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          * In the case of the P9 single vcpu per vcore case, the real
 904          * mode handler is not called but no other threads are in the
 905          * source vcore.
 906          */
 907
 908         spin_lock(&vcore->lock);
 909         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 910             vcore->vcore_state != VCORE_INACTIVE &&
 911             vcore->runner)
 912                 target = vcore->runner;
 913         spin_unlock(&vcore->lock);
 914
 915         return kvm_vcpu_yield_to(target);
 916 }
 917
 918 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 919 {
 920         int yield_count = 0;
 921         struct lppaca *lppaca;
 922
 923         spin_lock(&vcpu->arch.vpa_update_lock);
 924         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 925         if (lppaca)
 926                 yield_count = be32_to_cpu(lppaca->yield_count);
 927         spin_unlock(&vcpu->arch.vpa_update_lock);
 928         return yield_count;
 929 }
 930
 931 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 932 {
 933         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 934         unsigned long target, ret = H_SUCCESS;
 935         int yield_count;
 936         struct kvm_vcpu *tvcpu;
 937         int idx, rc;
 938
 939         if (req <= MAX_HCALL_OPCODE &&
 940             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 941                 return RESUME_HOST;
 942
 943         switch (req) {
 944         case H_CEDE:
 945                 break;
 946         case H_PROD:
 947                 target = kvmppc_get_gpr(vcpu, 4);
 948                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 949                 if (!tvcpu) {
 950                         ret = H_PARAMETER;
 951                         break;
 952                 }
 953                 tvcpu->arch.prodded = 1;
 954                 smp_mb();
 955                 if (tvcpu->arch.ceded)
 956                         kvmppc_fast_vcpu_kick_hv(tvcpu);
 957                 break;
 958         case H_CONFER:
 959                 target = kvmppc_get_gpr(vcpu, 4);
 960                 if (target == -1)
 961                         break;
 962                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 963                 if (!tvcpu) {
 964                         ret = H_PARAMETER;
 965                         break;
 966                 }
 967                 yield_count = kvmppc_get_gpr(vcpu, 5);
 968                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 969                         break;
 970                 kvm_arch_vcpu_yield_to(tvcpu);
 971                 break;
 972         case H_REGISTER_VPA:
 973                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 974                                         kvmppc_get_gpr(vcpu, 5),
 975                                         kvmppc_get_gpr(vcpu, 6));
 976                 break;
 977         case H_RTAS:
 978                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 979                         return RESUME_HOST;
 980
 981                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 982                 rc = kvmppc_rtas_hcall(vcpu);
 983                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 984
 985                 if (rc == -ENOENT)
 986                         return RESUME_HOST;
 987                 else if (rc == 0)
 988                         break;
 989
 990                 /* Send the error out to userspace via KVM_RUN */
 991                 return rc;
 992         case H_LOGICAL_CI_LOAD:
 993                 ret = kvmppc_h_logical_ci_load(vcpu);
 994                 if (ret == H_TOO_HARD)
 995                         return RESUME_HOST;
 996                 break;
 997         case H_LOGICAL_CI_STORE:
 998                 ret = kvmppc_h_logical_ci_store(vcpu);
 999                 if (ret == H_TOO_HARD)
1000                         return RESUME_HOST;
1001                 break;
1002         case H_SET_MODE:
1003                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
1004                                         kvmppc_get_gpr(vcpu, 5),
1005                                         kvmppc_get_gpr(vcpu, 6),
1006                                         kvmppc_get_gpr(vcpu, 7));
1007                 if (ret == H_TOO_HARD)
1008                         return RESUME_HOST;
1009                 break;
1010         case H_XIRR:
1011         case H_CPPR:
1012         case H_EOI:
1013         case H_IPI:
1014         case H_IPOLL:
1015         case H_XIRR_X:
1016                 if (kvmppc_xics_enabled(vcpu)) {
1017                         if (xics_on_xive()) {
1018                                 ret = H_NOT_AVAILABLE;
1019                                 return RESUME_GUEST;
1020                         }
1021                         ret = kvmppc_xics_hcall(vcpu, req);
1022                         break;
1023                 }
1024                 return RESUME_HOST;
1025         case H_SET_DABR:
1026                 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1027                 break;
1028         case H_SET_XDABR:
1029                 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1030                                                 kvmppc_get_gpr(vcpu, 5));
1031                 break;
1032 #ifdef CONFIG_SPAPR_TCE_IOMMU
1033         case H_GET_TCE:
1034                 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1035                                                 kvmppc_get_gpr(vcpu, 5));
1036                 if (ret == H_TOO_HARD)
1037                         return RESUME_HOST;
1038                 break;
1039         case H_PUT_TCE:
1040                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1041                                                 kvmppc_get_gpr(vcpu, 5),
1042                                                 kvmppc_get_gpr(vcpu, 6));
1043                 if (ret == H_TOO_HARD)
1044                         return RESUME_HOST;
1045                 break;
1046         case H_PUT_TCE_INDIRECT:
1047                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1048                                                 kvmppc_get_gpr(vcpu, 5),
1049                                                 kvmppc_get_gpr(vcpu, 6),
1050                                                 kvmppc_get_gpr(vcpu, 7));
1051                 if (ret == H_TOO_HARD)
1052                         return RESUME_HOST;
1053                 break;
1054         case H_STUFF_TCE:
1055                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1056                                                 kvmppc_get_gpr(vcpu, 5),
1057                                                 kvmppc_get_gpr(vcpu, 6),
1058                                                 kvmppc_get_gpr(vcpu, 7));
1059                 if (ret == H_TOO_HARD)
1060                         return RESUME_HOST;
1061                 break;
1062 #endif
1063         case H_RANDOM:
1064                 if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1065                         ret = H_HARDWARE;
1066                 break;
1067
1068         case H_SET_PARTITION_TABLE:
1069                 ret = H_FUNCTION;
1070                 if (nesting_enabled(vcpu->kvm))
1071                         ret = kvmhv_set_partition_table(vcpu);
1072                 break;
1073         case H_ENTER_NESTED:
1074                 ret = H_FUNCTION;
1075                 if (!nesting_enabled(vcpu->kvm))
1076                         break;
1077                 ret = kvmhv_enter_nested_guest(vcpu);
1078                 if (ret == H_INTERRUPT) {
1079                         kvmppc_set_gpr(vcpu, 3, 0);
1080                         vcpu->arch.hcall_needed = 0;
1081                         return -EINTR;
1082                 } else if (ret == H_TOO_HARD) {
1083                         kvmppc_set_gpr(vcpu, 3, 0);
1084                         vcpu->arch.hcall_needed = 0;
1085                         return RESUME_HOST;
1086                 }
1087                 break;
1088         case H_TLB_INVALIDATE:
1089                 ret = H_FUNCTION;
1090                 if (nesting_enabled(vcpu->kvm))
1091                         ret = kvmhv_do_nested_tlbie(vcpu);
1092                 break;
1093         case H_COPY_TOFROM_GUEST:
1094                 ret = H_FUNCTION;
1095                 if (nesting_enabled(vcpu->kvm))
1096                         ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1097                 break;
1098         case H_PAGE_INIT:
1099                 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1100                                          kvmppc_get_gpr(vcpu, 5),
1101                                          kvmppc_get_gpr(vcpu, 6));
1102                 break;
1103         case H_SVM_PAGE_IN:
1104                 ret = H_UNSUPPORTED;
1105                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1106                         ret = kvmppc_h_svm_page_in(vcpu->kvm,
1107                                                    kvmppc_get_gpr(vcpu, 4),
1108                                                    kvmppc_get_gpr(vcpu, 5),
1109                                                    kvmppc_get_gpr(vcpu, 6));
1110                 break;
1111         case H_SVM_PAGE_OUT:
1112                 ret = H_UNSUPPORTED;
1113                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1114                         ret = kvmppc_h_svm_page_out(vcpu->kvm,
1115                                                     kvmppc_get_gpr(vcpu, 4),
1116                                                     kvmppc_get_gpr(vcpu, 5),
1117                                                     kvmppc_get_gpr(vcpu, 6));
1118                 break;
1119         case H_SVM_INIT_START:
1120                 ret = H_UNSUPPORTED;
1121                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1122                         ret = kvmppc_h_svm_init_start(vcpu->kvm);
1123                 break;
1124         case H_SVM_INIT_DONE:
1125                 ret = H_UNSUPPORTED;
1126                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1127                         ret = kvmppc_h_svm_init_done(vcpu->kvm);
1128                 break;
1129         case H_SVM_INIT_ABORT:
1130                 /*
1131                  * Even if that call is made by the Ultravisor, the SSR1 value
1132                  * is the guest context one, with the secure bit clear as it has
1133                  * not yet been secured. So we can't check it here.
1134                  * Instead the kvm->arch.secure_guest flag is checked inside
1135                  * kvmppc_h_svm_init_abort().
1136                  */
1137                 ret = kvmppc_h_svm_init_abort(vcpu->kvm);
1138                 break;
1139
1140         default:
1141                 return RESUME_HOST;
1142         }
1143         kvmppc_set_gpr(vcpu, 3, ret);
1144         vcpu->arch.hcall_needed = 0;
1145         return RESUME_GUEST;
1146 }
1147
1148 /*
1149  * Handle H_CEDE in the P9 path where we don't call the real-mode hcall
1150  * handlers in book3s_hv_rmhandlers.S.
1151  *
1152  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1153  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1154  */
1155 static void kvmppc_cede(struct kvm_vcpu *vcpu)
1156 {
1157         vcpu->arch.shregs.msr |= MSR_EE;
1158         vcpu->arch.ceded = 1;
1159         smp_mb();
1160         if (vcpu->arch.prodded) {
1161                 vcpu->arch.prodded = 0;
1162                 smp_mb();
1163                 vcpu->arch.ceded = 0;
1164         }
1165 }
1166
1167 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1168 {
1169         switch (cmd) {
1170         case H_CEDE:
1171         case H_PROD:
1172         case H_CONFER:
1173         case H_REGISTER_VPA:
1174         case H_SET_MODE:
1175         case H_LOGICAL_CI_LOAD:
1176         case H_LOGICAL_CI_STORE:
1177 #ifdef CONFIG_KVM_XICS
1178         case H_XIRR:
1179         case H_CPPR:
1180         case H_EOI:
1181         case H_IPI:
1182         case H_IPOLL:
1183         case H_XIRR_X:
1184 #endif
1185         case H_PAGE_INIT:
1186                 return 1;
1187         }
1188
1189         /* See if it's in the real-mode table */
1190         return kvmppc_hcall_impl_hv_realmode(cmd);
1191 }
1192
1193 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1194 {
1195         u32 last_inst;
1196
1197         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1198                                         EMULATE_DONE) {
1199                 /*
1200                  * Fetch failed, so return to guest and
1201                  * try executing it again.
1202                  */
1203                 return RESUME_GUEST;
1204         }
1205
1206         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1207                 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1208                 vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1209                 return RESUME_HOST;
1210         } else {
1211                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1212                 return RESUME_GUEST;
1213         }
1214 }
1215
1216 static void do_nothing(void *x)
1217 {
1218 }
1219
1220 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1221 {
1222         int thr, cpu, pcpu, nthreads;
1223         struct kvm_vcpu *v;
1224         unsigned long dpdes;
1225
1226         nthreads = vcpu->kvm->arch.emul_smt_mode;
1227         dpdes = 0;
1228         cpu = vcpu->vcpu_id & ~(nthreads - 1);
1229         for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1230                 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1231                 if (!v)
1232                         continue;
1233                 /*
1234                  * If the vcpu is currently running on a physical cpu thread,
1235                  * interrupt it in order to pull it out of the guest briefly,
1236                  * which will update its vcore->dpdes value.
1237                  */
1238                 pcpu = READ_ONCE(v->cpu);
1239                 if (pcpu >= 0)
1240                         smp_call_function_single(pcpu, do_nothing, NULL, 1);
1241                 if (kvmppc_doorbell_pending(v))
1242                         dpdes |= 1 << thr;
1243         }
1244         return dpdes;
1245 }
1246
1247 /*
1248  * On POWER9, emulate doorbell-related instructions in order to
1249  * give the guest the illusion of running on a multi-threaded core.
1250  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1251  * and mfspr DPDES.
1252  */
1253 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1254 {
1255         u32 inst, rb, thr;
1256         unsigned long arg;
1257         struct kvm *kvm = vcpu->kvm;
1258         struct kvm_vcpu *tvcpu;
1259
1260         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1261                 return RESUME_GUEST;
1262         if (get_op(inst) != 31)
1263                 return EMULATE_FAIL;
1264         rb = get_rb(inst);
1265         thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1266         switch (get_xop(inst)) {
1267         case OP_31_XOP_MSGSNDP:
1268                 arg = kvmppc_get_gpr(vcpu, rb);
1269                 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1270                         break;
1271                 arg &= 0x7f;
1272                 if (arg >= kvm->arch.emul_smt_mode)
1273                         break;
1274                 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1275                 if (!tvcpu)
1276                         break;
1277                 if (!tvcpu->arch.doorbell_request) {
1278                         tvcpu->arch.doorbell_request = 1;
1279                         kvmppc_fast_vcpu_kick_hv(tvcpu);
1280                 }
1281                 break;
1282         case OP_31_XOP_MSGCLRP:
1283                 arg = kvmppc_get_gpr(vcpu, rb);
1284                 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1285                         break;
1286                 vcpu->arch.vcore->dpdes = 0;
1287                 vcpu->arch.doorbell_request = 0;
1288                 break;
1289         case OP_31_XOP_MFSPR:
1290                 switch (get_sprn(inst)) {
1291                 case SPRN_TIR:
1292                         arg = thr;
1293                         break;
1294                 case SPRN_DPDES:
1295                         arg = kvmppc_read_dpdes(vcpu);
1296                         break;
1297                 default:
1298                         return EMULATE_FAIL;
1299                 }
1300                 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1301                 break;
1302         default:
1303                 return EMULATE_FAIL;
1304         }
1305         kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1306         return RESUME_GUEST;
1307 }
1308
1309 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1310                                  struct task_struct *tsk)
1311 {
1312         struct kvm_run *run = vcpu->run;
1313         int r = RESUME_HOST;
1314
1315         vcpu->stat.sum_exits++;
1316
1317         /*
1318          * This can happen if an interrupt occurs in the last stages
1319          * of guest entry or the first stages of guest exit (i.e. after
1320          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1321          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1322          * That can happen due to a bug, or due to a machine check
1323          * occurring at just the wrong time.
1324          */
1325         if (vcpu->arch.shregs.msr & MSR_HV) {
1326                 printk(KERN_EMERG "KVM trap in HV mode!\n");
1327                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1328                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1329                         vcpu->arch.shregs.msr);
1330                 kvmppc_dump_regs(vcpu);
1331                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1332                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1333                 return RESUME_HOST;
1334         }
1335         run->exit_reason = KVM_EXIT_UNKNOWN;
1336         run->ready_for_interrupt_injection = 1;
1337         switch (vcpu->arch.trap) {
1338         /* We're good on these - the host merely wanted to get our attention */
1339         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1340                 vcpu->stat.dec_exits++;
1341                 r = RESUME_GUEST;
1342                 break;
1343         case BOOK3S_INTERRUPT_EXTERNAL:
1344         case BOOK3S_INTERRUPT_H_DOORBELL:
1345         case BOOK3S_INTERRUPT_H_VIRT:
1346                 vcpu->stat.ext_intr_exits++;
1347                 r = RESUME_GUEST;
1348                 break;
1349         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1350         case BOOK3S_INTERRUPT_HMI:
1351         case BOOK3S_INTERRUPT_PERFMON:
1352         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1353                 r = RESUME_GUEST;
1354                 break;
1355         case BOOK3S_INTERRUPT_MACHINE_CHECK: {
1356                 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1357                                               DEFAULT_RATELIMIT_BURST);
1358                 /*
1359                  * Print the MCE event to host console. Ratelimit so the guest
1360                  * can't flood the host log.
1361                  */
1362                 if (__ratelimit(&rs))
1363                         machine_check_print_event_info(&vcpu->arch.mce_evt,false, true);
1364
1365                 /*
1366                  * If the guest can do FWNMI, exit to userspace so it can
1367                  * deliver a FWNMI to the guest.
1368                  * Otherwise we synthesize a machine check for the guest
1369                  * so that it knows that the machine check occurred.
1370                  */
1371                 if (!vcpu->kvm->arch.fwnmi_enabled) {
1372                         ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1373                         kvmppc_core_queue_machine_check(vcpu, flags);
1374                         r = RESUME_GUEST;
1375                         break;
1376                 }
1377
1378                 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1379                 run->exit_reason = KVM_EXIT_NMI;
1380                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1381                 /* Clear out the old NMI status from run->flags */
1382                 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1383                 /* Now set the NMI status */
1384                 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1385                         run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1386                 else
1387                         run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1388
1389                 r = RESUME_HOST;
1390                 break;
1391         }
1392         case BOOK3S_INTERRUPT_PROGRAM:
1393         {
1394                 ulong flags;
1395                 /*
1396                  * Normally program interrupts are delivered directly
1397                  * to the guest by the hardware, but we can get here
1398                  * as a result of a hypervisor emulation interrupt
1399                  * (e40) getting turned into a 700 by BML RTAS.
1400                  */
1401                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1402                 kvmppc_core_queue_program(vcpu, flags);
1403                 r = RESUME_GUEST;
1404                 break;
1405         }
1406         case BOOK3S_INTERRUPT_SYSCALL:
1407         {
1408                 int i;
1409
1410                 if (unlikely(vcpu->arch.shregs.msr & MSR_PR)) {
1411                         /*
1412                          * Guest userspace executed sc 1. This can only be
1413                          * reached by the P9 path because the old path
1414                          * handles this case in realmode hcall handlers.
1415                          *
1416                          * Radix guests can not run PR KVM or nested HV hash
1417                          * guests which might run PR KVM, so this is always
1418                          * a privilege fault. Send a program check to guest
1419                          * kernel.
1420                          */
1421                         kvmppc_core_queue_program(vcpu, SRR1_PROGPRIV);
1422                         r = RESUME_GUEST;
1423                         break;
1424                 }
1425
1426                 /*
1427                  * hcall - gather args and set exit_reason. This will next be
1428                  * handled by kvmppc_pseries_do_hcall which may be able to deal
1429                  * with it and resume guest, or may punt to userspace.
1430                  */
1431                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1432                 for (i = 0; i < 9; ++i)
1433                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1434                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1435                 vcpu->arch.hcall_needed = 1;
1436                 r = RESUME_HOST;
1437                 break;
1438         }
1439         /*
1440          * We get these next two if the guest accesses a page which it thinks
1441          * it has mapped but which is not actually present, either because
1442          * it is for an emulated I/O device or because the corresonding
1443          * host page has been paged out.  Any other HDSI/HISI interrupts
1444          * have been handled already.
1445          */
1446         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1447                 r = RESUME_PAGE_FAULT;
1448                 break;
1449         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1450                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1451                 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1452                         DSISR_SRR1_MATCH_64S;
1453                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1454                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1455                 r = RESUME_PAGE_FAULT;
1456                 break;
1457         /*
1458          * This occurs if the guest executes an illegal instruction.
1459          * If the guest debug is disabled, generate a program interrupt
1460          * to the guest. If guest debug is enabled, we need to check
1461          * whether the instruction is a software breakpoint instruction.
1462          * Accordingly return to Guest or Host.
1463          */
1464         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1465                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1466                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1467                                 swab32(vcpu->arch.emul_inst) :
1468                                 vcpu->arch.emul_inst;
1469                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1470                         r = kvmppc_emulate_debug_inst(vcpu);
1471                 } else {
1472                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1473                         r = RESUME_GUEST;
1474                 }
1475                 break;
1476         /*
1477          * This occurs if the guest (kernel or userspace), does something that
1478          * is prohibited by HFSCR.
1479          * On POWER9, this could be a doorbell instruction that we need
1480          * to emulate.
1481          * Otherwise, we just generate a program interrupt to the guest.
1482          */
1483         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1484                 r = EMULATE_FAIL;
1485                 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1486                     cpu_has_feature(CPU_FTR_ARCH_300))
1487                         r = kvmppc_emulate_doorbell_instr(vcpu);
1488                 if (r == EMULATE_FAIL) {
1489                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1490                         r = RESUME_GUEST;
1491                 }
1492                 break;
1493
1494 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1495         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1496                 /*
1497                  * This occurs for various TM-related instructions that
1498                  * we need to emulate on POWER9 DD2.2.  We have already
1499                  * handled the cases where the guest was in real-suspend
1500                  * mode and was transitioning to transactional state.
1501                  */
1502                 r = kvmhv_p9_tm_emulation(vcpu);
1503                 break;
1504 #endif
1505
1506         case BOOK3S_INTERRUPT_HV_RM_HARD:
1507                 r = RESUME_PASSTHROUGH;
1508                 break;
1509         default:
1510                 kvmppc_dump_regs(vcpu);
1511                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1512                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1513                         vcpu->arch.shregs.msr);
1514                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1515                 r = RESUME_HOST;
1516                 break;
1517         }
1518
1519         return r;
1520 }
1521
1522 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1523 {
1524         int r;
1525         int srcu_idx;
1526
1527         vcpu->stat.sum_exits++;
1528
1529         /*
1530          * This can happen if an interrupt occurs in the last stages
1531          * of guest entry or the first stages of guest exit (i.e. after
1532          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1533          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1534          * That can happen due to a bug, or due to a machine check
1535          * occurring at just the wrong time.
