcbz */
5117 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5118                                      unsigned int inst, int *advance)
5119 {
5120         return EMULATE_FAIL;
5121 }
5122
5123 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5124                                         ulong spr_val)
5125 {
5126         return EMULATE_FAIL;
5127 }
5128
5129 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5130                                         ulong *spr_val)
5131 {
5132         return EMULATE_FAIL;
5133 }
5134
5135 static int kvmppc_core_check_processor_compat_hv(void)
5136 {
5137         if (cpu_has_feature(CPU_FTR_HVMODE) &&
5138             cpu_has_feature(CPU_FTR_ARCH_206))
5139                 return 0;
5140
5141         /* POWER9 in radix mode is capable of being a nested hypervisor. */
5142         if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5143                 return 0;
5144
5145         return -EIO;
5146 }
5147
5148 #ifdef CONFIG_KVM_XICS
5149
5150 void kvmppc_free_pimap(struct kvm *kvm)
5151 {
5152         kfree(kvm->arch.pimap);
5153 }
5154
5155 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5156 {
5157         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5158 }
5159
5160 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5161 {
5162         struct irq_desc *desc;
5163         struct kvmppc_irq_map *irq_map;
5164         struct kvmppc_passthru_irqmap *pimap;
5165         struct irq_chip *chip;
5166         int i, rc = 0;
5167
5168         if (!kvm_irq_bypass)
5169                 return 1;
5170
5171         desc = irq_to_desc(host_irq);
5172         if (!desc)
5173                 return -EIO;
5174
5175         mutex_lock(&kvm->lock);
5176
5177         pimap = kvm->arch.pimap;
5178         if (pimap == NULL) {
5179                 /* First call, allocate structure to hold IRQ map */
5180                 pimap = kvmppc_alloc_pimap();
5181                 if (pimap == NULL) {
5182                         mutex_unlock(&kvm->lock);
5183                         return -ENOMEM;
5184                 }
5185                 kvm->arch.pimap = pimap;
5186         }
5187
5188         /*
5189          * For now, we only support interrupts for which the EOI operation
5190          * is an OPAL call followed by a write to XIRR, since that's
5191          * what our real-mode EOI code does, or a XIVE interrupt
5192          */
5193         chip = irq_data_get_irq_chip(&desc->irq_data);
5194         if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5195                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5196                         host_irq, guest_gsi);
5197                 mutex_unlock(&kvm->lock);
5198                 return -ENOENT;
5199         }
5200
5201         /*
5202          * See if we already have an entry for this guest IRQ number.
5203          * If it's mapped to a hardware IRQ number, that's an error,
5204          * otherwise re-use this entry.
5205          */
5206         for (i = 0; i < pimap->n_mapped; i++) {
5207                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5208                         if (pimap->mapped[i].r_hwirq) {
5209                                 mutex_unlock(&kvm->lock);
5210                                 return -EINVAL;
5211                         }
5212                         break;
5213                 }
5214         }
5215
5216         if (i == KVMPPC_PIRQ_MAPPED) {
5217                 mutex_unlock(&kvm->lock);
5218                 return -EAGAIN;         /* table is full */
5219         }
5220
5221         irq_map = &pimap->mapped[i];
5222
5223         irq_map->v_hwirq = guest_gsi;
5224         irq_map->desc = desc;
5225
5226         /*
5227          * Order the above two stores before the next to serialize with
5228          * the KVM real mode handler.
