arch/powerpc/kvm/book3s_hv.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
   4  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
   5  *
   6  * Authors:
   7  *    Paul Mackerras <paulus@au1.ibm.com>
   8  *    Alexander Graf <agraf@suse.de>
   9  *    Kevin Wolf <mail@kevin-wolf.de>
  10  *
  11  * Description: KVM functions specific to running on Book 3S
  12  * processors in hypervisor mode (specifically POWER7 and later).
  13  *
  14  * This file is derived from arch/powerpc/kvm/book3s.c,
  15  * by Alexander Graf <agraf@suse.de>.
  16  */
  17
  18 #include <linux/kvm_host.h>
  19 #include <linux/kernel.h>
  20 #include <linux/err.h>
  21 #include <linux/slab.h>
  22 #include <linux/preempt.h>
  23 #include <linux/sched/signal.h>
  24 #include <linux/sched/stat.h>
  25 #include <linux/delay.h>
  26 #include <linux/export.h>
  27 #include <linux/fs.h>
  28 #include <linux/anon_inodes.h>
  29 #include <linux/cpu.h>
  30 #include <linux/cpumask.h>
  31 #include <linux/spinlock.h>
  32 #include <linux/page-flags.h>
  33 #include <linux/srcu.h>
  34 #include <linux/miscdevice.h>
  35 #include <linux/debugfs.h>
  36 #include <linux/gfp.h>
  37 #include <linux/vmalloc.h>
  38 #include <linux/highmem.h>
  39 #include <linux/hugetlb.h>
  40 #include <linux/kvm_irqfd.h>
  41 #include <linux/irqbypass.h>
  42 #include <linux/module.h>
  43 #include <linux/compiler.h>
  44 #include <linux/of.h>
  45
  46 #include <asm/ftrace.h>
  47 #include <asm/reg.h>
  48 #include <asm/ppc-opcode.h>
  49 #include <asm/asm-prototypes.h>
  50 #include <asm/archrandom.h>
  51 #include <asm/debug.h>
  52 #include <asm/disassemble.h>
  53 #include <asm/cputable.h>
  54 #include <asm/cacheflush.h>
  55 #include <linux/uaccess.h>
  56 #include <asm/io.h>
  57 #include <asm/kvm_ppc.h>
  58 #include <asm/kvm_book3s.h>
  59 #include <asm/mmu_context.h>
  60 #include <asm/lppaca.h>
  61 #include <asm/processor.h>
  62 #include <asm/cputhreads.h>
  63 #include <asm/page.h>
  64 #include <asm/hvcall.h>
  65 #include <asm/switch_to.h>
  66 #include <asm/smp.h>
  67 #include <asm/dbell.h>
  68 #include <asm/hmi.h>
  69 #include <asm/pnv-pci.h>
  70 #include <asm/mmu.h>
  71 #include <asm/opal.h>
  72 #include <asm/xics.h>
  73 #include <asm/xive.h>
  74 #include <asm/hw_breakpoint.h>
  75 #include <asm/kvm_book3s_uvmem.h>
  76 #include <asm/ultravisor.h>
  77 #include <asm/dtl.h>
  78
  79 #include "book3s.h"
  80
  81 #define CREATE_TRACE_POINTS
  82 #include "trace_hv.h"
  83
  84 /* #define EXIT_DEBUG */
  85 /* #define EXIT_DEBUG_SIMPLE */
  86 /* #define EXIT_DEBUG_INT */
  87
  88 /* Used to indicate that a guest page fault needs to be handled */
  89 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  90 /* Used to indicate that a guest passthrough interrupt needs to be handled */
  91 #define RESUME_PASSTHROUGH      (RESUME_GUEST | RESUME_FLAG_ARCH2)
  92
  93 /* Used as a "null" value for timebase values */
  94 #define TB_NIL  (~(u64)0)
  95
  96 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
  97
  98 static int dynamic_mt_modes = 6;
  99 module_param(dynamic_mt_modes, int, 0644);
 100 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
 101 static int target_smt_mode;
 102 module_param(target_smt_mode, int, 0644);
 103 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
 104
 105 static bool indep_threads_mode = true;
 106 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
 107 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
 108
 109 static bool one_vm_per_core;
 110 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
 111 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)");
 112
 113 #ifdef CONFIG_KVM_XICS
 114 static const struct kernel_param_ops module_param_ops = {
 115         .set = param_set_int,
 116         .get = param_get_int,
 117 };
 118
 119 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
 120 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
 121
 122 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
 123 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
 124 #endif
 125
 126 /* If set, guests are allowed to create and control nested guests */
 127 static bool nested = true;
 128 module_param(nested, bool, S_IRUGO | S_IWUSR);
 129 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
 130
 131 static inline bool nesting_enabled(struct kvm *kvm)
 132 {
 133         return kvm->arch.nested_enable && kvm_is_radix(kvm);
 134 }
 135
 136 /* If set, the threads on each CPU core have to be in the same MMU mode */
 137 static bool no_mixing_hpt_and_radix __read_mostly;
 138
 139 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
 140
 141 /*
 142  * RWMR values for POWER8.  These control the rate at which PURR
 143  * and SPURR count and should be set according to the number of
 144  * online threads in the vcore being run.
 145  */
 146 #define RWMR_RPA_P8_1THREAD     0x164520C62609AECAUL
 147 #define RWMR_RPA_P8_2THREAD     0x7FFF2908450D8DA9UL
 148 #define RWMR_RPA_P8_3THREAD     0x164520C62609AECAUL
 149 #define RWMR_RPA_P8_4THREAD     0x199A421245058DA9UL
 150 #define RWMR_RPA_P8_5THREAD     0x164520C62609AECAUL
 151 #define RWMR_RPA_P8_6THREAD     0x164520C62609AECAUL
 152 #define RWMR_RPA_P8_7THREAD     0x164520C62609AECAUL
 153 #define RWMR_RPA_P8_8THREAD     0x164520C62609AECAUL
 154
 155 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
 156         RWMR_RPA_P8_1THREAD,
 157         RWMR_RPA_P8_1THREAD,
 158         RWMR_RPA_P8_2THREAD,
 159         RWMR_RPA_P8_3THREAD,
 160         RWMR_RPA_P8_4THREAD,
 161         RWMR_RPA_P8_5THREAD,
 162         RWMR_RPA_P8_6THREAD,
 163         RWMR_RPA_P8_7THREAD,
 164         RWMR_RPA_P8_8THREAD,
 165 };
 166
 167 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
 168                 int *ip)
 169 {
 170         int i = *ip;
 171         struct kvm_vcpu *vcpu;
 172
 173         while (++i < MAX_SMT_THREADS) {
 174                 vcpu = READ_ONCE(vc->runnable_threads[i]);
 175                 if (vcpu) {
 176                         *ip = i;
 177                         return vcpu;
 178                 }
 179         }
 180         return NULL;
 181 }
 182
 183 /* Used to traverse the list of runnable threads for a given vcore */
 184 #define for_each_runnable_thread(i, vcpu, vc) \
 185         for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
 186
 187 static bool kvmppc_ipi_thread(int cpu)
 188 {
 189         unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
 190
 191         /* If we're a nested hypervisor, fall back to ordinary IPIs for now */
 192         if (kvmhv_on_pseries())
 193                 return false;
 194
 195         /* On POWER9 we can use msgsnd to IPI any cpu */
 196         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
 197                 msg |= get_hard_smp_processor_id(cpu);
 198                 smp_mb();
 199                 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 200                 return true;
 201         }
 202
 203         /* On POWER8 for IPIs to threads in the same core, use msgsnd */
 204         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
 205                 preempt_disable();
 206                 if (cpu_first_thread_sibling(cpu) ==
 207                     cpu_first_thread_sibling(smp_processor_id())) {
 208                         msg |= cpu_thread_in_core(cpu);
 209                         smp_mb();
 210                         __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 211                         preempt_enable();
 212                         return true;
 213                 }
 214                 preempt_enable();
 215         }
 216
 217 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
 218         if (cpu >= 0 && cpu < nr_cpu_ids) {
 219                 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
 220                         xics_wake_cpu(cpu);
 221                         return true;
 222                 }
 223                 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
 224                 return true;
 225         }
 226 #endif
 227
 228         return false;
 229 }
 230
 231 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
 232 {
 233         int cpu;
 234         struct rcuwait *waitp;
 235
 236         waitp = kvm_arch_vcpu_get_wait(vcpu);
 237         if (rcuwait_wake_up(waitp))
 238                 ++vcpu->stat.halt_wakeup;
 239
 240         cpu = READ_ONCE(vcpu->arch.thread_cpu);
 241         if (cpu >= 0 && kvmppc_ipi_thread(cpu))
 242                 return;
 243
 244         /* CPU points to the first thread of the core */
 245         cpu = vcpu->cpu;
 246         if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
 247                 smp_send_reschedule(cpu);
 248 }
 249
 250 /*
 251  * We use the vcpu_load/put functions to measure stolen time.
 252  * Stolen time is counted as time when either the vcpu is able to
 253  * run as part of a virtual core, but the task running the vcore
 254  * is preempted or sleeping, or when the vcpu needs something done
 255  * in the kernel by the task running the vcpu, but that task is
 256  * preempted or sleeping.  Those two things have to be counted
 257  * separately, since one of the vcpu tasks will take on the job
 258  * of running the core, and the other vcpu tasks in the vcore will
 259  * sleep waiting for it to do that, but that sleep shouldn't count
 260  * as stolen time.
 261  *
 262  * Hence we accumulate stolen time when the vcpu can run as part of
 263  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 264  * needs its task to do other things in the kernel (for example,
 265  * service a page fault) in busy_stolen.  We don't accumulate
 266  * stolen time for a vcore when it is inactive, or for a vcpu
 267  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 268  * a misnomer; it means that the vcpu task is not executing in
 269  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 270  * the kernel.  We don't have any way of dividing up that time
 271  * between time that the vcpu is genuinely stopped, time that
 272  * the task is actively working on behalf of the vcpu, and time
 273  * that the task is preempted, so we don't count any of it as
 274  * stolen.
 275  *
 276  * Updates to busy_stolen are protected by arch.tbacct_lock;
 277  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
 278  * lock.  The stolen times are measured in units of timebase ticks.
 279  * (Note that the != TB_NIL checks below are purely defensive;
 280  * they should never fail.)
 281  */
 282
 283 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
 284 {
 285         unsigned long flags;
 286
 287         spin_lock_irqsave(&vc->stoltb_lock, flags);
 288         vc->preempt_tb = mftb();
 289         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 290 }
 291
 292 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
 293 {
 294         unsigned long flags;
 295
 296         spin_lock_irqsave(&vc->stoltb_lock, flags);
 297         if (vc->preempt_tb != TB_NIL) {
 298                 vc->stolen_tb += mftb() - vc->preempt_tb;
 299                 vc->preempt_tb = TB_NIL;
 300         }
 301         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 302 }
 303
 304 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 305 {
 306         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 307         unsigned long flags;
 308
 309         /*
 310          * We can test vc->runner without taking the vcore lock,
 311          * because only this task ever sets vc->runner to this
 312          * vcpu, and once it is set to this vcpu, only this task
 313          * ever sets it to NULL.
 314          */
 315         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 316                 kvmppc_core_end_stolen(vc);
 317
 318         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 319         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 320             vcpu->arch.busy_preempt != TB_NIL) {
 321                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 322                 vcpu->arch.busy_preempt = TB_NIL;
 323         }
 324         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 325 }
 326
 327 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 328 {
 329         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 330         unsigned long flags;
 331
 332         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 333                 kvmppc_core_start_stolen(vc);
 334
 335         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 336         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 337                 vcpu->arch.busy_preempt = mftb();
 338         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 339 }
 340
 341 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 342 {
 343         vcpu->arch.pvr = pvr;
 344 }
 345
 346 /* Dummy value used in computing PCR value below */
 347 #define PCR_ARCH_31    (PCR_ARCH_300 << 1)
 348
 349 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 350 {
 351         unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
 352         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 353
 354         /* We can (emulate) our own architecture version and anything older */
 355         if (cpu_has_feature(CPU_FTR_ARCH_31))
 356                 host_pcr_bit = PCR_ARCH_31;
 357         else if (cpu_has_feature(CPU_FTR_ARCH_300))
 358                 host_pcr_bit = PCR_ARCH_300;
 359         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
 360                 host_pcr_bit = PCR_ARCH_207;
 361         else if (cpu_has_feature(CPU_FTR_ARCH_206))
 362                 host_pcr_bit = PCR_ARCH_206;
 363         else
 364                 host_pcr_bit = PCR_ARCH_205;
 365
 366         /* Determine lowest PCR bit needed to run guest in given PVR level */
 367         guest_pcr_bit = host_pcr_bit;
 368         if (arch_compat) {
 369                 switch (arch_compat) {
 370                 case PVR_ARCH_205:
 371                         guest_pcr_bit = PCR_ARCH_205;
 372                         break;
 373                 case PVR_ARCH_206:
 374                 case PVR_ARCH_206p:
 375                         guest_pcr_bit = PCR_ARCH_206;
 376                         break;
 377                 case PVR_ARCH_207:
 378                         guest_pcr_bit = PCR_ARCH_207;
 379                         break;
 380                 case PVR_ARCH_300:
 381                         guest_pcr_bit = PCR_ARCH_300;
 382                         break;
 383                 case PVR_ARCH_31:
 384                         guest_pcr_bit = PCR_ARCH_31;
 385                         break;
 386                 default:
 387                         return -EINVAL;
 388                 }
 389         }
 390
 391         /* Check requested PCR bits don't exceed our capabilities */
 392         if (guest_pcr_bit > host_pcr_bit)
 393                 return -EINVAL;
 394
 395         spin_lock(&vc->lock);
 396         vc->arch_compat = arch_compat;
 397         /*
 398          * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
 399          * Also set all reserved PCR bits
 400          */
 401         vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
 402         spin_unlock(&vc->lock);
 403
 404         return 0;
 405 }
 406
 407 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 408 {
 409         int r;
 410
 411         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 412         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 413                vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
 414         for (r = 0; r < 16; ++r)
 415                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 416                        r, kvmppc_get_gpr(vcpu, r),
 417                        r+16, kvmppc_get_gpr(vcpu, r+16));
 418         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 419                vcpu->arch.regs.ctr, vcpu->arch.regs.link);
 420         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 421                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 422         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 423                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 424         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 425                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 426         pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
 427                vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
 428         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 429         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 430                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 431         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 432         for (r = 0; r < vcpu->arch.slb_max; ++r)
 433                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 434                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 435         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 436                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 437                vcpu->arch.last_inst);
 438 }
 439
 440 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 441 {
 442         return kvm_get_vcpu_by_id(kvm, id);
 443 }
 444
 445 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 446 {
 447         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 448         vpa->yield_count = cpu_to_be32(1);
 449 }
 450
 451 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 452                    unsigned long addr, unsigned long len)
 453 {
 454         /* check address is cacheline aligned */
 455         if (addr & (L1_CACHE_BYTES - 1))
 456                 return -EINVAL;
 457         spin_lock(&vcpu->arch.vpa_update_lock);
 458         if (v->next_gpa != addr || v->len != len) {
 459                 v->next_gpa = addr;
 460                 v->len = addr ? len : 0;
 461                 v->update_pending = 1;
 462         }
 463         spin_unlock(&vcpu->arch.vpa_update_lock);
 464         return 0;
 465 }
 466
 467 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 468 struct reg_vpa {
 469         u32 dummy;
 470         union {
 471                 __be16 hword;
 472                 __be32 word;
 473         } length;
 474 };
 475
 476 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 477 {
 478         if (vpap->update_pending)
 479                 return vpap->next_gpa != 0;
 480         return vpap->pinned_addr != NULL;
 481 }
 482
 483 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 484                                        unsigned long flags,
 485                                        unsigned long vcpuid, unsigned long vpa)
 486 {
 487         struct kvm *kvm = vcpu->kvm;
 488         unsigned long len, nb;
 489         void *va;
 490         struct kvm_vcpu *tvcpu;
 491         int err;
 492         int subfunc;
 493         struct kvmppc_vpa *vpap;
 494
 495         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 496         if (!tvcpu)
 497                 return H_PARAMETER;
 498
 499         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 500         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 501             subfunc == H_VPA_REG_SLB) {
 502                 /* Registering new area - address must be cache-line aligned */
 503                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 504                         return H_PARAMETER;
 505
 506                 /* convert logical addr to kernel addr and read length */
 507                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 508                 if (va == NULL)
 509                         return H_PARAMETER;
 510                 if (subfunc == H_VPA_REG_VPA)
 511                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 512                 else
 513                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 514                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 515
 516                 /* Check length */
 517                 if (len > nb || len < sizeof(struct reg_vpa))
 518                         return H_PARAMETER;
 519         } else {
 520                 vpa = 0;
 521                 len = 0;
 522         }
 523
 524         err = H_PARAMETER;
 525         vpap = NULL;
 526         spin_lock(&tvcpu->arch.vpa_update_lock);
 527
 528         switch (subfunc) {
 529         case H_VPA_REG_VPA:             /* register VPA */
 530                 /*
 531                  * The size of our lppaca is 1kB because of the way we align
 532                  * it for the guest to avoid crossing a 4kB boundary. We only
 533                  * use 640 bytes of the structure though, so we should accept
 534                  * clients that set a size of 640.
 535                  */
 536                 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
 537                 if (len < sizeof(struct lppaca))
 538                         break;
 539                 vpap = &tvcpu->arch.vpa;
 540                 err = 0;
 541                 break;
 542
 543         case H_VPA_REG_DTL:             /* register DTL */
 544                 if (len < sizeof(struct dtl_entry))
 545                         break;
 546                 len -= len % sizeof(struct dtl_entry);
 547
 548                 /* Check that they have previously registered a VPA */
 549                 err = H_RESOURCE;
 550                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 551                         break;
 552
 553                 vpap = &tvcpu->arch.dtl;
 554                 err = 0;
 555                 break;
 556
 557         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 558                 /* Check that they have previously registered a VPA */
 559                 err = H_RESOURCE;
 560                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 561                         break;
 562
 563                 vpap = &tvcpu->arch.slb_shadow;
 564                 err = 0;
 565                 break;
 566
 567         case H_VPA_DEREG_VPA:           /* deregister VPA */
 568                 /* Check they don't still have a DTL or SLB buf registered */
 569                 err = H_RESOURCE;
 570                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 571                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 572                         break;
 573
 574                 vpap = &tvcpu->arch.vpa;
 575                 err = 0;
 576                 break;
 577
 578         case H_VPA_DEREG_DTL:           /* deregister DTL */
 579                 vpap = &tvcpu->arch.dtl;
 580                 err = 0;
 581                 break;
 582
 583         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 584                 vpap = &tvcpu->arch.slb_shadow;
 585                 err = 0;
 586                 break;
 587         }
 588
 589         if (vpap) {
 590                 vpap->next_gpa = vpa;
 591                 vpap->len = len;
 592                 vpap->update_pending = 1;
 593         }
 594
 595         spin_unlock(&tvcpu->arch.vpa_update_lock);
 596
 597         return err;
 598 }
 599
 600 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 601 {
 602         struct kvm *kvm = vcpu->kvm;
 603         void *va;
 604         unsigned long nb;
 605         unsigned long gpa;
 606
 607         /*
 608          * We need to pin the page pointed to by vpap->next_gpa,
 609          * but we can't call kvmppc_pin_guest_page under the lock
 610          * as it does get_user_pages() and down_read().  So we
 611          * have to drop the lock, pin the page, then get the lock
 612          * again and check that a new area didn't get registered
 613          * in the meantime.
 614          */
 615         for (;;) {
 616                 gpa = vpap->next_gpa;
 617                 spin_unlock(&vcpu->arch.vpa_update_lock);
 618                 va = NULL;
 619                 nb = 0;
 620                 if (gpa)
 621                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 622                 spin_lock(&vcpu->arch.vpa_update_lock);
 623                 if (gpa == vpap->next_gpa)
 624                         break;
 625                 /* sigh... unpin that one and try again */
 626                 if (va)
 627                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 628         }
 629
 630         vpap->update_pending = 0;
 631         if (va && nb < vpap->len) {
 632                 /*
 633                  * If it's now too short, it must be that userspace
 634                  * has changed the mappings underlying guest memory,
 635                  * so unregister the region.
 636                  */
 637                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 638                 va = NULL;
 639         }
 640         if (vpap->pinned_addr)
 641                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 642                                         vpap->dirty);
 643         vpap->gpa = gpa;
 644         vpap->pinned_addr = va;
 645         vpap->dirty = false;
 646         if (va)
 647                 vpap->pinned_end = va + vpap->len;
 648 }
 649
 650 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 651 {
 652         if (!(vcpu->arch.vpa.update_pending ||
 653               vcpu->arch.slb_shadow.update_pending ||
 654               vcpu->arch.dtl.update_pending))
 655                 return;
 656
 657         spin_lock(&vcpu->arch.vpa_update_lock);
 658         if (vcpu->arch.vpa.update_pending) {
 659                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 660                 if (vcpu->arch.vpa.pinned_addr)
 661                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 662         }
 663         if (vcpu->arch.dtl.update_pending) {
 664                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 665                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 666                 vcpu->arch.dtl_index = 0;
 667         }
 668         if (vcpu->arch.slb_shadow.update_pending)
 669                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 670         spin_unlock(&vcpu->arch.vpa_update_lock);
 671 }
 672
 673 /*
 674  * Return the accumulated stolen time for the vcore up until `now'.
 675  * The caller should hold the vcore lock.