1536          */
1537         if (vcpu->arch.shregs.msr & MSR_HV) {
1538                 pr_emerg("KVM trap in HV mode while nested!\n");
1539                 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1540                          vcpu->arch.trap, kvmppc_get_pc(vcpu),
1541                          vcpu->arch.shregs.msr);
1542                 kvmppc_dump_regs(vcpu);
1543                 return RESUME_HOST;
1544         }
1545         switch (vcpu->arch.trap) {
1546         /* We're good on these - the host merely wanted to get our attention */
1547         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1548                 vcpu->stat.dec_exits++;
1549                 r = RESUME_GUEST;
1550                 break;
1551         case BOOK3S_INTERRUPT_EXTERNAL:
1552                 vcpu->stat.ext_intr_exits++;
1553                 r = RESUME_HOST;
1554                 break;
1555         case BOOK3S_INTERRUPT_H_DOORBELL:
1556         case BOOK3S_INTERRUPT_H_VIRT:
1557                 vcpu->stat.ext_intr_exits++;
1558                 r = RESUME_GUEST;
1559                 break;
1560         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1561         case BOOK3S_INTERRUPT_HMI:
1562         case BOOK3S_INTERRUPT_PERFMON:
1563         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1564                 r = RESUME_GUEST;
1565                 break;
1566         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1567         {
1568                 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1569                                               DEFAULT_RATELIMIT_BURST);
1570                 /* Pass the machine check to the L1 guest */
1571                 r = RESUME_HOST;
1572                 /* Print the MCE event to host console. */
1573                 if (__ratelimit(&rs))
1574                         machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1575                 break;
1576         }
1577         /*
1578          * We get these next two if the guest accesses a page which it thinks
1579          * it has mapped but which is not actually present, either because
1580          * it is for an emulated I/O device or because the corresonding
1581          * host page has been paged out.
1582          */
1583         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1584                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1585                 r = kvmhv_nested_page_fault(vcpu);
1586                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1587                 break;
1588         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1589                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1590                 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1591                                          DSISR_SRR1_MATCH_64S;
1592                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1593                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1594                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1595                 r = kvmhv_nested_page_fault(vcpu);
1596                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1597                 break;
1598
1599 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1600         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1601                 /*
1602                  * This occurs for various TM-related instructions that
1603                  * we need to emulate on POWER9 DD2.2.  We have already
1604                  * handled the cases where the guest was in real-suspend
1605                  * mode and was transitioning to transactional state.
1606                  */
1607                 r = kvmhv_p9_tm_emulation(vcpu);
1608                 break;
1609 #endif
1610
1611         case BOOK3S_INTERRUPT_HV_RM_HARD:
1612                 vcpu->arch.trap = 0;
1613                 r = RESUME_GUEST;
1614                 if (!xics_on_xive())
1615                         kvmppc_xics_rm_complete(vcpu, 0);
1616                 break;
1617         default:
1618                 r = RESUME_HOST;
1619                 break;
1620         }
1621
1622         return r;
1623 }
1624
1625 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1626                                             struct kvm_sregs *sregs)
1627 {
1628         int i;
1629
1630         memset(sregs, 0, sizeof(struct kvm_sregs));
1631         sregs->pvr = vcpu->arch.pvr;
1632         for (i = 0; i < vcpu->arch.slb_max; i++) {
1633                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1634                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1635         }
1636
1637         return 0;
1638 }
1639
1640 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1641                                             struct kvm_sregs *sregs)
1642 {
1643         int i, j;
1644
1645         /* Only accept the same PVR as the host's, since we can't spoof it */
1646         if (sregs->pvr != vcpu->arch.pvr)
1647                 return -EINVAL;
1648
1649         j = 0;
1650         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1651                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1652                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1653                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1654                         ++j;
1655                 }
1656         }
1657         vcpu->arch.slb_max = j;
1658
1659         return 0;
1660 }
1661
1662 /*
1663  * Enforce limits on guest LPCR values based on hardware availability,
1664  * guest configuration, and possibly hypervisor support and security
1665  * concerns.
1666  */
1667 unsigned long kvmppc_filter_lpcr_hv(struct kvm *kvm, unsigned long lpcr)
1668 {
1669         /* LPCR_TC only applies to HPT guests */
1670         if (kvm_is_radix(kvm))
1671                 lpcr &= ~LPCR_TC;
1672
1673         /* On POWER8 and above, userspace can modify AIL */
1674         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
1675                 lpcr &= ~LPCR_AIL;
1676         if ((lpcr & LPCR_AIL) != LPCR_AIL_3)
1677                 lpcr &= ~LPCR_AIL; /* LPCR[AIL]=1/2 is disallowed */
1678
1679         /*
1680          * On POWER9, allow userspace to enable large decrementer for the
1681          * guest, whether or not the host has it enabled.
1682          */
1683         if (!cpu_has_feature(CPU_FTR_ARCH_300))
1684                 lpcr &= ~LPCR_LD;
1685
1686         return lpcr;
1687 }
1688
1689 static void verify_lpcr(struct kvm *kvm, unsigned long lpcr)
1690 {
1691         if (lpcr != kvmppc_filter_lpcr_hv(kvm, lpcr)) {
1692                 WARN_ONCE(1, "lpcr 0x%lx differs from filtered 0x%lx\n",
1693                           lpcr, kvmppc_filter_lpcr_hv(kvm, lpcr));
1694         }
1695 }
1696
1697 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1698                 bool preserve_top32)
1699 {
1700         struct kvm *kvm = vcpu->kvm;
1701         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1702         u64 mask;
1703
1704         spin_lock(&vc->lock);
1705
1706         /*
1707          * Userspace can only modify
1708          * DPFD (default prefetch depth), ILE (interrupt little-endian),
1709          * TC (translation control), AIL (alternate interrupt location),
1710          * LD (large decrementer).
1711          * These are subject to restrictions from kvmppc_filter_lcpr_hv().
1712          */
1713         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD;
1714
1715         /* Broken 32-bit version of LPCR must not clear top bits */
1716         if (preserve_top32)
1717                 mask &= 0xFFFFFFFF;
1718
1719         new_lpcr = kvmppc_filter_lpcr_hv(kvm,
1720                         (vc->lpcr & ~mask) | (new_lpcr & mask));
1721
1722         /*
1723          * If ILE (interrupt little-endian) has changed, update the
1724          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1725          */
1726         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1727                 struct kvm_vcpu *vcpu;
1728                 int i;
1729
1730                 kvm_for_each_vcpu(i, vcpu, kvm) {
1731                         if (vcpu->arch.vcore != vc)
1732                                 continue;
1733                         if (new_lpcr & LPCR_ILE)
1734                                 vcpu->arch.intr_msr |= MSR_LE;
1735                         else
1736                                 vcpu->arch.intr_msr &= ~MSR_LE;
1737                 }
1738         }
1739
1740         vc->lpcr = new_lpcr;
1741
1742         spin_unlock(&vc->lock);
1743 }
1744
1745 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1746                                  union kvmppc_one_reg *val)
1747 {
1748         int r = 0;
1749         long int i;
1750
1751         switch (id) {
1752         case KVM_REG_PPC_DEBUG_INST:
1753                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1754                 break;
1755         case KVM_REG_PPC_HIOR:
1756                 *val = get_reg_val(id, 0);
1757                 break;
1758         case KVM_REG_PPC_DABR:
1759                 *val = get_reg_val(id, vcpu->arch.dabr);
1760                 break;
1761         case KVM_REG_PPC_DABRX:
1762                 *val = get_reg_val(id, vcpu->arch.dabrx);
1763                 break;
1764         case KVM_REG_PPC_DSCR:
1765                 *val = get_reg_val(id, vcpu->arch.dscr);
1766                 break;
1767         case KVM_REG_PPC_PURR:
1768                 *val = get_reg_val(id, vcpu->arch.purr);
1769                 break;
1770         case KVM_REG_PPC_SPURR:
1771                 *val = get_reg_val(id, vcpu->arch.spurr);
1772                 break;
1773         case KVM_REG_PPC_AMR:
1774                 *val = get_reg_val(id, vcpu->arch.amr);
1775                 break;
1776         case KVM_REG_PPC_UAMOR:
1777                 *val = get_reg_val(id, vcpu->arch.uamor);
1778                 break;
1779         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
1780                 i = id - KVM_REG_PPC_MMCR0;
1781                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1782                 break;
1783         case KVM_REG_PPC_MMCR2:
1784                 *val = get_reg_val(id, vcpu->arch.mmcr[2]);
1785                 break;
1786         case KVM_REG_PPC_MMCRA:
1787                 *val = get_reg_val(id, vcpu->arch.mmcra);
1788                 break;
1789         case KVM_REG_PPC_MMCRS:
1790                 *val = get_reg_val(id, vcpu->arch.mmcrs);
1791                 break;
1792         case KVM_REG_PPC_MMCR3:
1793                 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
1794                 break;
1795         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1796                 i = id - KVM_REG_PPC_PMC1;
1797                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1798                 break;
1799         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1800                 i = id - KVM_REG_PPC_SPMC1;
1801                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1802                 break;
1803         case KVM_REG_PPC_SIAR:
1804                 *val = get_reg_val(id, vcpu->arch.siar);
1805                 break;
1806         case KVM_REG_PPC_SDAR:
1807                 *val = get_reg_val(id, vcpu->arch.sdar);
1808                 break;
1809         case KVM_REG_PPC_SIER:
1810                 *val = get_reg_val(id, vcpu->arch.sier[0]);
1811                 break;
1812         case KVM_REG_PPC_SIER2:
1813                 *val = get_reg_val(id, vcpu->arch.sier[1]);
1814                 break;
1815         case KVM_REG_PPC_SIER3:
1816                 *val = get_reg_val(id, vcpu->arch.sier[2]);
1817                 break;
1818         case KVM_REG_PPC_IAMR:
1819                 *val = get_reg_val(id, vcpu->arch.iamr);
1820                 break;
1821         case KVM_REG_PPC_PSPB:
1822                 *val = get_reg_val(id, vcpu->arch.pspb);
1823                 break;
1824         case KVM_REG_PPC_DPDES:
1825                 /*
1826                  * On POWER9, where we are emulating msgsndp etc.,
1827                  * we return 1 bit for each vcpu, which can come from
1828                  * either vcore->dpdes or doorbell_request.
1829                  * On POWER8, doorbell_request is 0.
1830                  */
1831                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes |
1832                                    vcpu->arch.doorbell_request);
1833                 break;
1834         case KVM_REG_PPC_VTB:
1835                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1836                 break;
1837         case KVM_REG_PPC_DAWR:
1838                 *val = get_reg_val(id, vcpu->arch.dawr0);
1839                 break;
1840         case KVM_REG_PPC_DAWRX:
1841                 *val = get_reg_val(id, vcpu->arch.dawrx0);
1842                 break;
1843         case KVM_REG_PPC_DAWR1:
1844                 *val = get_reg_val(id, vcpu->arch.dawr1);
1845                 break;
1846         case KVM_REG_PPC_DAWRX1:
1847                 *val = get_reg_val(id, vcpu->arch.dawrx1);
1848                 break;
1849         case KVM_REG_PPC_CIABR:
1850                 *val = get_reg_val(id, vcpu->arch.ciabr);
1851                 break;
1852         case KVM_REG_PPC_CSIGR:
1853                 *val = get_reg_val(id, vcpu->arch.csigr);
1854                 break;
1855         case KVM_REG_PPC_TACR:
1856                 *val = get_reg_val(id, vcpu->arch.tacr);
1857                 break;
1858         case KVM_REG_PPC_TCSCR:
1859                 *val = get_reg_val(id, vcpu->arch.tcscr);
1860                 break;
1861         case KVM_REG_PPC_PID:
1862                 *val = get_reg_val(id, vcpu->arch.pid);
1863                 break;
1864         case KVM_REG_PPC_ACOP:
1865                 *val = get_reg_val(id, vcpu->arch.acop);
1866                 break;
1867         case KVM_REG_PPC_WORT:
1868                 *val = get_reg_val(id, vcpu->arch.wort);
1869                 break;
1870         case KVM_REG_PPC_TIDR:
1871                 *val = get_reg_val(id, vcpu->arch.tid);
1872                 break;
1873         case KVM_REG_PPC_PSSCR:
1874                 *val = get_reg_val(id, vcpu->arch.psscr);
1875                 break;
1876         case KVM_REG_PPC_VPA_ADDR:
1877                 spin_lock(&vcpu->arch.vpa_update_lock);
1878                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1879                 spin_unlock(&vcpu->arch.vpa_update_lock);
1880                 break;
1881         case KVM_REG_PPC_VPA_SLB:
1882                 spin_lock(&vcpu->arch.vpa_update_lock);
1883                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1884                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1885                 spin_unlock(&vcpu->arch.vpa_update_lock);
1886                 break;
1887         case KVM_REG_PPC_VPA_DTL:
1888                 spin_lock(&vcpu->arch.vpa_update_lock);
1889                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1890                 val->vpaval.length = vcpu->arch.dtl.len;
1891                 spin_unlock(&vcpu->arch.vpa_update_lock);
1892                 break;
1893         case KVM_REG_PPC_TB_OFFSET:
1894                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1895                 break;
1896         case KVM_REG_PPC_LPCR:
1897         case KVM_REG_PPC_LPCR_64:
1898                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1899                 break;
1900         case KVM_REG_PPC_PPR:
1901                 *val = get_reg_val(id, vcpu->arch.ppr);
1902                 break;
1903 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1904         case KVM_REG_PPC_TFHAR:
1905                 *val = get_reg_val(id, vcpu->arch.tfhar);
1906                 break;
1907         case KVM_REG_PPC_TFIAR:
1908                 *val = get_reg_val(id, vcpu->arch.tfiar);
1909                 break;
1910         case KVM_REG_PPC_TEXASR:
1911                 *val = get_reg_val(id, vcpu->arch.texasr);
1912                 break;
1913         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1914                 i = id - KVM_REG_PPC_TM_GPR0;
1915                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1916                 break;
1917         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1918         {
1919                 int j;
1920                 i = id - KVM_REG_PPC_TM_VSR0;
1921                 if (i < 32)
1922                         for (j = 0; j < TS_FPRWIDTH; j++)
1923                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1924                 else {
1925                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1926                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1927                         else
1928                                 r = -ENXIO;
1929                 }
1930                 break;
1931         }
1932         case KVM_REG_PPC_TM_CR:
1933                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1934                 break;
1935         case KVM_REG_PPC_TM_XER:
1936                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1937                 break;
1938         case KVM_REG_PPC_TM_LR:
1939                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1940                 break;
1941         case KVM_REG_PPC_TM_CTR:
1942                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1943                 break;
1944         case KVM_REG_PPC_TM_FPSCR:
1945                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1946                 break;
1947         case KVM_REG_PPC_TM_AMR:
1948                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1949                 break;
1950         case KVM_REG_PPC_TM_PPR:
1951                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1952                 break;
1953         case KVM_REG_PPC_TM_VRSAVE:
1954                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1955                 break;
1956         case KVM_REG_PPC_TM_VSCR:
1957                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1958                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1959                 else
1960                         r = -ENXIO;
1961                 break;
1962         case KVM_REG_PPC_TM_DSCR:
1963                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1964                 break;
1965         case KVM_REG_PPC_TM_TAR:
1966                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1967                 break;
1968 #endif
1969         case KVM_REG_PPC_ARCH_COMPAT:
1970                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1971                 break;
1972         case KVM_REG_PPC_DEC_EXPIRY:
1973                 *val = get_reg_val(id, vcpu->arch.dec_expires +
1974                                    vcpu->arch.vcore->tb_offset);
1975                 break;
1976         case KVM_REG_PPC_ONLINE:
1977                 *val = get_reg_val(id, vcpu->arch.online);
1978                 break;
1979         case KVM_REG_PPC_PTCR:
1980                 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1981                 break;
1982         default:
1983                 r = -EINVAL;
1984                 break;
1985         }
1986
1987         return r;
1988 }
1989
1990 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1991                                  union kvmppc_one_reg *val)
1992 {
1993         int r = 0;
1994         long int i;
1995         unsigned long addr, len;
1996
1997         switch (id) {
1998         case KVM_REG_PPC_HIOR:
1999                 /* Only allow this to be set to zero */
2000                 if (set_reg_val(id, *val))
2001                         r = -EINVAL;
2002                 break;
2003         case KVM_REG_PPC_DABR:
2004                 vcpu->arch.dabr = set_reg_val(id, *val);
2005                 break;
2006         case KVM_REG_PPC_DABRX:
2007                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
2008                 break;
2009         case KVM_REG_PPC_DSCR:
2010                 vcpu->arch.dscr = set_reg_val(id, *val);
2011                 break;
2012         case KVM_REG_PPC_PURR:
2013                 vcpu->arch.purr = set_reg_val(id, *val);
2014                 break;
2015         case KVM_REG_PPC_SPURR:
2016                 vcpu->arch.spurr = set_reg_val(id, *val);
2017                 break;
2018         case KVM_REG_PPC_AMR:
2019                 vcpu->arch.amr = set_reg_val(id, *val);
2020                 break;
2021         case KVM_REG_PPC_UAMOR:
2022                 vcpu->arch.uamor = set_reg_val(id, *val);
2023                 break;
2024         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2025                 i = id - KVM_REG_PPC_MMCR0;
2026                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
2027                 break;
2028         case KVM_REG_PPC_MMCR2:
2029                 vcpu->arch.mmcr[2] = set_reg_val(id, *val);
2030                 break;
2031         case KVM_REG_PPC_MMCRA:
2032                 vcpu->arch.mmcra = set_reg_val(id, *val);
2033                 break;
2034         case KVM_REG_PPC_MMCRS:
2035                 vcpu->arch.mmcrs = set_reg_val(id, *val);
2036                 break;
2037         case KVM_REG_PPC_MMCR3:
2038                 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
2039                 break;
2040         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2041                 i = id - KVM_REG_PPC_PMC1;
2042                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
2043                 break;
2044         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2045                 i = id - KVM_REG_PPC_SPMC1;
2046                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
2047                 break;
2048         case KVM_REG_PPC_SIAR:
2049                 vcpu->arch.siar = set_reg_val(id, *val);
2050                 break;
2051         case KVM_REG_PPC_SDAR:
2052                 vcpu->arch.sdar = set_reg_val(id, *val);
2053                 break;
2054         case KVM_REG_PPC_SIER:
2055                 vcpu->arch.sier[0] = set_reg_val(id, *val);
2056                 break;
2057         case KVM_REG_PPC_SIER2:
2058                 vcpu->arch.sier[1] = set_reg_val(id, *val);
2059                 break;
2060         case KVM_REG_PPC_SIER3:
2061                 vcpu->arch.sier[2] = set_reg_val(id, *val);
2062                 break;
2063         case KVM_REG_PPC_IAMR:
2064                 vcpu->arch.iamr = set_reg_val(id, *val);
2065                 break;
2066         case KVM_REG_PPC_PSPB:
2067                 vcpu->arch.pspb = set_reg_val(id, *val);
2068                 break;
2069         case KVM_REG_PPC_DPDES:
2070                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
2071                 break;
2072         case KVM_REG_PPC_VTB:
2073                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
2074                 break;
2075         case KVM_REG_PPC_DAWR:
2076                 vcpu->arch.dawr0 = set_reg_val(id, *val);
2077                 break;
2078         case KVM_REG_PPC_DAWRX:
2079                 vcpu->arch.dawrx0 = set_reg_val(id, *val) & ~DAWRX_HYP;
2080                 break;
2081         case KVM_REG_PPC_DAWR1:
2082                 vcpu->arch.dawr1 = set_reg_val(id, *val);
2083                 break;
2084         case KVM_REG_PPC_DAWRX1:
2085                 vcpu->arch.dawrx1 = set_reg_val(id, *val) & ~DAWRX_HYP;
2086                 break;
2087         case KVM_REG_PPC_CIABR:
2088                 vcpu->arch.ciabr = set_reg_val(id, *val);
2089                 /* Don't allow setting breakpoints in hypervisor code */
2090                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
2091                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
2092                 break;
2093         case KVM_REG_PPC_CSIGR:
2094                 vcpu->arch.csigr = set_reg_val(id, *val);
2095                 break;
2096         case KVM_REG_PPC_TACR:
2097                 vcpu->arch.tacr = set_reg_val(id, *val);
2098                 break;
2099         case KVM_REG_PPC_TCSCR:
2100                 vcpu->arch.tcscr = set_reg_val(id, *val);
2101                 break;
2102         case KVM_REG_PPC_PID:
2103                 vcpu->arch.pid = set_reg_val(id, *val);
2104                 break;
2105         case KVM_REG_PPC_ACOP:
2106                 vcpu->arch.acop = set_reg_val(id, *val);
2107                 break;
2108         case KVM_REG_PPC_WORT:
2109                 vcpu->arch.wort = set_reg_val(id, *val);
2110                 break;
2111         case KVM_REG_PPC_TIDR:
2112                 vcpu->arch.tid = set_reg_val(id, *val);
2113                 break;
2114         case KVM_REG_PPC_PSSCR:
2115                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2116                 break;
2117         case KVM_REG_PPC_VPA_ADDR:
2118                 addr = set_reg_val(id, *val);
2119                 r = -EINVAL;
2120                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2121                               vcpu->arch.dtl.next_gpa))
2122                         break;
2123                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2124                 break;
2125         case KVM_REG_PPC_VPA_SLB:
2126                 addr = val->vpaval.addr;
2127                 len = val->vpaval.length;
2128                 r = -EINVAL;
2129                 if (addr && !vcpu->arch.vpa.next_gpa)
2130                         break;
2131                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2132                 break;
2133         case KVM_REG_PPC_VPA_DTL:
2134                 addr = val->vpaval.addr;
2135                 len = val->vpaval.length;
2136                 r = -EINVAL;
2137                 if (addr && (len < sizeof(struct dtl_entry) ||
2138                              !vcpu->arch.vpa.next_gpa))
2139                         break;
2140                 len -= len % sizeof(struct dtl_entry);
2141                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2142                 break;
2143         case KVM_REG_PPC_TB_OFFSET:
2144                 /* round up to multiple of 2^24 */
2145                 vcpu->arch.vcore->tb_offset =
2146                         ALIGN(set_reg_val(id, *val), 1UL << 24);
2147                 break;
2148         case KVM_REG_PPC_LPCR:
2149                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2150                 break;
2151         case KVM_REG_PPC_LPCR_64:
2152                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2153                 break;
2154         case KVM_REG_PPC_PPR:
2155                 vcpu->arch.ppr = set_reg_val(id, *val);
2156                 break;
2157 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2158         case KVM_REG_PPC_TFHAR:
2159                 vcpu->arch.tfhar = set_reg_val(id, *val);
2160                 break;
2161         case KVM_REG_PPC_TFIAR:
2162                 vcpu->arch.tfiar = set_reg_val(id, *val);
2163                 break;
2164         case KVM_REG_PPC_TEXASR:
2165                 vcpu->arch.texasr = set_reg_val(id, *val);
2166                 break;
2167         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2168                 i = id - KVM_REG_PPC_TM_GPR0;
2169                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2170                 break;
2171         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2172         {
2173                 int j;
2174                 i = id - KVM_REG_PPC_TM_VSR0;
2175                 if (i < 32)
2176                         for (j = 0; j < TS_FPRWIDTH; j++)
2177                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2178                 else
2179                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
2180                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2181                         else
2182                                 r = -ENXIO;
2183                 break;
2184         }
2185         case KVM_REG_PPC_TM_CR:
2186                 vcpu->arch.cr_tm = set_reg_val(id, *val);
2187                 break;
2188         case KVM_REG_PPC_TM_XER:
2189                 vcpu->arch.xer_tm = set_reg_val(id, *val);
2190                 break;
2191         case KVM_REG_PPC_TM_LR:
2192                 vcpu->arch.lr_tm = set_reg_val(id, *val);
2193                 break;
2194         case KVM_REG_PPC_TM_CTR:
2195                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2196                 break;
2197         case KVM_REG_PPC_TM_FPSCR:
2198                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2199                 break;
2200         case KVM_REG_PPC_TM_AMR:
2201                 vcpu->arch.amr_tm = set_reg_val(id, *val);
2202                 break;
2203         case KVM_REG_PPC_TM_PPR:
2204                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2205                 break;
2206         case KVM_REG_PPC_TM_VRSAVE:
2207                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2208                 break;
2209         case KVM_REG_PPC_TM_VSCR:
2210                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2211                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2212                 else
2213                         r = - ENXIO;
2214                 break;
2215         case KVM_REG_PPC_TM_DSCR:
2216                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2217                 break;
2218         case KVM_REG_PPC_TM_TAR:
2219                 vcpu->arch.tar_tm = set_reg_val(id, *val);
2220                 break;
2221 #endif
2222         case KVM_REG_PPC_ARCH_COMPAT:
2223                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2224                 break;
2225         case KVM_REG_PPC_DEC_EXPIRY:
2226                 vcpu->arch.dec_expires = set_reg_val(id, *val) -
2227                         vcpu->arch.vcore->tb_offset;
2228                 break;
2229         case KVM_REG_PPC_ONLINE:
2230                 i = set_reg_val(id, *val);
2231                 if (i && !vcpu->arch.online)
2232                         atomic_inc(&vcpu->arch.vcore->online_count);
2233                 else if (!i && vcpu->arch.online)
2234                         atomic_dec(&vcpu->arch.vcore->online_count);
2235                 vcpu->arch.online = i;
2236                 break;
2237         case KVM_REG_PPC_PTCR:
2238                 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2239                 break;
2240         default:
2241                 r = -EINVAL;
2242                 break;
2243         }
2244
2245         return r;
2246 }
2247
2248 /*
2249  * On POWER9, threads are independent and can be in different partitions.