5229          */
5230         smp_wmb();
5231         irq_map->r_hwirq = desc->irq_data.hwirq;
5232
5233         if (i == pimap->n_mapped)
5234                 pimap->n_mapped++;
5235
5236         if (xics_on_xive())
5237                 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5238         else
5239                 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5240         if (rc)
5241                 irq_map->r_hwirq = 0;
5242
5243         mutex_unlock(&kvm->lock);
5244
5245         return 0;
5246 }
5247
5248 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5249 {
5250         struct irq_desc *desc;
5251         struct kvmppc_passthru_irqmap *pimap;
5252         int i, rc = 0;
5253
5254         if (!kvm_irq_bypass)
5255                 return 0;
5256
5257         desc = irq_to_desc(host_irq);
5258         if (!desc)
5259                 return -EIO;
5260
5261         mutex_lock(&kvm->lock);
5262         if (!kvm->arch.pimap)
5263                 goto unlock;
5264
5265         pimap = kvm->arch.pimap;
5266
5267         for (i = 0; i < pimap->n_mapped; i++) {
5268                 if (guest_gsi == pimap->mapped[i].v_hwirq)
5269                         break;
5270         }
5271
5272         if (i == pimap->n_mapped) {
5273                 mutex_unlock(&kvm->lock);
5274                 return -ENODEV;
5275         }
5276
5277         if (xics_on_xive())
5278                 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5279         else
5280                 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5281
5282         /* invalidate the entry (what do do on error from the above ?) */
5283         pimap->mapped[i].r_hwirq = 0;
5284
5285         /*
5286          * We don't free this structure even when the count goes to
5287          * zero. The structure is freed when we destroy the VM.
5288          */
5289  unlock:
5290         mutex_unlock(&kvm->lock);
5291         return rc;
5292 }
5293
5294 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5295                                              struct irq_bypass_producer *prod)
5296 {
5297         int ret = 0;
5298         struct kvm_kernel_irqfd *irqfd =
5299                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5300
5301         irqfd->producer = prod;
5302
5303         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5304         if (ret)
5305                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5306                         prod->irq, irqfd->gsi, ret);
5307
5308         return ret;
5309 }
5310
5311 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5312                                               struct irq_bypass_producer *prod)
5313 {
5314         int ret;
5315         struct kvm_kernel_irqfd *irqfd =
5316                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5317
5318         irqfd->producer = NULL;
5319
5320         /*
5321          * When producer of consumer is unregistered, we change back to
5322          * default external interrupt handling mode - KVM real mode
5323          * will switch back to host.
5324          */
5325         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5326         if (ret)
5327                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5328                         prod->irq, irqfd->gsi, ret);
5329 }
5330 #endif
5331
5332 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5333                                  unsigned int ioctl, unsigned long arg)
5334 {
5335         struct kvm *kvm __maybe_unused = filp->private_data;
5336         void __user *argp = (void __user *)arg;
5337         long r;
5338
5339         switch (ioctl) {
5340
5341         case KVM_PPC_ALLOCATE_HTAB: {
5342                 u32 htab_order;
5343
5344                 /* If we're a nested hypervisor, we currently only support radix */
5345                 if (kvmhv_on_pseries()) {
5346                         r = -EOPNOTSUPP;
5347                         break;
5348                 }
5349
5350                 r = -EFAULT;
5351                 if (get_user(htab_order, (u32 __user *)argp))
5352                         break;
5353                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5354                 if (r)
5355                         break;
5356                 r = 0;
5357                 break;
5358         }
5359
5360         case KVM_PPC_GET_HTAB_FD: {
5361                 struct kvm_get_htab_fd ghf;
5362
5363                 r = -EFAULT;
5364                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5365                         break;
5366                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5367                 break;
5368         }
5369
5370         case KVM_PPC_RESIZE_HPT_PREPARE: {
5371                 struct kvm_ppc_resize_hpt rhpt;
5372
5373                 r = -EFAULT;
5374                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5375                         break;
5376
5377                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5378                 break;
5379         }
5380
5381         case KVM_PPC_RESIZE_HPT_COMMIT: {
5382                 struct kvm_ppc_resize_hpt rhpt;
5383
5384                 r = -EFAULT;
5385                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5386                         break;
5387
5388                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5389                 break;
5390         }
5391
5392         default:
5393                 r = -ENOTTY;
5394         }
5395
5396         return r;
5397 }
5398
5399 /*
5400  * List of hcall numbers to enable by default.