 676  */
 677 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 678 {
 679         u64 p;
 680         unsigned long flags;
 681
 682         spin_lock_irqsave(&vc->stoltb_lock, flags);
 683         p = vc->stolen_tb;
 684         if (vc->vcore_state != VCORE_INACTIVE &&
 685             vc->preempt_tb != TB_NIL)
 686                 p += now - vc->preempt_tb;
 687         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 688         return p;
 689 }
 690
 691 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 692                                     struct kvmppc_vcore *vc)
 693 {
 694         struct dtl_entry *dt;
 695         struct lppaca *vpa;
 696         unsigned long stolen;
 697         unsigned long core_stolen;
 698         u64 now;
 699         unsigned long flags;
 700
 701         dt = vcpu->arch.dtl_ptr;
 702         vpa = vcpu->arch.vpa.pinned_addr;
 703         now = mftb();
 704         core_stolen = vcore_stolen_time(vc, now);
 705         stolen = core_stolen - vcpu->arch.stolen_logged;
 706         vcpu->arch.stolen_logged = core_stolen;
 707         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 708         stolen += vcpu->arch.busy_stolen;
 709         vcpu->arch.busy_stolen = 0;
 710         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 711         if (!dt || !vpa)
 712                 return;
 713         memset(dt, 0, sizeof(struct dtl_entry));
 714         dt->dispatch_reason = 7;
 715         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 716         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 717         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 718         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 719         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 720         ++dt;
 721         if (dt == vcpu->arch.dtl.pinned_end)
 722                 dt = vcpu->arch.dtl.pinned_addr;
 723         vcpu->arch.dtl_ptr = dt;
 724         /* order writing *dt vs. writing vpa->dtl_idx */
 725         smp_wmb();
 726         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 727         vcpu->arch.dtl.dirty = true;
 728 }
 729
 730 /* See if there is a doorbell interrupt pending for a vcpu */
 731 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
 732 {
 733         int thr;
 734         struct kvmppc_vcore *vc;
 735
 736         if (vcpu->arch.doorbell_request)
 737                 return true;
 738         /*
 739          * Ensure that the read of vcore->dpdes comes after the read
 740          * of vcpu->doorbell_request.  This barrier matches the
 741          * smp_wmb() in kvmppc_guest_entry_inject().
 742          */
 743         smp_rmb();
 744         vc = vcpu->arch.vcore;
 745         thr = vcpu->vcpu_id - vc->first_vcpuid;
 746         return !!(vc->dpdes & (1 << thr));
 747 }
 748
 749 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 750 {
 751         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 752                 return true;
 753         if ((!vcpu->arch.vcore->arch_compat) &&
 754             cpu_has_feature(CPU_FTR_ARCH_207S))
 755                 return true;
 756         return false;
 757 }
 758
 759 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 760                              unsigned long resource, unsigned long value1,
 761                              unsigned long value2)
 762 {
 763         switch (resource) {
 764         case H_SET_MODE_RESOURCE_SET_CIABR:
 765                 if (!kvmppc_power8_compatible(vcpu))
 766                         return H_P2;
 767                 if (value2)
 768                         return H_P4;
 769                 if (mflags)
 770                         return H_UNSUPPORTED_FLAG_START;
 771                 /* Guests can't breakpoint the hypervisor */
 772                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 773                         return H_P3;
 774                 vcpu->arch.ciabr  = value1;
 775                 return H_SUCCESS;
 776         case H_SET_MODE_RESOURCE_SET_DAWR0:
 777                 if (!kvmppc_power8_compatible(vcpu))
 778                         return H_P2;
 779                 if (!ppc_breakpoint_available())
 780                         return H_P2;
 781                 if (mflags)
 782                         return H_UNSUPPORTED_FLAG_START;
 783                 if (value2 & DABRX_HYP)
 784                         return H_P4;
 785                 vcpu->arch.dawr0  = value1;
 786                 vcpu->arch.dawrx0 = value2;
 787                 return H_SUCCESS;
 788         case H_SET_MODE_RESOURCE_SET_DAWR1:
 789                 if (!kvmppc_power8_compatible(vcpu))
 790                         return H_P2;
 791                 if (!ppc_breakpoint_available())
 792                         return H_P2;
 793                 if (!cpu_has_feature(CPU_FTR_DAWR1))
 794                         return H_P2;
 795                 if (!vcpu->kvm->arch.dawr1_enabled)
 796                         return H_FUNCTION;
 797                 if (mflags)
 798                         return H_UNSUPPORTED_FLAG_START;
 799                 if (value2 & DABRX_HYP)
 800                         return H_P4;
 801                 vcpu->arch.dawr1  = value1;
 802                 vcpu->arch.dawrx1 = value2;
 803                 return H_SUCCESS;
 804         case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
 805                 /* KVM does not support mflags=2 (AIL=2) */
 806                 if (mflags != 0 && mflags != 3)
 807                         return H_UNSUPPORTED_FLAG_START;
 808                 return H_TOO_HARD;
 809         default:
 810                 return H_TOO_HARD;
 811         }
 812 }
 813
 814 /* Copy guest memory in place - must reside within a single memslot */
 815 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
 816                                   unsigned long len)
 817 {
 818         struct kvm_memory_slot *to_memslot = NULL;
 819         struct kvm_memory_slot *from_memslot = NULL;
 820         unsigned long to_addr, from_addr;
 821         int r;
 822
 823         /* Get HPA for from address */
 824         from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
 825         if (!from_memslot)
 826                 return -EFAULT;
 827         if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
 828                              << PAGE_SHIFT))
 829                 return -EINVAL;
 830         from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
 831         if (kvm_is_error_hva(from_addr))
 832                 return -EFAULT;
 833         from_addr |= (from & (PAGE_SIZE - 1));
 834
 835         /* Get HPA for to address */
 836         to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
 837         if (!to_memslot)
 838                 return -EFAULT;
 839         if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
 840                            << PAGE_SHIFT))
 841                 return -EINVAL;
 842         to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
 843         if (kvm_is_error_hva(to_addr))
 844                 return -EFAULT;
 845         to_addr |= (to & (PAGE_SIZE - 1));
 846
 847         /* Perform copy */
 848         r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
 849                              len);
 850         if (r)
 851                 return -EFAULT;
 852         mark_page_dirty(kvm, to >> PAGE_SHIFT);
 853         return 0;
 854 }
 855
 856 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
 857                                unsigned long dest, unsigned long src)
 858 {
 859         u64 pg_sz = SZ_4K;              /* 4K page size */
 860         u64 pg_mask = SZ_4K - 1;
 861         int ret;
 862
 863         /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
 864         if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
 865                       H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
 866                 return H_PARAMETER;
 867
 868         /* dest (and src if copy_page flag set) must be page aligned */
 869         if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
 870                 return H_PARAMETER;
 871
 872         /* zero and/or copy the page as determined by the flags */
 873         if (flags & H_COPY_PAGE) {
 874                 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
 875                 if (ret < 0)
 876                         return H_PARAMETER;
 877         } else if (flags & H_ZERO_PAGE) {
 878                 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
 879                 if (ret < 0)
 880                         return H_PARAMETER;
 881         }
 882
 883         /* We can ignore the remaining flags */
 884
 885         return H_SUCCESS;
 886 }
 887
 888 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 889 {
 890         struct kvmppc_vcore *vcore = target->arch.vcore;
 891
 892         /*
 893          * We expect to have been called by the real mode handler
 894          * (kvmppc_rm_h_confer()) which would have directly returned
 895          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 896          * have useful work to do and should not confer) so we don't
 897          * recheck that here.
 898          */
 899
 900         spin_lock(&vcore->lock);
 901         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 902             vcore->vcore_state != VCORE_INACTIVE &&
 903             vcore->runner)
 904                 target = vcore->runner;
 905         spin_unlock(&vcore->lock);
 906
 907         return kvm_vcpu_yield_to(target);
 908 }
 909
 910 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 911 {
 912         int yield_count = 0;
 913         struct lppaca *lppaca;
 914
 915         spin_lock(&vcpu->arch.vpa_update_lock);
 916         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 917         if (lppaca)
 918                 yield_count = be32_to_cpu(lppaca->yield_count);
 919         spin_unlock(&vcpu->arch.vpa_update_lock);
 920         return yield_count;
 921 }
 922
 923 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 924 {
 925         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 926         unsigned long target, ret = H_SUCCESS;
 927         int yield_count;
 928         struct kvm_vcpu *tvcpu;
 929         int idx, rc;
 930
 931         if (req <= MAX_HCALL_OPCODE &&
 932             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 933                 return RESUME_HOST;
 934
 935         switch (req) {
 936         case H_CEDE:
 937                 break;
 938         case H_PROD:
 939                 target = kvmppc_get_gpr(vcpu, 4);
 940                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 941                 if (!tvcpu) {
 942                         ret = H_PARAMETER;
 943                         break;
 944                 }
 945                 tvcpu->arch.prodded = 1;
 946                 smp_mb();
 947                 if (tvcpu->arch.ceded)
 948                         kvmppc_fast_vcpu_kick_hv(tvcpu);
 949                 break;
 950         case H_CONFER:
 951                 target = kvmppc_get_gpr(vcpu, 4);
 952                 if (target == -1)
 953                         break;
 954                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 955                 if (!tvcpu) {
 956                         ret = H_PARAMETER;
 957                         break;
 958                 }
 959                 yield_count = kvmppc_get_gpr(vcpu, 5);
 960                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 961                         break;
 962                 kvm_arch_vcpu_yield_to(tvcpu);
 963                 break;
 964         case H_REGISTER_VPA:
 965                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 966                                         kvmppc_get_gpr(vcpu, 5),
 967                                         kvmppc_get_gpr(vcpu, 6));
 968                 break;
 969         case H_RTAS:
 970                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 971                         return RESUME_HOST;
 972
 973                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 974                 rc = kvmppc_rtas_hcall(vcpu);
 975                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 976
 977                 if (rc == -ENOENT)
 978                         return RESUME_HOST;
 979                 else if (rc == 0)
 980                         break;
 981
 982                 /* Send the error out to userspace via KVM_RUN */
 983                 return rc;
 984         case H_LOGICAL_CI_LOAD:
 985                 ret = kvmppc_h_logical_ci_load(vcpu);
 986                 if (ret == H_TOO_HARD)
 987                         return RESUME_HOST;
 988                 break;
 989         case H_LOGICAL_CI_STORE:
 990                 ret = kvmppc_h_logical_ci_store(vcpu);
 991                 if (ret == H_TOO_HARD)
 992                         return RESUME_HOST;
 993                 break;
 994         case H_SET_MODE:
 995                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 996                                         kvmppc_get_gpr(vcpu, 5),
 997                                         kvmppc_get_gpr(vcpu, 6),
 998                                         kvmppc_get_gpr(vcpu, 7));
 999                 if (ret == H_TOO_HARD)
1000                         return RESUME_HOST;
1001                 break;
1002         case H_XIRR:
1003         case H_CPPR:
1004         case H_EOI:
1005         case H_IPI:
1006         case H_IPOLL:
1007         case H_XIRR_X:
1008                 if (kvmppc_xics_enabled(vcpu)) {
1009                         if (xics_on_xive()) {
1010                                 ret = H_NOT_AVAILABLE;
1011                                 return RESUME_GUEST;
1012                         }
1013                         ret = kvmppc_xics_hcall(vcpu, req);
1014                         break;
1015                 }
1016                 return RESUME_HOST;
1017         case H_SET_DABR:
1018                 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1019                 break;
1020         case H_SET_XDABR:
1021                 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1022                                                 kvmppc_get_gpr(vcpu, 5));
1023                 break;
1024 #ifdef CONFIG_SPAPR_TCE_IOMMU
1025         case H_GET_TCE:
1026                 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1027                                                 kvmppc_get_gpr(vcpu, 5));
1028                 if (ret == H_TOO_HARD)
1029                         return RESUME_HOST;
1030                 break;
1031         case H_PUT_TCE:
1032                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1033                                                 kvmppc_get_gpr(vcpu, 5),
1034                                                 kvmppc_get_gpr(vcpu, 6));
1035                 if (ret == H_TOO_HARD)
1036                         return RESUME_HOST;
1037                 break;
1038         case H_PUT_TCE_INDIRECT:
1039                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1040                                                 kvmppc_get_gpr(vcpu, 5),
1041                                                 kvmppc_get_gpr(vcpu, 6),
1042                                                 kvmppc_get_gpr(vcpu, 7));
1043                 if (ret == H_TOO_HARD)
1044                         return RESUME_HOST;
1045                 break;
1046         case H_STUFF_TCE:
1047                 ret = kvmppc_h_stuff_tce(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 #endif
1055         case H_RANDOM:
1056                 if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1057                         ret = H_HARDWARE;
1058                 break;
1059
1060         case H_SET_PARTITION_TABLE:
1061                 ret = H_FUNCTION;
1062                 if (nesting_enabled(vcpu->kvm))
1063                         ret = kvmhv_set_partition_table(vcpu);
1064                 break;
1065         case H_ENTER_NESTED:
1066                 ret = H_FUNCTION;
1067                 if (!nesting_enabled(vcpu->kvm))
1068                         break;
1069                 ret = kvmhv_enter_nested_guest(vcpu);
1070                 if (ret == H_INTERRUPT) {
1071                         kvmppc_set_gpr(vcpu, 3, 0);
1072                         vcpu->arch.hcall_needed = 0;
1073                         return -EINTR;
1074                 } else if (ret == H_TOO_HARD) {
1075                         kvmppc_set_gpr(vcpu, 3, 0);
1076                         vcpu->arch.hcall_needed = 0;
1077                         return RESUME_HOST;
1078                 }
1079                 break;
1080         case H_TLB_INVALIDATE:
1081                 ret = H_FUNCTION;
1082                 if (nesting_enabled(vcpu->kvm))
1083                         ret = kvmhv_do_nested_tlbie(vcpu);
1084                 break;
1085         case H_COPY_TOFROM_GUEST:
1086                 ret = H_FUNCTION;
1087                 if (nesting_enabled(vcpu->kvm))
1088                         ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1089                 break;
1090         case H_PAGE_INIT:
1091                 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1092                                          kvmppc_get_gpr(vcpu, 5),
1093                                          kvmppc_get_gpr(vcpu, 6));
1094                 break;
1095         case H_SVM_PAGE_IN:
1096                 ret = H_UNSUPPORTED;
1097                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1098                         ret = kvmppc_h_svm_page_in(vcpu->kvm,
1099                                                    kvmppc_get_gpr(vcpu, 4),
1100                                                    kvmppc_get_gpr(vcpu, 5),
1101                                                    kvmppc_get_gpr(vcpu, 6));
1102                 break;
1103         case H_SVM_PAGE_OUT:
1104                 ret = H_UNSUPPORTED;
1105                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1106                         ret = kvmppc_h_svm_page_out(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_INIT_START:
1112                 ret = H_UNSUPPORTED;
1113                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1114                         ret = kvmppc_h_svm_init_start(vcpu->kvm);
1115                 break;
1116         case H_SVM_INIT_DONE:
1117                 ret = H_UNSUPPORTED;
1118                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1119                         ret = kvmppc_h_svm_init_done(vcpu->kvm);
1120                 break;
1121         case H_SVM_INIT_ABORT:
1122                 /*
1123                  * Even if that call is made by the Ultravisor, the SSR1 value
1124                  * is the guest context one, with the secure bit clear as it has
1125                  * not yet been secured. So we can't check it here.
1126                  * Instead the kvm->arch.secure_guest flag is checked inside
1127                  * kvmppc_h_svm_init_abort().
1128                  */
1129                 ret = kvmppc_h_svm_init_abort(vcpu->kvm);
1130                 break;
1131
1132         default:
1133                 return RESUME_HOST;
1134         }
1135         kvmppc_set_gpr(vcpu, 3, ret);
1136         vcpu->arch.hcall_needed = 0;
1137         return RESUME_GUEST;
1138 }
1139
1140 /*
1141  * Handle H_CEDE in the nested virtualization case where we haven't
1142  * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1143  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1144  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1145  */
1146 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1147 {
1148         vcpu->arch.shregs.msr |= MSR_EE;
1149         vcpu->arch.ceded = 1;
1150         smp_mb();
1151         if (vcpu->arch.prodded) {
1152                 vcpu->arch.prodded = 0;
1153                 smp_mb();
1154                 vcpu->arch.ceded = 0;
1155         }
1156 }
1157
1158 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1159 {
1160         switch (cmd) {
1161         case H_CEDE:
1162         case H_PROD:
1163         case H_CONFER:
1164         case H_REGISTER_VPA:
1165         case H_SET_MODE:
1166         case H_LOGICAL_CI_LOAD:
1167         case H_LOGICAL_CI_STORE:
1168 #ifdef CONFIG_KVM_XICS
1169         case H_XIRR:
1170         case H_CPPR:
1171         case H_EOI:
1172         case H_IPI:
1173         case H_IPOLL:
1174         case H_XIRR_X:
1175 #endif
1176         case H_PAGE_INIT:
1177                 return 1;
1178         }
1179
1180         /* See if it's in the real-mode table */
1181         return kvmppc_hcall_impl_hv_realmode(cmd);
1182 }
1183
1184 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1185 {
1186         u32 last_inst;
1187
1188         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1189                                         EMULATE_DONE) {
1190                 /*
1191                  * Fetch failed, so return to guest and
1192                  * try executing it again.
1193                  */
1194                 return RESUME_GUEST;
1195         }
1196
1197         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1198                 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1199                 vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1200                 return RESUME_HOST;
1201         } else {
1202                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1203                 return RESUME_GUEST;
1204         }
1205 }
1206
1207 static void do_nothing(void *x)
1208 {
1209 }
1210
1211 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1212 {
1213         int thr, cpu, pcpu, nthreads;
1214         struct kvm_vcpu *v;
1215         unsigned long dpdes;
1216
1217         nthreads = vcpu->kvm->arch.emul_smt_mode;
1218         dpdes = 0;
1219         cpu = vcpu->vcpu_id & ~(nthreads - 1);
1220         for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1221                 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1222                 if (!v)
1223                         continue;
1224                 /*
1225                  * If the vcpu is currently running on a physical cpu thread,
1226                  * interrupt it in order to pull it out of the guest briefly,
1227                  * which will update its vcore->dpdes value.
1228                  */
1229                 pcpu = READ_ONCE(v->cpu);
1230                 if (pcpu >= 0)
1231                         smp_call_function_single(pcpu, do_nothing, NULL, 1);
1232                 if (kvmppc_doorbell_pending(v))
1233                         dpdes |= 1 << thr;
1234         }
1235         return dpdes;
1236 }
1237
1238 /*
1239  * On POWER9, emulate doorbell-related instructions in order to
1240  * give the guest the illusion of running on a multi-threaded core.
1241  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1242  * and mfspr DPDES.
1243  */
1244 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1245 {
1246         u32 inst, rb, thr;
1247         unsigned long arg;
1248         struct kvm *kvm = vcpu->kvm;
1249         struct kvm_vcpu *tvcpu;
1250
1251         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1252                 return RESUME_GUEST;
1253         if (get_op(inst) != 31)
1254                 return EMULATE_FAIL;
1255         rb = get_rb(inst);
1256         thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1257         switch (get_xop(inst)) {
1258         case OP_31_XOP_MSGSNDP:
1259                 arg = kvmppc_get_gpr(vcpu, rb);
1260                 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1261                         break;
1262                 arg &= 0x7f;
1263                 if (arg >= kvm->arch.emul_smt_mode)
1264                         break;
1265                 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1266                 if (!tvcpu)
1267                         break;
1268                 if (!tvcpu->arch.doorbell_request) {
1269                         tvcpu->arch.doorbell_request = 1;
1270                         kvmppc_fast_vcpu_kick_hv(tvcpu);
1271                 }
1272                 break;
1273         case OP_31_XOP_MSGCLRP:
1274                 arg = kvmppc_get_gpr(vcpu, rb);
1275                 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1276                         break;
1277                 vcpu->arch.vcore->dpdes = 0;
1278                 vcpu->arch.doorbell_request = 0;
1279                 break;
1280         case OP_31_XOP_MFSPR:
1281                 switch (get_sprn(inst)) {
1282                 case SPRN_TIR:
1283                         arg = thr;
1284                         break;
1285                 case SPRN_DPDES:
1286                         arg = kvmppc_read_dpdes(vcpu);
1287                         break;
1288                 default:
1289                         return EMULATE_FAIL;
1290                 }
1291                 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1292                 break;
1293         default:
1294                 return EMULATE_FAIL;
1295         }
1296         kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1297         return RESUME_GUEST;
1298 }
1299
1300 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1301                                  struct task_struct *tsk)
1302 {
1303         struct kvm_run *run = vcpu->run;
1304         int r = RESUME_HOST;
1305
1306         vcpu->stat.sum_exits++;
1307
1308         /*
1309          * This can happen if an interrupt occurs in the last stages
1310          * of guest entry or the first stages of guest exit (i.e. after
1311          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1312          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1313          * That can happen due to a bug, or due to a machine check
1314          * occurring at just the wrong time.
1315          */
1316         if (vcpu->arch.shregs.msr & MSR_HV) {
1317                 printk(KERN_EMERG "KVM trap in HV mode!\n");
1318                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1319                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1320                         vcpu->arch.shregs.msr);
1321                 kvmppc_dump_regs(vcpu);
1322                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1323                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1324                 return RESUME_HOST;
1325         }
1326         run->exit_reason = KVM_EXIT_UNKNOWN;
1327         run->ready_for_interrupt_injection = 1;
1328         switch (vcpu->arch.trap) {
1329         /* We're good on these - the host merely wanted to get our attention */
1330         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1331                 vcpu->stat.dec_exits++;
1332                 r = RESUME_GUEST;
1333                 break;
1334         case BOOK3S_INTERRUPT_EXTERNAL:
1335         case BOOK3S_INTERRUPT_H_DOORBELL:
1336         case BOOK3S_INTERRUPT_H_VIRT:
1337                 vcpu->stat.ext_intr_exits++;
1338                 r = RESUME_GUEST;
1339                 break;
1340         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1341         case BOOK3S_INTERRUPT_HMI:
1342         case BOOK3S_INTERRUPT_PERFMON:
1343         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1344                 r = RESUME_GUEST;
1345                 break;
1346         case BOOK3S_INTERRUPT_MACHINE_CHECK: {
1347                 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1348                                               DEFAULT_RATELIMIT_BURST);
1349                 /*
1350                  * Print the MCE event to host console. Ratelimit so the guest
1351                  * can't flood the host log.