2250  * Therefore we consider each thread to be a subcore.
2251  * There is a restriction that all threads have to be in the same
2252  * MMU mode (radix or HPT), unfortunately, but since we only support
2253  * HPT guests on a HPT host so far, that isn't an impediment yet.
2254  */
2255 static int threads_per_vcore(struct kvm *kvm)
2256 {
2257         if (kvm->arch.threads_indep)
2258                 return 1;
2259         return threads_per_subcore;
2260 }
2261
2262 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2263 {
2264         struct kvmppc_vcore *vcore;
2265
2266         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2267
2268         if (vcore == NULL)
2269                 return NULL;
2270
2271         spin_lock_init(&vcore->lock);
2272         spin_lock_init(&vcore->stoltb_lock);
2273         rcuwait_init(&vcore->wait);
2274         vcore->preempt_tb = TB_NIL;
2275         vcore->lpcr = kvm->arch.lpcr;
2276         vcore->first_vcpuid = id;
2277         vcore->kvm = kvm;
2278         INIT_LIST_HEAD(&vcore->preempt_list);
2279
2280         return vcore;
2281 }
2282
2283 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2284 static struct debugfs_timings_element {
2285         const char *name;
2286         size_t offset;
2287 } timings[] = {
2288         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
2289         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
2290         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
2291         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
2292         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
2293 };
2294
2295 #define N_TIMINGS       (ARRAY_SIZE(timings))
2296
2297 struct debugfs_timings_state {
2298         struct kvm_vcpu *vcpu;
2299         unsigned int    buflen;
2300         char            buf[N_TIMINGS * 100];
2301 };
2302
2303 static int debugfs_timings_open(struct inode *inode, struct file *file)
2304 {
2305         struct kvm_vcpu *vcpu = inode->i_private;
2306         struct debugfs_timings_state *p;
2307
2308         p = kzalloc(sizeof(*p), GFP_KERNEL);
2309         if (!p)
2310                 return -ENOMEM;
2311
2312         kvm_get_kvm(vcpu->kvm);
2313         p->vcpu = vcpu;
2314         file->private_data = p;
2315
2316         return nonseekable_open(inode, file);
2317 }
2318
2319 static int debugfs_timings_release(struct inode *inode, struct file *file)
2320 {
2321         struct debugfs_timings_state *p = file->private_data;
2322
2323         kvm_put_kvm(p->vcpu->kvm);
2324         kfree(p);
2325         return 0;
2326 }
2327
2328 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2329                                     size_t len, loff_t *ppos)
2330 {
2331         struct debugfs_timings_state *p = file->private_data;
2332         struct kvm_vcpu *vcpu = p->vcpu;
2333         char *s, *buf_end;
2334         struct kvmhv_tb_accumulator tb;
2335         u64 count;
2336         loff_t pos;
2337         ssize_t n;
2338         int i, loops;
2339         bool ok;
2340
2341         if (!p->buflen) {
2342                 s = p->buf;
2343                 buf_end = s + sizeof(p->buf);
2344                 for (i = 0; i < N_TIMINGS; ++i) {
2345                         struct kvmhv_tb_accumulator *acc;
2346
2347                         acc = (struct kvmhv_tb_accumulator *)
2348                                 ((unsigned long)vcpu + timings[i].offset);
2349                         ok = false;
2350                         for (loops = 0; loops < 1000; ++loops) {
2351                                 count = acc->seqcount;
2352                                 if (!(count & 1)) {
2353                                         smp_rmb();
2354                                         tb = *acc;
2355                                         smp_rmb();
2356                                         if (count == acc->seqcount) {
2357                                                 ok = true;
2358                                                 break;
2359                                         }
2360                                 }
2361                                 udelay(1);
2362                         }
2363                         if (!ok)
2364                                 snprintf(s, buf_end - s, "%s: stuck\n",
2365                                         timings[i].name);
2366                         else
2367                                 snprintf(s, buf_end - s,
2368                                         "%s: %llu %llu %llu %llu\n",
2369                                         timings[i].name, count / 2,
2370                                         tb_to_ns(tb.tb_total),
2371                                         tb_to_ns(tb.tb_min),
2372                                         tb_to_ns(tb.tb_max));
2373                         s += strlen(s);
2374                 }
2375                 p->buflen = s - p->buf;
2376         }
2377
2378         pos = *ppos;
2379         if (pos >= p->buflen)
2380                 return 0;
2381         if (len > p->buflen - pos)
2382                 len = p->buflen - pos;
2383         n = copy_to_user(buf, p->buf + pos, len);
2384         if (n) {
2385                 if (n == len)
2386                         return -EFAULT;
2387                 len -= n;
2388         }
2389         *ppos = pos + len;
2390         return len;
2391 }
2392
2393 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2394                                      size_t len, loff_t *ppos)
2395 {
2396         return -EACCES;
2397 }
2398
2399 static const struct file_operations debugfs_timings_ops = {
2400         .owner   = THIS_MODULE,
2401         .open    = debugfs_timings_open,
2402         .release = debugfs_timings_release,
2403         .read    = debugfs_timings_read,
2404         .write   = debugfs_timings_write,
2405         .llseek  = generic_file_llseek,
2406 };
2407
2408 /* Create a debugfs directory for the vcpu */
2409 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2410 {
2411         char buf[16];
2412         struct kvm *kvm = vcpu->kvm;
2413
2414         snprintf(buf, sizeof(buf), "vcpu%u", id);
2415         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2416         debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir, vcpu,
2417                             &debugfs_timings_ops);
2418 }
2419
2420 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2421 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2422 {
2423 }
2424 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2425
2426 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2427 {
2428         int err;
2429         int core;
2430         struct kvmppc_vcore *vcore;
2431         struct kvm *kvm;
2432         unsigned int id;
2433
2434         kvm = vcpu->kvm;
2435         id = vcpu->vcpu_id;
2436
2437         vcpu->arch.shared = &vcpu->arch.shregs;
2438 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2439         /*
2440          * The shared struct is never shared on HV,
2441          * so we can always use host endianness
2442          */
2443 #ifdef __BIG_ENDIAN__
2444         vcpu->arch.shared_big_endian = true;
2445 #else
2446         vcpu->arch.shared_big_endian = false;
2447 #endif
2448 #endif
2449         vcpu->arch.mmcr[0] = MMCR0_FC;
2450         vcpu->arch.ctrl = CTRL_RUNLATCH;
2451         /* default to host PVR, since we can't spoof it */
2452         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2453         spin_lock_init(&vcpu->arch.vpa_update_lock);
2454         spin_lock_init(&vcpu->arch.tbacct_lock);
2455         vcpu->arch.busy_preempt = TB_NIL;
2456         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2457
2458         /*
2459          * Set the default HFSCR for the guest from the host value.
2460          * This value is only used on POWER9.
2461          * On POWER9, we want to virtualize the doorbell facility, so we
2462          * don't set the HFSCR_MSGP bit, and that causes those instructions
2463          * to trap and then we emulate them.
2464          */
2465         vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2466                 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP | HFSCR_PREFIX;
2467         if (cpu_has_feature(CPU_FTR_HVMODE)) {
2468                 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2469                 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2470                         vcpu->arch.hfscr |= HFSCR_TM;
2471         }
2472         if (cpu_has_feature(CPU_FTR_TM_COMP))
2473                 vcpu->arch.hfscr |= HFSCR_TM;
2474
2475         kvmppc_mmu_book3s_hv_init(vcpu);
2476
2477         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2478
2479         init_waitqueue_head(&vcpu->arch.cpu_run);
2480
2481         mutex_lock(&kvm->lock);
2482         vcore = NULL;
2483         err = -EINVAL;
2484         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2485                 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2486                         pr_devel("KVM: VCPU ID too high\n");
2487                         core = KVM_MAX_VCORES;
2488                 } else {
2489                         BUG_ON(kvm->arch.smt_mode != 1);
2490                         core = kvmppc_pack_vcpu_id(kvm, id);
2491                 }
2492         } else {
2493                 core = id / kvm->arch.smt_mode;
2494         }
2495         if (core < KVM_MAX_VCORES) {
2496                 vcore = kvm->arch.vcores[core];
2497                 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2498                         pr_devel("KVM: collision on id %u", id);
2499                         vcore = NULL;
2500                 } else if (!vcore) {
2501                         /*
2502                          * Take mmu_setup_lock for mutual exclusion
2503                          * with kvmppc_update_lpcr().
2504                          */
2505                         err = -ENOMEM;
2506                         vcore = kvmppc_vcore_create(kvm,
2507                                         id & ~(kvm->arch.smt_mode - 1));
2508                         mutex_lock(&kvm->arch.mmu_setup_lock);
2509                         kvm->arch.vcores[core] = vcore;
2510                         kvm->arch.online_vcores++;
2511                         mutex_unlock(&kvm->arch.mmu_setup_lock);
2512                 }
2513         }
2514         mutex_unlock(&kvm->lock);
2515
2516         if (!vcore)
2517                 return err;
2518
2519         spin_lock(&vcore->lock);
2520         ++vcore->num_threads;
2521         spin_unlock(&vcore->lock);
2522         vcpu->arch.vcore = vcore;
2523         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2524         vcpu->arch.thread_cpu = -1;
2525         vcpu->arch.prev_cpu = -1;
2526
2527         vcpu->arch.cpu_type = KVM_CPU_3S_64;
2528         kvmppc_sanity_check(vcpu);
2529
2530         debugfs_vcpu_init(vcpu, id);
2531
2532         return 0;
2533 }
2534
2535 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2536                               unsigned long flags)
2537 {
2538         int err;
2539         int esmt = 0;
2540
2541         if (flags)
2542                 return -EINVAL;
2543         if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2544                 return -EINVAL;
2545         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2546                 /*
2547                  * On POWER8 (or POWER7), the threading mode is "strict",
2548                  * so we pack smt_mode vcpus per vcore.
2549                  */
2550                 if (smt_mode > threads_per_subcore)
2551                         return -EINVAL;
2552         } else {
2553                 /*
2554                  * On POWER9, the threading mode is "loose",
2555                  * so each vcpu gets its own vcore.
2556                  */
2557                 esmt = smt_mode;
2558                 smt_mode = 1;
2559         }
2560         mutex_lock(&kvm->lock);
2561         err = -EBUSY;
2562         if (!kvm->arch.online_vcores) {
2563                 kvm->arch.smt_mode = smt_mode;
2564                 kvm->arch.emul_smt_mode = esmt;
2565                 err = 0;
2566         }
2567         mutex_unlock(&kvm->lock);
2568
2569         return err;
2570 }
2571
2572 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2573 {
2574         if (vpa->pinned_addr)
2575                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2576                                         vpa->dirty);
2577 }
2578
2579 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2580 {
2581         spin_lock(&vcpu->arch.vpa_update_lock);
2582         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2583         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2584         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2585         spin_unlock(&vcpu->arch.vpa_update_lock);
2586 }
2587
2588 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2589 {
2590         /* Indicate we want to get back into the guest */
2591         return 1;
2592 }
2593
2594 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2595 {
2596         unsigned long dec_nsec, now;
2597
2598         now = get_tb();
2599         if (now > vcpu->arch.dec_expires) {
2600                 /* decrementer has already gone negative */
2601                 kvmppc_core_queue_dec(vcpu);
2602                 kvmppc_core_prepare_to_enter(vcpu);
2603                 return;
2604         }
2605         dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2606         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2607         vcpu->arch.timer_running = 1;
2608 }
2609
2610 extern int __kvmppc_vcore_entry(void);
2611
2612 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2613                                    struct kvm_vcpu *vcpu)
2614 {
2615         u64 now;
2616
2617         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2618                 return;
2619         spin_lock_irq(&vcpu->arch.tbacct_lock);
2620         now = mftb();
2621         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2622                 vcpu->arch.stolen_logged;
2623         vcpu->arch.busy_preempt = now;
2624         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2625         spin_unlock_irq(&vcpu->arch.tbacct_lock);
2626         --vc->n_runnable;
2627         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2628 }
2629
2630 static int kvmppc_grab_hwthread(int cpu)
2631 {
2632         struct paca_struct *tpaca;
2633         long timeout = 10000;
2634
2635         tpaca = paca_ptrs[cpu];
2636
2637         /* Ensure the thread won't go into the kernel if it wakes */
2638         tpaca->kvm_hstate.kvm_vcpu = NULL;
2639         tpaca->kvm_hstate.kvm_vcore = NULL;
2640         tpaca->kvm_hstate.napping = 0;
2641         smp_wmb();
2642         tpaca->kvm_hstate.hwthread_req = 1;
2643
2644         /*
2645          * If the thread is already executing in the kernel (e.g. handling
2646          * a stray interrupt), wait for it to get back to nap mode.
2647          * The smp_mb() is to ensure that our setting of hwthread_req
2648          * is visible before we look at hwthread_state, so if this
2649          * races with the code at system_reset_pSeries and the thread
2650          * misses our setting of hwthread_req, we are sure to see its
2651          * setting of hwthread_state, and vice versa.
2652          */
2653         smp_mb();
2654         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2655                 if (--timeout <= 0) {
2656                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
2657                         return -EBUSY;
2658                 }
2659                 udelay(1);
2660         }
2661         return 0;
2662 }
2663
2664 static void kvmppc_release_hwthread(int cpu)
2665 {
2666         struct paca_struct *tpaca;
2667
2668         tpaca = paca_ptrs[cpu];
2669         tpaca->kvm_hstate.hwthread_req = 0;
2670         tpaca->kvm_hstate.kvm_vcpu = NULL;
2671         tpaca->kvm_hstate.kvm_vcore = NULL;
2672         tpaca->kvm_hstate.kvm_split_mode = NULL;
2673 }
2674
2675 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2676 {
2677         struct kvm_nested_guest *nested = vcpu->arch.nested;
2678         cpumask_t *cpu_in_guest;
2679         int i;
2680
2681         cpu = cpu_first_thread_sibling(cpu);
2682         if (nested) {
2683                 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2684                 cpu_in_guest = &nested->cpu_in_guest;
2685         } else {
2686                 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2687                 cpu_in_guest = &kvm->arch.cpu_in_guest;
2688         }
2689         /*
2690          * Make sure setting of bit in need_tlb_flush precedes
2691          * testing of cpu_in_guest bits.  The matching barrier on
2692          * the other side is the first smp_mb() in kvmppc_run_core().
2693          */
2694         smp_mb();
2695         for (i = 0; i < threads_per_core; ++i)
2696                 if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2697                         smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2698 }
2699
2700 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2701 {
2702         struct kvm_nested_guest *nested = vcpu->arch.nested;
2703         struct kvm *kvm = vcpu->kvm;
2704         int prev_cpu;
2705
2706         if (!cpu_has_feature(CPU_FTR_HVMODE))
2707                 return;
2708
2709         if (nested)
2710                 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2711         else
2712                 prev_cpu = vcpu->arch.prev_cpu;
2713
2714         /*
2715          * With radix, the guest can do TLB invalidations itself,
2716          * and it could choose to use the local form (tlbiel) if
2717          * it is invalidating a translation that has only ever been
2718          * used on one vcpu.  However, that doesn't mean it has
2719          * only ever been used on one physical cpu, since vcpus
2720          * can move around between pcpus.  To cope with this, when
2721          * a vcpu moves from one pcpu to another, we need to tell
2722          * any vcpus running on the same core as this vcpu previously
2723          * ran to flush the TLB.  The TLB is shared between threads,
2724          * so we use a single bit in .need_tlb_flush for all 4 threads.