5401  * For compatibility with old userspace, we enable by default
5402  * all hcalls that were implemented before the hcall-enabling
5403  * facility was added.  Note this list should not include H_RTAS.
5404  */
5405 static unsigned int default_hcall_list[] = {
5406         H_REMOVE,
5407         H_ENTER,
5408         H_READ,
5409         H_PROTECT,
5410         H_BULK_REMOVE,
5411 #ifdef CONFIG_SPAPR_TCE_IOMMU
5412         H_GET_TCE,
5413         H_PUT_TCE,
5414 #endif
5415         H_SET_DABR,
5416         H_SET_XDABR,
5417         H_CEDE,
5418         H_PROD,
5419         H_CONFER,
5420         H_REGISTER_VPA,
5421 #ifdef CONFIG_KVM_XICS
5422         H_EOI,
5423         H_CPPR,
5424         H_IPI,
5425         H_IPOLL,
5426         H_XIRR,
5427         H_XIRR_X,
5428 #endif
5429         0
5430 };
5431
5432 static void init_default_hcalls(void)
5433 {
5434         int i;
5435         unsigned int hcall;
5436
5437         for (i = 0; default_hcall_list[i]; ++i) {
5438                 hcall = default_hcall_list[i];
5439                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5440                 __set_bit(hcall / 4, default_enabled_hcalls);
5441         }
5442 }
5443
5444 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5445 {
5446         unsigned long lpcr;
5447         int radix;
5448         int err;
5449
5450         /* If not on a POWER9, reject it */
5451         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5452                 return -ENODEV;
5453
5454         /* If any unknown flags set, reject it */
5455         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5456                 return -EINVAL;
5457
5458         /* GR (guest radix) bit in process_table field must match */
5459         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5460         if (!!(cfg->process_table & PATB_GR) != radix)
5461                 return -EINVAL;
5462
5463         /* Process table size field must be reasonable, i.e. <= 24 */
5464         if ((cfg->process_table & PRTS_MASK) > 24)
5465                 return -EINVAL;
5466
5467         /* We can change a guest to/from radix now, if the host is radix */
5468         if (radix && !radix_enabled())
5469                 return -EINVAL;
5470
5471         /* If we're a nested hypervisor, we currently only support radix */
5472         if (kvmhv_on_pseries() && !radix)
5473                 return -EINVAL;
5474
5475         mutex_lock(&kvm->arch.mmu_setup_lock);
5476         if (radix != kvm_is_radix(kvm)) {
5477                 if (kvm->arch.mmu_ready) {
5478                         kvm->arch.mmu_ready = 0;
5479                         /* order mmu_ready vs. vcpus_running */
5480                         smp_mb();
5481                         if (atomic_read(&kvm->arch.vcpus_running)) {
5482                                 kvm->arch.mmu_ready = 1;
5483                                 err = -EBUSY;
5484                                 goto out_unlock;
5485                         }
5486                 }
5487                 if (radix)
5488                         err = kvmppc_switch_mmu_to_radix(kvm);
5489                 else
5490                         err = kvmppc_switch_mmu_to_hpt(kvm);
5491                 if (err)
5492                         goto out_unlock;
5493         }
5494
5495         kvm->arch.process_table = cfg->process_table;
5496         kvmppc_setup_partition_table(kvm);
5497
5498         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5499         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5500         err = 0;
5501
5502  out_unlock:
5503         mutex_unlock(&kvm->arch.mmu_setup_lock);
5504         return err;
5505 }
5506
5507 static int kvmhv_enable_nested(struct kvm *kvm)
5508 {
5509         if (!nested)
5510                 return -EPERM;
5511         if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5512                 return -ENODEV;
5513
5514         /* kvm == NULL means the caller is testing if the capability exists */
5515         if (kvm)
5516                 kvm->arch.nested_enable = true;
5517         return 0;
5518 }
5519
5520 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5521                                  int size)