1352                  */
1353                 if (__ratelimit(&rs))
1354                         machine_check_print_event_info(&vcpu->arch.mce_evt,false, true);
1355
1356                 /*
1357                  * If the guest can do FWNMI, exit to userspace so it can
1358                  * deliver a FWNMI to the guest.
1359                  * Otherwise we synthesize a machine check for the guest
1360                  * so that it knows that the machine check occurred.
1361                  */
1362                 if (!vcpu->kvm->arch.fwnmi_enabled) {
1363                         ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1364                         kvmppc_core_queue_machine_check(vcpu, flags);
1365                         r = RESUME_GUEST;
1366                         break;
1367                 }
1368
1369                 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1370                 run->exit_reason = KVM_EXIT_NMI;
1371                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1372                 /* Clear out the old NMI status from run->flags */
1373                 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1374                 /* Now set the NMI status */
1375                 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1376                         run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1377                 else
1378                         run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1379
1380                 r = RESUME_HOST;
1381                 break;
1382         }
1383         case BOOK3S_INTERRUPT_PROGRAM:
1384         {
1385                 ulong flags;
1386                 /*
1387                  * Normally program interrupts are delivered directly
1388                  * to the guest by the hardware, but we can get here
1389                  * as a result of a hypervisor emulation interrupt
1390                  * (e40) getting turned into a 700 by BML RTAS.
1391                  */
1392                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1393                 kvmppc_core_queue_program(vcpu, flags);
1394                 r = RESUME_GUEST;
1395                 break;
1396         }
1397         case BOOK3S_INTERRUPT_SYSCALL:
1398         {
1399                 /* hcall - punt to userspace */
1400                 int i;
1401
1402                 /* hypercall with MSR_PR has already been handled in rmode,
1403                  * and never reaches here.
1404                  */
1405
1406                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1407                 for (i = 0; i < 9; ++i)
1408                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1409                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1410                 vcpu->arch.hcall_needed = 1;
1411                 r = RESUME_HOST;
1412                 break;
1413         }
1414         /*
1415          * We get these next two if the guest accesses a page which it thinks
1416          * it has mapped but which is not actually present, either because
1417          * it is for an emulated I/O device or because the corresonding
1418          * host page has been paged out.  Any other HDSI/HISI interrupts
1419          * have been handled already.
1420          */
1421         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1422                 r = RESUME_PAGE_FAULT;
1423                 break;
1424         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1425                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1426                 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1427                         DSISR_SRR1_MATCH_64S;
1428                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1429                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1430                 r = RESUME_PAGE_FAULT;
1431                 break;
1432         /*
1433          * This occurs if the guest executes an illegal instruction.
1434          * If the guest debug is disabled, generate a program interrupt
1435          * to the guest. If guest debug is enabled, we need to check
1436          * whether the instruction is a software breakpoint instruction.
1437          * Accordingly return to Guest or Host.
1438          */
1439         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1440                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1441                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1442                                 swab32(vcpu->arch.emul_inst) :
1443                                 vcpu->arch.emul_inst;
1444                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1445                         r = kvmppc_emulate_debug_inst(vcpu);
1446                 } else {
1447                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1448                         r = RESUME_GUEST;
1449                 }
1450                 break;
1451         /*
1452          * This occurs if the guest (kernel or userspace), does something that
1453          * is prohibited by HFSCR.
1454          * On POWER9, this could be a doorbell instruction that we need
1455          * to emulate.
1456          * Otherwise, we just generate a program interrupt to the guest.
1457          */
1458         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1459                 r = EMULATE_FAIL;
1460                 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1461                     cpu_has_feature(CPU_FTR_ARCH_300))
1462                         r = kvmppc_emulate_doorbell_instr(vcpu);
1463                 if (r == EMULATE_FAIL) {
1464                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1465                         r = RESUME_GUEST;
1466                 }
1467                 break;
1468
1469 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1470         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1471                 /*
1472                  * This occurs for various TM-related instructions that
1473                  * we need to emulate on POWER9 DD2.2.  We have already
1474                  * handled the cases where the guest was in real-suspend
1475                  * mode and was transitioning to transactional state.
1476                  */
1477                 r = kvmhv_p9_tm_emulation(vcpu);
1478                 break;
1479 #endif
1480
1481         case BOOK3S_INTERRUPT_HV_RM_HARD:
1482                 r = RESUME_PASSTHROUGH;
1483                 break;
1484         default:
1485                 kvmppc_dump_regs(vcpu);
1486                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1487                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1488                         vcpu->arch.shregs.msr);
1489                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1490                 r = RESUME_HOST;
1491                 break;
1492         }
1493
1494         return r;
1495 }
1496
1497 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1498 {
1499         int r;
1500         int srcu_idx;
1501
1502         vcpu->stat.sum_exits++;
1503
1504         /*
1505          * This can happen if an interrupt occurs in the last stages
1506          * of guest entry or the first stages of guest exit (i.e. after
1507          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1508          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1509          * That can happen due to a bug, or due to a machine check
1510          * occurring at just the wrong time.
1511          */
1512         if (vcpu->arch.shregs.msr & MSR_HV) {
1513                 pr_emerg("KVM trap in HV mode while nested!\n");
1514                 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1515                          vcpu->arch.trap, kvmppc_get_pc(vcpu),
1516                          vcpu->arch.shregs.msr);
1517                 kvmppc_dump_regs(vcpu);
1518                 return RESUME_HOST;
1519         }
1520         switch (vcpu->arch.trap) {
1521         /* We're good on these - the host merely wanted to get our attention */
1522         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1523                 vcpu->stat.dec_exits++;
1524                 r = RESUME_GUEST;
1525                 break;
1526         case BOOK3S_INTERRUPT_EXTERNAL:
1527                 vcpu->stat.ext_intr_exits++;
1528                 r = RESUME_HOST;
1529                 break;
1530         case BOOK3S_INTERRUPT_H_DOORBELL:
1531         case BOOK3S_INTERRUPT_H_VIRT:
1532                 vcpu->stat.ext_intr_exits++;
1533                 r = RESUME_GUEST;
1534                 break;
1535         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1536         case BOOK3S_INTERRUPT_HMI:
1537         case BOOK3S_INTERRUPT_PERFMON:
1538         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1539                 r = RESUME_GUEST;
1540                 break;
1541         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1542         {
1543                 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1544                                               DEFAULT_RATELIMIT_BURST);
1545                 /* Pass the machine check to the L1 guest */
1546                 r = RESUME_HOST;
1547                 /* Print the MCE event to host console. */
1548                 if (__ratelimit(&rs))
1549                         machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1550                 break;
1551         }
1552         /*
1553          * We get these next two if the guest accesses a page which it thinks
1554          * it has mapped but which is not actually present, either because
1555          * it is for an emulated I/O device or because the corresonding
1556          * host page has been paged out.
1557          */
1558         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1559                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1560                 r = kvmhv_nested_page_fault(vcpu);
1561                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1562                 break;
1563         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1564                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1565                 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1566                                          DSISR_SRR1_MATCH_64S;
1567                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1568                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1569                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1570                 r = kvmhv_nested_page_fault(vcpu);
1571                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1572                 break;
1573
1574 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1575         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1576                 /*
1577                  * This occurs for various TM-related instructions that
1578                  * we need to emulate on POWER9 DD2.2.  We have already
1579                  * handled the cases where the guest was in real-suspend
1580                  * mode and was transitioning to transactional state.
1581                  */
1582                 r = kvmhv_p9_tm_emulation(vcpu);
1583                 break;
1584 #endif
1585
1586         case BOOK3S_INTERRUPT_HV_RM_HARD:
1587                 vcpu->arch.trap = 0;
1588                 r = RESUME_GUEST;
1589                 if (!xics_on_xive())
1590                         kvmppc_xics_rm_complete(vcpu, 0);
1591                 break;
1592         default:
1593                 r = RESUME_HOST;
1594                 break;
1595         }
1596
1597         return r;
1598 }
1599
1600 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1601                                             struct kvm_sregs *sregs)
1602 {
1603         int i;
1604
1605         memset(sregs, 0, sizeof(struct kvm_sregs));
1606         sregs->pvr = vcpu->arch.pvr;
1607         for (i = 0; i < vcpu->arch.slb_max; i++) {
1608                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1609                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1610         }
1611
1612         return 0;
1613 }
1614
1615 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1616                                             struct kvm_sregs *sregs)
1617 {
1618         int i, j;
1619
1620         /* Only accept the same PVR as the host's, since we can't spoof it */
1621         if (sregs->pvr != vcpu->arch.pvr)
1622                 return -EINVAL;
1623
1624         j = 0;
1625         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1626                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1627                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1628                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1629                         ++j;
1630                 }
1631         }
1632         vcpu->arch.slb_max = j;
1633
1634         return 0;
1635 }
1636
1637 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1638                 bool preserve_top32)
1639 {
1640         struct kvm *kvm = vcpu->kvm;
1641         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1642         u64 mask;
1643
1644         spin_lock(&vc->lock);
1645         /*
1646          * If ILE (interrupt little-endian) has changed, update the
1647          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1648          */
1649         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1650                 struct kvm_vcpu *vcpu;
1651                 int i;
1652
1653                 kvm_for_each_vcpu(i, vcpu, kvm) {
1654                         if (vcpu->arch.vcore != vc)
1655                                 continue;
1656                         if (new_lpcr & LPCR_ILE)
1657                                 vcpu->arch.intr_msr |= MSR_LE;
1658                         else
1659                                 vcpu->arch.intr_msr &= ~MSR_LE;
1660                 }
1661         }
1662
1663         /*
1664          * Userspace can only modify DPFD (default prefetch depth),
1665          * ILE (interrupt little-endian) and TC (translation control).
1666          * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1667          */
1668         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1669         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1670                 mask |= LPCR_AIL;
1671         /*
1672          * On POWER9, allow userspace to enable large decrementer for the
1673          * guest, whether or not the host has it enabled.
1674          */
1675         if (cpu_has_feature(CPU_FTR_ARCH_300))
1676                 mask |= LPCR_LD;
1677
1678         /* Broken 32-bit version of LPCR must not clear top bits */
1679         if (preserve_top32)
1680                 mask &= 0xFFFFFFFF;
1681         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1682         spin_unlock(&vc->lock);
1683 }
1684
1685 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1686                                  union kvmppc_one_reg *val)
1687 {
1688         int r = 0;
1689         long int i;
1690
1691         switch (id) {
1692         case KVM_REG_PPC_DEBUG_INST:
1693                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1694                 break;
1695         case KVM_REG_PPC_HIOR:
1696                 *val = get_reg_val(id, 0);
1697                 break;
1698         case KVM_REG_PPC_DABR:
1699                 *val = get_reg_val(id, vcpu->arch.dabr);
1700                 break;
1701         case KVM_REG_PPC_DABRX:
1702                 *val = get_reg_val(id, vcpu->arch.dabrx);
1703                 break;
1704         case KVM_REG_PPC_DSCR:
1705                 *val = get_reg_val(id, vcpu->arch.dscr);
1706                 break;
1707         case KVM_REG_PPC_PURR:
1708                 *val = get_reg_val(id, vcpu->arch.purr);
1709                 break;
1710         case KVM_REG_PPC_SPURR:
1711                 *val = get_reg_val(id, vcpu->arch.spurr);
1712                 break;
1713         case KVM_REG_PPC_AMR:
1714                 *val = get_reg_val(id, vcpu->arch.amr);
1715                 break;
1716         case KVM_REG_PPC_UAMOR:
1717                 *val = get_reg_val(id, vcpu->arch.uamor);
1718                 break;
1719         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
1720                 i = id - KVM_REG_PPC_MMCR0;
1721                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1722                 break;
1723         case KVM_REG_PPC_MMCR2:
1724                 *val = get_reg_val(id, vcpu->arch.mmcr[2]);
1725                 break;
1726         case KVM_REG_PPC_MMCRA:
1727                 *val = get_reg_val(id, vcpu->arch.mmcra);
1728                 break;
1729         case KVM_REG_PPC_MMCRS:
1730                 *val = get_reg_val(id, vcpu->arch.mmcrs);
1731                 break;
1732         case KVM_REG_PPC_MMCR3:
1733                 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
1734                 break;
1735         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1736                 i = id - KVM_REG_PPC_PMC1;
1737                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1738                 break;
1739         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1740                 i = id - KVM_REG_PPC_SPMC1;
1741                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1742                 break;
1743         case KVM_REG_PPC_SIAR:
1744                 *val = get_reg_val(id, vcpu->arch.siar);
1745                 break;
1746         case KVM_REG_PPC_SDAR:
1747                 *val = get_reg_val(id, vcpu->arch.sdar);
1748                 break;
1749         case KVM_REG_PPC_SIER:
1750                 *val = get_reg_val(id, vcpu->arch.sier[0]);
1751                 break;
1752         case KVM_REG_PPC_SIER2:
1753                 *val = get_reg_val(id, vcpu->arch.sier[1]);
1754                 break;
1755         case KVM_REG_PPC_SIER3:
1756                 *val = get_reg_val(id, vcpu->arch.sier[2]);
1757                 break;
1758         case KVM_REG_PPC_IAMR:
1759                 *val = get_reg_val(id, vcpu->arch.iamr);
1760                 break;
1761         case KVM_REG_PPC_PSPB:
1762                 *val = get_reg_val(id, vcpu->arch.pspb);
1763                 break;
1764         case KVM_REG_PPC_DPDES:
1765                 /*
1766                  * On POWER9, where we are emulating msgsndp etc.,
1767                  * we return 1 bit for each vcpu, which can come from
1768                  * either vcore->dpdes or doorbell_request.
1769                  * On POWER8, doorbell_request is 0.
1770                  */
1771                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes |
1772                                    vcpu->arch.doorbell_request);
1773                 break;
1774         case KVM_REG_PPC_VTB:
1775                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1776                 break;
1777         case KVM_REG_PPC_DAWR:
1778                 *val = get_reg_val(id, vcpu->arch.dawr0);
1779                 break;
1780         case KVM_REG_PPC_DAWRX:
1781                 *val = get_reg_val(id, vcpu->arch.dawrx0);
1782                 break;
1783         case KVM_REG_PPC_DAWR1:
1784                 *val = get_reg_val(id, vcpu->arch.dawr1);
1785                 break;
1786         case KVM_REG_PPC_DAWRX1:
1787                 *val = get_reg_val(id, vcpu->arch.dawrx1);
1788                 break;
1789         case KVM_REG_PPC_CIABR:
1790                 *val = get_reg_val(id, vcpu->arch.ciabr);
1791                 break;
1792         case KVM_REG_PPC_CSIGR:
1793                 *val = get_reg_val(id, vcpu->arch.csigr);
1794                 break;
1795         case KVM_REG_PPC_TACR:
1796                 *val = get_reg_val(id, vcpu->arch.tacr);
1797                 break;
1798         case KVM_REG_PPC_TCSCR:
1799                 *val = get_reg_val(id, vcpu->arch.tcscr);
1800                 break;
1801         case KVM_REG_PPC_PID:
1802                 *val = get_reg_val(id, vcpu->arch.pid);
1803                 break;
1804         case KVM_REG_PPC_ACOP:
1805                 *val = get_reg_val(id, vcpu->arch.acop);
1806                 break;
1807         case KVM_REG_PPC_WORT:
1808                 *val = get_reg_val(id, vcpu->arch.wort);
1809                 break;
1810         case KVM_REG_PPC_TIDR:
1811                 *val = get_reg_val(id, vcpu->arch.tid);
1812                 break;
1813         case KVM_REG_PPC_PSSCR:
1814                 *val = get_reg_val(id, vcpu->arch.psscr);
1815                 break;
1816         case KVM_REG_PPC_VPA_ADDR:
1817                 spin_lock(&vcpu->arch.vpa_update_lock);
1818                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1819                 spin_unlock(&vcpu->arch.vpa_update_lock);
1820                 break;
1821         case KVM_REG_PPC_VPA_SLB:
1822                 spin_lock(&vcpu->arch.vpa_update_lock);
1823                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1824                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1825                 spin_unlock(&vcpu->arch.vpa_update_lock);
1826                 break;
1827         case KVM_REG_PPC_VPA_DTL:
1828                 spin_lock(&vcpu->arch.vpa_update_lock);
1829                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1830                 val->vpaval.length = vcpu->arch.dtl.len;
1831                 spin_unlock(&vcpu->arch.vpa_update_lock);
1832                 break;
1833         case KVM_REG_PPC_TB_OFFSET:
1834                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1835                 break;
1836         case KVM_REG_PPC_LPCR:
1837         case KVM_REG_PPC_LPCR_64:
1838                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1839                 break;
1840         case KVM_REG_PPC_PPR:
1841                 *val = get_reg_val(id, vcpu->arch.ppr);
1842                 break;
1843 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1844         case KVM_REG_PPC_TFHAR:
1845                 *val = get_reg_val(id, vcpu->arch.tfhar);
1846                 break;
1847         case KVM_REG_PPC_TFIAR:
1848                 *val = get_reg_val(id, vcpu->arch.tfiar);
1849                 break;
1850         case KVM_REG_PPC_TEXASR:
1851                 *val = get_reg_val(id, vcpu->arch.texasr);
1852                 break;
1853         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1854                 i = id - KVM_REG_PPC_TM_GPR0;
1855                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1856                 break;
1857         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1858         {
1859                 int j;
1860                 i = id - KVM_REG_PPC_TM_VSR0;
1861                 if (i < 32)
1862                         for (j = 0; j < TS_FPRWIDTH; j++)
1863                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1864                 else {
1865                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1866                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1867                         else
1868                                 r = -ENXIO;
1869                 }
1870                 break;
1871         }
1872         case KVM_REG_PPC_TM_CR:
1873                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1874                 break;
1875         case KVM_REG_PPC_TM_XER:
1876                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1877                 break;
1878         case KVM_REG_PPC_TM_LR:
1879                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1880                 break;
1881         case KVM_REG_PPC_TM_CTR:
1882                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1883                 break;
1884         case KVM_REG_PPC_TM_FPSCR:
1885                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1886                 break;
1887         case KVM_REG_PPC_TM_AMR:
1888                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1889                 break;
1890         case KVM_REG_PPC_TM_PPR:
1891                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1892                 break;
1893         case KVM_REG_PPC_TM_VRSAVE:
1894                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1895                 break;
1896         case KVM_REG_PPC_TM_VSCR:
1897                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1898                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1899                 else
1900                         r = -ENXIO;
1901                 break;
1902         case KVM_REG_PPC_TM_DSCR:
1903                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1904                 break;
1905         case KVM_REG_PPC_TM_TAR:
1906                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1907                 break;
1908 #endif
1909         case KVM_REG_PPC_ARCH_COMPAT:
1910                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1911                 break;
1912         case KVM_REG_PPC_DEC_EXPIRY:
1913                 *val = get_reg_val(id, vcpu->arch.dec_expires +
1914                                    vcpu->arch.vcore->tb_offset);
1915                 break;
1916         case KVM_REG_PPC_ONLINE:
1917                 *val = get_reg_val(id, vcpu->arch.online);
1918                 break;
1919         case KVM_REG_PPC_PTCR:
1920                 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1921                 break;
1922         default:
1923                 r = -EINVAL;
1924                 break;
1925         }
1926
1927         return r;
1928 }
1929
1930 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1931                                  union kvmppc_one_reg *val)
1932 {
1933         int r = 0;
1934         long int i;
1935         unsigned long addr, len;
1936
1937         switch (id) {
1938         case KVM_REG_PPC_HIOR:
1939                 /* Only allow this to be set to zero */
1940                 if (set_reg_val(id, *val))
1941                         r = -EINVAL;
1942                 break;
1943         case KVM_REG_PPC_DABR:
1944                 vcpu->arch.dabr = set_reg_val(id, *val);
1945                 break;
1946         case KVM_REG_PPC_DABRX:
1947                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1948                 break;
1949         case KVM_REG_PPC_DSCR:
1950                 vcpu->arch.dscr = set_reg_val(id, *val);
1951                 break;
1952         case KVM_REG_PPC_PURR:
1953                 vcpu->arch.purr = set_reg_val(id, *val);
1954                 break;
1955         case KVM_REG_PPC_SPURR:
1956                 vcpu->arch.spurr = set_reg_val(id, *val);
1957                 break;
1958         case KVM_REG_PPC_AMR:
1959                 vcpu->arch.amr = set_reg_val(id, *val);
1960                 break;
1961         case KVM_REG_PPC_UAMOR:
1962                 vcpu->arch.uamor = set_reg_val(id, *val);
1963                 break;
1964         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
1965                 i = id - KVM_REG_PPC_MMCR0;
1966                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1967                 break;
1968         case KVM_REG_PPC_MMCR2:
1969                 vcpu->arch.mmcr[2] = set_reg_val(id, *val);
1970                 break;
1971         case KVM_REG_PPC_MMCRA:
1972                 vcpu->arch.mmcra = set_reg_val(id, *val);
1973                 break;
1974         case KVM_REG_PPC_MMCRS:
1975                 vcpu->arch.mmcrs = set_reg_val(id, *val);
1976                 break;
1977         case KVM_REG_PPC_MMCR3:
1978                 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
1979                 break;
1980         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1981                 i = id - KVM_REG_PPC_PMC1;
1982                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1983                 break;
1984         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1985                 i = id - KVM_REG_PPC_SPMC1;
1986                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1987                 break;
1988         case KVM_REG_PPC_SIAR:
1989                 vcpu->arch.siar = set_reg_val(id, *val);
1990                 break;
1991         case KVM_REG_PPC_SDAR:
1992                 vcpu->arch.sdar = set_reg_val(id, *val);
1993                 break;
1994         case KVM_REG_PPC_SIER:
1995                 vcpu->arch.sier[0] = set_reg_val(id, *val);
1996                 break;
1997         case KVM_REG_PPC_SIER2:
1998                 vcpu->arch.sier[1] = set_reg_val(id, *val);
1999                 break;
2000         case KVM_REG_PPC_SIER3:
2001                 vcpu->arch.sier[2] = set_reg_val(id, *val);
2002                 break;
2003         case KVM_REG_PPC_IAMR:
2004                 vcpu->arch.iamr = set_reg_val(id, *val);
2005                 break;
2006         case KVM_REG_PPC_PSPB:
2007                 vcpu->arch.pspb = set_reg_val(id, *val);
2008                 break;
2009         case KVM_REG_PPC_DPDES:
2010                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
2011                 break;
2012         case KVM_REG_PPC_VTB:
2013                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
2014                 break;
2015         case KVM_REG_PPC_DAWR:
2016                 vcpu->arch.dawr0 = set_reg_val(id, *val);
2017                 break;
2018         case KVM_REG_PPC_DAWRX:
2019                 vcpu->arch.dawrx0 = set_reg_val(id, *val) & ~DAWRX_HYP;
2020                 break;
2021         case KVM_REG_PPC_DAWR1:
2022                 vcpu->arch.dawr1 = set_reg_val(id, *val);
2023                 break;
2024         case KVM_REG_PPC_DAWRX1:
2025                 vcpu->arch.dawrx1 = set_reg_val(id, *val) & ~DAWRX_HYP;
2026                 break;
2027         case KVM_REG_PPC_CIABR:
2028                 vcpu->arch.ciabr = set_reg_val(id, *val);
2029                 /* Don't allow setting breakpoints in hypervisor code */
2030                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
2031                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
2032                 break;
2033         case KVM_REG_PPC_CSIGR:
2034                 vcpu->arch.csigr = set_reg_val(id, *val);
2035                 break;
2036         case KVM_REG_PPC_TACR:
2037                 vcpu->arch.tacr = set_reg_val(id, *val);
2038                 break;
2039         case KVM_REG_PPC_TCSCR:
2040                 vcpu->arch.tcscr = set_reg_val(id, *val);
2041                 break;
2042         case KVM_REG_PPC_PID:
2043                 vcpu->arch.pid = set_reg_val(id, *val);
2044                 break;
2045         case KVM_REG_PPC_ACOP:
2046                 vcpu->arch.acop = set_reg_val(id, *val);
2047                 break;
2048         case KVM_REG_PPC_WORT:
2049                 vcpu->arch.wort = set_reg_val(id, *val);
2050                 break;
2051         case KVM_REG_PPC_TIDR:
2052                 vcpu->arch.tid = set_reg_val(id, *val);
2053                 break;
2054         case KVM_REG_PPC_PSSCR:
2055                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2056                 break;
2057         case KVM_REG_PPC_VPA_ADDR:
2058                 addr = set_reg_val(id, *val);
2059                 r = -EINVAL;
2060                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2061                               vcpu->arch.dtl.next_gpa))
2062                         break;
2063                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2064                 break;
2065         case KVM_REG_PPC_VPA_SLB:
2066                 addr = val->vpaval.addr;
2067                 len = val->vpaval.length;
2068                 r = -EINVAL;
2069                 if (addr && !vcpu->arch.vpa.next_gpa)
2070                         break;
2071                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2072                 break;
2073         case KVM_REG_PPC_VPA_DTL:
2074                 addr = val->vpaval.addr;
2075                 len = val->vpaval.length;
2076                 r = -EINVAL;
2077                 if (addr && (len < sizeof(struct dtl_entry) ||
2078                              !vcpu->arch.vpa.next_gpa))
2079                         break;
2080                 len -= len % sizeof(struct dtl_entry);
2081                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2082                 break;
2083         case KVM_REG_PPC_TB_OFFSET:
2084                 /* round up to multiple of 2^24 */
2085                 vcpu->arch.vcore->tb_offset =
2086                         ALIGN(set_reg_val(id, *val), 1UL << 24);
2087                 break;
2088         case KVM_REG_PPC_LPCR:
2089                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2090                 break;
2091         case KVM_REG_PPC_LPCR_64:
2092                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2093                 break;
2094         case KVM_REG_PPC_PPR:
2095                 vcpu->arch.ppr = set_reg_val(id, *val);
2096                 break;
2097 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2098         case KVM_REG_PPC_TFHAR:
2099                 vcpu->arch.tfhar = set_reg_val(id, *val);
2100                 break;
2101         case KVM_REG_PPC_TFIAR:
2102                 vcpu->arch.tfiar = set_reg_val(id, *val);
2103                 break;
2104         case KVM_REG_PPC_TEXASR:
2105                 vcpu->arch.texasr = set_reg_val(id, *val);
2106                 break;
2107         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2108                 i = id - KVM_REG_PPC_TM_GPR0;
2109                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2110                 break;
2111         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2112         {
2113                 int j;
2114                 i = id - KVM_REG_PPC_TM_VSR0;
2115                 if (i < 32)
2116                         for (j = 0; j < TS_FPRWIDTH; j++)
2117                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2118                 else
2119                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
2120                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2121                         else
2122                                 r = -ENXIO;
2123                 break;
2124         }
2125         case KVM_REG_PPC_TM_CR:
2126                 vcpu->arch.cr_tm = set_reg_val(id, *val);
2127                 break;
2128         case KVM_REG_PPC_TM_XER:
2129                 vcpu->arch.xer_tm = set_reg_val(id, *val);
2130                 break;
2131         case KVM_REG_PPC_TM_LR:
2132                 vcpu->arch.lr_tm = set_reg_val(id, *val);
2133                 break;
2134         case KVM_REG_PPC_TM_CTR:
2135                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2136                 break;
2137         case KVM_REG_PPC_TM_FPSCR:
2138                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2139                 break;
2140         case KVM_REG_PPC_TM_AMR:
2141                 vcpu->arch.amr_tm = set_reg_val(id, *val);
2142                 break;
2143         case KVM_REG_PPC_TM_PPR:
2144                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2145                 break;
2146         case KVM_REG_PPC_TM_VRSAVE:
2147                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2148                 break;
2149         case KVM_REG_PPC_TM_VSCR:
2150                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2151                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2152                 else
2153                         r = - ENXIO;
2154                 break;
2155         case KVM_REG_PPC_TM_DSCR:
2156                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2157                 break;
2158         case KVM_REG_PPC_TM_TAR:
2159                 vcpu->arch.tar_tm = set_reg_val(id, *val);
2160                 break;
2161 #endif
2162         case KVM_REG_PPC_ARCH_COMPAT:
2163                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2164                 break;
2165         case KVM_REG_PPC_DEC_EXPIRY:
2166                 vcpu->arch.dec_expires = set_reg_val(id, *val) -
2167                         vcpu->arch.vcore->tb_offset;
2168                 break;
2169         case KVM_REG_PPC_ONLINE:
2170                 i = set_reg_val(id, *val);
2171                 if (i && !vcpu->arch.online)
2172                         atomic_inc(&vcpu->arch.vcore->online_count);
2173                 else if (!i && vcpu->arch.online)
2174                         atomic_dec(&vcpu->arch.vcore->online_count);
2175                 vcpu->arch.online = i;
2176                 break;
2177         case KVM_REG_PPC_PTCR:
2178                 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2179                 break;
2180         default:
2181                 r = -EINVAL;
2182                 break;
2183         }
2184
2185         return r;
2186 }
2187
2188 /*
2189  * On POWER9, threads are independent and can be in different partitions.
2190  * Therefore we consider each thread to be a subcore.
2191  * There is a restriction that all threads have to be in the same
2192  * MMU mode (radix or HPT), unfortunately, but since we only support
2193  * HPT guests on a HPT host so far, that isn't an impediment yet.
2194  */
2195 static int threads_per_vcore(struct kvm *kvm)
2196 {
2197         if (kvm->arch.threads_indep)
2198                 return 1;
2199         return threads_per_subcore;
2200 }
2201
2202 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2203 {
2204         struct kvmppc_vcore *vcore;
2205
2206         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2207
2208         if (vcore == NULL)
2209                 return NULL;
2210
2211         spin_lock_init(&vcore->lock);
2212         spin_lock_init(&vcore->stoltb_lock);
2213         rcuwait_init(&vcore->wait);
2214         vcore->preempt_tb = TB_NIL;
2215         vcore->lpcr = kvm->arch.lpcr;
2216         vcore->first_vcpuid = id;
2217         vcore->kvm = kvm;
2218         INIT_LIST_HEAD(&vcore->preempt_list);
2219
2220         return vcore;
2221 }
2222
2223 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2224 static struct debugfs_timings_element {
2225         const char *name;
2226         size_t offset;
2227 } timings[] = {
2228         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
2229         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
2230         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
2231         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
2232         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
2233 };
2234
2235 #define N_TIMINGS       (ARRAY_SIZE(timings))
2236
2237 struct debugfs_timings_state {
2238         struct kvm_vcpu *vcpu;
2239         unsigned int    buflen;
2240         char            buf[N_TIMINGS * 100];
2241 };
2242
2243 static int debugfs_timings_open(struct inode *inode, struct file *file)
2244 {
2245         struct kvm_vcpu *vcpu = inode->i_private;
2246         struct debugfs_timings_state *p;
2247
2248         p = kzalloc(sizeof(*p), GFP_KERNEL);
2249         if (!p)
2250                 return -ENOMEM;
2251
2252         kvm_get_kvm(vcpu->kvm);
2253         p->vcpu = vcpu;
2254         file->private_data = p;
2255
2256         return nonseekable_open(inode, file);
2257 }
2258
2259 static int debugfs_timings_release(struct inode *inode, struct file *file)
2260 {
2261         struct debugfs_timings_state *p = file->private_data;
2262
2263         kvm_put_kvm(p->vcpu->kvm);
2264         kfree(p);
2265         return 0;
2266 }
2267
2268 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2269                                     size_t len, loff_t *ppos)
2270 {
2271         struct debugfs_timings_state *p = file->private_data;
2272         struct kvm_vcpu *vcpu = p->vcpu;
2273         char *s, *buf_end;
2274         struct kvmhv_tb_accumulator tb;
2275         u64 count;
2276         loff_t pos;
2277         ssize_t n;
2278         int i, loops;
2279         bool ok;
2280
2281         if (!p->buflen) {
2282                 s = p->buf;
2283                 buf_end = s + sizeof(p->buf);
2284                 for (i = 0; i < N_TIMINGS; ++i) {
2285                         struct kvmhv_tb_accumulator *acc;
2286
2287                         acc = (struct kvmhv_tb_accumulator *)
2288                                 ((unsigned long)vcpu + timings[i].offset);
2289                         ok = false;
2290                         for (loops = 0; loops < 1000; ++loops) {
2291                                 count = acc->seqcount;
2292                                 if (!(count & 1)) {
2293                                         smp_rmb();
2294                                         tb = *acc;
2295                                         smp_rmb();
2296                                         if (count == acc->seqcount) {
2297                                                 ok = true;
2298                                                 break;
2299                                         }
2300                                 }
2301                                 udelay(1);
2302                         }
2303                         if (!ok)
2304                                 snprintf(s, buf_end - s, "%s: stuck\n",
2305                                         timings[i].name);
2306                         else
2307                                 snprintf(s, buf_end - s,
2308                                         "%s: %llu %llu %llu %llu\n",
2309                                         timings[i].name, count / 2,
2310                                         tb_to_ns(tb.tb_total),
2311                                         tb_to_ns(tb.tb_min),
2312                                         tb_to_ns(tb.tb_max));
2313                         s += strlen(s);
2314                 }
2315                 p->buflen = s - p->buf;
2316         }
2317
2318         pos = *ppos;
2319         if (pos >= p->buflen)
2320                 return 0;
2321         if (len > p->buflen - pos)
2322                 len = p->buflen - pos;
2323         n = copy_to_user(buf, p->buf + pos, len);
2324         if (n) {
2325                 if (n == len)
2326                         return -EFAULT;
2327                 len -= n;
2328         }
2329         *ppos = pos + len;
2330         return len;
2331 }
2332
2333 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2334                                      size_t len, loff_t *ppos)
2335 {
2336         return -EACCES;
2337 }
2338
2339 static const struct file_operations debugfs_timings_ops = {
2340         .owner   = THIS_MODULE,
2341         .open    = debugfs_timings_open,
2342         .release = debugfs_timings_release,
2343         .read    = debugfs_timings_read,
2344         .write   = debugfs_timings_write,
2345         .llseek  = generic_file_llseek,
2346 };
2347
2348 /* Create a debugfs directory for the vcpu */
2349 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2350 {
2351         char buf[16];
2352         struct kvm *kvm = vcpu->kvm;
2353
2354         snprintf(buf, sizeof(buf), "vcpu%u", id);
2355         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2356         debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir, vcpu,
2357                             &debugfs_timings_ops);
2358 }
2359
2360 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2361 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2362 {
2363 }
2364 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2365
2366 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2367 {
2368         int err;
2369         int core;
2370         struct kvmppc_vcore *vcore;
2371         struct kvm *kvm;
2372         unsigned int id;
2373
2374         kvm = vcpu->kvm;
2375         id = vcpu->vcpu_id;
2376
2377         vcpu->arch.shared = &vcpu->arch.shregs;
2378 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2379         /*
2380          * The shared struct is never shared on HV,
2381          * so we can always use host endianness
2382          */
2383 #ifdef __BIG_ENDIAN__
2384         vcpu->arch.shared_big_endian = true;
2385 #else
2386         vcpu->arch.shared_big_endian = false;
2387 #endif
2388 #endif
2389         vcpu->arch.mmcr[0] = MMCR0_FC;
2390         vcpu->arch.ctrl = CTRL_RUNLATCH;
2391         /* default to host PVR, since we can't spoof it */
2392         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2393         spin_lock_init(&vcpu->arch.vpa_update_lock);
2394         spin_lock_init(&vcpu->arch.tbacct_lock);
2395         vcpu->arch.busy_preempt = TB_NIL;
2396         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2397
2398         /*
2399          * Set the default HFSCR for the guest from the host value.
2400          * This value is only used on POWER9.
2401          * On POWER9, we want to virtualize the doorbell facility, so we
2402          * don't set the HFSCR_MSGP bit, and that causes those instructions
2403          * to trap and then we emulate them.
2404          */
2405         vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2406                 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP | HFSCR_PREFIX;
2407         if (cpu_has_feature(CPU_FTR_HVMODE)) {
2408                 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2409                 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2410                         vcpu->arch.hfscr |= HFSCR_TM;
2411         }
2412         if (cpu_has_feature(CPU_FTR_TM_COMP))
2413                 vcpu->arch.hfscr |= HFSCR_TM;
2414
2415         kvmppc_mmu_book3s_hv_init(vcpu);
2416
2417         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2418
2419         init_waitqueue_head(&vcpu->arch.cpu_run);
2420
2421         mutex_lock(&kvm->lock);
2422         vcore = NULL;
2423         err = -EINVAL;
2424         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2425                 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2426                         pr_devel("KVM: VCPU ID too high\n");
2427                         core = KVM_MAX_VCORES;
2428                 } else {
2429                         BUG_ON(kvm->arch.smt_mode != 1);
2430                         core = kvmppc_pack_vcpu_id(kvm, id);
2431                 }
2432         } else {
2433                 core = id / kvm->arch.smt_mode;
2434         }
2435         if (core < KVM_MAX_VCORES) {
2436                 vcore = kvm->arch.vcores[core];
2437                 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2438                         pr_devel("KVM: collision on id %u", id);
2439                         vcore = NULL;
2440                 } else if (!vcore) {
2441                         /*
2442                          * Take mmu_setup_lock for mutual exclusion
2443                          * with kvmppc_update_lpcr().
2444                          */
2445                         err = -ENOMEM;
2446                         vcore = kvmppc_vcore_create(kvm,
2447                                         id & ~(kvm->arch.smt_mode - 1));
2448                         mutex_lock(&kvm->arch.mmu_setup_lock);
2449                         kvm->arch.vcores[core] = vcore;
2450                         kvm->arch.online_vcores++;
2451                         mutex_unlock(&kvm->arch.mmu_setup_lock);
2452                 }
2453         }
2454         mutex_unlock(&kvm->lock);
2455
2456         if (!vcore)
2457                 return err;
2458
2459         spin_lock(&vcore->lock);
2460         ++vcore->num_threads;
2461         spin_unlock(&vcore->lock);
2462         vcpu->arch.vcore = vcore;
2463         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2464         vcpu->arch.thread_cpu = -1;
2465         vcpu->arch.prev_cpu = -1;
2466
2467         vcpu->arch.cpu_type = KVM_CPU_3S_64;
2468         kvmppc_sanity_check(vcpu);
2469
2470         debugfs_vcpu_init(vcpu, id);
2471
2472         return 0;
2473 }
2474
2475 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2476                               unsigned long flags)
2477 {
2478         int err;
2479         int esmt = 0;
2480
2481         if (flags)
2482                 return -EINVAL;
2483         if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2484                 return -EINVAL;
2485         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2486                 /*
2487                  * On POWER8 (or POWER7), the threading mode is "strict",
2488                  * so we pack smt_mode vcpus per vcore.
2489                  */
2490                 if (smt_mode > threads_per_subcore)
2491                         return -EINVAL;
2492         } else {
2493                 /*
2494                  * On POWER9, the threading mode is "loose",
2495                  * so each vcpu gets its own vcore.
2496                  */
2497                 esmt = smt_mode;
2498                 smt_mode = 1;
2499         }
2500         mutex_lock(&kvm->lock);
2501         err = -EBUSY;
2502         if (!kvm->arch.online_vcores) {
2503                 kvm->arch.smt_mode = smt_mode;
2504                 kvm->arch.emul_smt_mode = esmt;
2505                 err = 0;
2506         }
2507         mutex_unlock(&kvm->lock);
2508
2509         return err;
2510 }
2511
2512 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2513 {
2514         if (vpa->pinned_addr)
2515                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2516                                         vpa->dirty);
2517 }
2518
2519 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2520 {
2521         spin_lock(&vcpu->arch.vpa_update_lock);
2522         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2523         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2524         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2525         spin_unlock(&vcpu->arch.vpa_update_lock);
2526 }
2527
2528 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2529 {
2530         /* Indicate we want to get back into the guest */
2531         return 1;
2532 }
2533
2534 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2535 {
2536         unsigned long dec_nsec, now;
2537
2538         now = get_tb();
2539         if (now > vcpu->arch.dec_expires) {
2540                 /* decrementer has already gone negative */
2541                 kvmppc_core_queue_dec(vcpu);
2542                 kvmppc_core_prepare_to_enter(vcpu);
2543                 return;
2544         }
2545         dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2546         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2547         vcpu->arch.timer_running = 1;
2548 }
2549
2550 extern int __kvmppc_vcore_entry(void);
2551
2552 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2553                                    struct kvm_vcpu *vcpu)
2554 {
2555         u64 now;
2556
2557         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2558                 return;
2559         spin_lock_irq(&vcpu->arch.tbacct_lock);
2560         now = mftb();
2561         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2562                 vcpu->arch.stolen_logged;
2563         vcpu->arch.busy_preempt = now;
2564         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2565         spin_unlock_irq(&vcpu->arch.tbacct_lock);
2566         --vc->n_runnable;
2567         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2568 }
2569
2570 static int kvmppc_grab_hwthread(int cpu)
2571 {
2572         struct paca_struct *tpaca;
2573         long timeout = 10000;
2574
2575         tpaca = paca_ptrs[cpu];
2576
2577         /* Ensure the thread won't go into the kernel if it wakes */
2578         tpaca->kvm_hstate.kvm_vcpu = NULL;
2579         tpaca->kvm_hstate.kvm_vcore = NULL;
2580         tpaca->kvm_hstate.napping = 0;
2581         smp_wmb();
2582         tpaca->kvm_hstate.hwthread_req = 1;
2583
2584         /*
2585          * If the thread is already executing in the kernel (e.g. handling
2586          * a stray interrupt), wait for it to get back to nap mode.
2587          * The smp_mb() is to ensure that our setting of hwthread_req
2588          * is visible before we look at hwthread_state, so if this
2589          * races with the code at system_reset_pSeries and the thread
2590          * misses our setting of hwthread_req, we are sure to see its
2591          * setting of hwthread_state, and vice versa.
2592          */
2593         smp_mb();
2594         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2595                 if (--timeout <= 0) {
2596                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
2597                         return -EBUSY;
2598                 }
2599                 udelay(1);
2600         }
2601         return 0;
2602 }
2603
2604 static void kvmppc_release_hwthread(int cpu)
2605 {
2606         struct paca_struct *tpaca;
2607
2608         tpaca = paca_ptrs[cpu];
2609         tpaca->kvm_hstate.hwthread_req = 0;
2610         tpaca->kvm_hstate.kvm_vcpu = NULL;
2611         tpaca->kvm_hstate.kvm_vcore = NULL;
2612         tpaca->kvm_hstate.kvm_split_mode = NULL;
2613 }
2614
2615 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2616 {
2617         struct kvm_nested_guest *nested = vcpu->arch.nested;
2618         cpumask_t *cpu_in_guest;
2619         int i;
2620
2621         cpu = cpu_first_thread_sibling(cpu);
2622         if (nested) {
2623                 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2624                 cpu_in_guest = &nested->cpu_in_guest;
2625         } else {
2626                 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2627                 cpu_in_guest = &kvm->arch.cpu_in_guest;
2628         }
2629         /*
2630          * Make sure setting of bit in need_tlb_flush precedes
2631          * testing of cpu_in_guest bits.  The matching barrier on
2632          * the other side is the first smp_mb() in kvmppc_run_core().
2633          */
2634         smp_mb();
2635         for (i = 0; i < threads_per_core; ++i)
2636                 if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2637                         smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2638 }
2639
2640 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2641 {
2642         struct kvm_nested_guest *nested = vcpu->arch.nested;
2643         struct kvm *kvm = vcpu->kvm;
2644         int prev_cpu;
2645
2646         if (!cpu_has_feature(CPU_FTR_HVMODE))
2647                 return;
2648
2649         if (nested)
2650                 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2651         else
2652                 prev_cpu = vcpu->arch.prev_cpu;
2653
2654         /*
2655          * With radix, the guest can do TLB invalidations itself,
2656          * and it could choose to use the local form (tlbiel) if
2657          * it is invalidating a translation that has only ever been
2658          * used on one vcpu.  However, that doesn't mean it has
2659          * only ever been used on one physical cpu, since vcpus
2660          * can move around between pcpus.  To cope with this, when
2661          * a vcpu moves from one pcpu to another, we need to tell
2662          * any vcpus running on the same core as this vcpu previously
2663          * ran to flush the TLB.  The TLB is shared between threads,
2664          * so we use a single bit in .need_tlb_flush for all 4 threads.