2725          */
2726         if (prev_cpu != pcpu) {
2727                 if (prev_cpu >= 0 &&
2728                     cpu_first_thread_sibling(prev_cpu) !=
2729                     cpu_first_thread_sibling(pcpu))
2730                         radix_flush_cpu(kvm, prev_cpu, vcpu);
2731                 if (nested)
2732                         nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2733                 else
2734                         vcpu->arch.prev_cpu = pcpu;
2735         }
2736 }
2737
2738 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2739 {
2740         int cpu;
2741         struct paca_struct *tpaca;
2742         struct kvm *kvm = vc->kvm;
2743
2744         cpu = vc->pcpu;
2745         if (vcpu) {
2746                 if (vcpu->arch.timer_running) {
2747                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2748                         vcpu->arch.timer_running = 0;
2749                 }
2750                 cpu += vcpu->arch.ptid;
2751                 vcpu->cpu = vc->pcpu;
2752                 vcpu->arch.thread_cpu = cpu;
2753                 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2754         }
2755         tpaca = paca_ptrs[cpu];
2756         tpaca->kvm_hstate.kvm_vcpu = vcpu;
2757         tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2758         tpaca->kvm_hstate.fake_suspend = 0;
2759         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2760         smp_wmb();
2761         tpaca->kvm_hstate.kvm_vcore = vc;
2762         if (cpu != smp_processor_id())
2763                 kvmppc_ipi_thread(cpu);
2764 }
2765
2766 static void kvmppc_wait_for_nap(int n_threads)
2767 {
2768         int cpu = smp_processor_id();
2769         int i, loops;
2770
2771         if (n_threads <= 1)
2772                 return;
2773         for (loops = 0; loops < 1000000; ++loops) {
2774                 /*
2775                  * Check if all threads are finished.
2776                  * We set the vcore pointer when starting a thread
2777                  * and the thread clears it when finished, so we look
2778                  * for any threads that still have a non-NULL vcore ptr.
2779                  */
2780                 for (i = 1; i < n_threads; ++i)
2781                         if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2782                                 break;
2783                 if (i == n_threads) {
2784                         HMT_medium();
2785                         return;
2786                 }
2787                 HMT_low();
2788         }
2789         HMT_medium();
2790         for (i = 1; i < n_threads; ++i)
2791                 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2792                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2793 }
2794
2795 /*
2796  * Check that we are on thread 0 and that any other threads in
2797  * this core are off-line.  Then grab the threads so they can't
2798  * enter the kernel.
2799  */
2800 static int on_primary_thread(void)
2801 {
2802         int cpu = smp_processor_id();
2803         int thr;
2804
2805         /* Are we on a primary subcore? */
2806         if (cpu_thread_in_subcore(cpu))
2807                 return 0;
2808
2809         thr = 0;
2810         while (++thr < threads_per_subcore)
2811                 if (cpu_online(cpu + thr))
2812                         return 0;
2813
2814         /* Grab all hw threads so they can't go into the kernel */
2815         for (thr = 1; thr < threads_per_subcore; ++thr) {
2816                 if (kvmppc_grab_hwthread(cpu + thr)) {
2817                         /* Couldn't grab one; let the others go */
2818                         do {
2819                                 kvmppc_release_hwthread(cpu + thr);
2820                         } while (--thr > 0);
2821                         return 0;
2822                 }
2823         }
2824         return 1;
2825 }
2826
2827 /*
2828  * A list of virtual cores for each physical CPU.
2829  * These are vcores that could run but their runner VCPU tasks are
2830  * (or may be) preempted.
2831  */
2832 struct preempted_vcore_list {
2833         struct list_head        list;
2834         spinlock_t              lock;
2835 };
2836
2837 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2838
2839 static void init_vcore_lists(void)
2840 {
2841         int cpu;
2842
2843         for_each_possible_cpu(cpu) {
2844                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2845                 spin_lock_init(&lp->lock);
2846                 INIT_LIST_HEAD(&lp->list);
2847         }
2848 }
2849
2850 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2851 {
2852         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2853
2854         vc->vcore_state = VCORE_PREEMPT;
2855         vc->pcpu = smp_processor_id();
2856         if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2857                 spin_lock(&lp->lock);
2858                 list_add_tail(&vc->preempt_list, &lp->list);
2859                 spin_unlock(&lp->lock);
2860         }
2861
2862         /* Start accumulating stolen time */
2863         kvmppc_core_start_stolen(vc);
2864 }
2865
2866 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2867 {
2868         struct preempted_vcore_list *lp;
2869
2870         kvmppc_core_end_stolen(vc);
2871         if (!list_empty(&vc->preempt_list)) {
2872                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2873                 spin_lock(&lp->lock);
2874                 list_del_init(&vc->preempt_list);
2875                 spin_unlock(&lp->lock);
2876         }
2877         vc->vcore_state = VCORE_INACTIVE;
2878 }
2879
2880 /*
2881  * This stores information about the virtual cores currently
2882  * assigned to a physical core.
2883  */
2884 struct core_info {
2885         int             n_subcores;
2886         int             max_subcore_threads;
2887         int             total_threads;
2888         int             subcore_threads[MAX_SUBCORES];
2889         struct kvmppc_vcore *vc[MAX_SUBCORES];
2890 };
2891
2892 /*
2893  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2894  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2895  */
2896 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2897
2898 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2899 {
2900         memset(cip, 0, sizeof(*cip));
2901         cip->n_subcores = 1;
2902         cip->max_subcore_threads = vc->num_threads;
2903         cip->total_threads = vc->num_threads;
2904         cip->subcore_threads[0] = vc->num_threads;
2905         cip->vc[0] = vc;
2906 }
2907
2908 static bool subcore_config_ok(int n_subcores, int n_threads)
2909 {
2910         /*
2911          * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2912          * split-core mode, with one thread per subcore.
2913          */
2914         if (cpu_has_feature(CPU_FTR_ARCH_300))
2915                 return n_subcores <= 4 && n_threads == 1;
2916
2917         /* On POWER8, can only dynamically split if unsplit to begin with */
2918         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2919                 return false;
2920         if (n_subcores > MAX_SUBCORES)
2921                 return false;
2922         if (n_subcores > 1) {
2923                 if (!(dynamic_mt_modes & 2))
2924                         n_subcores = 4;
2925                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2926                         return false;
2927         }
2928
2929         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2930 }
2931
2932 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2933 {
2934         vc->entry_exit_map = 0;
2935         vc->in_guest = 0;
2936         vc->napping_threads = 0;
2937         vc->conferring_threads = 0;
2938         vc->tb_offset_applied = 0;
2939 }
2940
2941 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2942 {
2943         int n_threads = vc->num_threads;
2944         int sub;
2945
2946         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2947                 return false;
2948
2949         /* In one_vm_per_core mode, require all vcores to be from the same vm */
2950         if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2951                 return false;
2952
2953         if (n_threads < cip->max_subcore_threads)
2954                 n_threads = cip->max_subcore_threads;
2955         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2956                 return false;
2957         cip->max_subcore_threads = n_threads;
2958
2959         sub = cip->n_subcores;
2960         ++cip->n_subcores;
2961         cip->total_threads += vc->num_threads;
2962         cip->subcore_threads[sub] = vc->num_threads;
2963         cip->vc[sub] = vc;
2964         init_vcore_to_run(vc);
2965         list_del_init(&vc->preempt_list);
2966
2967         return true;
2968 }
2969
2970 /*
2971  * Work out whether it is possible to piggyback the execution of
2972  * vcore *pvc onto the execution of the other vcores described in *cip.
2973  */
2974 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2975                           int target_threads)
2976 {
2977         if (cip->total_threads + pvc->num_threads > target_threads)
2978                 return false;
2979
2980         return can_dynamic_split(pvc, cip);
2981 }
2982
2983 static void prepare_threads(struct kvmppc_vcore *vc)
2984 {
2985         int i;
2986         struct kvm_vcpu *vcpu;
2987
2988         for_each_runnable_thread(i, vcpu, vc) {
2989                 if (signal_pending(vcpu->arch.run_task))
2990                         vcpu->arch.ret = -EINTR;
2991                 else if (no_mixing_hpt_and_radix &&
2992                          kvm_is_radix(vc->kvm) != radix_enabled())
2993                         vcpu->arch.ret = -EINVAL;
2994                 else if (vcpu->arch.vpa.update_pending ||
2995                          vcpu->arch.slb_shadow.update_pending ||
2996                          vcpu->arch.dtl.update_pending)
2997                         vcpu->arch.ret = RESUME_GUEST;
2998                 else
2999                         continue;
3000                 kvmppc_remove_runnable(vc, vcpu);
3001                 wake_up(&vcpu->arch.cpu_run);
3002         }
3003 }
3004
3005 static void collect_piggybacks(struct core_info *cip, int target_threads)
3006 {
3007         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3008         struct kvmppc_vcore *pvc, *vcnext;
3009
3010         spin_lock(&lp->lock);
3011         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
3012                 if (!spin_trylock(&pvc->lock))
3013                         continue;
3014                 prepare_threads(pvc);
3015                 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
3016                         list_del_init(&pvc->preempt_list);
3017                         if (pvc->runner == NULL) {
3018                                 pvc->vcore_state = VCORE_INACTIVE;
3019                                 kvmppc_core_end_stolen(pvc);
3020                         }
3021                         spin_unlock(&pvc->lock);
3022                         continue;
3023                 }
3024                 if (!can_piggyback(pvc, cip, target_threads)) {
3025                         spin_unlock(&pvc->lock);
3026                         continue;
3027                 }
3028                 kvmppc_core_end_stolen(pvc);
3029                 pvc->vcore_state = VCORE_PIGGYBACK;
3030                 if (cip->total_threads >= target_threads)
3031                         break;
3032         }
3033         spin_unlock(&lp->lock);
3034 }
3035
3036 static bool recheck_signals_and_mmu(struct core_info *cip)
3037 {
3038         int sub, i;
3039         struct kvm_vcpu *vcpu;
3040         struct kvmppc_vcore *vc;
3041
3042         for (sub = 0; sub < cip->n_subcores; ++sub) {
3043                 vc = cip->vc[sub];
3044                 if (!vc->kvm->arch.mmu_ready)
3045                         return true;
3046                 for_each_runnable_thread(i, vcpu, vc)
3047                         if (signal_pending(vcpu->arch.run_task))
3048                                 return true;
3049         }
3050         return false;
3051 }
3052
3053 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
3054 {
3055         int still_running = 0, i;
3056         u64 now;
3057         long ret;
3058         struct kvm_vcpu *vcpu;
3059
3060         spin_lock(&vc->lock);
3061         now = get_tb();
3062         for_each_runnable_thread(i, vcpu, vc) {
3063                 /*
3064                  * It's safe to unlock the vcore in the loop here, because
3065                  * for_each_runnable_thread() is safe against removal of
3066                  * the vcpu, and the vcore state is VCORE_EXITING here,
3067                  * so any vcpus becoming runnable will have their arch.trap
3068                  * set to zero and can't actually run in the guest.
3069                  */
3070                 spin_unlock(&vc->lock);
3071                 /* cancel pending dec exception if dec is positive */
3072                 if (now < vcpu->arch.dec_expires &&
3073                     kvmppc_core_pending_dec(vcpu))
3074                         kvmppc_core_dequeue_dec(vcpu);
3075
3076                 trace_kvm_guest_exit(vcpu);
3077
3078                 ret = RESUME_GUEST;
3079                 if (vcpu->arch.trap)
3080                         ret = kvmppc_handle_exit_hv(vcpu,
3081                                                     vcpu->arch.run_task);
3082
3083                 vcpu->arch.ret = ret;
3084                 vcpu->arch.trap = 0;
3085
3086                 spin_lock(&vc->lock);
3087                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
3088                         if (vcpu->arch.pending_exceptions)
3089                                 kvmppc_core_prepare_to_enter(vcpu);
3090                         if (vcpu->arch.ceded)
3091                                 kvmppc_set_timer(vcpu);
3092                         else
3093                                 ++still_running;
3094                 } else {
3095                         kvmppc_remove_runnable(vc, vcpu);
3096                         wake_up(&vcpu->arch.cpu_run);
3097                 }
3098         }
3099         if (!is_master) {
3100                 if (still_running > 0) {
3101                         kvmppc_vcore_preempt(vc);
3102                 } else if (vc->runner) {
3103                         vc->vcore_state = VCORE_PREEMPT;
3104                         kvmppc_core_start_stolen(vc);
3105                 } else {
3106                         vc->vcore_state = VCORE_INACTIVE;
3107                 }
3108                 if (vc->n_runnable > 0 && vc->runner == NULL) {
3109                         /* make sure there's a candidate runner awake */
3110                         i = -1;
3111                         vcpu = next_runnable_thread(vc, &i);
3112                         wake_up(&vcpu->arch.cpu_run);
3113                 }
3114         }
3115         spin_unlock(&vc->lock);
3116 }
3117
3118 /*
3119  * Clear core from the list of active host cores as we are about to
3120  * enter the guest. Only do this if it is the primary thread of the
3121  * core (not if a subcore) that is entering the guest.
3122  */
3123 static inline int kvmppc_clear_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          * Memory barrier can be omitted here as we will do a smp_wmb()
3131          * later in kvmppc_start_thread and we need ensure that state is
3132          * visible to other CPUs only after we enter guest.
3133          */
3134         core = cpu >> threads_shift;
3135         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3136         return 0;
3137 }
3138
3139 /*
3140  * Advertise this core as an active host core since we exited the guest
3141  * Only need to do this if it is the primary thread of the core that is
3142  * exiting.
3143  */
3144 static inline int kvmppc_set_host_core(unsigned int cpu)
3145 {
3146         int core;
3147
3148         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3149                 return 0;
3150
3151         /*
3152          * Memory barrier can be omitted here because we do a spin_unlock
3153          * immediately after this which provides the memory barrier.
3154          */
3155         core = cpu >> threads_shift;
3156         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3157         return 0;
3158 }
3159
3160 static void set_irq_happened(int trap)
3161 {
3162         switch (trap) {
3163         case BOOK3S_INTERRUPT_EXTERNAL:
3164                 local_paca->irq_happened |= PACA_IRQ_EE;
3165                 break;
3166         case BOOK3S_INTERRUPT_H_DOORBELL:
3167                 local_paca->irq_happened |= PACA_IRQ_DBELL;
3168                 break;
3169         case BOOK3S_INTERRUPT_HMI:
3170                 local_paca->irq_happened |= PACA_IRQ_HMI;
3171                 break;
3172         case BOOK3S_INTERRUPT_SYSTEM_RESET:
3173                 replay_system_reset();
3174                 break;
3175         }
3176 }
3177
3178 /*
3179  * Run a set of guest threads on a physical core.
3180  * Called with vc->lock held.
3181  */
3182 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3183 {
3184         struct kvm_vcpu *vcpu;
3185         int i;
3186         int srcu_idx;
3187         struct core_info core_info;
3188         struct kvmppc_vcore *pvc;
3189         struct kvm_split_mode split_info, *sip;
3190         int split, subcore_size, active;
3191         int sub;
3192         bool thr0_done;
3193         unsigned long cmd_bit, stat_bit;
3194         int pcpu, thr;
3195         int target_threads;
3196         int controlled_threads;
3197         int trap;
3198         bool is_power8;
3199
3200         /*
3201          * Remove from the list any threads that have a signal pending
3202          * or need a VPA update done
3203          */
3204         prepare_threads(vc);
3205
3206         /* if the runner is no longer runnable, let the caller pick a new one */
3207         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3208                 return;
3209
3210         /*
3211          * Initialize *vc.
3212          */
3213         init_vcore_to_run(vc);
3214         vc->preempt_tb = TB_NIL;
3215
3216         /*
3217          * Number of threads that we will be controlling: the same as
3218          * the number of threads per subcore, except on POWER9,
3219          * where it's 1 because the threads are (mostly) independent.
3220          */
3221         controlled_threads = threads_per_vcore(vc->kvm);
3222
3223         /*
3224          * Make sure we are running on primary threads, and that secondary
3225          * threads are offline.  Also check if the number of threads in this
3226          * guest are greater than the current system threads per guest.
3227          * On POWER9, we need to be not in independent-threads mode if
3228          * this is a HPT guest on a radix host machine where the
3229          * CPU threads may not be in different MMU modes.
3230          */
3231         if ((controlled_threads > 1) &&
3232             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
3233                 for_each_runnable_thread(i, vcpu, vc) {
3234                         vcpu->arch.ret = -EBUSY;
3235                         kvmppc_remove_runnable(vc, vcpu);
3236                         wake_up(&vcpu->arch.cpu_run);
3237                 }
3238                 goto out;
3239         }
3240
3241         /*
3242          * See if we could run any other vcores on the physical core
3243          * along with this one.
3244          */
3245         init_core_info(&core_info, vc);
3246         pcpu = smp_processor_id();
3247         target_threads = controlled_threads;
3248         if (target_smt_mode && target_smt_mode < target_threads)
3249                 target_threads = target_smt_mode;
3250         if (vc->num_threads < target_threads)
3251                 collect_piggybacks(&core_info, target_threads);
3252
3253         /*
3254          * On radix, arrange for TLB flushing if necessary.
3255          * This has to be done before disabling interrupts since
3256          * it uses smp_call_function().
3257          */
3258         pcpu = smp_processor_id();
3259         if (kvm_is_radix(vc->kvm)) {
3260                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3261                         for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3262                                 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3263         }
3264
3265         /*
3266          * Hard-disable interrupts, and check resched flag and signals.
3267          * If we need to reschedule or deliver a signal, clean up
3268          * and return without going into the guest(s).
3269          * If the mmu_ready flag has been cleared, don't go into the
3270          * guest because that means a HPT resize operation is in progress.
3271          */
3272         local_irq_disable();
3273         hard_irq_disable();
3274         if (lazy_irq_pending() || need_resched() ||
3275             recheck_signals_and_mmu(&core_info)) {
3276                 local_irq_enable();
3277                 vc->vcore_state = VCORE_INACTIVE;
3278                 /* Unlock all except the primary vcore */
3279                 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3280                         pvc = core_info.vc[sub];
3281                         /* Put back on to the preempted vcores list */
3282                         kvmppc_vcore_preempt(pvc);
3283                         spin_unlock(&pvc->lock);
3284                 }
3285                 for (i = 0; i < controlled_threads; ++i)
3286                         kvmppc_release_hwthread(pcpu + i);
3287                 return;
3288         }
3289
3290         kvmppc_clear_host_core(pcpu);
3291
3292         /* Decide on micro-threading (split-core) mode */
3293         subcore_size = threads_per_subcore;
3294         cmd_bit = stat_bit = 0;
3295         split = core_info.n_subcores;
3296         sip = NULL;
3297         is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3298                 && !cpu_has_feature(CPU_FTR_ARCH_300);
3299
3300         if (split > 1) {
3301                 sip = &split_info;
3302                 memset(&split_info, 0, sizeof(split_info));
3303                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3304                         split_info.vc[sub] = core_info.vc[sub];
3305
3306                 if (is_power8) {
3307                         if (split == 2 && (dynamic_mt_modes & 2)) {
3308                                 cmd_bit = HID0_POWER8_1TO2LPAR;
3309                                 stat_bit = HID0_POWER8_2LPARMODE;
3310                         } else {
3311                                 split = 4;
3312                                 cmd_bit = HID0_POWER8_1TO4LPAR;
3313                                 stat_bit = HID0_POWER8_4LPARMODE;
3314                         }
3315                         subcore_size = MAX_SMT_THREADS / split;
3316                         split_info.rpr = mfspr(SPRN_RPR);
3317                         split_info.pmmar = mfspr(SPRN_PMMAR);
3318                         split_info.ldbar = mfspr(SPRN_LDBAR);
3319                         split_info.subcore_size = subcore_size;
3320                 } else {
3321                         split_info.subcore_size = 1;
3322                 }
3323
3324                 /* order writes to split_info before kvm_split_mode pointer */
3325                 smp_wmb();
3326         }
3327
3328         for (thr = 0; thr < controlled_threads; ++thr) {
3329                 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3330
3331                 paca->kvm_hstate.napping = 0;
3332                 paca->kvm_hstate.kvm_split_mode = sip;
3333         }
3334
3335         /* Initiate micro-threading (split-core) on POWER8 if required */
3336         if (cmd_bit) {
3337                 unsigned long hid0 = mfspr(SPRN_HID0);
3338
3339                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3340                 mb();
3341                 mtspr(SPRN_HID0, hid0);
3342                 isync();
3343                 for (;;) {
3344                         hid0 = mfspr(SPRN_HID0);
3345                         if (hid0 & stat_bit)
3346                                 break;
3347                         cpu_relax();
3348                 }
3349         }
3350
3351         /*
3352          * On POWER8, set RWMR register.
3353          * Since it only affects PURR and SPURR, it doesn't affect
3354          * the host, so we don't save/restore the host value.
3355          */
3356         if (is_power8) {
3357                 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3358                 int n_online = atomic_read(&vc->online_count);
3359
3360                 /*
3361                  * Use the 8-thread value if we're doing split-core
3362                  * or if the vcore's online count looks bogus.
3363                  */
3364                 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3365                     n_online >= 1 && n_online <= MAX_SMT_THREADS)
3366                         rwmr_val = p8_rwmr_values[n_online];
3367                 mtspr(SPRN_RWMR, rwmr_val);
3368         }
3369
3370         /* Start all the threads */
3371         active = 0;
3372         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3373                 thr = is_power8 ? subcore_thread_map[sub] : sub;
3374                 thr0_done = false;
3375                 active |= 1 << thr;
3376                 pvc = core_info.vc[sub];
3377                 pvc->pcpu = pcpu + thr;
3378                 for_each_runnable_thread(i, vcpu, pvc) {
3379                         kvmppc_start_thread(vcpu, pvc);
3380                         kvmppc_create_dtl_entry(vcpu, pvc);
3381                         trace_kvm_guest_enter(vcpu);
3382                         if (!vcpu->arch.ptid)
3383                                 thr0_done = true;
3384                         active |= 1 << (thr + vcpu->arch.ptid);
3385                 }
3386                 /*
3387                  * We need to start the first thread of each subcore
3388                  * even if it doesn't have a vcpu.