5522 {
5523         int rc = -EINVAL;
5524
5525         if (kvmhv_vcpu_is_radix(vcpu)) {
5526                 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5527
5528                 if (rc > 0)
5529                         rc = -EINVAL;
5530         }
5531
5532         /* For now quadrants are the only way to access nested guest memory */
5533         if (rc && vcpu->arch.nested)
5534                 rc = -EAGAIN;
5535
5536         return rc;
5537 }
5538
5539 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5540                                 int size)
5541 {
5542         int rc = -EINVAL;
5543
5544         if (kvmhv_vcpu_is_radix(vcpu)) {
5545                 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5546
5547                 if (rc > 0)
5548                         rc = -EINVAL;
5549         }
5550
5551         /* For now quadrants are the only way to access nested guest memory */
5552         if (rc && vcpu->arch.nested)
5553                 rc = -EAGAIN;
5554
5555         return rc;
5556 }
5557
5558 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5559 {
5560         unpin_vpa(kvm, vpa);
5561         vpa->gpa = 0;
5562         vpa->pinned_addr = NULL;
5563         vpa->dirty = false;
5564         vpa->update_pending = 0;
5565 }
5566
5567 /*
5568  * Enable a guest to become a secure VM, or test whether
5569  * that could be enabled.
5570  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
5571  * tested (kvm == NULL) or enabled (kvm != NULL).
5572  */
5573 static int kvmhv_enable_svm(struct kvm *kvm)
5574 {
5575         if (!kvmppc_uvmem_available())
5576                 return -EINVAL;
5577         if (kvm)
5578                 kvm->arch.svm_enabled = 1;
5579         return 0;
5580 }
5581
5582 /*
5583  *  IOCTL handler to turn off secure mode of guest
5584  *
5585  * - Release all device pages
5586  * - Issue ucall to terminate the guest on the UV side
5587  * - Unpin the VPA pages.
5588  * - Reinit the partition scoped page tables
5589  */
5590 static int kvmhv_svm_off(struct kvm *kvm)
5591 {
5592         struct kvm_vcpu *vcpu;
5593         int mmu_was_ready;
5594         int srcu_idx;
5595         int ret = 0;
5596         int i;
5597
5598         if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5599                 return ret;
5600
5601         mutex_lock(&kvm->arch.mmu_setup_lock);
5602         mmu_was_ready = kvm->arch.mmu_ready;
5603         if (kvm->arch.mmu_ready) {
5604                 kvm->arch.mmu_ready = 0;
5605                 /* order mmu_ready vs. vcpus_running */
5606                 smp_mb();
5607                 if (atomic_read(&kvm->arch.vcpus_running)) {
5608                         kvm->arch.mmu_ready = 1;
5609                         ret = -EBUSY;
5610                         goto out;
5611                 }
5612         }
5613
5614         srcu_idx = srcu_read_lock(&kvm->srcu);
5615         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5616                 struct kvm_memory_slot *memslot;
5617                 struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5618
5619                 if (!slots)
5620                         continue;
5621
5622                 kvm_for_each_memslot(memslot, slots) {
5623                         kvmppc_uvmem_drop_pages(memslot, kvm, true);
5624                         uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5625                 }
5626         }
5627         srcu_read_unlock(&kvm->srcu, srcu_idx);
5628
5629         ret = uv_svm_terminate(kvm->arch.lpid);
5630         if (ret != U_SUCCESS) {
5631                 ret = -EINVAL;
5632                 goto out;
5633         }
5634
5635         /*
5636          * When secure guest is reset, all the guest pages are sent
5637          * to UV via UV_PAGE_IN before the non-boot vcpus get a
5638          * chance to run and unpin their VPA pages. Unpinning of all
5639          * VPA pages is done here explicitly so that VPA pages
5640          * can be migrated to the secure side.
5641          *
5642          * This is required to for the secure SMP guest to reboot
5643          * correctly.