2665          */
2666         if (prev_cpu != pcpu) {
2667                 if (prev_cpu >= 0 &&
2668                     cpu_first_thread_sibling(prev_cpu) !=
2669                     cpu_first_thread_sibling(pcpu))
2670                         radix_flush_cpu(kvm, prev_cpu, vcpu);
2671                 if (nested)
2672                         nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2673                 else
2674                         vcpu->arch.prev_cpu = pcpu;
2675         }
2676 }
2677
2678 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2679 {
2680         int cpu;
2681         struct paca_struct *tpaca;
2682         struct kvm *kvm = vc->kvm;
2683
2684         cpu = vc->pcpu;
2685         if (vcpu) {
2686                 if (vcpu->arch.timer_running) {
2687                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2688                         vcpu->arch.timer_running = 0;
2689                 }
2690                 cpu += vcpu->arch.ptid;
2691                 vcpu->cpu = vc->pcpu;
2692                 vcpu->arch.thread_cpu = cpu;
2693                 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2694         }
2695         tpaca = paca_ptrs[cpu];
2696         tpaca->kvm_hstate.kvm_vcpu = vcpu;
2697         tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2698         tpaca->kvm_hstate.fake_suspend = 0;
2699         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2700         smp_wmb();
2701         tpaca->kvm_hstate.kvm_vcore = vc;
2702         if (cpu != smp_processor_id())
2703                 kvmppc_ipi_thread(cpu);
2704 }
2705
2706 static void kvmppc_wait_for_nap(int n_threads)
2707 {
2708         int cpu = smp_processor_id();
2709         int i, loops;
2710
2711         if (n_threads <= 1)
2712                 return;
2713         for (loops = 0; loops < 1000000; ++loops) {
2714                 /*
2715                  * Check if all threads are finished.
2716                  * We set the vcore pointer when starting a thread
2717                  * and the thread clears it when finished, so we look
2718                  * for any threads that still have a non-NULL vcore ptr.
2719                  */
2720                 for (i = 1; i < n_threads; ++i)
2721                         if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2722                                 break;
2723                 if (i == n_threads) {
2724                         HMT_medium();
2725                         return;
2726                 }
2727                 HMT_low();
2728         }
2729         HMT_medium();
2730         for (i = 1; i < n_threads; ++i)
2731                 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2732                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2733 }
2734
2735 /*
2736  * Check that we are on thread 0 and that any other threads in
2737  * this core are off-line.  Then grab the threads so they can't
2738  * enter the kernel.
2739  */
2740 static int on_primary_thread(void)
2741 {
2742         int cpu = smp_processor_id();
2743         int thr;
2744
2745         /* Are we on a primary subcore? */
2746         if (cpu_thread_in_subcore(cpu))
2747                 return 0;
2748
2749         thr = 0;
2750         while (++thr < threads_per_subcore)
2751                 if (cpu_online(cpu + thr))
2752                         return 0;
2753
2754         /* Grab all hw threads so they can't go into the kernel */
2755         for (thr = 1; thr < threads_per_subcore; ++thr) {
2756                 if (kvmppc_grab_hwthread(cpu + thr)) {
2757                         /* Couldn't grab one; let the others go */
2758                         do {
2759                                 kvmppc_release_hwthread(cpu + thr);
2760                         } while (--thr > 0);
2761                         return 0;
2762                 }
2763         }
2764         return 1;
2765 }
2766
2767 /*
2768  * A list of virtual cores for each physical CPU.
2769  * These are vcores that could run but their runner VCPU tasks are
2770  * (or may be) preempted.
2771  */
2772 struct preempted_vcore_list {
2773         struct list_head        list;
2774         spinlock_t              lock;
2775 };
2776
2777 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2778
2779 static void init_vcore_lists(void)
2780 {
2781         int cpu;
2782
2783         for_each_possible_cpu(cpu) {
2784                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2785                 spin_lock_init(&lp->lock);
2786                 INIT_LIST_HEAD(&lp->list);
2787         }
2788 }
2789
2790 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2791 {
2792         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2793
2794         vc->vcore_state = VCORE_PREEMPT;
2795         vc->pcpu = smp_processor_id();
2796         if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2797                 spin_lock(&lp->lock);
2798                 list_add_tail(&vc->preempt_list, &lp->list);
2799                 spin_unlock(&lp->lock);
2800         }
2801
2802         /* Start accumulating stolen time */
2803         kvmppc_core_start_stolen(vc);
2804 }
2805
2806 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2807 {
2808         struct preempted_vcore_list *lp;
2809
2810         kvmppc_core_end_stolen(vc);
2811         if (!list_empty(&vc->preempt_list)) {
2812                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2813                 spin_lock(&lp->lock);
2814                 list_del_init(&vc->preempt_list);
2815                 spin_unlock(&lp->lock);
2816         }
2817         vc->vcore_state = VCORE_INACTIVE;
2818 }
2819
2820 /*
2821  * This stores information about the virtual cores currently
2822  * assigned to a physical core.
2823  */
2824 struct core_info {
2825         int             n_subcores;
2826         int             max_subcore_threads;
2827         int             total_threads;
2828         int             subcore_threads[MAX_SUBCORES];
2829         struct kvmppc_vcore *vc[MAX_SUBCORES];
2830 };
2831
2832 /*
2833  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2834  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2835  */
2836 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2837
2838 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2839 {
2840         memset(cip, 0, sizeof(*cip));
2841         cip->n_subcores = 1;
2842         cip->max_subcore_threads = vc->num_threads;
2843         cip->total_threads = vc->num_threads;
2844         cip->subcore_threads[0] = vc->num_threads;
2845         cip->vc[0] = vc;
2846 }
2847
2848 static bool subcore_config_ok(int n_subcores, int n_threads)
2849 {
2850         /*
2851          * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2852          * split-core mode, with one thread per subcore.
2853          */
2854         if (cpu_has_feature(CPU_FTR_ARCH_300))
2855                 return n_subcores <= 4 && n_threads == 1;
2856
2857         /* On POWER8, can only dynamically split if unsplit to begin with */
2858         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2859                 return false;
2860         if (n_subcores > MAX_SUBCORES)
2861                 return false;
2862         if (n_subcores > 1) {
2863                 if (!(dynamic_mt_modes & 2))
2864                         n_subcores = 4;
2865                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2866                         return false;
2867         }
2868
2869         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2870 }
2871
2872 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2873 {
2874         vc->entry_exit_map = 0;
2875         vc->in_guest = 0;
2876         vc->napping_threads = 0;
2877         vc->conferring_threads = 0;
2878         vc->tb_offset_applied = 0;
2879 }
2880
2881 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2882 {
2883         int n_threads = vc->num_threads;
2884         int sub;
2885
2886         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2887                 return false;
2888
2889         /* In one_vm_per_core mode, require all vcores to be from the same vm */
2890         if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2891                 return false;
2892
2893         if (n_threads < cip->max_subcore_threads)
2894                 n_threads = cip->max_subcore_threads;
2895         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2896                 return false;
2897         cip->max_subcore_threads = n_threads;
2898
2899         sub = cip->n_subcores;
2900         ++cip->n_subcores;
2901         cip->total_threads += vc->num_threads;
2902         cip->subcore_threads[sub] = vc->num_threads;
2903         cip->vc[sub] = vc;
2904         init_vcore_to_run(vc);
2905         list_del_init(&vc->preempt_list);
2906
2907         return true;
2908 }
2909
2910 /*
2911  * Work out whether it is possible to piggyback the execution of
2912  * vcore *pvc onto the execution of the other vcores described in *cip.
2913  */
2914 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2915                           int target_threads)
2916 {
2917         if (cip->total_threads + pvc->num_threads > target_threads)
2918                 return false;
2919
2920         return can_dynamic_split(pvc, cip);
2921 }
2922
2923 static void prepare_threads(struct kvmppc_vcore *vc)
2924 {
2925         int i;
2926         struct kvm_vcpu *vcpu;
2927
2928         for_each_runnable_thread(i, vcpu, vc) {
2929                 if (signal_pending(vcpu->arch.run_task))
2930                         vcpu->arch.ret = -EINTR;
2931                 else if (no_mixing_hpt_and_radix &&
2932                          kvm_is_radix(vc->kvm) != radix_enabled())
2933                         vcpu->arch.ret = -EINVAL;
2934                 else if (vcpu->arch.vpa.update_pending ||
2935                          vcpu->arch.slb_shadow.update_pending ||
2936                          vcpu->arch.dtl.update_pending)
2937                         vcpu->arch.ret = RESUME_GUEST;
2938                 else
2939                         continue;
2940                 kvmppc_remove_runnable(vc, vcpu);
2941                 wake_up(&vcpu->arch.cpu_run);
2942         }
2943 }
2944
2945 static void collect_piggybacks(struct core_info *cip, int target_threads)
2946 {
2947         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2948         struct kvmppc_vcore *pvc, *vcnext;
2949
2950         spin_lock(&lp->lock);
2951         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2952                 if (!spin_trylock(&pvc->lock))
2953                         continue;
2954                 prepare_threads(pvc);
2955                 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
2956                         list_del_init(&pvc->preempt_list);
2957                         if (pvc->runner == NULL) {
2958                                 pvc->vcore_state = VCORE_INACTIVE;
2959                                 kvmppc_core_end_stolen(pvc);
2960                         }
2961                         spin_unlock(&pvc->lock);
2962                         continue;
2963                 }
2964                 if (!can_piggyback(pvc, cip, target_threads)) {
2965                         spin_unlock(&pvc->lock);
2966                         continue;
2967                 }
2968                 kvmppc_core_end_stolen(pvc);
2969                 pvc->vcore_state = VCORE_PIGGYBACK;
2970                 if (cip->total_threads >= target_threads)
2971                         break;
2972         }
2973         spin_unlock(&lp->lock);
2974 }
2975
2976 static bool recheck_signals_and_mmu(struct core_info *cip)
2977 {
2978         int sub, i;
2979         struct kvm_vcpu *vcpu;
2980         struct kvmppc_vcore *vc;
2981
2982         for (sub = 0; sub < cip->n_subcores; ++sub) {
2983                 vc = cip->vc[sub];
2984                 if (!vc->kvm->arch.mmu_ready)
2985                         return true;
2986                 for_each_runnable_thread(i, vcpu, vc)
2987                         if (signal_pending(vcpu->arch.run_task))
2988                                 return true;
2989         }
2990         return false;
2991 }
2992
2993 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2994 {
2995         int still_running = 0, i;
2996         u64 now;
2997         long ret;
2998         struct kvm_vcpu *vcpu;
2999
3000         spin_lock(&vc->lock);
3001         now = get_tb();
3002         for_each_runnable_thread(i, vcpu, vc) {
3003                 /*
3004                  * It's safe to unlock the vcore in the loop here, because
3005                  * for_each_runnable_thread() is safe against removal of
3006                  * the vcpu, and the vcore state is VCORE_EXITING here,
3007                  * so any vcpus becoming runnable will have their arch.trap
3008                  * set to zero and can't actually run in the guest.
3009                  */
3010                 spin_unlock(&vc->lock);
3011                 /* cancel pending dec exception if dec is positive */
3012                 if (now < vcpu->arch.dec_expires &&
3013                     kvmppc_core_pending_dec(vcpu))
3014                         kvmppc_core_dequeue_dec(vcpu);
3015
3016                 trace_kvm_guest_exit(vcpu);
3017
3018                 ret = RESUME_GUEST;
3019                 if (vcpu->arch.trap)
3020                         ret = kvmppc_handle_exit_hv(vcpu,
3021                                                     vcpu->arch.run_task);
3022
3023                 vcpu->arch.ret = ret;
3024                 vcpu->arch.trap = 0;
3025
3026                 spin_lock(&vc->lock);
3027                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
3028                         if (vcpu->arch.pending_exceptions)
3029                                 kvmppc_core_prepare_to_enter(vcpu);
3030                         if (vcpu->arch.ceded)
3031                                 kvmppc_set_timer(vcpu);
3032                         else
3033                                 ++still_running;
3034                 } else {
3035                         kvmppc_remove_runnable(vc, vcpu);
3036                         wake_up(&vcpu->arch.cpu_run);
3037                 }
3038         }
3039         if (!is_master) {
3040                 if (still_running > 0) {
3041                         kvmppc_vcore_preempt(vc);
3042                 } else if (vc->runner) {
3043                         vc->vcore_state = VCORE_PREEMPT;
3044                         kvmppc_core_start_stolen(vc);
3045                 } else {
3046                         vc->vcore_state = VCORE_INACTIVE;
3047                 }
3048                 if (vc->n_runnable > 0 && vc->runner == NULL) {
3049                         /* make sure there's a candidate runner awake */
3050                         i = -1;
3051                         vcpu = next_runnable_thread(vc, &i);
3052                         wake_up(&vcpu->arch.cpu_run);
3053                 }
3054         }
3055         spin_unlock(&vc->lock);
3056 }
3057
3058 /*
3059  * Clear core from the list of active host cores as we are about to
3060  * enter the guest. Only do this if it is the primary thread of the
3061  * core (not if a subcore) that is entering the guest.
3062  */
3063 static inline int kvmppc_clear_host_core(unsigned int cpu)
3064 {
3065         int core;
3066
3067         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3068                 return 0;
3069         /*
3070          * Memory barrier can be omitted here as we will do a smp_wmb()
3071          * later in kvmppc_start_thread and we need ensure that state is
3072          * visible to other CPUs only after we enter guest.
3073          */
3074         core = cpu >> threads_shift;
3075         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3076         return 0;
3077 }
3078
3079 /*
3080  * Advertise this core as an active host core since we exited the guest
3081  * Only need to do this if it is the primary thread of the core that is
3082  * exiting.
3083  */
3084 static inline int kvmppc_set_host_core(unsigned int cpu)
3085 {
3086         int core;
3087
3088         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3089                 return 0;
3090
3091         /*
3092          * Memory barrier can be omitted here because we do a spin_unlock
3093          * immediately after this which provides the memory barrier.
3094          */
3095         core = cpu >> threads_shift;
3096         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3097         return 0;
3098 }
3099
3100 static void set_irq_happened(int trap)
3101 {
3102         switch (trap) {
3103         case BOOK3S_INTERRUPT_EXTERNAL:
3104                 local_paca->irq_happened |= PACA_IRQ_EE;
3105                 break;
3106         case BOOK3S_INTERRUPT_H_DOORBELL:
3107                 local_paca->irq_happened |= PACA_IRQ_DBELL;
3108                 break;
3109         case BOOK3S_INTERRUPT_HMI:
3110                 local_paca->irq_happened |= PACA_IRQ_HMI;
3111                 break;
3112         case BOOK3S_INTERRUPT_SYSTEM_RESET:
3113                 replay_system_reset();
3114                 break;
3115         }
3116 }
3117
3118 /*
3119  * Run a set of guest threads on a physical core.
3120  * Called with vc->lock held.
3121  */
3122 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3123 {
3124         struct kvm_vcpu *vcpu;
3125         int i;
3126         int srcu_idx;
3127         struct core_info core_info;
3128         struct kvmppc_vcore *pvc;
3129         struct kvm_split_mode split_info, *sip;
3130         int split, subcore_size, active;
3131         int sub;
3132         bool thr0_done;
3133         unsigned long cmd_bit, stat_bit;
3134         int pcpu, thr;
3135         int target_threads;
3136         int controlled_threads;
3137         int trap;
3138         bool is_power8;
3139
3140         /*
3141          * Remove from the list any threads that have a signal pending
3142          * or need a VPA update done
3143          */
3144         prepare_threads(vc);
3145
3146         /* if the runner is no longer runnable, let the caller pick a new one */
3147         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3148                 return;
3149
3150         /*
3151          * Initialize *vc.
3152          */
3153         init_vcore_to_run(vc);
3154         vc->preempt_tb = TB_NIL;
3155
3156         /*
3157          * Number of threads that we will be controlling: the same as
3158          * the number of threads per subcore, except on POWER9,
3159          * where it's 1 because the threads are (mostly) independent.
3160          */
3161         controlled_threads = threads_per_vcore(vc->kvm);
3162
3163         /*
3164          * Make sure we are running on primary threads, and that secondary
3165          * threads are offline.  Also check if the number of threads in this
3166          * guest are greater than the current system threads per guest.
3167          * On POWER9, we need to be not in independent-threads mode if
3168          * this is a HPT guest on a radix host machine where the
3169          * CPU threads may not be in different MMU modes.
3170          */
3171         if ((controlled_threads > 1) &&
3172             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
3173                 for_each_runnable_thread(i, vcpu, vc) {
3174                         vcpu->arch.ret = -EBUSY;
3175                         kvmppc_remove_runnable(vc, vcpu);
3176                         wake_up(&vcpu->arch.cpu_run);
3177                 }
3178                 goto out;
3179         }
3180
3181         /*
3182          * See if we could run any other vcores on the physical core
3183          * along with this one.
3184          */
3185         init_core_info(&core_info, vc);
3186         pcpu = smp_processor_id();
3187         target_threads = controlled_threads;
3188         if (target_smt_mode && target_smt_mode < target_threads)
3189                 target_threads = target_smt_mode;
3190         if (vc->num_threads < target_threads)
3191                 collect_piggybacks(&core_info, target_threads);
3192
3193         /*
3194          * On radix, arrange for TLB flushing if necessary.
3195          * This has to be done before disabling interrupts since
3196          * it uses smp_call_function().
3197          */
3198         pcpu = smp_processor_id();
3199         if (kvm_is_radix(vc->kvm)) {
3200                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3201                         for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3202                                 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3203         }
3204
3205         /*
3206          * Hard-disable interrupts, and check resched flag and signals.
3207          * If we need to reschedule or deliver a signal, clean up
3208          * and return without going into the guest(s).
3209          * If the mmu_ready flag has been cleared, don't go into the
3210          * guest because that means a HPT resize operation is in progress.
3211          */
3212         local_irq_disable();
3213         hard_irq_disable();
3214         if (lazy_irq_pending() || need_resched() ||
3215             recheck_signals_and_mmu(&core_info)) {
3216                 local_irq_enable();
3217                 vc->vcore_state = VCORE_INACTIVE;
3218                 /* Unlock all except the primary vcore */
3219                 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3220                         pvc = core_info.vc[sub];
3221                         /* Put back on to the preempted vcores list */
3222                         kvmppc_vcore_preempt(pvc);
3223                         spin_unlock(&pvc->lock);
3224                 }
3225                 for (i = 0; i < controlled_threads; ++i)
3226                         kvmppc_release_hwthread(pcpu + i);
3227                 return;
3228         }
3229
3230         kvmppc_clear_host_core(pcpu);
3231
3232         /* Decide on micro-threading (split-core) mode */
3233         subcore_size = threads_per_subcore;
3234         cmd_bit = stat_bit = 0;
3235         split = core_info.n_subcores;
3236         sip = NULL;
3237         is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3238                 && !cpu_has_feature(CPU_FTR_ARCH_300);
3239
3240         if (split > 1) {
3241                 sip = &split_info;
3242                 memset(&split_info, 0, sizeof(split_info));
3243                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3244                         split_info.vc[sub] = core_info.vc[sub];
3245
3246                 if (is_power8) {
3247                         if (split == 2 && (dynamic_mt_modes & 2)) {
3248                                 cmd_bit = HID0_POWER8_1TO2LPAR;
3249                                 stat_bit = HID0_POWER8_2LPARMODE;
3250                         } else {
3251                                 split = 4;
3252                                 cmd_bit = HID0_POWER8_1TO4LPAR;
3253                                 stat_bit = HID0_POWER8_4LPARMODE;
3254                         }
3255                         subcore_size = MAX_SMT_THREADS / split;
3256                         split_info.rpr = mfspr(SPRN_RPR);
3257                         split_info.pmmar = mfspr(SPRN_PMMAR);
3258                         split_info.ldbar = mfspr(SPRN_LDBAR);
3259                         split_info.subcore_size = subcore_size;
3260                 } else {
3261                         split_info.subcore_size = 1;
3262                 }
3263
3264                 /* order writes to split_info before kvm_split_mode pointer */
3265                 smp_wmb();
3266         }
3267
3268         for (thr = 0; thr < controlled_threads; ++thr) {
3269                 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3270
3271                 paca->kvm_hstate.napping = 0;
3272                 paca->kvm_hstate.kvm_split_mode = sip;
3273         }
3274
3275         /* Initiate micro-threading (split-core) on POWER8 if required */
3276         if (cmd_bit) {
3277                 unsigned long hid0 = mfspr(SPRN_HID0);
3278
3279                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3280                 mb();
3281                 mtspr(SPRN_HID0, hid0);
3282                 isync();
3283                 for (;;) {
3284                         hid0 = mfspr(SPRN_HID0);
3285                         if (hid0 & stat_bit)
3286                                 break;
3287                         cpu_relax();
3288                 }
3289         }
3290
3291         /*
3292          * On POWER8, set RWMR register.
3293          * Since it only affects PURR and SPURR, it doesn't affect
3294          * the host, so we don't save/restore the host value.
3295          */
3296         if (is_power8) {
3297                 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3298                 int n_online = atomic_read(&vc->online_count);
3299
3300                 /*
3301                  * Use the 8-thread value if we're doing split-core
3302                  * or if the vcore's online count looks bogus.