3389                  */
3390                 if (!thr0_done)
3391                         kvmppc_start_thread(NULL, pvc);
3392         }
3393
3394         /*
3395          * Ensure that split_info.do_nap is set after setting
3396          * the vcore pointer in the PACA of the secondaries.
3397          */
3398         smp_mb();
3399
3400         /*
3401          * When doing micro-threading, poke the inactive threads as well.
3402          * This gets them to the nap instruction after kvm_do_nap,
3403          * which reduces the time taken to unsplit later.
3404          */
3405         if (cmd_bit) {
3406                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
3407                 for (thr = 1; thr < threads_per_subcore; ++thr)
3408                         if (!(active & (1 << thr)))
3409                                 kvmppc_ipi_thread(pcpu + thr);
3410         }
3411
3412         vc->vcore_state = VCORE_RUNNING;
3413         preempt_disable();
3414
3415         trace_kvmppc_run_core(vc, 0);
3416
3417         for (sub = 0; sub < core_info.n_subcores; ++sub)
3418                 spin_unlock(&core_info.vc[sub]->lock);
3419
3420         guest_enter_irqoff();
3421
3422         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3423
3424         this_cpu_disable_ftrace();
3425
3426         /*
3427          * Interrupts will be enabled once we get into the guest,
3428          * so tell lockdep that we're about to enable interrupts.
3429          */
3430         trace_hardirqs_on();
3431
3432         trap = __kvmppc_vcore_entry();
3433
3434         trace_hardirqs_off();
3435
3436         this_cpu_enable_ftrace();
3437
3438         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3439
3440         set_irq_happened(trap);
3441
3442         spin_lock(&vc->lock);
3443         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3444         vc->vcore_state = VCORE_EXITING;
3445
3446         /* wait for secondary threads to finish writing their state to memory */
3447         kvmppc_wait_for_nap(controlled_threads);
3448
3449         /* Return to whole-core mode if we split the core earlier */
3450         if (cmd_bit) {
3451                 unsigned long hid0 = mfspr(SPRN_HID0);
3452                 unsigned long loops = 0;
3453
3454                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3455                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3456                 mb();
3457                 mtspr(SPRN_HID0, hid0);
3458                 isync();
3459                 for (;;) {
3460                         hid0 = mfspr(SPRN_HID0);
3461                         if (!(hid0 & stat_bit))
3462                                 break;
3463                         cpu_relax();
3464                         ++loops;
3465                 }
3466                 split_info.do_nap = 0;
3467         }
3468
3469         kvmppc_set_host_core(pcpu);
3470
3471         guest_exit_irqoff();
3472
3473         local_irq_enable();
3474
3475         /* Let secondaries go back to the offline loop */
3476         for (i = 0; i < controlled_threads; ++i) {
3477                 kvmppc_release_hwthread(pcpu + i);
3478                 if (sip && sip->napped[i])
3479                         kvmppc_ipi_thread(pcpu + i);
3480                 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3481         }
3482
3483         spin_unlock(&vc->lock);
3484
3485         /* make sure updates to secondary vcpu structs are visible now */
3486         smp_mb();
3487
3488         preempt_enable();
3489
3490         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3491                 pvc = core_info.vc[sub];
3492                 post_guest_process(pvc, pvc == vc);
3493         }
3494
3495         spin_lock(&vc->lock);
3496
3497  out:
3498         vc->vcore_state = VCORE_INACTIVE;
3499         trace_kvmppc_run_core(vc, 1);
3500 }
3501
3502 static void switch_mmu_to_guest_radix(struct kvm *kvm, struct kvm_vcpu *vcpu, u64 lpcr)
3503 {
3504         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3505         struct kvm_nested_guest *nested = vcpu->arch.nested;
3506         u32 lpid;
3507
3508         lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
3509
3510         /*
3511          * All the isync()s are overkill but trivially follow the ISA
3512          * requirements. Some can likely be replaced with justification
3513          * comment for why they are not needed.
3514          */
3515         isync();
3516         mtspr(SPRN_LPID, lpid);
3517         isync();
3518         mtspr(SPRN_LPCR, lpcr);
3519         isync();
3520         mtspr(SPRN_PID, vcpu->arch.pid);
3521         isync();
3522
3523         /* TLBIEL must have LPIDR set, so set guest LPID before flushing. */
3524         kvmppc_check_need_tlb_flush(kvm, vc->pcpu, nested);
3525 }
3526
3527 static void switch_mmu_to_host_radix(struct kvm *kvm, u32 pid)
3528 {
3529         isync();
3530         mtspr(SPRN_PID, pid);
3531         isync();
3532         mtspr(SPRN_LPID, kvm->arch.host_lpid);
3533         isync();
3534         mtspr(SPRN_LPCR, kvm->arch.host_lpcr);
3535         isync();
3536 }
3537
3538 /*
3539  * Load up hypervisor-mode registers on P9.
3540  */
3541 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3542                                      unsigned long lpcr)
3543 {
3544         struct kvm *kvm = vcpu->kvm;
3545         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3546         s64 hdec;
3547         u64 tb, purr, spurr;
3548         int trap;
3549         unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3550         unsigned long host_ciabr = mfspr(SPRN_CIABR);
3551         unsigned long host_dawr0 = mfspr(SPRN_DAWR0);
3552         unsigned long host_dawrx0 = mfspr(SPRN_DAWRX0);
3553         unsigned long host_psscr = mfspr(SPRN_PSSCR);
3554         unsigned long host_pidr = mfspr(SPRN_PID);
3555         unsigned long host_dawr1 = 0;
3556         unsigned long host_dawrx1 = 0;
3557
3558         if (cpu_has_feature(CPU_FTR_DAWR1)) {
3559                 host_dawr1 = mfspr(SPRN_DAWR1);
3560                 host_dawrx1 = mfspr(SPRN_DAWRX1);
3561         }
3562
3563         hdec = time_limit - mftb();
3564         if (hdec < 0)
3565                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3566
3567         if (vc->tb_offset) {
3568                 u64 new_tb = mftb() + vc->tb_offset;
3569                 mtspr(SPRN_TBU40, new_tb);
3570                 tb = mftb();
3571                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3572                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3573                 vc->tb_offset_applied = vc->tb_offset;
3574         }
3575
3576         if (vc->pcr)
3577                 mtspr(SPRN_PCR, vc->pcr | PCR_MASK);
3578         mtspr(SPRN_DPDES, vc->dpdes);
3579         mtspr(SPRN_VTB, vc->vtb);
3580
3581         local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3582         local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3583         mtspr(SPRN_PURR, vcpu->arch.purr);
3584         mtspr(SPRN_SPURR, vcpu->arch.spurr);
3585
3586         if (dawr_enabled()) {
3587                 mtspr(SPRN_DAWR0, vcpu->arch.dawr0);
3588                 mtspr(SPRN_DAWRX0, vcpu->arch.dawrx0);
3589                 if (cpu_has_feature(CPU_FTR_DAWR1)) {
3590                         mtspr(SPRN_DAWR1, vcpu->arch.dawr1);
3591                         mtspr(SPRN_DAWRX1, vcpu->arch.dawrx1);
3592                 }
3593         }
3594         mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3595         mtspr(SPRN_IC, vcpu->arch.ic);
3596
3597         mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3598               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3599
3600         mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3601
3602         mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3603         mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3604         mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3605         mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3606
3607         mtspr(SPRN_AMOR, ~0UL);
3608
3609         switch_mmu_to_guest_radix(kvm, vcpu, lpcr);
3610
3611         /*
3612          * P9 suppresses the HDEC exception when LPCR[HDICE] = 0,
3613          * so set guest LPCR (with HDICE) before writing HDEC.
3614          */
3615         mtspr(SPRN_HDEC, hdec);
3616
3617         mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3618         mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3619
3620         trap = __kvmhv_vcpu_entry_p9(vcpu);
3621
3622         /* Advance host PURR/SPURR by the amount used by guest */
3623         purr = mfspr(SPRN_PURR);
3624         spurr = mfspr(SPRN_SPURR);
3625         mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3626               purr - vcpu->arch.purr);
3627         mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3628               spurr - vcpu->arch.spurr);
3629         vcpu->arch.purr = purr;
3630         vcpu->arch.spurr = spurr;
3631
3632         vcpu->arch.ic = mfspr(SPRN_IC);
3633         vcpu->arch.pid = mfspr(SPRN_PID);
3634         vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3635
3636         vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3637         vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3638         vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3639         vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3640
3641         /* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3642         mtspr(SPRN_PSSCR, host_psscr |
3643               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3644         mtspr(SPRN_HFSCR, host_hfscr);
3645         mtspr(SPRN_CIABR, host_ciabr);
3646         mtspr(SPRN_DAWR0, host_dawr0);
3647         mtspr(SPRN_DAWRX0, host_dawrx0);
3648         if (cpu_has_feature(CPU_FTR_DAWR1)) {
3649                 mtspr(SPRN_DAWR1, host_dawr1);
3650                 mtspr(SPRN_DAWRX1, host_dawrx1);
3651         }
3652
3653         /*
3654          * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3655          * case we interrupted the guest between a tlbie and a ptesync.
3656          */
3657         asm volatile("eieio; tlbsync; ptesync");
3658
3659         /*
3660          * cp_abort is required if the processor supports local copy-paste
3661          * to clear the copy buffer that was under control of the guest.
3662          */
3663         if (cpu_has_feature(CPU_FTR_ARCH_31))
3664                 asm volatile(PPC_CP_ABORT);
3665
3666         vc->dpdes = mfspr(SPRN_DPDES);
3667         vc->vtb = mfspr(SPRN_VTB);
3668         mtspr(SPRN_DPDES, 0);
3669         if (vc->pcr)
3670                 mtspr(SPRN_PCR, PCR_MASK);
3671
3672         if (vc->tb_offset_applied) {
3673                 u64 new_tb = mftb() - vc->tb_offset_applied;
3674                 mtspr(SPRN_TBU40, new_tb);
3675                 tb = mftb();
3676                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3677                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3678                 vc->tb_offset_applied = 0;
3679         }
3680
3681         mtspr(SPRN_HDEC, 0x7fffffff);
3682
3683         switch_mmu_to_host_radix(kvm, host_pidr);
3684
3685         return trap;
3686 }
3687
3688 static inline bool hcall_is_xics(unsigned long req)
3689 {
3690         return req == H_EOI || req == H_CPPR || req == H_IPI ||
3691                 req == H_IPOLL || req == H_XIRR || req == H_XIRR_X;
3692 }
3693
3694 /*
3695  * Virtual-mode guest entry for POWER9 and later when the host and
3696  * guest are both using the radix MMU.  The LPIDR has already been set.
3697  */
3698 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3699                          unsigned long lpcr)
3700 {
3701         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3702         unsigned long host_dscr = mfspr(SPRN_DSCR);
3703         unsigned long host_tidr = mfspr(SPRN_TIDR);
3704         unsigned long host_iamr = mfspr(SPRN_IAMR);
3705         unsigned long host_amr = mfspr(SPRN_AMR);
3706         unsigned long host_fscr = mfspr(SPRN_FSCR);
3707         s64 dec;
3708         u64 tb;
3709         int trap, save_pmu;
3710
3711         dec = mfspr(SPRN_DEC);
3712         tb = mftb();
3713         if (dec < 0)
3714                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3715         local_paca->kvm_hstate.dec_expires = dec + tb;
3716         if (local_paca->kvm_hstate.dec_expires < time_limit)
3717                 time_limit = local_paca->kvm_hstate.dec_expires;
3718
3719         vcpu->arch.ceded = 0;
3720
3721         kvmhv_save_host_pmu();          /* saves it to PACA kvm_hstate */
3722
3723         kvmppc_subcore_enter_guest();
3724
3725         vc->entry_exit_map = 1;
3726         vc->in_guest = 1;
3727
3728         if (vcpu->arch.vpa.pinned_addr) {
3729                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3730                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3731                 lp->yield_count = cpu_to_be32(yield_count);
3732                 vcpu->arch.vpa.dirty = 1;
3733         }
3734
3735         if (cpu_has_feature(CPU_FTR_TM) ||
3736             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3737                 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3738
3739         kvmhv_load_guest_pmu(vcpu);
3740
3741         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3742         load_fp_state(&vcpu->arch.fp);
3743 #ifdef CONFIG_ALTIVEC
3744         load_vr_state(&vcpu->arch.vr);
3745 #endif
3746         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3747
3748         mtspr(SPRN_DSCR, vcpu->arch.dscr);
3749         mtspr(SPRN_IAMR, vcpu->arch.iamr);
3750         mtspr(SPRN_PSPB, vcpu->arch.pspb);
3751         mtspr(SPRN_FSCR, vcpu->arch.fscr);
3752         mtspr(SPRN_TAR, vcpu->arch.tar);
3753         mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3754         mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3755         mtspr(SPRN_BESCR, vcpu->arch.bescr);
3756         mtspr(SPRN_WORT, vcpu->arch.wort);
3757         mtspr(SPRN_TIDR, vcpu->arch.tid);
3758         mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3759         mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3760         mtspr(SPRN_AMR, vcpu->arch.amr);
3761         mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3762
3763         if (!(vcpu->arch.ctrl & 1))
3764                 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3765
3766         /*
3767          * When setting DEC, we must always deal with irq_work_raise via NMI vs
3768          * setting DEC. The problem occurs right as we switch into guest mode
3769          * if a NMI hits and sets pending work and sets DEC, then that will
3770          * apply to the guest and not bring us back to the host.
3771          *
3772          * irq_work_raise could check a flag (or possibly LPCR[HDICE] for
3773          * example) and set HDEC to 1? That wouldn't solve the nested hv
3774          * case which needs to abort the hcall or zero the time limit.
3775          *
3776          * XXX: Another day's problem.
3777          */
3778         mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3779
3780         if (kvmhv_on_pseries()) {
3781                 /*
3782                  * We need to save and restore the guest visible part of the
3783                  * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3784                  * doesn't do this for us. Note only required if pseries since
3785                  * this is done in kvmhv_load_hv_regs_and_go() below otherwise.
3786                  */
3787                 unsigned long host_psscr;
3788                 /* call our hypervisor to load up HV regs and go */
3789                 struct hv_guest_state hvregs;
3790
3791                 host_psscr = mfspr(SPRN_PSSCR_PR);
3792                 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3793                 kvmhv_save_hv_regs(vcpu, &hvregs);
3794                 hvregs.lpcr = lpcr;
3795                 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3796                 hvregs.version = HV_GUEST_STATE_VERSION;
3797                 if (vcpu->arch.nested) {
3798                         hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3799                         hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3800                 } else {
3801                         hvregs.lpid = vcpu->kvm->arch.lpid;
3802                         hvregs.vcpu_token = vcpu->vcpu_id;
3803                 }
3804                 hvregs.hdec_expiry = time_limit;
3805                 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3806                                           __pa(&vcpu->arch.regs));
3807                 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3808                 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3809                 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3810                 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3811                 vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3812                 mtspr(SPRN_PSSCR_PR, host_psscr);
3813
3814                 /* H_CEDE has to be handled now, not later */
3815                 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3816                     kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3817                         kvmppc_cede(vcpu);
3818                         kvmppc_set_gpr(vcpu, 3, 0);
3819                         trap = 0;
3820                 }
3821         } else {
3822                 kvmppc_xive_push_vcpu(vcpu);
3823                 trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3824                 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3825                     !(vcpu->arch.shregs.msr & MSR_PR)) {
3826                         unsigned long req = kvmppc_get_gpr(vcpu, 3);
3827
3828                         /* H_CEDE has to be handled now, not later */
3829                         if (req == H_CEDE) {
3830                                 kvmppc_cede(vcpu);
3831                                 kvmppc_xive_rearm_escalation(vcpu); /* may un-cede */
3832                                 kvmppc_set_gpr(vcpu, 3, 0);
3833                                 trap = 0;
3834
3835                         /* XICS hcalls must be handled before xive is pulled */
3836                         } else if (hcall_is_xics(req)) {
3837                                 int ret;
3838
3839                                 ret = kvmppc_xive_xics_hcall(vcpu, req);
3840                                 if (ret != H_TOO_HARD) {
3841                                         kvmppc_set_gpr(vcpu, 3, ret);
3842                                         trap = 0;
3843                                 }
3844                         }
3845                 }
3846                 kvmppc_xive_pull_vcpu(vcpu);
3847         }
3848
3849         vcpu->arch.slb_max = 0;
3850         dec = mfspr(SPRN_DEC);
3851         if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
3852                 dec = (s32) dec;
3853         tb = mftb();
3854         vcpu->arch.dec_expires = dec + tb;
3855         vcpu->cpu = -1;
3856         vcpu->arch.thread_cpu = -1;
3857         /* Save guest CTRL register, set runlatch to 1 */
3858         vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3859         if (!(vcpu->arch.ctrl & 1))
3860                 mtspr(SPRN_CTRLT, vcpu->arch.ctrl | 1);
3861
3862         vcpu->arch.iamr = mfspr(SPRN_IAMR);
3863         vcpu->arch.pspb = mfspr(SPRN_PSPB);
3864         vcpu->arch.fscr = mfspr(SPRN_FSCR);
3865         vcpu->arch.tar = mfspr(SPRN_TAR);
3866         vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3867         vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3868         vcpu->arch.bescr = mfspr(SPRN_BESCR);
3869         vcpu->arch.wort = mfspr(SPRN_WORT);
3870         vcpu->arch.tid = mfspr(SPRN_TIDR);
3871         vcpu->arch.amr = mfspr(SPRN_AMR);
3872         vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3873         vcpu->arch.dscr = mfspr(SPRN_DSCR);
3874
3875         mtspr(SPRN_PSPB, 0);
3876         mtspr(SPRN_WORT, 0);
3877         mtspr(SPRN_UAMOR, 0);
3878         mtspr(SPRN_DSCR, host_dscr);
3879         mtspr(SPRN_TIDR, host_tidr);
3880         mtspr(SPRN_IAMR, host_iamr);
3881
3882         if (host_amr != vcpu->arch.amr)
3883                 mtspr(SPRN_AMR, host_amr);
3884
3885         if (host_fscr != vcpu->arch.fscr)
3886                 mtspr(SPRN_FSCR, host_fscr);
3887
3888         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3889         store_fp_state(&vcpu->arch.fp);
3890 #ifdef CONFIG_ALTIVEC
3891         store_vr_state(&vcpu->arch.vr);
3892 #endif
3893         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3894
3895         if (cpu_has_feature(CPU_FTR_TM) ||
3896             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3897                 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3898
3899         save_pmu = 1;
3900         if (vcpu->arch.vpa.pinned_addr) {
3901                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3902                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3903                 lp->yield_count = cpu_to_be32(yield_count);
3904                 vcpu->arch.vpa.dirty = 1;
3905                 save_pmu = lp->pmcregs_in_use;
3906         }
3907         /* Must save pmu if this guest is capable of running nested guests */
3908         save_pmu |= nesting_enabled(vcpu->kvm);
3909
3910         kvmhv_save_guest_pmu(vcpu, save_pmu);
3911
3912         vc->entry_exit_map = 0x101;
3913         vc->in_guest = 0;
3914
3915         mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3916         /* We may have raced with new irq work */
3917         if (test_irq_work_pending())
3918                 set_dec(1);
3919         mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3920
3921         kvmhv_load_host_pmu();
3922
3923         kvmppc_subcore_exit_guest();
3924
3925         return trap;
3926 }
3927
3928 /*
3929  * Wait for some other vcpu thread to execute us, and
3930  * wake us up when we need to handle something in the host.
3931  */
3932 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3933                                  struct kvm_vcpu *vcpu, int wait_state)
3934 {
3935         DEFINE_WAIT(wait);
3936
3937         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3938         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3939                 spin_unlock(&vc->lock);
3940                 schedule();
3941                 spin_lock(&vc->lock);
3942         }
3943         finish_wait(&vcpu->arch.cpu_run, &wait);
3944 }
3945
3946 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3947 {
3948         if (!halt_poll_ns_grow)
3949                 return;
3950
3951         vc->halt_poll_ns *= halt_poll_ns_grow;
3952         if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3953                 vc->halt_poll_ns = halt_poll_ns_grow_start;
3954 }
3955
3956 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3957 {
3958         if (halt_poll_ns_shrink == 0)
3959                 vc->halt_poll_ns = 0;
3960         else
3961                 vc->halt_poll_ns /= halt_poll_ns_shrink;
3962 }
3963
3964 #ifdef CONFIG_KVM_XICS
3965 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3966 {
3967         if (!xics_on_xive())
3968                 return false;
3969         return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3970                 vcpu->arch.xive_saved_state.cppr;
3971 }
3972 #else
3973 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3974 {
3975         return false;
3976 }
3977 #endif /* CONFIG_KVM_XICS */
3978
3979 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3980 {
3981         if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3982             kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3983                 return true;
3984
3985         return false;
3986 }
3987
3988 /*
3989  * Check to see if any of the runnable vcpus on the vcore have pending
3990  * exceptions or are no longer ceded
3991  */
3992 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3993 {
3994         struct kvm_vcpu *vcpu;
3995         int i;
3996
3997         for_each_runnable_thread(i, vcpu, vc) {
3998                 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3999                         return 1;
4000         }
4001
4002         return 0;
4003 }
4004
4005 /*
4006  * All the vcpus in this vcore are idle, so wait for a decrementer
4007  * or external interrupt to one of the vcpus.  vc->lock is held.