5644          */
5645         kvm_for_each_vcpu(i, vcpu, kvm) {
5646                 spin_lock(&vcpu->arch.vpa_update_lock);
5647                 unpin_vpa_reset(kvm, &vcpu->arch.dtl);
5648                 unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
5649                 unpin_vpa_reset(kvm, &vcpu->arch.vpa);
5650                 spin_unlock(&vcpu->arch.vpa_update_lock);
5651         }
5652
5653         kvmppc_setup_partition_table(kvm);
5654         kvm->arch.secure_guest = 0;
5655         kvm->arch.mmu_ready = mmu_was_ready;
5656 out:
5657         mutex_unlock(&kvm->arch.mmu_setup_lock);
5658         return ret;
5659 }
5660
5661 static int kvmhv_enable_dawr1(struct kvm *kvm)
5662 {
5663         if (!cpu_has_feature(CPU_FTR_DAWR1))
5664                 return -ENODEV;
5665
5666         /* kvm == NULL means the caller is testing if the capability exists */
5667         if (kvm)
5668                 kvm->arch.dawr1_enabled = true;
5669         return 0;
5670 }
5671
5672 static bool kvmppc_hash_v3_possible(void)
5673 {
5674         if (radix_enabled() && no_mixing_hpt_and_radix)
5675                 return false;
5676
5677         return cpu_has_feature(CPU_FTR_ARCH_300) &&
5678                 cpu_has_feature(CPU_FTR_HVMODE);
5679 }
5680
5681 static struct kvmppc_ops kvm_ops_hv = {
5682         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5683         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5684         .get_one_reg = kvmppc_get_one_reg_hv,
5685         .set_one_reg = kvmppc_set_one_reg_hv,
5686         .vcpu_load   = kvmppc_core_vcpu_load_hv,
5687         .vcpu_put    = kvmppc_core_vcpu_put_hv,
5688         .inject_interrupt = kvmppc_inject_interrupt_hv,
5689         .set_msr     = kvmppc_set_msr_hv,
5690         .vcpu_run    = kvmppc_vcpu_run_hv,
5691         .vcpu_create = kvmppc_core_vcpu_create_hv,
5692         .vcpu_free   = kvmppc_core_vcpu_free_hv,
5693         .check_requests = kvmppc_core_check_requests_hv,
5694         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5695         .flush_memslot  = kvmppc_core_flush_memslot_hv,
5696         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5697         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5698         .unmap_gfn_range = kvm_unmap_gfn_range_hv,
5699         .age_gfn = kvm_age_gfn_hv,
5700         .test_age_gfn = kvm_test_age_gfn_hv,
5701         .set_spte_gfn = kvm_set_spte_gfn_hv,
5702         .free_memslot = kvmppc_core_free_memslot_hv,
5703         .init_vm =  kvmppc_core_init_vm_hv,
5704         .destroy_vm = kvmppc_core_destroy_vm_hv,
5705         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5706         .emulate_op = kvmppc_core_emulate_op_hv,
5707         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5708         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5709         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5710         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5711         .hcall_implemented = kvmppc_hcall_impl_hv,
5712 #ifdef CONFIG_KVM_XICS
5713         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5714         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5715 #endif
5716         .configure_mmu = kvmhv_configure_mmu,
5717         .get_rmmu_info = kvmhv_get_rmmu_info,
5718         .set_smt_mode = kvmhv_set_smt_mode,
5719         .enable_nested = kvmhv_enable_nested,
5720         .load_from_eaddr = kvmhv_load_from_eaddr,
5721         .store_to_eaddr = kvmhv_store_to_eaddr,
5722         .enable_svm = kvmhv_enable_svm,
5723         .svm_off = kvmhv_svm_off,
5724         .enable_dawr1 = kvmhv_enable_dawr1,
5725         .hash_v3_possible = kvmppc_hash_v3_possible,
5726 };
5727
5728 static int kvm_init_subcore_bitmap(void)
5729 {
5730         int i, j;
5731         int nr_cores = cpu_nr_cores();
5732         struct sibling_subcore_state *sibling_subcore_state;
5733
5734         for (i = 0; i < nr_cores; i++) {
5735                 int first_cpu = i * threads_per_core;
5736                 int node = cpu_to_node(first_cpu);