3303                  */
3304                 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3305                     n_online >= 1 && n_online <= MAX_SMT_THREADS)
3306                         rwmr_val = p8_rwmr_values[n_online];
3307                 mtspr(SPRN_RWMR, rwmr_val);
3308         }
3309
3310         /* Start all the threads */
3311         active = 0;
3312         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3313                 thr = is_power8 ? subcore_thread_map[sub] : sub;
3314                 thr0_done = false;
3315                 active |= 1 << thr;
3316                 pvc = core_info.vc[sub];
3317                 pvc->pcpu = pcpu + thr;
3318                 for_each_runnable_thread(i, vcpu, pvc) {
3319                         kvmppc_start_thread(vcpu, pvc);
3320                         kvmppc_create_dtl_entry(vcpu, pvc);
3321                         trace_kvm_guest_enter(vcpu);
3322                         if (!vcpu->arch.ptid)
3323                                 thr0_done = true;
3324                         active |= 1 << (thr + vcpu->arch.ptid);
3325                 }
3326                 /*
3327                  * We need to start the first thread of each subcore
3328                  * even if it doesn't have a vcpu.
3329                  */
3330                 if (!thr0_done)
3331                         kvmppc_start_thread(NULL, pvc);
3332         }
3333
3334         /*
3335          * Ensure that split_info.do_nap is set after setting
3336          * the vcore pointer in the PACA of the secondaries.
3337          */
3338         smp_mb();
3339
3340         /*
3341          * When doing micro-threading, poke the inactive threads as well.
3342          * This gets them to the nap instruction after kvm_do_nap,
3343          * which reduces the time taken to unsplit later.
3344          */
3345         if (cmd_bit) {
3346                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
3347                 for (thr = 1; thr < threads_per_subcore; ++thr)
3348                         if (!(active & (1 << thr)))
3349                                 kvmppc_ipi_thread(pcpu + thr);
3350         }
3351
3352         vc->vcore_state = VCORE_RUNNING;
3353         preempt_disable();
3354
3355         trace_kvmppc_run_core(vc, 0);
3356
3357         for (sub = 0; sub < core_info.n_subcores; ++sub)
3358                 spin_unlock(&core_info.vc[sub]->lock);
3359
3360         guest_enter_irqoff();
3361
3362         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3363
3364         this_cpu_disable_ftrace();
3365
3366         /*
3367          * Interrupts will be enabled once we get into the guest,
3368          * so tell lockdep that we're about to enable interrupts.
3369          */
3370         trace_hardirqs_on();
3371
3372         trap = __kvmppc_vcore_entry();
3373
3374         trace_hardirqs_off();
3375
3376         this_cpu_enable_ftrace();
3377
3378         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3379
3380         set_irq_happened(trap);
3381
3382         spin_lock(&vc->lock);
3383         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3384         vc->vcore_state = VCORE_EXITING;
3385
3386         /* wait for secondary threads to finish writing their state to memory */
3387         kvmppc_wait_for_nap(controlled_threads);
3388
3389         /* Return to whole-core mode if we split the core earlier */
3390         if (cmd_bit) {
3391                 unsigned long hid0 = mfspr(SPRN_HID0);
3392                 unsigned long loops = 0;
3393
3394                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3395                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3396                 mb();
3397                 mtspr(SPRN_HID0, hid0);
3398                 isync();
3399                 for (;;) {
3400                         hid0 = mfspr(SPRN_HID0);
3401                         if (!(hid0 & stat_bit))
3402                                 break;
3403                         cpu_relax();
3404                         ++loops;
3405                 }
3406                 split_info.do_nap = 0;
3407         }
3408
3409         kvmppc_set_host_core(pcpu);
3410
3411         local_irq_enable();
3412         guest_exit();
3413
3414         /* Let secondaries go back to the offline loop */
3415         for (i = 0; i < controlled_threads; ++i) {
3416                 kvmppc_release_hwthread(pcpu + i);
3417                 if (sip && sip->napped[i])
3418                         kvmppc_ipi_thread(pcpu + i);
3419                 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3420         }
3421
3422         spin_unlock(&vc->lock);
3423
3424         /* make sure updates to secondary vcpu structs are visible now */
3425         smp_mb();
3426
3427         preempt_enable();
3428
3429         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3430                 pvc = core_info.vc[sub];
3431                 post_guest_process(pvc, pvc == vc);
3432         }
3433
3434         spin_lock(&vc->lock);
3435
3436  out:
3437         vc->vcore_state = VCORE_INACTIVE;
3438         trace_kvmppc_run_core(vc, 1);
3439 }
3440
3441 /*
3442  * Load up hypervisor-mode registers on P9.
3443  */
3444 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3445                                      unsigned long lpcr)
3446 {
3447         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3448         s64 hdec;
3449         u64 tb, purr, spurr;
3450         int trap;
3451         unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3452         unsigned long host_ciabr = mfspr(SPRN_CIABR);
3453         unsigned long host_dawr0 = mfspr(SPRN_DAWR0);
3454         unsigned long host_dawrx0 = mfspr(SPRN_DAWRX0);
3455         unsigned long host_psscr = mfspr(SPRN_PSSCR);
3456         unsigned long host_pidr = mfspr(SPRN_PID);
3457         unsigned long host_dawr1 = 0;
3458         unsigned long host_dawrx1 = 0;
3459
3460         if (cpu_has_feature(CPU_FTR_DAWR1)) {
3461                 host_dawr1 = mfspr(SPRN_DAWR1);
3462                 host_dawrx1 = mfspr(SPRN_DAWRX1);
3463         }
3464
3465         /*
3466          * P8 and P9 suppress the HDEC exception when LPCR[HDICE] = 0,
3467          * so set HDICE before writing HDEC.
3468          */
3469         mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr | LPCR_HDICE);
3470         isync();
3471
3472         hdec = time_limit - mftb();
3473         if (hdec < 0) {
3474                 mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3475                 isync();
3476                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3477         }
3478         mtspr(SPRN_HDEC, hdec);
3479
3480         if (vc->tb_offset) {
3481                 u64 new_tb = mftb() + vc->tb_offset;
3482                 mtspr(SPRN_TBU40, new_tb);
3483                 tb = mftb();
3484                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3485                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3486                 vc->tb_offset_applied = vc->tb_offset;
3487         }
3488
3489         if (vc->pcr)
3490                 mtspr(SPRN_PCR, vc->pcr | PCR_MASK);
3491         mtspr(SPRN_DPDES, vc->dpdes);
3492         mtspr(SPRN_VTB, vc->vtb);
3493
3494         local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3495         local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3496         mtspr(SPRN_PURR, vcpu->arch.purr);
3497         mtspr(SPRN_SPURR, vcpu->arch.spurr);
3498
3499         if (dawr_enabled()) {
3500                 mtspr(SPRN_DAWR0, vcpu->arch.dawr0);
3501                 mtspr(SPRN_DAWRX0, vcpu->arch.dawrx0);
3502                 if (cpu_has_feature(CPU_FTR_DAWR1)) {
3503                         mtspr(SPRN_DAWR1, vcpu->arch.dawr1);
3504                         mtspr(SPRN_DAWRX1, vcpu->arch.dawrx1);
3505                 }
3506         }
3507         mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3508         mtspr(SPRN_IC, vcpu->arch.ic);
3509         mtspr(SPRN_PID, vcpu->arch.pid);
3510
3511         mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3512               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3513
3514         mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3515
3516         mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3517         mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3518         mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3519         mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3520
3521         mtspr(SPRN_AMOR, ~0UL);
3522
3523         mtspr(SPRN_LPCR, lpcr);
3524         isync();
3525
3526         kvmppc_xive_push_vcpu(vcpu);
3527
3528         mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3529         mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3530
3531         trap = __kvmhv_vcpu_entry_p9(vcpu);
3532
3533         /* Advance host PURR/SPURR by the amount used by guest */
3534         purr = mfspr(SPRN_PURR);
3535         spurr = mfspr(SPRN_SPURR);
3536         mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3537               purr - vcpu->arch.purr);
3538         mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3539               spurr - vcpu->arch.spurr);
3540         vcpu->arch.purr = purr;
3541         vcpu->arch.spurr = spurr;
3542
3543         vcpu->arch.ic = mfspr(SPRN_IC);
3544         vcpu->arch.pid = mfspr(SPRN_PID);
3545         vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3546
3547         vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3548         vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3549         vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3550         vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3551
3552         /* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3553         mtspr(SPRN_PSSCR, host_psscr |
3554               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3555         mtspr(SPRN_HFSCR, host_hfscr);
3556         mtspr(SPRN_CIABR, host_ciabr);
3557         mtspr(SPRN_DAWR0, host_dawr0);
3558         mtspr(SPRN_DAWRX0, host_dawrx0);
3559         if (cpu_has_feature(CPU_FTR_DAWR1)) {
3560                 mtspr(SPRN_DAWR1, host_dawr1);
3561                 mtspr(SPRN_DAWRX1, host_dawrx1);
3562         }
3563         mtspr(SPRN_PID, host_pidr);
3564
3565         /*
3566          * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3567          * case we interrupted the guest between a tlbie and a ptesync.
3568          */
3569         asm volatile("eieio; tlbsync; ptesync");
3570
3571         /*
3572          * cp_abort is required if the processor supports local copy-paste
3573          * to clear the copy buffer that was under control of the guest.
3574          */
3575         if (cpu_has_feature(CPU_FTR_ARCH_31))
3576                 asm volatile(PPC_CP_ABORT);
3577
3578         mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid);    /* restore host LPID */
3579         isync();
3580
3581         vc->dpdes = mfspr(SPRN_DPDES);
3582         vc->vtb = mfspr(SPRN_VTB);
3583         mtspr(SPRN_DPDES, 0);
3584         if (vc->pcr)
3585                 mtspr(SPRN_PCR, PCR_MASK);
3586
3587         if (vc->tb_offset_applied) {
3588                 u64 new_tb = mftb() - vc->tb_offset_applied;
3589                 mtspr(SPRN_TBU40, new_tb);
3590                 tb = mftb();
3591                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3592                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3593                 vc->tb_offset_applied = 0;
3594         }
3595
3596         mtspr(SPRN_HDEC, 0x7fffffff);
3597         mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3598
3599         return trap;
3600 }
3601
3602 /*
3603  * Virtual-mode guest entry for POWER9 and later when the host and
3604  * guest are both using the radix MMU.  The LPIDR has already been set.
3605  */
3606 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3607                          unsigned long lpcr)
3608 {
3609         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3610         unsigned long host_dscr = mfspr(SPRN_DSCR);
3611         unsigned long host_tidr = mfspr(SPRN_TIDR);
3612         unsigned long host_iamr = mfspr(SPRN_IAMR);
3613         unsigned long host_amr = mfspr(SPRN_AMR);
3614         unsigned long host_fscr = mfspr(SPRN_FSCR);
3615         s64 dec;
3616         u64 tb;
3617         int trap, save_pmu;
3618
3619         dec = mfspr(SPRN_DEC);
3620         tb = mftb();
3621         if (dec < 0)
3622                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3623         local_paca->kvm_hstate.dec_expires = dec + tb;
3624         if (local_paca->kvm_hstate.dec_expires < time_limit)
3625                 time_limit = local_paca->kvm_hstate.dec_expires;
3626
3627         vcpu->arch.ceded = 0;
3628
3629         kvmhv_save_host_pmu();          /* saves it to PACA kvm_hstate */
3630
3631         kvmppc_subcore_enter_guest();
3632
3633         vc->entry_exit_map = 1;
3634         vc->in_guest = 1;
3635
3636         if (vcpu->arch.vpa.pinned_addr) {
3637                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3638                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3639                 lp->yield_count = cpu_to_be32(yield_count);
3640                 vcpu->arch.vpa.dirty = 1;
3641         }
3642
3643         if (cpu_has_feature(CPU_FTR_TM) ||
3644             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3645                 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3646
3647         kvmhv_load_guest_pmu(vcpu);
3648
3649         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3650         load_fp_state(&vcpu->arch.fp);
3651 #ifdef CONFIG_ALTIVEC
3652         load_vr_state(&vcpu->arch.vr);
3653 #endif
3654         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3655
3656         mtspr(SPRN_DSCR, vcpu->arch.dscr);
3657         mtspr(SPRN_IAMR, vcpu->arch.iamr);
3658         mtspr(SPRN_PSPB, vcpu->arch.pspb);
3659         mtspr(SPRN_FSCR, vcpu->arch.fscr);
3660         mtspr(SPRN_TAR, vcpu->arch.tar);
3661         mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3662         mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3663         mtspr(SPRN_BESCR, vcpu->arch.bescr);
3664         mtspr(SPRN_WORT, vcpu->arch.wort);
3665         mtspr(SPRN_TIDR, vcpu->arch.tid);
3666         mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3667         mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3668         mtspr(SPRN_AMR, vcpu->arch.amr);
3669         mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3670
3671         if (!(vcpu->arch.ctrl & 1))
3672                 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3673
3674         mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3675
3676         if (kvmhv_on_pseries()) {
3677                 /*
3678                  * We need to save and restore the guest visible part of the
3679                  * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3680                  * doesn't do this for us. Note only required if pseries since
3681                  * this is done in kvmhv_load_hv_regs_and_go() below otherwise.
3682                  */
3683                 unsigned long host_psscr;
3684                 /* call our hypervisor to load up HV regs and go */
3685                 struct hv_guest_state hvregs;
3686
3687                 host_psscr = mfspr(SPRN_PSSCR_PR);
3688                 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3689                 kvmhv_save_hv_regs(vcpu, &hvregs);
3690                 hvregs.lpcr = lpcr;
3691                 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3692                 hvregs.version = HV_GUEST_STATE_VERSION;
3693                 if (vcpu->arch.nested) {
3694                         hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3695                         hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3696                 } else {
3697                         hvregs.lpid = vcpu->kvm->arch.lpid;
3698                         hvregs.vcpu_token = vcpu->vcpu_id;
3699                 }
3700                 hvregs.hdec_expiry = time_limit;
3701                 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3702                                           __pa(&vcpu->arch.regs));
3703                 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3704                 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3705                 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3706                 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3707                 vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3708                 mtspr(SPRN_PSSCR_PR, host_psscr);
3709
3710                 /* H_CEDE has to be handled now, not later */
3711                 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3712                     kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3713                         kvmppc_nested_cede(vcpu);
3714                         kvmppc_set_gpr(vcpu, 3, 0);
3715                         trap = 0;
3716                 }
3717         } else {
3718                 trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3719         }
3720
3721         vcpu->arch.slb_max = 0;
3722         dec = mfspr(SPRN_DEC);
3723         if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
3724                 dec = (s32) dec;
3725         tb = mftb();
3726         vcpu->arch.dec_expires = dec + tb;
3727         vcpu->cpu = -1;
3728         vcpu->arch.thread_cpu = -1;
3729         vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3730
3731         vcpu->arch.iamr = mfspr(SPRN_IAMR);
3732         vcpu->arch.pspb = mfspr(SPRN_PSPB);
3733         vcpu->arch.fscr = mfspr(SPRN_FSCR);
3734         vcpu->arch.tar = mfspr(SPRN_TAR);
3735         vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3736         vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3737         vcpu->arch.bescr = mfspr(SPRN_BESCR);
3738         vcpu->arch.wort = mfspr(SPRN_WORT);
3739         vcpu->arch.tid = mfspr(SPRN_TIDR);
3740         vcpu->arch.amr = mfspr(SPRN_AMR);
3741         vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3742         vcpu->arch.dscr = mfspr(SPRN_DSCR);
3743
3744         mtspr(SPRN_PSPB, 0);
3745         mtspr(SPRN_WORT, 0);
3746         mtspr(SPRN_UAMOR, 0);
3747         mtspr(SPRN_DSCR, host_dscr);
3748         mtspr(SPRN_TIDR, host_tidr);
3749         mtspr(SPRN_IAMR, host_iamr);
3750         mtspr(SPRN_PSPB, 0);
3751
3752         if (host_amr != vcpu->arch.amr)
3753                 mtspr(SPRN_AMR, host_amr);
3754
3755         if (host_fscr != vcpu->arch.fscr)
3756                 mtspr(SPRN_FSCR, host_fscr);
3757
3758         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3759         store_fp_state(&vcpu->arch.fp);
3760 #ifdef CONFIG_ALTIVEC
3761         store_vr_state(&vcpu->arch.vr);
3762 #endif
3763         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3764
3765         if (cpu_has_feature(CPU_FTR_TM) ||
3766             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3767                 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3768
3769         save_pmu = 1;
3770         if (vcpu->arch.vpa.pinned_addr) {
3771                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3772                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3773                 lp->yield_count = cpu_to_be32(yield_count);
3774                 vcpu->arch.vpa.dirty = 1;
3775                 save_pmu = lp->pmcregs_in_use;
3776         }
3777         /* Must save pmu if this guest is capable of running nested guests */
3778         save_pmu |= nesting_enabled(vcpu->kvm);
3779
3780         kvmhv_save_guest_pmu(vcpu, save_pmu);
3781
3782         vc->entry_exit_map = 0x101;
3783         vc->in_guest = 0;
3784
3785         mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3786         mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3787
3788         kvmhv_load_host_pmu();
3789
3790         kvmppc_subcore_exit_guest();
3791
3792         return trap;
3793 }
3794
3795 /*
3796  * Wait for some other vcpu thread to execute us, and
3797  * wake us up when we need to handle something in the host.
3798  */
3799 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3800                                  struct kvm_vcpu *vcpu, int wait_state)
3801 {
3802         DEFINE_WAIT(wait);
3803
3804         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3805         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3806                 spin_unlock(&vc->lock);
3807                 schedule();
3808                 spin_lock(&vc->lock);
3809         }
3810         finish_wait(&vcpu->arch.cpu_run, &wait);
3811 }
3812
3813 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3814 {
3815         if (!halt_poll_ns_grow)
3816                 return;
3817
3818         vc->halt_poll_ns *= halt_poll_ns_grow;
3819         if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3820                 vc->halt_poll_ns = halt_poll_ns_grow_start;
3821 }
3822
3823 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3824 {
3825         if (halt_poll_ns_shrink == 0)
3826                 vc->halt_poll_ns = 0;
3827         else
3828                 vc->halt_poll_ns /= halt_poll_ns_shrink;
3829 }
3830
3831 #ifdef CONFIG_KVM_XICS
3832 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3833 {
3834         if (!xics_on_xive())
3835                 return false;
3836         return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3837                 vcpu->arch.xive_saved_state.cppr;
3838 }
3839 #else
3840 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3841 {
3842         return false;
3843 }
3844 #endif /* CONFIG_KVM_XICS */
3845
3846 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3847 {
3848         if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3849             kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3850                 return true;
3851
3852         return false;
3853 }
3854
3855 /*
3856  * Check to see if any of the runnable vcpus on the vcore have pending
3857  * exceptions or are no longer ceded
3858  */
3859 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3860 {
3861         struct kvm_vcpu *vcpu;
3862         int i;
3863
3864         for_each_runnable_thread(i, vcpu, vc) {
3865                 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3866                         return 1;
3867         }
3868
3869         return 0;
3870 }
3871
3872 /*
3873  * All the vcpus in this vcore are idle, so wait for a decrementer
3874  * or external interrupt to one of the vcpus.  vc->lock is held.
3875  */
3876 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3877 {
3878         ktime_t cur, start_poll, start_wait;
3879         int do_sleep = 1;
3880         u64 block_ns;
3881
3882         /* Poll for pending exceptions and ceded state */
3883         cur = start_poll = ktime_get();
3884         if (vc->halt_poll_ns) {
3885                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3886                 ++vc->runner->stat.halt_attempted_poll;
3887
3888                 vc->vcore_state = VCORE_POLLING;
3889                 spin_unlock(&vc->lock);
3890
3891                 do {
3892                         if (kvmppc_vcore_check_block(vc)) {
3893                                 do_sleep = 0;
3894                                 break;
3895                         }
3896                         cur = ktime_get();
3897                 } while (single_task_running() && ktime_before(cur, stop));
3898
3899                 spin_lock(&vc->lock);
3900                 vc->vcore_state = VCORE_INACTIVE;
3901
3902                 if (!do_sleep) {
3903                         ++vc->runner->stat.halt_successful_poll;
3904                         goto out;
3905                 }
3906         }
3907
3908         prepare_to_rcuwait(&vc->wait);
3909         set_current_state(TASK_INTERRUPTIBLE);
3910         if (kvmppc_vcore_check_block(vc)) {
3911                 finish_rcuwait(&vc->wait);
3912                 do_sleep = 0;
3913                 /* If we polled, count this as a successful poll */
3914                 if (vc->halt_poll_ns)
3915                         ++vc->runner->stat.halt_successful_poll;
3916                 goto out;
3917         }
3918
3919         start_wait = ktime_get();
3920
3921         vc->vcore_state = VCORE_SLEEPING;
3922         trace_kvmppc_vcore_blocked(vc, 0);
3923         spin_unlock(&vc->lock);
3924         schedule();
3925         finish_rcuwait(&vc->wait);
3926         spin_lock(&vc->lock);
3927         vc->vcore_state = VCORE_INACTIVE;
3928         trace_kvmppc_vcore_blocked(vc, 1);
3929         ++vc->runner->stat.halt_successful_wait;
3930
3931         cur = ktime_get();
3932
3933 out:
3934         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3935
3936         /* Attribute wait time */
3937         if (do_sleep) {
3938                 vc->runner->stat.halt_wait_ns +=
3939                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
3940                 /* Attribute failed poll time */
3941                 if (vc->halt_poll_ns)
3942                         vc->runner->stat.halt_poll_fail_ns +=
3943                                 ktime_to_ns(start_wait) -
3944                                 ktime_to_ns(start_poll);
3945         } else {
3946                 /* Attribute successful poll time */
3947                 if (vc->halt_poll_ns)
3948                         vc->runner->stat.halt_poll_success_ns +=
3949                                 ktime_to_ns(cur) -
3950                                 ktime_to_ns(start_poll);
3951         }
3952
3953         /* Adjust poll time */
3954         if (halt_poll_ns) {
3955                 if (block_ns <= vc->halt_poll_ns)
3956                         ;
3957                 /* We slept and blocked for longer than the max halt time */
3958                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3959                         shrink_halt_poll_ns(vc);
3960                 /* We slept and our poll time is too small */
3961                 else if (vc->halt_poll_ns < halt_poll_ns &&
3962                                 block_ns < halt_poll_ns)
3963                         grow_halt_poll_ns(vc);
3964                 if (vc->halt_poll_ns > halt_poll_ns)
3965                         vc->halt_poll_ns = halt_poll_ns;
3966         } else
3967                 vc->halt_poll_ns = 0;
3968
3969         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3970 }
3971
3972 /*
3973  * This never fails for a radix guest, as none of the operations it does
3974  * for a radix guest can fail or have a way to report failure.