4008  */
4009 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
4010 {
4011         ktime_t cur, start_poll, start_wait;
4012         int do_sleep = 1;
4013         u64 block_ns;
4014
4015         /* Poll for pending exceptions and ceded state */
4016         cur = start_poll = ktime_get();
4017         if (vc->halt_poll_ns) {
4018                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
4019                 ++vc->runner->stat.halt_attempted_poll;
4020
4021                 vc->vcore_state = VCORE_POLLING;
4022                 spin_unlock(&vc->lock);
4023
4024                 do {
4025                         if (kvmppc_vcore_check_block(vc)) {
4026                                 do_sleep = 0;
4027                                 break;
4028                         }
4029                         cur = ktime_get();
4030                 } while (single_task_running() && ktime_before(cur, stop));
4031
4032                 spin_lock(&vc->lock);
4033                 vc->vcore_state = VCORE_INACTIVE;
4034
4035                 if (!do_sleep) {
4036                         ++vc->runner->stat.halt_successful_poll;
4037                         goto out;
4038                 }
4039         }
4040
4041         prepare_to_rcuwait(&vc->wait);
4042         set_current_state(TASK_INTERRUPTIBLE);
4043         if (kvmppc_vcore_check_block(vc)) {
4044                 finish_rcuwait(&vc->wait);
4045                 do_sleep = 0;
4046                 /* If we polled, count this as a successful poll */
4047                 if (vc->halt_poll_ns)
4048                         ++vc->runner->stat.halt_successful_poll;
4049                 goto out;
4050         }
4051
4052         start_wait = ktime_get();
4053
4054         vc->vcore_state = VCORE_SLEEPING;
4055         trace_kvmppc_vcore_blocked(vc, 0);
4056         spin_unlock(&vc->lock);
4057         schedule();
4058         finish_rcuwait(&vc->wait);
4059         spin_lock(&vc->lock);
4060         vc->vcore_state = VCORE_INACTIVE;
4061         trace_kvmppc_vcore_blocked(vc, 1);
4062         ++vc->runner->stat.halt_successful_wait;
4063
4064         cur = ktime_get();
4065
4066 out:
4067         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
4068
4069         /* Attribute wait time */
4070         if (do_sleep) {
4071                 vc->runner->stat.halt_wait_ns +=
4072                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
4073                 /* Attribute failed poll time */
4074                 if (vc->halt_poll_ns)
4075                         vc->runner->stat.halt_poll_fail_ns +=
4076                                 ktime_to_ns(start_wait) -
4077                                 ktime_to_ns(start_poll);
4078         } else {
4079                 /* Attribute successful poll time */
4080                 if (vc->halt_poll_ns)
4081                         vc->runner->stat.halt_poll_success_ns +=
4082                                 ktime_to_ns(cur) -
4083                                 ktime_to_ns(start_poll);
4084         }
4085
4086         /* Adjust poll time */
4087         if (halt_poll_ns) {
4088                 if (block_ns <= vc->halt_poll_ns)
4089                         ;
4090                 /* We slept and blocked for longer than the max halt time */
4091                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
4092                         shrink_halt_poll_ns(vc);
4093                 /* We slept and our poll time is too small */
4094                 else if (vc->halt_poll_ns < halt_poll_ns &&
4095                                 block_ns < halt_poll_ns)
4096                         grow_halt_poll_ns(vc);
4097                 if (vc->halt_poll_ns > halt_poll_ns)
4098                         vc->halt_poll_ns = halt_poll_ns;
4099         } else
4100                 vc->halt_poll_ns = 0;
4101
4102         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
4103 }
4104
4105 /*
4106  * This never fails for a radix guest, as none of the operations it does
4107  * for a radix guest can fail or have a way to report failure.
4108  * kvmhv_run_single_vcpu() relies on this fact.
4109  */
4110 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
4111 {
4112         int r = 0;
4113         struct kvm *kvm = vcpu->kvm;
4114
4115         mutex_lock(&kvm->arch.mmu_setup_lock);
4116         if (!kvm->arch.mmu_ready) {
4117                 if (!kvm_is_radix(kvm))
4118                         r = kvmppc_hv_setup_htab_rma(vcpu);
4119                 if (!r) {
4120                         if (cpu_has_feature(CPU_FTR_ARCH_300))
4121                                 kvmppc_setup_partition_table(kvm);
4122                         kvm->arch.mmu_ready = 1;
4123                 }
4124         }
4125         mutex_unlock(&kvm->arch.mmu_setup_lock);
4126         return r;
4127 }
4128
4129 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
4130 {
4131         struct kvm_run *run = vcpu->run;
4132         int n_ceded, i, r;
4133         struct kvmppc_vcore *vc;
4134         struct kvm_vcpu *v;
4135
4136         trace_kvmppc_run_vcpu_enter(vcpu);
4137
4138         run->exit_reason = 0;
4139         vcpu->arch.ret = RESUME_GUEST;
4140         vcpu->arch.trap = 0;
4141         kvmppc_update_vpas(vcpu);
4142
4143         /*
4144          * Synchronize with other threads in this virtual core
4145          */
4146         vc = vcpu->arch.vcore;
4147         spin_lock(&vc->lock);
4148         vcpu->arch.ceded = 0;
4149         vcpu->arch.run_task = current;
4150         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4151         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4152         vcpu->arch.busy_preempt = TB_NIL;
4153         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4154         ++vc->n_runnable;
4155
4156         /*
4157          * This happens the first time this is called for a vcpu.
4158          * If the vcore is already running, we may be able to start
4159          * this thread straight away and have it join in.
4160          */
4161         if (!signal_pending(current)) {
4162                 if ((vc->vcore_state == VCORE_PIGGYBACK ||
4163                      vc->vcore_state == VCORE_RUNNING) &&
4164                            !VCORE_IS_EXITING(vc)) {
4165                         kvmppc_create_dtl_entry(vcpu, vc);
4166                         kvmppc_start_thread(vcpu, vc);
4167                         trace_kvm_guest_enter(vcpu);
4168                 } else if (vc->vcore_state == VCORE_SLEEPING) {
4169                         rcuwait_wake_up(&vc->wait);
4170                 }
4171
4172         }
4173
4174         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4175                !signal_pending(current)) {
4176                 /* See if the MMU is ready to go */
4177                 if (!vcpu->kvm->arch.mmu_ready) {
4178                         spin_unlock(&vc->lock);
4179                         r = kvmhv_setup_mmu(vcpu);
4180                         spin_lock(&vc->lock);
4181                         if (r) {
4182                                 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4183                                 run->fail_entry.
4184                                         hardware_entry_failure_reason = 0;
4185                                 vcpu->arch.ret = r;
4186                                 break;
4187                         }
4188                 }
4189
4190                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4191                         kvmppc_vcore_end_preempt(vc);
4192
4193                 if (vc->vcore_state != VCORE_INACTIVE) {
4194                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4195                         continue;
4196                 }
4197                 for_each_runnable_thread(i, v, vc) {
4198                         kvmppc_core_prepare_to_enter(v);
4199                         if (signal_pending(v->arch.run_task)) {
4200                                 kvmppc_remove_runnable(vc, v);
4201                                 v->stat.signal_exits++;
4202                                 v->run->exit_reason = KVM_EXIT_INTR;
4203                                 v->arch.ret = -EINTR;
4204                                 wake_up(&v->arch.cpu_run);
4205                         }
4206                 }
4207                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4208                         break;
4209                 n_ceded = 0;
4210                 for_each_runnable_thread(i, v, vc) {
4211                         if (!kvmppc_vcpu_woken(v))
4212                                 n_ceded += v->arch.ceded;
4213                         else
4214                                 v->arch.ceded = 0;
4215                 }
4216                 vc->runner = vcpu;
4217                 if (n_ceded == vc->n_runnable) {
4218                         kvmppc_vcore_blocked(vc);
4219                 } else if (need_resched()) {
4220                         kvmppc_vcore_preempt(vc);
4221                         /* Let something else run */
4222                         cond_resched_lock(&vc->lock);
4223                         if (vc->vcore_state == VCORE_PREEMPT)
4224                                 kvmppc_vcore_end_preempt(vc);
4225                 } else {
4226                         kvmppc_run_core(vc);
4227                 }
4228                 vc->runner = NULL;
4229         }
4230
4231         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4232                (vc->vcore_state == VCORE_RUNNING ||
4233                 vc->vcore_state == VCORE_EXITING ||
4234                 vc->vcore_state == VCORE_PIGGYBACK))
4235                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4236
4237         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4238                 kvmppc_vcore_end_preempt(vc);
4239
4240         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4241                 kvmppc_remove_runnable(vc, vcpu);
4242                 vcpu->stat.signal_exits++;
4243                 run->exit_reason = KVM_EXIT_INTR;
4244                 vcpu->arch.ret = -EINTR;
4245         }
4246
4247         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4248                 /* Wake up some vcpu to run the core */
4249                 i = -1;
4250                 v = next_runnable_thread(vc, &i);
4251                 wake_up(&v->arch.cpu_run);
4252         }
4253
4254         trace_kvmppc_run_vcpu_exit(vcpu);
4255         spin_unlock(&vc->lock);
4256         return vcpu->arch.ret;
4257 }
4258
4259 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4260                           unsigned long lpcr)
4261 {
4262         struct kvm_run *run = vcpu->run;
4263         int trap, r, pcpu;
4264         int srcu_idx;
4265         struct kvmppc_vcore *vc;
4266         struct kvm *kvm = vcpu->kvm;
4267         struct kvm_nested_guest *nested = vcpu->arch.nested;
4268
4269         trace_kvmppc_run_vcpu_enter(vcpu);
4270
4271         run->exit_reason = 0;
4272         vcpu->arch.ret = RESUME_GUEST;
4273         vcpu->arch.trap = 0;
4274
4275         vc = vcpu->arch.vcore;
4276         vcpu->arch.ceded = 0;
4277         vcpu->arch.run_task = current;
4278         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4279         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4280         vcpu->arch.busy_preempt = TB_NIL;
4281         vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4282         vc->runnable_threads[0] = vcpu;
4283         vc->n_runnable = 1;
4284         vc->runner = vcpu;
4285
4286         /* See if the MMU is ready to go */
4287         if (!kvm->arch.mmu_ready)
4288                 kvmhv_setup_mmu(vcpu);
4289
4290         if (need_resched())
4291                 cond_resched();
4292
4293         kvmppc_update_vpas(vcpu);
4294
4295         init_vcore_to_run(vc);
4296         vc->preempt_tb = TB_NIL;
4297
4298         preempt_disable();
4299         pcpu = smp_processor_id();
4300         vc->pcpu = pcpu;
4301         kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4302
4303         local_irq_disable();
4304         hard_irq_disable();
4305         if (signal_pending(current))
4306                 goto sigpend;
4307         if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4308                 goto out;
4309
4310         if (!nested) {
4311                 kvmppc_core_prepare_to_enter(vcpu);
4312                 if (vcpu->arch.doorbell_request) {
4313                         vc->dpdes = 1;
4314                         smp_wmb();
4315                         vcpu->arch.doorbell_request = 0;
4316                 }
4317                 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4318                              &vcpu->arch.pending_exceptions))
4319                         lpcr |= LPCR_MER;
4320         } else if (vcpu->arch.pending_exceptions ||
4321                    vcpu->arch.doorbell_request ||
4322                    xive_interrupt_pending(vcpu)) {
4323                 vcpu->arch.ret = RESUME_HOST;
4324                 goto out;
4325         }
4326
4327         kvmppc_clear_host_core(pcpu);
4328
4329         local_paca->kvm_hstate.napping = 0;
4330         local_paca->kvm_hstate.kvm_split_mode = NULL;
4331         kvmppc_start_thread(vcpu, vc);
4332         kvmppc_create_dtl_entry(vcpu, vc);
4333         trace_kvm_guest_enter(vcpu);
4334
4335         vc->vcore_state = VCORE_RUNNING;
4336         trace_kvmppc_run_core(vc, 0);
4337
4338         guest_enter_irqoff();
4339
4340         srcu_idx = srcu_read_lock(&kvm->srcu);
4341
4342         this_cpu_disable_ftrace();
4343
4344         /* Tell lockdep that we're about to enable interrupts */
4345         trace_hardirqs_on();
4346
4347         trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4348         vcpu->arch.trap = trap;
4349
4350         trace_hardirqs_off();
4351
4352         this_cpu_enable_ftrace();
4353
4354         srcu_read_unlock(&kvm->srcu, srcu_idx);
4355
4356         set_irq_happened(trap);
4357
4358         kvmppc_set_host_core(pcpu);
4359
4360         guest_exit_irqoff();
4361
4362         local_irq_enable();
4363
4364         cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4365
4366         preempt_enable();
4367
4368         /*
4369          * cancel pending decrementer exception if DEC is now positive, or if
4370          * entering a nested guest in which case the decrementer is now owned
4371          * by L2 and the L1 decrementer is provided in hdec_expires
4372          */
4373         if (kvmppc_core_pending_dec(vcpu) &&
4374                         ((get_tb() < vcpu->arch.dec_expires) ||
4375                          (trap == BOOK3S_INTERRUPT_SYSCALL &&
4376                           kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4377                 kvmppc_core_dequeue_dec(vcpu);
4378
4379         trace_kvm_guest_exit(vcpu);
4380         r = RESUME_GUEST;
4381         if (trap) {
4382                 if (!nested)
4383                         r = kvmppc_handle_exit_hv(vcpu, current);
4384                 else
4385                         r = kvmppc_handle_nested_exit(vcpu);
4386         }
4387         vcpu->arch.ret = r;
4388
4389         if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4390             !kvmppc_vcpu_woken(vcpu)) {
4391                 kvmppc_set_timer(vcpu);
4392                 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4393                         if (signal_pending(current)) {
4394                                 vcpu->stat.signal_exits++;
4395                                 run->exit_reason = KVM_EXIT_INTR;
4396                                 vcpu->arch.ret = -EINTR;
4397                                 break;
4398                         }
4399                         spin_lock(&vc->lock);
4400                         kvmppc_vcore_blocked(vc);
4401                         spin_unlock(&vc->lock);
4402                 }
4403         }
4404         vcpu->arch.ceded = 0;
4405
4406         vc->vcore_state = VCORE_INACTIVE;
4407         trace_kvmppc_run_core(vc, 1);
4408
4409  done:
4410         kvmppc_remove_runnable(vc, vcpu);
4411         trace_kvmppc_run_vcpu_exit(vcpu);
4412
4413         return vcpu->arch.ret;
4414
4415  sigpend:
4416         vcpu->stat.signal_exits++;
4417         run->exit_reason = KVM_EXIT_INTR;
4418         vcpu->arch.ret = -EINTR;
4419  out:
4420         local_irq_enable();
4421         preempt_enable();
4422         goto done;
4423 }
4424
4425 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
4426 {
4427         struct kvm_run *run = vcpu->run;
4428         int r;
4429         int srcu_idx;
4430         unsigned long ebb_regs[3] = {}; /* shut up GCC */
4431         unsigned long user_tar = 0;
4432         unsigned int user_vrsave;
4433         struct kvm *kvm;
4434
4435         if (!vcpu->arch.sane) {
4436                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4437                 return -EINVAL;
4438         }
4439
4440         /*
4441          * Don't allow entry with a suspended transaction, because
4442          * the guest entry/exit code will lose it.
4443          * If the guest has TM enabled, save away their TM-related SPRs
4444          * (they will get restored by the TM unavailable interrupt).
4445          */
4446 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4447         if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4448             (current->thread.regs->msr & MSR_TM)) {
4449                 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4450                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4451                         run->fail_entry.hardware_entry_failure_reason = 0;
4452                         return -EINVAL;
4453                 }
4454                 /* Enable TM so we can read the TM SPRs */
4455                 mtmsr(mfmsr() | MSR_TM);
4456                 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4457                 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4458                 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4459                 current->thread.regs->msr &= ~MSR_TM;
4460         }
4461 #endif
4462
4463         /*
4464          * Force online to 1 for the sake of old userspace which doesn't
4465          * set it.
4466          */
4467         if (!vcpu->arch.online) {
4468                 atomic_inc(&vcpu->arch.vcore->online_count);
4469                 vcpu->arch.online = 1;
4470         }
4471
4472         kvmppc_core_prepare_to_enter(vcpu);
4473
4474         /* No need to go into the guest when all we'll do is come back out */
4475         if (signal_pending(current)) {
4476                 run->exit_reason = KVM_EXIT_INTR;
4477                 return -EINTR;
4478         }
4479
4480         kvm = vcpu->kvm;
4481         atomic_inc(&kvm->arch.vcpus_running);
4482         /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4483         smp_mb();
4484
4485         flush_all_to_thread(current);
4486
4487         /* Save userspace EBB and other register values */
4488         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4489                 ebb_regs[0] = mfspr(SPRN_EBBHR);
4490                 ebb_regs[1] = mfspr(SPRN_EBBRR);
4491                 ebb_regs[2] = mfspr(SPRN_BESCR);
4492                 user_tar = mfspr(SPRN_TAR);
4493         }
4494         user_vrsave = mfspr(SPRN_VRSAVE);
4495
4496         vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4497         vcpu->arch.pgdir = kvm->mm->pgd;
4498         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4499
4500         do {
4501                 /*
4502                  * The TLB prefetch bug fixup is only in the kvmppc_run_vcpu
4503                  * path, which also handles hash and dependent threads mode.
4504                  */
4505                 if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4506                     !cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
4507                         r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
4508                                                   vcpu->arch.vcore->lpcr);
4509                 else
4510                         r = kvmppc_run_vcpu(vcpu);
4511
4512                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL) {
4513                         if (WARN_ON_ONCE(vcpu->arch.shregs.msr & MSR_PR)) {
4514                                 /*
4515                                  * These should have been caught reflected
4516                                  * into the guest by now. Final sanity check:
4517                                  * don't allow userspace to execute hcalls in
4518                                  * the hypervisor.
4519                                  */
4520                                 r = RESUME_GUEST;
4521                                 continue;
4522                         }
4523                         trace_kvm_hcall_enter(vcpu);
4524                         r = kvmppc_pseries_do_hcall(vcpu);
4525                         trace_kvm_hcall_exit(vcpu, r);
4526                         kvmppc_core_prepare_to_enter(vcpu);
4527                 } else if (r == RESUME_PAGE_FAULT) {
4528                         srcu_idx = srcu_read_lock(&kvm->srcu);
4529                         r = kvmppc_book3s_hv_page_fault(vcpu,
4530                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4531                         srcu_read_unlock(&kvm->srcu, srcu_idx);
4532                 } else if (r == RESUME_PASSTHROUGH) {
4533                         if (WARN_ON(xics_on_xive()))
4534                                 r = H_SUCCESS;
4535                         else
4536                                 r = kvmppc_xics_rm_complete(vcpu, 0);
4537                 }
4538         } while (is_kvmppc_resume_guest(r));
4539
4540         /* Restore userspace EBB and other register values */
4541         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4542                 mtspr(SPRN_EBBHR, ebb_regs[0]);
4543                 mtspr(SPRN_EBBRR, ebb_regs[1]);
4544                 mtspr(SPRN_BESCR, ebb_regs[2]);
4545                 mtspr(SPRN_TAR, user_tar);
4546         }
4547         mtspr(SPRN_VRSAVE, user_vrsave);
4548
4549         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4550         atomic_dec(&kvm->arch.vcpus_running);
4551         return r;
4552 }
4553
4554 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4555                                      int shift, int sllp)
4556 {
4557         (*sps)->page_shift = shift;
4558         (*sps)->slb_enc = sllp;
4559         (*sps)->enc[0].page_shift = shift;
4560         (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4561         /*
4562          * Add 16MB MPSS support (may get filtered out by userspace)
4563          */
4564         if (shift != 24) {
4565                 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4566                 if (penc != -1) {
4567                         (*sps)->enc[1].page_shift = 24;
4568                         (*sps)->enc[1].pte_enc = penc;
4569                 }
4570         }
4571         (*sps)++;
4572 }
4573
4574 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4575                                          struct kvm_ppc_smmu_info *info)
4576 {
4577         struct kvm_ppc_one_seg_page_size *sps;
4578
4579         /*
4580          * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4581          * POWER7 doesn't support keys for instruction accesses,
4582          * POWER8 and POWER9 do.
4583          */
4584         info->data_keys = 32;
4585         info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4586
4587         /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4588         info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4589         info->slb_size = 32;
4590
4591         /* We only support these sizes for now, and no muti-size segments */
4592         sps = &info->sps[0];
4593         kvmppc_add_seg_page_size(&sps, 12, 0);
4594         kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4595         kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4596
4597         /* If running as a nested hypervisor, we don't support HPT guests */
4598         if (kvmhv_on_pseries())
4599                 info->flags |= KVM_PPC_NO_HASH;
4600
4601         return 0;
4602 }
4603
4604 /*
4605  * Get (and clear) the dirty memory log for a memory slot.