5737
5738                 /* Ignore if it is already allocated. */
5739                 if (paca_ptrs[first_cpu]->sibling_subcore_state)
5740                         continue;
5741
5742                 sibling_subcore_state =
5743                         kzalloc_node(sizeof(struct sibling_subcore_state),
5744                                                         GFP_KERNEL, node);
5745                 if (!sibling_subcore_state)
5746                         return -ENOMEM;
5747
5748
5749                 for (j = 0; j < threads_per_core; j++) {
5750                         int cpu = first_cpu + j;
5751
5752                         paca_ptrs[cpu]->sibling_subcore_state =
5753                                                 sibling_subcore_state;
5754                 }
5755         }
5756         return 0;
5757 }
5758
5759 static int kvmppc_radix_possible(void)
5760 {
5761         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5762 }
5763
5764 static int kvmppc_book3s_init_hv(void)
5765 {
5766         int r;
5767
5768         if (!tlbie_capable) {
5769                 pr_err("KVM-HV: Host does not support TLBIE\n");
5770                 return -ENODEV;
5771         }
5772
5773         /*
5774          * FIXME!! Do we need to check on all cpus ?
5775          */
5776         r = kvmppc_core_check_processor_compat_hv();
5777         if (r < 0)
5778                 return -ENODEV;
5779
5780         r = kvmhv_nested_init();
5781         if (r)
5782                 return r;
5783
5784         r = kvm_init_subcore_bitmap();
5785         if (r)
5786                 return r;
5787
5788         /*
5789          * We need a way of accessing the XICS interrupt controller,
5790          * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5791          * indirectly, via OPAL.
5792          */
5793 #ifdef CONFIG_SMP
5794         if (!xics_on_xive() && !kvmhv_on_pseries() &&
5795             !local_paca->kvm_hstate.xics_phys) {
5796                 struct device_node *np;
5797
5798                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5799                 if (!np) {
5800                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5801                         return -ENODEV;
5802                 }
5803                 /* presence of intc confirmed - node can be dropped again */
5804                 of_node_put(np);
5805         }
5806 #endif
5807
5808         kvm_ops_hv.owner = THIS_MODULE;
5809         kvmppc_hv_ops = &kvm_ops_hv;
5810
5811         init_default_hcalls();
5812
5813         init_vcore_lists();
5814
5815         r = kvmppc_mmu_hv_init();
5816         if (r)
5817                 return r;
5818
5819         if (kvmppc_radix_possible())
5820                 r = kvmppc_radix_init();
5821
5822         /*
5823          * POWER9 chips before version 2.02 can't have some threads in
5824          * HPT mode and some in radix mode on the same core.
5825          */
5826         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5827                 unsigned int pvr = mfspr(SPRN_PVR);
5828                 if ((pvr >> 16) == PVR_POWER9 &&
5829                     (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5830                      ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5831                         no_mixing_hpt_and_radix = true;
5832         }
5833
5834         r = kvmppc_uvmem_init();
5835         if (r < 0)
5836                 pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
5837
5838         return r;
5839 }
5840
5841 static void kvmppc_book3s_exit_hv(void)
5842 {
5843         kvmppc_uvmem_free();
5844         kvmppc_free_host_rm_ops();
5845         if (kvmppc_radix_possible())
5846                 kvmppc_radix_exit();
5847         kvmppc_hv_ops = NULL;
5848         kvmhv_nested_exit();
5849 }
5850
5851 module_init(kvmppc_book3s_init_hv);
5852 module_exit(kvmppc_book3s_exit_hv);
5853 MODULE_LICENSE("GPL");
5854 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5855 MODULE_ALIAS("devname:kvm");