3975  * kvmhv_run_single_vcpu() relies on this fact.
3976  */
3977 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3978 {
3979         int r = 0;
3980         struct kvm *kvm = vcpu->kvm;
3981
3982         mutex_lock(&kvm->arch.mmu_setup_lock);
3983         if (!kvm->arch.mmu_ready) {
3984                 if (!kvm_is_radix(kvm))
3985                         r = kvmppc_hv_setup_htab_rma(vcpu);
3986                 if (!r) {
3987                         if (cpu_has_feature(CPU_FTR_ARCH_300))
3988                                 kvmppc_setup_partition_table(kvm);
3989                         kvm->arch.mmu_ready = 1;
3990                 }
3991         }
3992         mutex_unlock(&kvm->arch.mmu_setup_lock);
3993         return r;
3994 }
3995
3996 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
3997 {
3998         struct kvm_run *run = vcpu->run;
3999         int n_ceded, i, r;
4000         struct kvmppc_vcore *vc;
4001         struct kvm_vcpu *v;
4002
4003         trace_kvmppc_run_vcpu_enter(vcpu);
4004
4005         run->exit_reason = 0;
4006         vcpu->arch.ret = RESUME_GUEST;
4007         vcpu->arch.trap = 0;
4008         kvmppc_update_vpas(vcpu);
4009
4010         /*
4011          * Synchronize with other threads in this virtual core
4012          */
4013         vc = vcpu->arch.vcore;
4014         spin_lock(&vc->lock);
4015         vcpu->arch.ceded = 0;
4016         vcpu->arch.run_task = current;
4017         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4018         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4019         vcpu->arch.busy_preempt = TB_NIL;
4020         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4021         ++vc->n_runnable;
4022
4023         /*
4024          * This happens the first time this is called for a vcpu.
4025          * If the vcore is already running, we may be able to start
4026          * this thread straight away and have it join in.
4027          */
4028         if (!signal_pending(current)) {
4029                 if ((vc->vcore_state == VCORE_PIGGYBACK ||
4030                      vc->vcore_state == VCORE_RUNNING) &&
4031                            !VCORE_IS_EXITING(vc)) {
4032                         kvmppc_create_dtl_entry(vcpu, vc);
4033                         kvmppc_start_thread(vcpu, vc);
4034                         trace_kvm_guest_enter(vcpu);
4035                 } else if (vc->vcore_state == VCORE_SLEEPING) {
4036                         rcuwait_wake_up(&vc->wait);
4037                 }
4038
4039         }
4040
4041         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4042                !signal_pending(current)) {
4043                 /* See if the MMU is ready to go */
4044                 if (!vcpu->kvm->arch.mmu_ready) {
4045                         spin_unlock(&vc->lock);
4046                         r = kvmhv_setup_mmu(vcpu);
4047                         spin_lock(&vc->lock);
4048                         if (r) {
4049                                 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4050                                 run->fail_entry.
4051                                         hardware_entry_failure_reason = 0;
4052                                 vcpu->arch.ret = r;
4053                                 break;
4054                         }
4055                 }
4056
4057                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4058                         kvmppc_vcore_end_preempt(vc);
4059
4060                 if (vc->vcore_state != VCORE_INACTIVE) {
4061                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4062                         continue;
4063                 }
4064                 for_each_runnable_thread(i, v, vc) {
4065                         kvmppc_core_prepare_to_enter(v);
4066                         if (signal_pending(v->arch.run_task)) {
4067                                 kvmppc_remove_runnable(vc, v);
4068                                 v->stat.signal_exits++;
4069                                 v->run->exit_reason = KVM_EXIT_INTR;
4070                                 v->arch.ret = -EINTR;
4071                                 wake_up(&v->arch.cpu_run);
4072                         }
4073                 }
4074                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4075                         break;
4076                 n_ceded = 0;
4077                 for_each_runnable_thread(i, v, vc) {
4078                         if (!kvmppc_vcpu_woken(v))
4079                                 n_ceded += v->arch.ceded;
4080                         else
4081                                 v->arch.ceded = 0;
4082                 }
4083                 vc->runner = vcpu;
4084                 if (n_ceded == vc->n_runnable) {
4085                         kvmppc_vcore_blocked(vc);
4086                 } else if (need_resched()) {
4087                         kvmppc_vcore_preempt(vc);
4088                         /* Let something else run */
4089                         cond_resched_lock(&vc->lock);
4090                         if (vc->vcore_state == VCORE_PREEMPT)
4091                                 kvmppc_vcore_end_preempt(vc);
4092                 } else {
4093                         kvmppc_run_core(vc);
4094                 }
4095                 vc->runner = NULL;
4096         }
4097
4098         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4099                (vc->vcore_state == VCORE_RUNNING ||
4100                 vc->vcore_state == VCORE_EXITING ||
4101                 vc->vcore_state == VCORE_PIGGYBACK))
4102                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4103
4104         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4105                 kvmppc_vcore_end_preempt(vc);
4106
4107         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4108                 kvmppc_remove_runnable(vc, vcpu);
4109                 vcpu->stat.signal_exits++;
4110                 run->exit_reason = KVM_EXIT_INTR;
4111                 vcpu->arch.ret = -EINTR;
4112         }
4113
4114         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4115                 /* Wake up some vcpu to run the core */
4116                 i = -1;
4117                 v = next_runnable_thread(vc, &i);
4118                 wake_up(&v->arch.cpu_run);
4119         }
4120
4121         trace_kvmppc_run_vcpu_exit(vcpu);
4122         spin_unlock(&vc->lock);
4123         return vcpu->arch.ret;
4124 }
4125
4126 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4127                           unsigned long lpcr)
4128 {
4129         struct kvm_run *run = vcpu->run;
4130         int trap, r, pcpu;
4131         int srcu_idx, lpid;
4132         struct kvmppc_vcore *vc;
4133         struct kvm *kvm = vcpu->kvm;
4134         struct kvm_nested_guest *nested = vcpu->arch.nested;
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
4142         vc = vcpu->arch.vcore;
4143         vcpu->arch.ceded = 0;
4144         vcpu->arch.run_task = current;
4145         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4146         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4147         vcpu->arch.busy_preempt = TB_NIL;
4148         vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4149         vc->runnable_threads[0] = vcpu;
4150         vc->n_runnable = 1;
4151         vc->runner = vcpu;
4152
4153         /* See if the MMU is ready to go */
4154         if (!kvm->arch.mmu_ready)
4155                 kvmhv_setup_mmu(vcpu);
4156
4157         if (need_resched())
4158                 cond_resched();
4159
4160         kvmppc_update_vpas(vcpu);
4161
4162         init_vcore_to_run(vc);
4163         vc->preempt_tb = TB_NIL;
4164
4165         preempt_disable();
4166         pcpu = smp_processor_id();
4167         vc->pcpu = pcpu;
4168         kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4169
4170         local_irq_disable();
4171         hard_irq_disable();
4172         if (signal_pending(current))
4173                 goto sigpend;
4174         if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4175                 goto out;
4176
4177         if (!nested) {
4178                 kvmppc_core_prepare_to_enter(vcpu);
4179                 if (vcpu->arch.doorbell_request) {
4180                         vc->dpdes = 1;
4181                         smp_wmb();
4182                         vcpu->arch.doorbell_request = 0;
4183                 }
4184                 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4185                              &vcpu->arch.pending_exceptions))
4186                         lpcr |= LPCR_MER;
4187         } else if (vcpu->arch.pending_exceptions ||
4188                    vcpu->arch.doorbell_request ||
4189                    xive_interrupt_pending(vcpu)) {
4190                 vcpu->arch.ret = RESUME_HOST;
4191                 goto out;
4192         }
4193
4194         kvmppc_clear_host_core(pcpu);
4195
4196         local_paca->kvm_hstate.napping = 0;
4197         local_paca->kvm_hstate.kvm_split_mode = NULL;
4198         kvmppc_start_thread(vcpu, vc);
4199         kvmppc_create_dtl_entry(vcpu, vc);
4200         trace_kvm_guest_enter(vcpu);
4201
4202         vc->vcore_state = VCORE_RUNNING;
4203         trace_kvmppc_run_core(vc, 0);
4204
4205         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4206                 lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4207                 mtspr(SPRN_LPID, lpid);
4208                 isync();
4209                 kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4210         }
4211
4212         guest_enter_irqoff();
4213
4214         srcu_idx = srcu_read_lock(&kvm->srcu);
4215
4216         this_cpu_disable_ftrace();
4217
4218         /* Tell lockdep that we're about to enable interrupts */
4219         trace_hardirqs_on();
4220
4221         trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4222         vcpu->arch.trap = trap;
4223
4224         trace_hardirqs_off();
4225
4226         this_cpu_enable_ftrace();
4227
4228         srcu_read_unlock(&kvm->srcu, srcu_idx);
4229
4230         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4231                 mtspr(SPRN_LPID, kvm->arch.host_lpid);
4232                 isync();
4233         }
4234
4235         set_irq_happened(trap);
4236
4237         kvmppc_set_host_core(pcpu);
4238
4239         local_irq_enable();
4240         guest_exit();
4241
4242         cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4243
4244         preempt_enable();
4245
4246         /*
4247          * cancel pending decrementer exception if DEC is now positive, or if
4248          * entering a nested guest in which case the decrementer is now owned
4249          * by L2 and the L1 decrementer is provided in hdec_expires
4250          */
4251         if (kvmppc_core_pending_dec(vcpu) &&
4252                         ((get_tb() < vcpu->arch.dec_expires) ||
4253                          (trap == BOOK3S_INTERRUPT_SYSCALL &&
4254                           kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4255                 kvmppc_core_dequeue_dec(vcpu);
4256
4257         trace_kvm_guest_exit(vcpu);
4258         r = RESUME_GUEST;
4259         if (trap) {
4260                 if (!nested)
4261                         r = kvmppc_handle_exit_hv(vcpu, current);
4262                 else
4263                         r = kvmppc_handle_nested_exit(vcpu);
4264         }
4265         vcpu->arch.ret = r;
4266
4267         if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4268             !kvmppc_vcpu_woken(vcpu)) {
4269                 kvmppc_set_timer(vcpu);
4270                 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4271                         if (signal_pending(current)) {
4272                                 vcpu->stat.signal_exits++;
4273                                 run->exit_reason = KVM_EXIT_INTR;
4274                                 vcpu->arch.ret = -EINTR;
4275                                 break;
4276                         }
4277                         spin_lock(&vc->lock);
4278                         kvmppc_vcore_blocked(vc);
4279                         spin_unlock(&vc->lock);
4280                 }
4281         }
4282         vcpu->arch.ceded = 0;
4283
4284         vc->vcore_state = VCORE_INACTIVE;
4285         trace_kvmppc_run_core(vc, 1);
4286
4287  done:
4288         kvmppc_remove_runnable(vc, vcpu);
4289         trace_kvmppc_run_vcpu_exit(vcpu);
4290
4291         return vcpu->arch.ret;
4292
4293  sigpend:
4294         vcpu->stat.signal_exits++;
4295         run->exit_reason = KVM_EXIT_INTR;
4296         vcpu->arch.ret = -EINTR;
4297  out:
4298         local_irq_enable();
4299         preempt_enable();
4300         goto done;
4301 }
4302
4303 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
4304 {
4305         struct kvm_run *run = vcpu->run;
4306         int r;
4307         int srcu_idx;
4308         unsigned long ebb_regs[3] = {}; /* shut up GCC */
4309         unsigned long user_tar = 0;
4310         unsigned int user_vrsave;
4311         struct kvm *kvm;
4312
4313         if (!vcpu->arch.sane) {
4314                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4315                 return -EINVAL;
4316         }
4317
4318         /*
4319          * Don't allow entry with a suspended transaction, because
4320          * the guest entry/exit code will lose it.
4321          * If the guest has TM enabled, save away their TM-related SPRs
4322          * (they will get restored by the TM unavailable interrupt).
4323          */
4324 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4325         if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4326             (current->thread.regs->msr & MSR_TM)) {
4327                 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4328                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4329                         run->fail_entry.hardware_entry_failure_reason = 0;
4330                         return -EINVAL;
4331                 }
4332                 /* Enable TM so we can read the TM SPRs */
4333                 mtmsr(mfmsr() | MSR_TM);
4334                 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4335                 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4336                 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4337                 current->thread.regs->msr &= ~MSR_TM;
4338         }
4339 #endif
4340
4341         /*
4342          * Force online to 1 for the sake of old userspace which doesn't
4343          * set it.
4344          */
4345         if (!vcpu->arch.online) {
4346                 atomic_inc(&vcpu->arch.vcore->online_count);
4347                 vcpu->arch.online = 1;
4348         }
4349
4350         kvmppc_core_prepare_to_enter(vcpu);
4351
4352         /* No need to go into the guest when all we'll do is come back out */
4353         if (signal_pending(current)) {
4354                 run->exit_reason = KVM_EXIT_INTR;
4355                 return -EINTR;
4356         }
4357
4358         kvm = vcpu->kvm;
4359         atomic_inc(&kvm->arch.vcpus_running);
4360         /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4361         smp_mb();
4362
4363         flush_all_to_thread(current);
4364
4365         /* Save userspace EBB and other register values */
4366         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4367                 ebb_regs[0] = mfspr(SPRN_EBBHR);
4368                 ebb_regs[1] = mfspr(SPRN_EBBRR);
4369                 ebb_regs[2] = mfspr(SPRN_BESCR);
4370                 user_tar = mfspr(SPRN_TAR);
4371         }
4372         user_vrsave = mfspr(SPRN_VRSAVE);
4373
4374         vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4375         vcpu->arch.pgdir = kvm->mm->pgd;
4376         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4377
4378         do {
4379                 /*
4380                  * The TLB prefetch bug fixup is only in the kvmppc_run_vcpu
4381                  * path, which also handles hash and dependent threads mode.
4382                  */
4383                 if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4384                     !cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
4385                         r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
4386                                                   vcpu->arch.vcore->lpcr);
4387                 else
4388                         r = kvmppc_run_vcpu(vcpu);
4389
4390                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4391                     !(vcpu->arch.shregs.msr & MSR_PR)) {
4392                         trace_kvm_hcall_enter(vcpu);
4393                         r = kvmppc_pseries_do_hcall(vcpu);
4394                         trace_kvm_hcall_exit(vcpu, r);
4395                         kvmppc_core_prepare_to_enter(vcpu);
4396                 } else if (r == RESUME_PAGE_FAULT) {
4397                         srcu_idx = srcu_read_lock(&kvm->srcu);
4398                         r = kvmppc_book3s_hv_page_fault(vcpu,
4399                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4400                         srcu_read_unlock(&kvm->srcu, srcu_idx);
4401                 } else if (r == RESUME_PASSTHROUGH) {
4402                         if (WARN_ON(xics_on_xive()))
4403                                 r = H_SUCCESS;
4404                         else
4405                                 r = kvmppc_xics_rm_complete(vcpu, 0);
4406                 }
4407         } while (is_kvmppc_resume_guest(r));
4408
4409         /* Restore userspace EBB and other register values */
4410         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4411                 mtspr(SPRN_EBBHR, ebb_regs[0]);
4412                 mtspr(SPRN_EBBRR, ebb_regs[1]);
4413                 mtspr(SPRN_BESCR, ebb_regs[2]);
4414                 mtspr(SPRN_TAR, user_tar);
4415                 mtspr(SPRN_FSCR, current->thread.fscr);
4416         }
4417         mtspr(SPRN_VRSAVE, user_vrsave);
4418
4419         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4420         atomic_dec(&kvm->arch.vcpus_running);
4421         return r;
4422 }
4423
4424 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4425                                      int shift, int sllp)
4426 {
4427         (*sps)->page_shift = shift;
4428         (*sps)->slb_enc = sllp;
4429         (*sps)->enc[0].page_shift = shift;
4430         (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4431         /*
4432          * Add 16MB MPSS support (may get filtered out by userspace)
4433          */
4434         if (shift != 24) {
4435                 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4436                 if (penc != -1) {
4437                         (*sps)->enc[1].page_shift = 24;
4438                         (*sps)->enc[1].pte_enc = penc;
4439                 }
4440         }
4441         (*sps)++;
4442 }
4443
4444 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4445                                          struct kvm_ppc_smmu_info *info)
4446 {
4447         struct kvm_ppc_one_seg_page_size *sps;
4448
4449         /*
4450          * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4451          * POWER7 doesn't support keys for instruction accesses,
4452          * POWER8 and POWER9 do.
4453          */
4454         info->data_keys = 32;
4455         info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4456
4457         /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4458         info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4459         info->slb_size = 32;
4460
4461         /* We only support these sizes for now, and no muti-size segments */
4462         sps = &info->sps[0];
4463         kvmppc_add_seg_page_size(&sps, 12, 0);
4464         kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4465         kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4466
4467         /* If running as a nested hypervisor, we don't support HPT guests */
4468         if (kvmhv_on_pseries())
4469                 info->flags |= KVM_PPC_NO_HASH;
4470
4471         return 0;
4472 }
4473
4474 /*
4475  * Get (and clear) the dirty memory log for a memory slot.
4476  */
4477 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4478                                          struct kvm_dirty_log *log)
4479 {
4480         struct kvm_memslots *slots;
4481         struct kvm_memory_slot *memslot;
4482         int i, r;
4483         unsigned long n;
4484         unsigned long *buf, *p;
4485         struct kvm_vcpu *vcpu;
4486
4487         mutex_lock(&kvm->slots_lock);
4488
4489         r = -EINVAL;
4490         if (log->slot >= KVM_USER_MEM_SLOTS)
4491                 goto out;
4492
4493         slots = kvm_memslots(kvm);
4494         memslot = id_to_memslot(slots, log->slot);
4495         r = -ENOENT;
4496         if (!memslot || !memslot->dirty_bitmap)
4497                 goto out;
4498
4499         /*
4500          * Use second half of bitmap area because both HPT and radix
4501          * accumulate bits in the first half.
4502          */
4503         n = kvm_dirty_bitmap_bytes(memslot);
4504         buf = memslot->dirty_bitmap + n / sizeof(long);
4505         memset(buf, 0, n);
4506
4507         if (kvm_is_radix(kvm))
4508                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4509         else
4510                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4511         if (r)
4512                 goto out;
4513
4514         /*
4515          * We accumulate dirty bits in the first half of the
4516          * memslot's dirty_bitmap area, for when pages are paged
4517          * out or modified by the host directly.  Pick up these
4518          * bits and add them to the map.
4519          */
4520         p = memslot->dirty_bitmap;
4521         for (i = 0; i < n / sizeof(long); ++i)
4522                 buf[i] |= xchg(&p[i], 0);
4523
4524         /* Harvest dirty bits from VPA and DTL updates */
4525         /* Note: we never modify the SLB shadow buffer areas */
4526         kvm_for_each_vcpu(i, vcpu, kvm) {
4527                 spin_lock(&vcpu->arch.vpa_update_lock);
4528                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4529                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4530                 spin_unlock(&vcpu->arch.vpa_update_lock);
4531         }
4532
4533         r = -EFAULT;
4534         if (copy_to_user(log->dirty_bitmap, buf, n))
4535                 goto out;
4536
4537         r = 0;
4538 out:
4539         mutex_unlock(&kvm->slots_lock);
4540         return r;
4541 }
4542
4543 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
4544 {
4545         vfree(slot->arch.rmap);
4546         slot->arch.rmap = NULL;
4547 }
4548
4549 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4550                                         struct kvm_memory_slot *slot,
4551                                         const struct kvm_userspace_memory_region *mem,
4552                                         enum kvm_mr_change change)
4553 {
4554         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4555
4556         if (change == KVM_MR_CREATE) {
4557                 slot->arch.rmap = vzalloc(array_size(npages,
4558                                           sizeof(*slot->arch.rmap)));
4559                 if (!slot->arch.rmap)
4560                         return -ENOMEM;
4561         }
4562
4563         return 0;
4564 }
4565
4566 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4567                                 const struct kvm_userspace_memory_region *mem,
4568                                 const struct kvm_memory_slot *old,
4569                                 const struct kvm_memory_slot *new,
4570                                 enum kvm_mr_change change)
4571 {
4572         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4573
4574         /*
4575          * If we are making a new memslot, it might make
4576          * some address that was previously cached as emulated
4577          * MMIO be no longer emulated MMIO, so invalidate
4578          * all the caches of emulated MMIO translations.
4579          */
4580         if (npages)
4581                 atomic64_inc(&kvm->arch.mmio_update);
4582
4583         /*
4584          * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4585          * have already called kvm_arch_flush_shadow_memslot() to
4586          * flush shadow mappings.  For KVM_MR_CREATE we have no
4587          * previous mappings.  So the only case to handle is
4588          * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4589          * has been changed.
4590          * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4591          * to get rid of any THP PTEs in the partition-scoped page tables
4592          * so we can track dirtiness at the page level; we flush when
4593          * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4594          * using THP PTEs.
4595          */
4596         if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4597             ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4598                 kvmppc_radix_flush_memslot(kvm, old);
4599         /*
4600          * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4601          */
4602         if (!kvm->arch.secure_guest)
4603                 return;
4604
4605         switch (change) {
4606         case KVM_MR_CREATE:
4607                 /*
4608                  * @TODO kvmppc_uvmem_memslot_create() can fail and
4609                  * return error. Fix this.
4610                  */
4611                 kvmppc_uvmem_memslot_create(kvm, new);
4612                 break;
4613         case KVM_MR_DELETE:
4614                 kvmppc_uvmem_memslot_delete(kvm, old);
4615                 break;
4616         default:
4617                 /* TODO: Handle KVM_MR_MOVE */
4618                 break;
4619         }
4620 }
4621
4622 /*
4623  * Update LPCR values in kvm->arch and in vcores.
4624  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4625  * of kvm->arch.lpcr update).