4606  */
4607 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4608                                          struct kvm_dirty_log *log)
4609 {
4610         struct kvm_memslots *slots;
4611         struct kvm_memory_slot *memslot;
4612         int i, r;
4613         unsigned long n;
4614         unsigned long *buf, *p;
4615         struct kvm_vcpu *vcpu;
4616
4617         mutex_lock(&kvm->slots_lock);
4618
4619         r = -EINVAL;
4620         if (log->slot >= KVM_USER_MEM_SLOTS)
4621                 goto out;
4622
4623         slots = kvm_memslots(kvm);
4624         memslot = id_to_memslot(slots, log->slot);
4625         r = -ENOENT;
4626         if (!memslot || !memslot->dirty_bitmap)
4627                 goto out;
4628
4629         /*
4630          * Use second half of bitmap area because both HPT and radix
4631          * accumulate bits in the first half.
4632          */
4633         n = kvm_dirty_bitmap_bytes(memslot);
4634         buf = memslot->dirty_bitmap + n / sizeof(long);
4635         memset(buf, 0, n);
4636
4637         if (kvm_is_radix(kvm))
4638                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4639         else
4640                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4641         if (r)
4642                 goto out;
4643
4644         /*
4645          * We accumulate dirty bits in the first half of the
4646          * memslot's dirty_bitmap area, for when pages are paged
4647          * out or modified by the host directly.  Pick up these
4648          * bits and add them to the map.
4649          */
4650         p = memslot->dirty_bitmap;
4651         for (i = 0; i < n / sizeof(long); ++i)
4652                 buf[i] |= xchg(&p[i], 0);
4653
4654         /* Harvest dirty bits from VPA and DTL updates */
4655         /* Note: we never modify the SLB shadow buffer areas */
4656         kvm_for_each_vcpu(i, vcpu, kvm) {
4657                 spin_lock(&vcpu->arch.vpa_update_lock);
4658                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4659                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4660                 spin_unlock(&vcpu->arch.vpa_update_lock);
4661         }
4662
4663         r = -EFAULT;
4664         if (copy_to_user(log->dirty_bitmap, buf, n))
4665                 goto out;
4666
4667         r = 0;
4668 out:
4669         mutex_unlock(&kvm->slots_lock);
4670         return r;
4671 }
4672
4673 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
4674 {
4675         vfree(slot->arch.rmap);
4676         slot->arch.rmap = NULL;
4677 }
4678
4679 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4680                                         struct kvm_memory_slot *slot,
4681                                         const struct kvm_userspace_memory_region *mem,
4682                                         enum kvm_mr_change change)
4683 {
4684         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4685
4686         if (change == KVM_MR_CREATE) {
4687                 slot->arch.rmap = vzalloc(array_size(npages,
4688                                           sizeof(*slot->arch.rmap)));
4689                 if (!slot->arch.rmap)
4690                         return -ENOMEM;
4691         }
4692
4693         return 0;
4694 }
4695
4696 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4697                                 const struct kvm_userspace_memory_region *mem,
4698                                 const struct kvm_memory_slot *old,
4699                                 const struct kvm_memory_slot *new,
4700                                 enum kvm_mr_change change)
4701 {
4702         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4703
4704         /*
4705          * If we are making a new memslot, it might make
4706          * some address that was previously cached as emulated
4707          * MMIO be no longer emulated MMIO, so invalidate
4708          * all the caches of emulated MMIO translations.
4709          */
4710         if (npages)
4711                 atomic64_inc(&kvm->arch.mmio_update);
4712
4713         /*
4714          * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4715          * have already called kvm_arch_flush_shadow_memslot() to
4716          * flush shadow mappings.  For KVM_MR_CREATE we have no
4717          * previous mappings.  So the only case to handle is
4718          * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4719          * has been changed.
4720          * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4721          * to get rid of any THP PTEs in the partition-scoped page tables
4722          * so we can track dirtiness at the page level; we flush when
4723          * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4724          * using THP PTEs.
4725          */
4726         if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4727             ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4728                 kvmppc_radix_flush_memslot(kvm, old);
4729         /*
4730          * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4731          */
4732         if (!kvm->arch.secure_guest)
4733                 return;
4734
4735         switch (change) {
4736         case KVM_MR_CREATE:
4737                 /*
4738                  * @TODO kvmppc_uvmem_memslot_create() can fail and
4739                  * return error. Fix this.
4740                  */
4741                 kvmppc_uvmem_memslot_create(kvm, new);
4742                 break;
4743         case KVM_MR_DELETE:
4744                 kvmppc_uvmem_memslot_delete(kvm, old);
4745                 break;
4746         default:
4747                 /* TODO: Handle KVM_MR_MOVE */
4748                 break;
4749         }
4750 }
4751
4752 /*
4753  * Update LPCR values in kvm->arch and in vcores.
4754  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4755  * of kvm->arch.lpcr update).
4756  */
4757 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4758 {
4759         long int i;
4760         u32 cores_done = 0;
4761
4762         if ((kvm->arch.lpcr & mask) == lpcr)
4763                 return;
4764
4765         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4766
4767         for (i = 0; i < KVM_MAX_VCORES; ++i) {
4768                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4769                 if (!vc)
4770                         continue;
4771
4772                 spin_lock(&vc->lock);
4773                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4774                 verify_lpcr(kvm, vc->lpcr);
4775                 spin_unlock(&vc->lock);
4776                 if (++cores_done >= kvm->arch.online_vcores)
4777                         break;
4778         }
4779 }
4780
4781 void kvmppc_setup_partition_table(struct kvm *kvm)
4782 {
4783         unsigned long dw0, dw1;
4784
4785         if (!kvm_is_radix(kvm)) {
4786                 /* PS field - page size for VRMA */
4787                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4788                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4789                 /* HTABSIZE and HTABORG fields */
4790                 dw0 |= kvm->arch.sdr1;
4791
4792                 /* Second dword as set by userspace */
4793                 dw1 = kvm->arch.process_table;
4794         } else {
4795                 dw0 = PATB_HR | radix__get_tree_size() |
4796                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4797                 dw1 = PATB_GR | kvm->arch.process_table;
4798         }
4799         kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4800 }
4801
4802 /*
4803  * Set up HPT (hashed page table) and RMA (real-mode area).
4804  * Must be called with kvm->arch.mmu_setup_lock held.
4805  */
4806 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4807 {
4808         int err = 0;
4809         struct kvm *kvm = vcpu->kvm;
4810         unsigned long hva;
4811         struct kvm_memory_slot *memslot;
4812         struct vm_area_struct *vma;
4813         unsigned long lpcr = 0, senc;
4814         unsigned long psize, porder;
4815         int srcu_idx;
4816
4817         /* Allocate hashed page table (if not done already) and reset it */
4818         if (!kvm->arch.hpt.virt) {
4819                 int order = KVM_DEFAULT_HPT_ORDER;
4820                 struct kvm_hpt_info info;
4821
4822                 err = kvmppc_allocate_hpt(&info, order);
4823                 /* If we get here, it means userspace didn't specify a
4824                  * size explicitly.  So, try successively smaller
4825                  * sizes if the default failed. */
4826                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4827                         err  = kvmppc_allocate_hpt(&info, order);
4828
4829                 if (err < 0) {
4830                         pr_err("KVM: Couldn't alloc HPT\n");
4831                         goto out;
4832                 }
4833
4834                 kvmppc_set_hpt(kvm, &info);
4835         }
4836
4837         /* Look up the memslot for guest physical address 0 */
4838         srcu_idx = srcu_read_lock(&kvm->srcu);
4839         memslot = gfn_to_memslot(kvm, 0);
4840
4841         /* We must have some memory at 0 by now */
4842         err = -EINVAL;
4843         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4844                 goto out_srcu;
4845
4846         /* Look up the VMA for the start of this memory slot */
4847         hva = memslot->userspace_addr;
4848         mmap_read_lock(kvm->mm);
4849         vma = find_vma(kvm->mm, hva);
4850         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4851                 goto up_out;
4852
4853         psize = vma_kernel_pagesize(vma);
4854
4855         mmap_read_unlock(kvm->mm);
4856
4857         /* We can handle 4k, 64k or 16M pages in the VRMA */
4858         if (psize >= 0x1000000)
4859                 psize = 0x1000000;
4860         else if (psize >= 0x10000)
4861                 psize = 0x10000;
4862         else
4863                 psize = 0x1000;
4864         porder = __ilog2(psize);
4865
4866         senc = slb_pgsize_encoding(psize);
4867         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4868                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4869         /* Create HPTEs in the hash page table for the VRMA */
4870         kvmppc_map_vrma(vcpu, memslot, porder);
4871
4872         /* Update VRMASD field in the LPCR */
4873         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4874                 /* the -4 is to account for senc values starting at 0x10 */
4875                 lpcr = senc << (LPCR_VRMASD_SH - 4);
4876                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4877         }
4878
4879         /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4880         smp_wmb();
4881         err = 0;
4882  out_srcu:
4883         srcu_read_unlock(&kvm->srcu, srcu_idx);
4884  out:
4885         return err;
4886
4887  up_out:
4888         mmap_read_unlock(kvm->mm);
4889         goto out_srcu;
4890 }
4891
4892 /*
4893  * Must be called with kvm->arch.mmu_setup_lock held and
4894  * mmu_ready = 0 and no vcpus running.
4895  */
4896 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4897 {
4898         if (nesting_enabled(kvm))
4899                 kvmhv_release_all_nested(kvm);
4900         kvmppc_rmap_reset(kvm);
4901         kvm->arch.process_table = 0;
4902         /* Mutual exclusion with kvm_unmap_gfn_range etc. */
4903         spin_lock(&kvm->mmu_lock);
4904         kvm->arch.radix = 0;
4905         spin_unlock(&kvm->mmu_lock);
4906         kvmppc_free_radix(kvm);
4907         kvmppc_update_lpcr(kvm, LPCR_VPM1,
4908                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4909         return 0;
4910 }
4911
4912 /*
4913  * Must be called with kvm->arch.mmu_setup_lock held and
4914  * mmu_ready = 0 and no vcpus running.
4915  */
4916 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4917 {
4918         int err;
4919
4920         err = kvmppc_init_vm_radix(kvm);
4921         if (err)
4922                 return err;
4923         kvmppc_rmap_reset(kvm);
4924         /* Mutual exclusion with kvm_unmap_gfn_range etc. */
4925         spin_lock(&kvm->mmu_lock);
4926         kvm->arch.radix = 1;
4927         spin_unlock(&kvm->mmu_lock);
4928         kvmppc_free_hpt(&kvm->arch.hpt);
4929         kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4930                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4931         return 0;
4932 }
4933
4934 #ifdef CONFIG_KVM_XICS
4935 /*
4936  * Allocate a per-core structure for managing state about which cores are
4937  * running in the host versus the guest and for exchanging data between
4938  * real mode KVM and CPU running in the host.
4939  * This is only done for the first VM.
4940  * The allocated structure stays even if all VMs have stopped.
4941  * It is only freed when the kvm-hv module is unloaded.
4942  * It's OK for this routine to fail, we just don't support host
4943  * core operations like redirecting H_IPI wakeups.
4944  */
4945 void kvmppc_alloc_host_rm_ops(void)
4946 {
4947         struct kvmppc_host_rm_ops *ops;
4948         unsigned long l_ops;
4949         int cpu, core;
4950         int size;
4951
4952         /* Not the first time here ? */
4953         if (kvmppc_host_rm_ops_hv != NULL)
4954                 return;
4955
4956         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4957         if (!ops)
4958                 return;
4959
4960         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4961         ops->rm_core = kzalloc(size, GFP_KERNEL);
4962
4963         if (!ops->rm_core) {
4964                 kfree(ops);
4965                 return;
4966         }
4967
4968         cpus_read_lock();
4969
4970         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4971                 if (!cpu_online(cpu))
4972                         continue;
4973
4974                 core = cpu >> threads_shift;
4975                 ops->rm_core[core].rm_state.in_host = 1;
4976         }
4977
4978         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4979
4980         /*
4981          * Make the contents of the kvmppc_host_rm_ops structure visible
4982          * to other CPUs before we assign it to the global variable.
4983          * Do an atomic assignment (no locks used here), but if someone
4984          * beats us to it, just free our copy and return.
4985          */
4986         smp_wmb();
4987         l_ops = (unsigned long) ops;
4988
4989         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4990                 cpus_read_unlock();
4991                 kfree(ops->rm_core);
4992                 kfree(ops);
4993                 return;
4994         }
4995
4996         cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4997                                              "ppc/kvm_book3s:prepare",
4998                                              kvmppc_set_host_core,
4999                                              kvmppc_clear_host_core);
5000         cpus_read_unlock();
5001 }
5002
5003 void kvmppc_free_host_rm_ops(void)
5004 {
5005         if (kvmppc_host_rm_ops_hv) {
5006                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
5007                 kfree(kvmppc_host_rm_ops_hv->rm_core);
5008                 kfree(kvmppc_host_rm_ops_hv);
5009                 kvmppc_host_rm_ops_hv = NULL;
5010         }
5011 }
5012 #endif
5013
5014 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
5015 {
5016         unsigned long lpcr, lpid;
5017         char buf[32];
5018         int ret;
5019
5020         mutex_init(&kvm->arch.uvmem_lock);
5021         INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
5022         mutex_init(&kvm->arch.mmu_setup_lock);
5023
5024         /* Allocate the guest's logical partition ID */
5025
5026         lpid = kvmppc_alloc_lpid();
5027         if ((long)lpid < 0)
5028                 return -ENOMEM;
5029         kvm->arch.lpid = lpid;
5030
5031         kvmppc_alloc_host_rm_ops();
5032
5033         kvmhv_vm_nested_init(kvm);
5034
5035         /*
5036          * Since we don't flush the TLB when tearing down a VM,
5037          * and this lpid might have previously been used,
5038          * make sure we flush on each core before running the new VM.
5039          * On POWER9, the tlbie in mmu_partition_table_set_entry()
5040          * does this flush for us.
5041          */
5042         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5043                 cpumask_setall(&kvm->arch.need_tlb_flush);
5044
5045         /* Start out with the default set of hcalls enabled */
5046         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
5047                sizeof(kvm->arch.enabled_hcalls));
5048
5049         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5050                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
5051
5052         /* Init LPCR for virtual RMA mode */
5053         if (cpu_has_feature(CPU_FTR_HVMODE)) {
5054                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
5055                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
5056                 lpcr &= LPCR_PECE | LPCR_LPES;
5057         } else {
5058                 lpcr = 0;
5059         }
5060         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
5061                 LPCR_VPM0 | LPCR_VPM1;
5062         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
5063                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
5064         /* On POWER8 turn on online bit to enable PURR/SPURR */
5065         if (cpu_has_feature(CPU_FTR_ARCH_207S))
5066                 lpcr |= LPCR_ONL;
5067         /*
5068          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
5069          * Set HVICE bit to enable hypervisor virtualization interrupts.
5070          * Set HEIC to prevent OS interrupts to go to hypervisor (should
5071          * be unnecessary but better safe than sorry in case we re-enable
5072          * EE in HV mode with this LPCR still set)
5073          */
5074         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5075                 lpcr &= ~LPCR_VPM0;
5076                 lpcr |= LPCR_HVICE | LPCR_HEIC;
5077
5078                 /*
5079                  * If xive is enabled, we route 0x500 interrupts directly
5080                  * to the guest.
5081                  */
5082                 if (xics_on_xive())
5083                         lpcr |= LPCR_LPES;
5084         }
5085
5086         /*
5087          * If the host uses radix, the guest starts out as radix.
5088          */
5089         if (radix_enabled()) {
5090                 kvm->arch.radix = 1;
5091                 kvm->arch.mmu_ready = 1;
5092                 lpcr &= ~LPCR_VPM1;
5093                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5094                 ret = kvmppc_init_vm_radix(kvm);
5095                 if (ret) {
5096                         kvmppc_free_lpid(kvm->arch.lpid);
5097                         return ret;
5098                 }
5099                 kvmppc_setup_partition_table(kvm);
5100         }
5101
5102         verify_lpcr(kvm, lpcr);
5103         kvm->arch.lpcr = lpcr;
5104
5105         /* Initialization for future HPT resizes */
5106         kvm->arch.resize_hpt = NULL;
5107
5108         /*
5109          * Work out how many sets the TLB has, for the use of
5110          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
5111          */
5112         if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5113                 /*
5114                  * P10 will flush all the congruence class with a single tlbiel
5115                  */
5116                 kvm->arch.tlb_sets = 1;
5117         } else if (radix_enabled())
5118                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
5119         else if (cpu_has_feature(CPU_FTR_ARCH_300))
5120                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
5121         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
5122                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
5123         else
5124                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
5125
5126         /*
5127          * Track that we now have a HV mode VM active. This blocks secondary
5128          * CPU threads from coming online.
5129          * On POWER9, we only need to do this if the "indep_threads_mode"
5130          * module parameter has been set to N.
5131          */
5132         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5133                 if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
5134                         pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
5135                         kvm->arch.threads_indep = true;
5136                 } else {
5137                         kvm->arch.threads_indep = indep_threads_mode;
5138                 }
5139         }
5140         if (!kvm->arch.threads_indep)
5141                 kvm_hv_vm_activated();
5142
5143         /*
5144          * Initialize smt_mode depending on processor.
5145          * POWER8 and earlier have to use "strict" threading, where
5146          * all vCPUs in a vcore have to run on the same (sub)core,
5147          * whereas on POWER9 the threads can each run a different
5148          * guest.
5149          */
5150         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5151                 kvm->arch.smt_mode = threads_per_subcore;
5152         else
5153                 kvm->arch.smt_mode = 1;
5154         kvm->arch.emul_smt_mode = 1;
5155
5156         /*
5157          * Create a debugfs directory for the VM
5158          */
5159         snprintf(buf, sizeof(buf), "vm%d", current->pid);
5160         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
5161         kvmppc_mmu_debugfs_init(kvm);
5162         if (radix_enabled())
5163                 kvmhv_radix_debugfs_init(kvm);
5164
5165         return 0;
5166 }
5167
5168 static void kvmppc_free_vcores(struct kvm *kvm)
5169 {
5170         long int i;
5171
5172         for (i = 0; i < KVM_MAX_VCORES; ++i)
5173                 kfree(kvm->arch.vcores[i]);
5174         kvm->arch.online_vcores = 0;
5175 }
5176
5177 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5178 {
5179         debugfs_remove_recursive(kvm->arch.debugfs_dir);
5180
5181         if (!kvm->arch.threads_indep)
5182                 kvm_hv_vm_deactivated();
5183
5184         kvmppc_free_vcores(kvm);
5185
5186
5187         if (kvm_is_radix(kvm))
5188                 kvmppc_free_radix(kvm);
5189         else
5190                 kvmppc_free_hpt(&kvm->arch.hpt);
5191
5192         /* Perform global invalidation and return lpid to the pool */
5193         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5194                 if (nesting_enabled(kvm))
5195                         kvmhv_release_all_nested(kvm);
5196                 kvm->arch.process_table = 0;
5197                 if (kvm->arch.secure_guest)
5198                         uv_svm_terminate(kvm->arch.lpid);
5199                 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5200         }
5201
5202         kvmppc_free_lpid(kvm->arch.lpid);
5203
5204         kvmppc_free_pimap(kvm);
5205 }
5206
5207 /* We don't need to emulate any privileged instructions or dcbz */
5208 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5209                                      unsigned int inst, int *advance)
5210 {
5211         return EMULATE_FAIL;
5212 }
5213
5214 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5215                                         ulong spr_val)
5216 {
5217         return EMULATE_FAIL;
5218 }
5219
5220 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5221                                         ulong *spr_val)
5222 {
5223         return EMULATE_FAIL;
5224 }
5225
5226 static int kvmppc_core_check_processor_compat_hv(void)
5227 {
5228         if (cpu_has_feature(CPU_FTR_HVMODE) &&
5229             cpu_has_feature(CPU_FTR_ARCH_206))
5230                 return 0;
5231
5232         /* POWER9 in radix mode is capable of being a nested hypervisor. */
5233         if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5234                 return 0;
5235
5236         return -EIO;
5237 }
5238
5239 #ifdef CONFIG_KVM_XICS
5240
5241 void kvmppc_free_pimap(struct kvm *kvm)
5242 {
5243         kfree(kvm->arch.pimap);
5244 }
5245
5246 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5247 {
5248         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5249 }
5250
5251 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5252 {
5253         struct irq_desc *desc;
5254         struct kvmppc_irq_map *irq_map;
5255         struct kvmppc_passthru_irqmap *pimap;
5256         struct irq_chip *chip;
5257         int i, rc = 0;
5258
5259         if (!kvm_irq_bypass)
5260                 return 1;
5261
5262         desc = irq_to_desc(host_irq);
5263         if (!desc)
5264                 return -EIO;
5265
5266         mutex_lock(&kvm->lock);
5267
5268         pimap = kvm->arch.pimap;
5269         if (pimap == NULL) {
5270                 /* First call, allocate structure to hold IRQ map */
5271                 pimap = kvmppc_alloc_pimap();
5272                 if (pimap == NULL) {
5273                         mutex_unlock(&kvm->lock);
5274                         return -ENOMEM;
5275                 }
5276                 kvm->arch.pimap = pimap;
5277         }
5278
5279         /*
5280          * For now, we only support interrupts for which the EOI operation
5281          * is an OPAL call followed by a write to XIRR, since that's
5282          * what our real-mode EOI code does, or a XIVE interrupt
5283          */
5284         chip = irq_data_get_irq_chip(&desc->irq_data);
5285         if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5286                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5287                         host_irq, guest_gsi);
5288                 mutex_unlock(&kvm->lock);
5289                 return -ENOENT;
5290         }
5291
5292         /*
5293          * See if we already have an entry for this guest IRQ number.
5294          * If it's mapped to a hardware IRQ number, that's an error,
5295          * otherwise re-use this entry.
5296          */
5297         for (i = 0; i < pimap->n_mapped; i++) {
5298                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5299                         if (pimap->mapped[i].r_hwirq) {
5300                                 mutex_unlock(&kvm->lock);
5301                                 return -EINVAL;
5302                         }
5303                         break;
5304                 }
5305         }
5306
5307         if (i == KVMPPC_PIRQ_MAPPED) {
5308                 mutex_unlock(&kvm->lock);
5309                 return -EAGAIN;         /* table is full */
5310         }
5311
5312         irq_map = &pimap->mapped[i];
5313
5314         irq_map->v_hwirq = guest_gsi;
5315         irq_map->desc = desc;
5316
5317         /*
5318          * Order the above two stores before the next to serialize with
5319          * the KVM real mode handler.
5320          */
5321         smp_wmb();
5322         irq_map->r_hwirq = desc->irq_data.hwirq;
5323
5324         if (i == pimap->n_mapped)
5325                 pimap->n_mapped++;
5326
5327         if (xics_on_xive())
5328                 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5329         else
5330                 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5331         if (rc)
5332                 irq_map->r_hwirq = 0;
5333
5334         mutex_unlock(&kvm->lock);
5335
5336         return 0;
5337 }
5338
5339 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5340 {
5341         struct irq_desc *desc;
5342         struct kvmppc_passthru_irqmap *pimap;
5343         int i, rc = 0;
5344
5345         if (!kvm_irq_bypass)
5346                 return 0;
5347
5348         desc = irq_to_desc(host_irq);
5349         if (!desc)
5350                 return -EIO;
5351
5352         mutex_lock(&kvm->lock);
5353         if (!kvm->arch.pimap)
5354                 goto unlock;
5355
5356         pimap = kvm->arch.pimap;
5357
5358         for (i = 0; i < pimap->n_mapped; i++) {
5359                 if (guest_gsi == pimap->mapped[i].v_hwirq)
5360                         break;
5361         }
5362
5363         if (i == pimap->n_mapped) {
5364                 mutex_unlock(&kvm->lock);
5365                 return -ENODEV;
5366         }
5367
5368         if (xics_on_xive())
5369                 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5370         else
5371                 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5372
5373         /* invalidate the entry (what do do on error from the above ?) */
5374         pimap->mapped[i].r_hwirq = 0;
5375
5376         /*
5377          * We don't free this structure even when the count goes to
5378          * zero. The structure is freed when we destroy the VM.
5379          */
5380  unlock:
5381         mutex_unlock(&kvm->lock);
5382         return rc;
5383 }
5384
5385 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5386                                              struct irq_bypass_producer *prod)
5387 {
5388         int ret = 0;
5389         struct kvm_kernel_irqfd *irqfd =
5390                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5391
5392         irqfd->producer = prod;
5393
5394         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5395         if (ret)
5396                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5397                         prod->irq, irqfd->gsi, ret);
5398
5399         return ret;
5400 }
5401
5402 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5403                                               struct irq_bypass_producer *prod)
5404 {
5405         int ret;
5406         struct kvm_kernel_irqfd *irqfd =
5407                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5408
5409         irqfd->producer = NULL;
5410
5411         /*
5412          * When producer of consumer is unregistered, we change back to
5413          * default external interrupt handling mode - KVM real mode
5414          * will switch back to host.
5415          */
5416         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5417         if (ret)
5418                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5419                         prod->irq, irqfd->gsi, ret);
5420 }
5421 #endif
5422
5423 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5424                                  unsigned int ioctl, unsigned long arg)
5425 {
5426         struct kvm *kvm __maybe_unused = filp->private_data;
5427         void __user *argp = (void __user *)arg;
5428         long r;
5429
5430         switch (ioctl) {
5431
5432         case KVM_PPC_ALLOCATE_HTAB: {
5433                 u32 htab_order;
5434
5435                 /* If we're a nested hypervisor, we currently only support radix */
5436                 if (kvmhv_on_pseries()) {
5437                         r = -EOPNOTSUPP;
5438                         break;
5439                 }
5440
5441                 r = -EFAULT;
5442                 if (get_user(htab_order, (u32 __user *)argp))
5443                         break;
5444                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5445                 if (r)
5446                         break;
5447                 r = 0;
5448                 break;
5449         }
5450
5451         case KVM_PPC_GET_HTAB_FD: {
5452                 struct kvm_get_htab_fd ghf;
5453
5454                 r = -EFAULT;
5455                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5456                         break;
5457                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5458                 break;
5459         }
5460
5461         case KVM_PPC_RESIZE_HPT_PREPARE: {
5462                 struct kvm_ppc_resize_hpt rhpt;
5463
5464                 r = -EFAULT;
5465                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5466                         break;
5467
5468                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5469                 break;
5470         }
5471
5472         case KVM_PPC_RESIZE_HPT_COMMIT: {
5473                 struct kvm_ppc_resize_hpt rhpt;
5474
5475                 r = -EFAULT;
5476                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5477                         break;
5478
5479                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5480                 break;
5481         }
5482
5483         default:
5484                 r = -ENOTTY;
5485         }
5486
5487         return r;
5488 }
5489
5490 /*
5491  * List of hcall numbers to enable by default.
5492  * For compatibility with old userspace, we enable by default
5493  * all hcalls that were implemented before the hcall-enabling
5494  * facility was added.  Note this list should not include H_RTAS.
5495  */
5496 static unsigned int default_hcall_list[] = {
5497         H_REMOVE,
5498         H_ENTER,
5499         H_READ,
5500         H_PROTECT,
5501         H_BULK_REMOVE,
5502 #ifdef CONFIG_SPAPR_TCE_IOMMU
5503         H_GET_TCE,
5504         H_PUT_TCE,
5505 #endif
5506         H_SET_DABR,
5507         H_SET_XDABR,
5508         H_CEDE,
5509         H_PROD,
5510         H_CONFER,
5511         H_REGISTER_VPA,
5512 #ifdef CONFIG_KVM_XICS
5513         H_EOI,
5514         H_CPPR,
5515         H_IPI,
5516         H_IPOLL,
5517         H_XIRR,
5518         H_XIRR_X,
5519 #endif
5520         0
5521 };
5522
5523 static void init_default_hcalls(void)
5524 {
5525         int i;
5526         unsigned int hcall;
5527
5528         for (i = 0; default_hcall_list[i]; ++i) {
5529                 hcall = default_hcall_list[i];
5530                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5531                 __set_bit(hcall / 4, default_enabled_hcalls);
5532         }
5533 }
5534
5535 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5536 {
5537         unsigned long lpcr;
5538         int radix;
5539         int err;
5540
5541         /* If not on a POWER9, reject it */
5542         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5543                 return -ENODEV;
5544
5545         /* If any unknown flags set, reject it */
5546         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5547                 return -EINVAL;
5548
5549         /* GR (guest radix) bit in process_table field must match */
5550         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5551         if (!!(cfg->process_table & PATB_GR) != radix)
5552                 return -EINVAL;
5553
5554         /* Process table size field must be reasonable, i.e. <= 24 */
5555         if ((cfg->process_table & PRTS_MASK) > 24)
5556                 return -EINVAL;
5557
5558         /* We can change a guest to/from radix now, if the host is radix */
5559         if (radix && !radix_enabled())
5560                 return -EINVAL;
5561
5562         /* If we're a nested hypervisor, we currently only support radix */
5563         if (kvmhv_on_pseries() && !radix)
5564                 return -EINVAL;
5565
5566         mutex_lock(&kvm->arch.mmu_setup_lock);
5567         if (radix != kvm_is_radix(kvm)) {
5568                 if (kvm->arch.mmu_ready) {
5569                         kvm->arch.mmu_ready = 0;
5570                         /* order mmu_ready vs. vcpus_running */
5571                         smp_mb();
5572                         if (atomic_read(&kvm->arch.vcpus_running)) {
5573                                 kvm->arch.mmu_ready = 1;
5574                                 err = -EBUSY;
5575                                 goto out_unlock;
5576                         }
5577                 }
5578                 if (radix)
5579                         err = kvmppc_switch_mmu_to_radix(kvm);
5580                 else
5581                         err = kvmppc_switch_mmu_to_hpt(kvm);
5582                 if (err)
5583                         goto out_unlock;
5584         }
5585
5586         kvm->arch.process_table = cfg->process_table;
5587         kvmppc_setup_partition_table(kvm);
5588
5589         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5590         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5591         err = 0;
5592
5593  out_unlock:
5594         mutex_unlock(&kvm->arch.mmu_setup_lock);
5595         return err;
5596 }
5597
5598 static int kvmhv_enable_nested(struct kvm *kvm)
5599 {
5600         if (!nested)
5601                 return -EPERM;
5602         if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5603                 return -ENODEV;
5604
5605         /* kvm == NULL means the caller is testing if the capability exists */
5606         if (kvm)
5607                 kvm->arch.nested_enable = true;
5608         return 0;
5609 }
5610
5611 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5612                                  int size)
5613 {
5614         int rc = -EINVAL;
5615
5616         if (kvmhv_vcpu_is_radix(vcpu)) {
5617                 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5618
5619                 if (rc > 0)
5620                         rc = -EINVAL;
5621         }
5622
5623         /* For now quadrants are the only way to access nested guest memory */
5624         if (rc && vcpu->arch.nested)
5625                 rc = -EAGAIN;
5626
5627         return rc;
5628 }
5629
5630 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5631                                 int size)
5632 {
5633         int rc = -EINVAL;
5634
5635         if (kvmhv_vcpu_is_radix(vcpu)) {
5636                 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5637
5638                 if (rc > 0)
5639                         rc = -EINVAL;
5640         }
5641
5642         /* For now quadrants are the only way to access nested guest memory */
5643         if (rc && vcpu->arch.nested)
5644                 rc = -EAGAIN;
5645
5646         return rc;
5647 }
5648
5649 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5650 {
5651         unpin_vpa(kvm, vpa);
5652         vpa->gpa = 0;
5653         vpa->pinned_addr = NULL;
5654         vpa->dirty = false;
5655         vpa->update_pending = 0;
5656 }
5657
5658 /*
5659  * Enable a guest to become a secure VM, or test whether
5660  * that could be enabled.
5661  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
5662  * tested (kvm == NULL) or enabled (kvm != NULL).
5663  */
5664 static int kvmhv_enable_svm(struct kvm *kvm)
5665 {
5666         if (!kvmppc_uvmem_available())
5667                 return -EINVAL;
5668         if (kvm)
5669                 kvm->arch.svm_enabled = 1;
5670         return 0;
5671 }
5672
5673 /*
5674  *  IOCTL handler to turn off secure mode of guest
5675  *
5676  * - Release all device pages
5677  * - Issue ucall to terminate the guest on the UV side
5678  * - Unpin the VPA pages.
5679  * - Reinit the partition scoped page tables
5680  */
5681 static int kvmhv_svm_off(struct kvm *kvm)
5682 {
5683         struct kvm_vcpu *vcpu;
5684         int mmu_was_ready;
5685         int srcu_idx;
5686         int ret = 0;
5687         int i;
5688
5689         if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5690                 return ret;
5691
5692         mutex_lock(&kvm->arch.mmu_setup_lock);
5693         mmu_was_ready = kvm->arch.mmu_ready;
5694         if (kvm->arch.mmu_ready) {
5695                 kvm->arch.mmu_ready = 0;
5696                 /* order mmu_ready vs. vcpus_running */
5697                 smp_mb();
5698                 if (atomic_read(&kvm->arch.vcpus_running)) {
5699                         kvm->arch.mmu_ready = 1;
5700                         ret = -EBUSY;
5701                         goto out;
5702                 }
5703         }
5704
5705         srcu_idx = srcu_read_lock(&kvm->srcu);
5706         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5707                 struct kvm_memory_slot *memslot;
5708                 struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5709
5710                 if (!slots)
5711                         continue;
5712
5713                 kvm_for_each_memslot(memslot, slots) {
5714                         kvmppc_uvmem_drop_pages(memslot, kvm, true);
5715                         uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5716                 }
5717         }
5718         srcu_read_unlock(&kvm->srcu, srcu_idx);
5719
5720         ret = uv_svm_terminate(kvm->arch.lpid);
5721         if (ret != U_SUCCESS) {
5722                 ret = -EINVAL;
5723                 goto out;
5724         }
5725
5726         /*
5727          * When secure guest is reset, all the guest pages are sent
5728          * to UV via UV_PAGE_IN before the non-boot vcpus get a
5729          * chance to run and unpin their VPA pages. Unpinning of all
5730          * VPA pages is done here explicitly so that VPA pages
5731          * can be migrated to the secure side.
5732          *
5733          * This is required to for the secure SMP guest to reboot
5734          * correctly.
5735          */
5736         kvm_for_each_vcpu(i, vcpu, kvm) {
5737                 spin_lock(&vcpu->arch.vpa_update_lock);
5738                 unpin_vpa_reset(kvm, &vcpu->arch.dtl);
5739                 unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
5740                 unpin_vpa_reset(kvm, &vcpu->arch.vpa);
5741                 spin_unlock(&vcpu->arch.vpa_update_lock);
5742         }
5743
5744         kvmppc_setup_partition_table(kvm);
5745         kvm->arch.secure_guest = 0;
5746         kvm->arch.mmu_ready = mmu_was_ready;
5747 out:
5748         mutex_unlock(&kvm->arch.mmu_setup_lock);
5749         return ret;
5750 }
5751
5752 static int kvmhv_enable_dawr1(struct kvm *kvm)
5753 {
5754         if (!cpu_has_feature(CPU_FTR_DAWR1))
5755                 return -ENODEV;
5756
5757         /* kvm == NULL means the caller is testing if the capability exists */
5758         if (kvm)
5759                 kvm->arch.dawr1_enabled = true;
5760         return 0;
5761 }
5762
5763 static bool kvmppc_hash_v3_possible(void)
5764 {
5765         if (radix_enabled() && no_mixing_hpt_and_radix)
5766                 return false;
5767
5768         return cpu_has_feature(CPU_FTR_ARCH_300) &&
5769                 cpu_has_feature(CPU_FTR_HVMODE);
5770 }
5771
5772 static struct kvmppc_ops kvm_ops_hv = {
5773         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5774         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5775         .get_one_reg = kvmppc_get_one_reg_hv,
5776         .set_one_reg = kvmppc_set_one_reg_hv,
5777         .vcpu_load   = kvmppc_core_vcpu_load_hv,
5778         .vcpu_put    = kvmppc_core_vcpu_put_hv,
5779         .inject_interrupt = kvmppc_inject_interrupt_hv,
5780         .set_msr     = kvmppc_set_msr_hv,
5781         .vcpu_run    = kvmppc_vcpu_run_hv,
5782         .vcpu_create = kvmppc_core_vcpu_create_hv,
5783         .vcpu_free   = kvmppc_core_vcpu_free_hv,
5784         .check_requests = kvmppc_core_check_requests_hv,
5785         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5786         .flush_memslot  = kvmppc_core_flush_memslot_hv,
5787         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5788         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5789         .unmap_gfn_range = kvm_unmap_gfn_range_hv,
5790         .age_gfn = kvm_age_gfn_hv,
5791         .test_age_gfn = kvm_test_age_gfn_hv,
5792         .set_spte_gfn = kvm_set_spte_gfn_hv,
5793         .free_memslot = kvmppc_core_free_memslot_hv,
5794         .init_vm =  kvmppc_core_init_vm_hv,
5795         .destroy_vm = kvmppc_core_destroy_vm_hv,
5796         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5797         .emulate_op = kvmppc_core_emulate_op_hv,
5798         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5799         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5800         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5801         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5802         .hcall_implemented = kvmppc_hcall_impl_hv,
5803 #ifdef CONFIG_KVM_XICS
5804         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5805         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5806 #endif
5807         .configure_mmu = kvmhv_configure_mmu,
5808         .get_rmmu_info = kvmhv_get_rmmu_info,
5809         .set_smt_mode = kvmhv_set_smt_mode,
5810         .enable_nested = kvmhv_enable_nested,
5811         .load_from_eaddr = kvmhv_load_from_eaddr,
5812         .store_to_eaddr = kvmhv_store_to_eaddr,
5813         .enable_svm = kvmhv_enable_svm,
5814         .svm_off = kvmhv_svm_off,
5815         .enable_dawr1 = kvmhv_enable_dawr1,
5816         .hash_v3_possible = kvmppc_hash_v3_possible,
5817 };
5818
5819 static int kvm_init_subcore_bitmap(void)
5820 {
5821         int i, j;
5822         int nr_cores = cpu_nr_cores();
5823         struct sibling_subcore_state *sibling_subcore_state;
5824
5825         for (i = 0; i < nr_cores; i++) {
5826                 int first_cpu = i * threads_per_core;
5827                 int node = cpu_to_node(first_cpu);
5828
5829                 /* Ignore if it is already allocated. */
5830                 if (paca_ptrs[first_cpu]->sibling_subcore_state)
5831                         continue;
5832
5833                 sibling_subcore_state =
5834                         kzalloc_node(sizeof(struct sibling_subcore_state),
5835                                                         GFP_KERNEL, node);
5836                 if (!sibling_subcore_state)
5837                         return -ENOMEM;
5838
5839
5840                 for (j = 0; j < threads_per_core; j++) {
5841                         int cpu = first_cpu + j;
5842
5843                         paca_ptrs[cpu]->sibling_subcore_state =
5844                                                 sibling_subcore_state;
5845                 }
5846         }
5847         return 0;
5848 }
5849
5850 static int kvmppc_radix_possible(void)
5851 {
5852         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5853 }
5854
5855 static int kvmppc_book3s_init_hv(void)
5856 {
5857         int r;
5858
5859         if (!tlbie_capable) {
5860                 pr_err("KVM-HV: Host does not support TLBIE\n");
5861                 return -ENODEV;
5862         }
5863
5864         /*
5865          * FIXME!! Do we need to check on all cpus ?
5866          */
5867         r = kvmppc_core_check_processor_compat_hv();
5868         if (r < 0)
5869                 return -ENODEV;
5870
5871         r = kvmhv_nested_init();
5872         if (r)
5873                 return r;
5874
5875         r = kvm_init_subcore_bitmap();
5876         if (r)
5877                 return r;
5878
5879         /*
5880          * We need a way of accessing the XICS interrupt controller,
5881          * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5882          * indirectly, via OPAL.
5883          */
5884 #ifdef CONFIG_SMP
5885         if (!xics_on_xive() && !kvmhv_on_pseries() &&
5886             !local_paca->kvm_hstate.xics_phys) {
5887                 struct device_node *np;
5888
5889                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5890                 if (!np) {
5891                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5892                         return -ENODEV;
5893                 }
5894                 /* presence of intc confirmed - node can be dropped again */
5895                 of_node_put(np);
5896         }
5897 #endif
5898
5899         kvm_ops_hv.owner = THIS_MODULE;
5900         kvmppc_hv_ops = &kvm_ops_hv;
5901
5902         init_default_hcalls();
5903
5904         init_vcore_lists();
5905
5906         r = kvmppc_mmu_hv_init();
5907         if (r)
5908                 return r;
5909
5910         if (kvmppc_radix_possible())
5911                 r = kvmppc_radix_init();
5912
5913         /*
5914          * POWER9 chips before version 2.02 can't have some threads in
5915          * HPT mode and some in radix mode on the same core.
5916          */
5917         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5918                 unsigned int pvr = mfspr(SPRN_PVR);
5919                 if ((pvr >> 16) == PVR_POWER9 &&
5920                     (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5921                      ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5922                         no_mixing_hpt_and_radix = true;
5923         }
5924
5925         r = kvmppc_uvmem_init();
5926         if (r < 0)
5927                 pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
5928
5929         return r;
5930 }
5931
5932 static void kvmppc_book3s_exit_hv(void)
5933 {
5934         kvmppc_uvmem_free();
5935         kvmppc_free_host_rm_ops();
5936         if (kvmppc_radix_possible())
5937                 kvmppc_radix_exit();
5938         kvmppc_hv_ops = NULL;
5939         kvmhv_nested_exit();
5940 }
5941
5942 module_init(kvmppc_book3s_init_hv);
5943 module_exit(kvmppc_book3s_exit_hv);
5944 MODULE_LICENSE("GPL");
5945 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5946 MODULE_ALIAS("devname:kvm");