4626  */
4627 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4628 {
4629         long int i;
4630         u32 cores_done = 0;
4631
4632         if ((kvm->arch.lpcr & mask) == lpcr)
4633                 return;
4634
4635         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4636
4637         for (i = 0; i < KVM_MAX_VCORES; ++i) {
4638                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4639                 if (!vc)
4640                         continue;
4641                 spin_lock(&vc->lock);
4642                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4643                 spin_unlock(&vc->lock);
4644                 if (++cores_done >= kvm->arch.online_vcores)
4645                         break;
4646         }
4647 }
4648
4649 void kvmppc_setup_partition_table(struct kvm *kvm)
4650 {
4651         unsigned long dw0, dw1;
4652
4653         if (!kvm_is_radix(kvm)) {
4654                 /* PS field - page size for VRMA */
4655                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4656                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4657                 /* HTABSIZE and HTABORG fields */
4658                 dw0 |= kvm->arch.sdr1;
4659
4660                 /* Second dword as set by userspace */
4661                 dw1 = kvm->arch.process_table;
4662         } else {
4663                 dw0 = PATB_HR | radix__get_tree_size() |
4664                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4665                 dw1 = PATB_GR | kvm->arch.process_table;
4666         }
4667         kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4668 }
4669
4670 /*
4671  * Set up HPT (hashed page table) and RMA (real-mode area).
4672  * Must be called with kvm->arch.mmu_setup_lock held.
4673  */
4674 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4675 {
4676         int err = 0;
4677         struct kvm *kvm = vcpu->kvm;
4678         unsigned long hva;
4679         struct kvm_memory_slot *memslot;
4680         struct vm_area_struct *vma;
4681         unsigned long lpcr = 0, senc;
4682         unsigned long psize, porder;
4683         int srcu_idx;
4684
4685         /* Allocate hashed page table (if not done already) and reset it */
4686         if (!kvm->arch.hpt.virt) {
4687                 int order = KVM_DEFAULT_HPT_ORDER;
4688                 struct kvm_hpt_info info;
4689
4690                 err = kvmppc_allocate_hpt(&info, order);
4691                 /* If we get here, it means userspace didn't specify a
4692                  * size explicitly.  So, try successively smaller
4693                  * sizes if the default failed. */
4694                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4695                         err  = kvmppc_allocate_hpt(&info, order);
4696
4697                 if (err < 0) {
4698                         pr_err("KVM: Couldn't alloc HPT\n");
4699                         goto out;
4700                 }
4701
4702                 kvmppc_set_hpt(kvm, &info);
4703         }
4704
4705         /* Look up the memslot for guest physical address 0 */
4706         srcu_idx = srcu_read_lock(&kvm->srcu);
4707         memslot = gfn_to_memslot(kvm, 0);
4708
4709         /* We must have some memory at 0 by now */
4710         err = -EINVAL;
4711         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4712                 goto out_srcu;
4713
4714         /* Look up the VMA for the start of this memory slot */
4715         hva = memslot->userspace_addr;
4716         mmap_read_lock(kvm->mm);
4717         vma = find_vma(kvm->mm, hva);
4718         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4719                 goto up_out;
4720
4721         psize = vma_kernel_pagesize(vma);
4722
4723         mmap_read_unlock(kvm->mm);
4724
4725         /* We can handle 4k, 64k or 16M pages in the VRMA */
4726         if (psize >= 0x1000000)
4727                 psize = 0x1000000;
4728         else if (psize >= 0x10000)
4729                 psize = 0x10000;
4730         else
4731                 psize = 0x1000;
4732         porder = __ilog2(psize);
4733
4734         senc = slb_pgsize_encoding(psize);
4735         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4736                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4737         /* Create HPTEs in the hash page table for the VRMA */
4738         kvmppc_map_vrma(vcpu, memslot, porder);
4739
4740         /* Update VRMASD field in the LPCR */
4741         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4742                 /* the -4 is to account for senc values starting at 0x10 */
4743                 lpcr = senc << (LPCR_VRMASD_SH - 4);
4744                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4745         }
4746
4747         /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4748         smp_wmb();
4749         err = 0;
4750  out_srcu:
4751         srcu_read_unlock(&kvm->srcu, srcu_idx);
4752  out:
4753         return err;
4754
4755  up_out:
4756         mmap_read_unlock(kvm->mm);
4757         goto out_srcu;
4758 }
4759
4760 /*
4761  * Must be called with kvm->arch.mmu_setup_lock held and
4762  * mmu_ready = 0 and no vcpus running.
4763  */
4764 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4765 {
4766         if (nesting_enabled(kvm))
4767                 kvmhv_release_all_nested(kvm);
4768         kvmppc_rmap_reset(kvm);
4769         kvm->arch.process_table = 0;
4770         /* Mutual exclusion with kvm_unmap_hva_range etc. */
4771         spin_lock(&kvm->mmu_lock);
4772         kvm->arch.radix = 0;
4773         spin_unlock(&kvm->mmu_lock);
4774         kvmppc_free_radix(kvm);
4775         kvmppc_update_lpcr(kvm, LPCR_VPM1,
4776                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4777         return 0;
4778 }
4779
4780 /*
4781  * Must be called with kvm->arch.mmu_setup_lock held and
4782  * mmu_ready = 0 and no vcpus running.
4783  */
4784 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4785 {
4786         int err;
4787
4788         err = kvmppc_init_vm_radix(kvm);
4789         if (err)
4790                 return err;
4791         kvmppc_rmap_reset(kvm);
4792         /* Mutual exclusion with kvm_unmap_hva_range etc. */
4793         spin_lock(&kvm->mmu_lock);
4794         kvm->arch.radix = 1;
4795         spin_unlock(&kvm->mmu_lock);
4796         kvmppc_free_hpt(&kvm->arch.hpt);
4797         kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4798                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4799         return 0;
4800 }
4801
4802 #ifdef CONFIG_KVM_XICS
4803 /*
4804  * Allocate a per-core structure for managing state about which cores are
4805  * running in the host versus the guest and for exchanging data between
4806  * real mode KVM and CPU running in the host.
4807  * This is only done for the first VM.
4808  * The allocated structure stays even if all VMs have stopped.
4809  * It is only freed when the kvm-hv module is unloaded.
4810  * It's OK for this routine to fail, we just don't support host
4811  * core operations like redirecting H_IPI wakeups.
4812  */
4813 void kvmppc_alloc_host_rm_ops(void)
4814 {
4815         struct kvmppc_host_rm_ops *ops;
4816         unsigned long l_ops;
4817         int cpu, core;
4818         int size;
4819
4820         /* Not the first time here ? */
4821         if (kvmppc_host_rm_ops_hv != NULL)
4822                 return;
4823
4824         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4825         if (!ops)
4826                 return;
4827
4828         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4829         ops->rm_core = kzalloc(size, GFP_KERNEL);
4830
4831         if (!ops->rm_core) {
4832                 kfree(ops);
4833                 return;
4834         }
4835
4836         cpus_read_lock();
4837
4838         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4839                 if (!cpu_online(cpu))
4840                         continue;
4841
4842                 core = cpu >> threads_shift;
4843                 ops->rm_core[core].rm_state.in_host = 1;
4844         }
4845
4846         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4847
4848         /*
4849          * Make the contents of the kvmppc_host_rm_ops structure visible
4850          * to other CPUs before we assign it to the global variable.
4851          * Do an atomic assignment (no locks used here), but if someone
4852          * beats us to it, just free our copy and return.
4853          */
4854         smp_wmb();
4855         l_ops = (unsigned long) ops;
4856
4857         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4858                 cpus_read_unlock();
4859                 kfree(ops->rm_core);
4860                 kfree(ops);
4861                 return;
4862         }
4863
4864         cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4865                                              "ppc/kvm_book3s:prepare",
4866                                              kvmppc_set_host_core,
4867                                              kvmppc_clear_host_core);
4868         cpus_read_unlock();
4869 }
4870
4871 void kvmppc_free_host_rm_ops(void)
4872 {
4873         if (kvmppc_host_rm_ops_hv) {
4874                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4875                 kfree(kvmppc_host_rm_ops_hv->rm_core);
4876                 kfree(kvmppc_host_rm_ops_hv);
4877                 kvmppc_host_rm_ops_hv = NULL;
4878         }
4879 }
4880 #endif
4881
4882 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4883 {
4884         unsigned long lpcr, lpid;
4885         char buf[32];
4886         int ret;
4887
4888         mutex_init(&kvm->arch.uvmem_lock);
4889         INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
4890         mutex_init(&kvm->arch.mmu_setup_lock);
4891
4892         /* Allocate the guest's logical partition ID */
4893
4894         lpid = kvmppc_alloc_lpid();
4895         if ((long)lpid < 0)
4896                 return -ENOMEM;
4897         kvm->arch.lpid = lpid;
4898
4899         kvmppc_alloc_host_rm_ops();
4900
4901         kvmhv_vm_nested_init(kvm);
4902
4903         /*
4904          * Since we don't flush the TLB when tearing down a VM,
4905          * and this lpid might have previously been used,
4906          * make sure we flush on each core before running the new VM.
4907          * On POWER9, the tlbie in mmu_partition_table_set_entry()
4908          * does this flush for us.
4909          */
4910         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4911                 cpumask_setall(&kvm->arch.need_tlb_flush);
4912
4913         /* Start out with the default set of hcalls enabled */
4914         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4915                sizeof(kvm->arch.enabled_hcalls));
4916
4917         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4918                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4919
4920         /* Init LPCR for virtual RMA mode */
4921         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4922                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
4923                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4924                 lpcr &= LPCR_PECE | LPCR_LPES;
4925         } else {
4926                 lpcr = 0;
4927         }
4928         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4929                 LPCR_VPM0 | LPCR_VPM1;
4930         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4931                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4932         /* On POWER8 turn on online bit to enable PURR/SPURR */
4933         if (cpu_has_feature(CPU_FTR_ARCH_207S))
4934                 lpcr |= LPCR_ONL;
4935         /*
4936          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4937          * Set HVICE bit to enable hypervisor virtualization interrupts.
4938          * Set HEIC to prevent OS interrupts to go to hypervisor (should
4939          * be unnecessary but better safe than sorry in case we re-enable
4940          * EE in HV mode with this LPCR still set)
4941          */
4942         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4943                 lpcr &= ~LPCR_VPM0;
4944                 lpcr |= LPCR_HVICE | LPCR_HEIC;
4945
4946                 /*
4947                  * If xive is enabled, we route 0x500 interrupts directly
4948                  * to the guest.
4949                  */
4950                 if (xics_on_xive())
4951                         lpcr |= LPCR_LPES;
4952         }
4953
4954         /*
4955          * If the host uses radix, the guest starts out as radix.
4956          */
4957         if (radix_enabled()) {
4958                 kvm->arch.radix = 1;
4959                 kvm->arch.mmu_ready = 1;
4960                 lpcr &= ~LPCR_VPM1;
4961                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4962                 ret = kvmppc_init_vm_radix(kvm);
4963                 if (ret) {
4964                         kvmppc_free_lpid(kvm->arch.lpid);
4965                         return ret;
4966                 }
4967                 kvmppc_setup_partition_table(kvm);
4968         }
4969
4970         kvm->arch.lpcr = lpcr;
4971
4972         /* Initialization for future HPT resizes */
4973         kvm->arch.resize_hpt = NULL;
4974
4975         /*
4976          * Work out how many sets the TLB has, for the use of
4977          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4978          */
4979         if (cpu_has_feature(CPU_FTR_ARCH_31)) {
4980                 /*
4981                  * P10 will flush all the congruence class with a single tlbiel
4982                  */
4983                 kvm->arch.tlb_sets = 1;
4984         } else if (radix_enabled())
4985                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
4986         else if (cpu_has_feature(CPU_FTR_ARCH_300))
4987                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
4988         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4989                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
4990         else
4991                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
4992
4993         /*
4994          * Track that we now have a HV mode VM active. This blocks secondary
4995          * CPU threads from coming online.
4996          * On POWER9, we only need to do this if the "indep_threads_mode"
4997          * module parameter has been set to N.
4998          */
4999         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5000                 if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
5001                         pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
5002                         kvm->arch.threads_indep = true;
5003                 } else {
5004                         kvm->arch.threads_indep = indep_threads_mode;
5005                 }
5006         }
5007         if (!kvm->arch.threads_indep)
5008                 kvm_hv_vm_activated();
5009
5010         /*
5011          * Initialize smt_mode depending on processor.
5012          * POWER8 and earlier have to use "strict" threading, where
5013          * all vCPUs in a vcore have to run on the same (sub)core,
5014          * whereas on POWER9 the threads can each run a different
5015          * guest.
5016          */
5017         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5018                 kvm->arch.smt_mode = threads_per_subcore;
5019         else
5020                 kvm->arch.smt_mode = 1;
5021         kvm->arch.emul_smt_mode = 1;
5022
5023         /*
5024          * Create a debugfs directory for the VM
5025          */
5026         snprintf(buf, sizeof(buf), "vm%d", current->pid);
5027         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
5028         kvmppc_mmu_debugfs_init(kvm);
5029         if (radix_enabled())
5030                 kvmhv_radix_debugfs_init(kvm);
5031
5032         return 0;
5033 }
5034
5035 static void kvmppc_free_vcores(struct kvm *kvm)
5036 {
5037         long int i;
5038
5039         for (i = 0; i < KVM_MAX_VCORES; ++i)
5040                 kfree(kvm->arch.vcores[i]);
5041         kvm->arch.online_vcores = 0;
5042 }
5043
5044 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5045 {
5046         debugfs_remove_recursive(kvm->arch.debugfs_dir);
5047
5048         if (!kvm->arch.threads_indep)
5049                 kvm_hv_vm_deactivated();
5050
5051         kvmppc_free_vcores(kvm);
5052
5053
5054         if (kvm_is_radix(kvm))
5055                 kvmppc_free_radix(kvm);
5056         else
5057                 kvmppc_free_hpt(&kvm->arch.hpt);
5058
5059         /* Perform global invalidation and return lpid to the pool */
5060         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5061                 if (nesting_enabled(kvm))
5062                         kvmhv_release_all_nested(kvm);
5063                 kvm->arch.process_table = 0;
5064                 if (kvm->arch.secure_guest)
5065                         uv_svm_terminate(kvm->arch.lpid);
5066                 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5067         }
5068
5069         kvmppc_free_lpid(kvm->arch.lpid);
5070
5071         kvmppc_free_pimap(kvm);
5072 }
5073
5074 /* We don't need to emulate any privileged instructions or dcbz */
5075 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5076                                      unsigned int inst, int *advance)
5077 {
5078         return EMULATE_FAIL;
5079 }
5080
5081 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5082                                         ulong spr_val)
5083 {
5084         return EMULATE_FAIL;
5085 }
5086
5087 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5088                                         ulong *spr_val)
5089 {
5090         return EMULATE_FAIL;
5091 }
5092
5093 static int kvmppc_core_check_processor_compat_hv(void)
5094 {
5095         if (cpu_has_feature(CPU_FTR_HVMODE) &&
5096             cpu_has_feature(CPU_FTR_ARCH_206))
5097                 return 0;
5098
5099         /* POWER9 in radix mode is capable of being a nested hypervisor. */
5100         if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5101                 return 0;
5102
5103         return -EIO;
5104 }
5105
5106 #ifdef CONFIG_KVM_XICS
5107
5108 void kvmppc_free_pimap(struct kvm *kvm)
5109 {
5110         kfree(kvm->arch.pimap);
5111 }
5112
5113 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5114 {
5115         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5116 }
5117
5118 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5119 {
5120         struct irq_desc *desc;
5121         struct kvmppc_irq_map *irq_map;
5122         struct kvmppc_passthru_irqmap *pimap;
5123         struct irq_chip *chip;
5124         int i, rc = 0;
5125
5126         if (!kvm_irq_bypass)
5127                 return 1;
5128
5129         desc = irq_to_desc(host_irq);
5130         if (!desc)
5131                 return -EIO;
5132
5133         mutex_lock(&kvm->lock);
5134
5135         pimap = kvm->arch.pimap;
5136         if (pimap == NULL) {
5137                 /* First call, allocate structure to hold IRQ map */
5138                 pimap = kvmppc_alloc_pimap();
5139                 if (pimap == NULL) {
5140                         mutex_unlock(&kvm->lock);
5141                         return -ENOMEM;
5142                 }
5143                 kvm->arch.pimap = pimap;
5144         }
5145
5146         /*
5147          * For now, we only support interrupts for which the EOI operation
5148          * is an OPAL call followed by a write to XIRR, since that's
5149          * what our real-mode EOI code does, or a XIVE interrupt
5150          */
5151         chip = irq_data_get_irq_chip(&desc->irq_data);
5152         if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5153                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5154                         host_irq, guest_gsi);
5155                 mutex_unlock(&kvm->lock);
5156                 return -ENOENT;
5157         }
5158
5159         /*
5160          * See if we already have an entry for this guest IRQ number.
5161          * If it's mapped to a hardware IRQ number, that's an error,
5162          * otherwise re-use this entry.
5163          */
5164         for (i = 0; i < pimap->n_mapped; i++) {
5165                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5166                         if (pimap->mapped[i].r_hwirq) {
5167                                 mutex_unlock(&kvm->lock);
5168                                 return -EINVAL;
5169                         }
5170                         break;
5171                 }
5172         }
5173
5174         if (i == KVMPPC_PIRQ_MAPPED) {
5175                 mutex_unlock(&kvm->lock);
5176                 return -EAGAIN;         /* table is full */
5177         }
5178
5179         irq_map = &pimap->mapped[i];
5180
5181         irq_map->v_hwirq = guest_gsi;
5182         irq_map->desc = desc;
5183
5184         /*
5185          * Order the above two stores before the next to serialize with
5186          * the KVM real mode handler.
5187          */
5188         smp_wmb();
5189         irq_map->r_hwirq = desc->irq_data.hwirq;
5190
5191         if (i == pimap->n_mapped)
5192                 pimap->n_mapped++;
5193
5194         if (xics_on_xive())
5195                 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5196         else
5197                 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5198         if (rc)
5199                 irq_map->r_hwirq = 0;
5200
5201         mutex_unlock(&kvm->lock);
5202
5203         return 0;
5204 }
5205
5206 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5207 {
5208         struct irq_desc *desc;
5209         struct kvmppc_passthru_irqmap *pimap;
5210         int i, rc = 0;
5211
5212         if (!kvm_irq_bypass)
5213                 return 0;
5214
5215         desc = irq_to_desc(host_irq);
5216         if (!desc)
5217                 return -EIO;
5218
5219         mutex_lock(&kvm->lock);
5220         if (!kvm->arch.pimap)
5221                 goto unlock;
5222
5223         pimap = kvm->arch.pimap;
5224
5225         for (i = 0; i < pimap->n_mapped; i++) {
5226                 if (guest_gsi == pimap->mapped[i].v_hwirq)
5227                         break;
5228         }
5229
5230         if (i == pimap->n_mapped) {
5231                 mutex_unlock(&kvm->lock);
5232                 return -ENODEV;
5233         }
5234
5235         if (xics_on_xive())
5236                 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5237         else
5238                 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5239
5240         /* invalidate the entry (what do do on error from the above ?) */
5241         pimap->mapped[i].r_hwirq = 0;
5242
5243         /*
5244          * We don't free this structure even when the count goes to
5245          * zero. The structure is freed when we destroy the VM.
5246          */
5247  unlock:
5248         mutex_unlock(&kvm->lock);
5249         return rc;
5250 }
5251
5252 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5253                                              struct irq_bypass_producer *prod)
5254 {
5255         int ret = 0;
5256         struct kvm_kernel_irqfd *irqfd =
5257                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5258
5259         irqfd->producer = prod;
5260
5261         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5262         if (ret)
5263                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5264                         prod->irq, irqfd->gsi, ret);
5265
5266         return ret;
5267 }
5268
5269 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5270                                               struct irq_bypass_producer *prod)
5271 {
5272         int ret;
5273         struct kvm_kernel_irqfd *irqfd =
5274                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5275
5276         irqfd->producer = NULL;
5277
5278         /*
5279          * When producer of consumer is unregistered, we change back to
5280          * default external interrupt handling mode - KVM real mode
5281          * will switch back to host.
5282          */
5283         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5284         if (ret)
5285                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5286                         prod->irq, irqfd->gsi, ret);
5287 }
5288 #endif
5289
5290 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5291                                  unsigned int ioctl, unsigned long arg)
5292 {
5293         struct kvm *kvm __maybe_unused = filp->private_data;
5294         void __user *argp = (void __user *)arg;
5295         long r;
5296
5297         switch (ioctl) {
5298
5299         case KVM_PPC_ALLOCATE_HTAB: {
5300                 u32 htab_order;
5301
5302                 /* If we're a nested hypervisor, we currently only support radix */
5303                 if (kvmhv_on_pseries()) {
5304                         r = -EOPNOTSUPP;
5305                         break;
5306                 }
5307
5308                 r = -EFAULT;
5309                 if (get_user(htab_order, (u32 __user *)argp))
5310                         break;
5311                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5312                 if (r)
5313                         break;
5314                 r = 0;
5315                 break;
5316         }
5317
5318         case KVM_PPC_GET_HTAB_FD: {
5319                 struct kvm_get_htab_fd ghf;
5320
5321                 r = -EFAULT;
5322                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5323                         break;
5324                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5325                 break;
5326         }
5327
5328         case KVM_PPC_RESIZE_HPT_PREPARE: {
5329                 struct kvm_ppc_resize_hpt rhpt;
5330
5331                 r = -EFAULT;
5332                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5333                         break;
5334
5335                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5336                 break;
5337         }
5338
5339         case KVM_PPC_RESIZE_HPT_COMMIT: {
5340                 struct kvm_ppc_resize_hpt rhpt;
5341
5342                 r = -EFAULT;
5343                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5344                         break;
5345
5346                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5347                 break;
5348         }
5349
5350         default:
5351                 r = -ENOTTY;
5352         }
5353
5354         return r;
5355 }
5356
5357 /*
5358  * List of hcall numbers to enable by default.
5359  * For compatibility with old userspace, we enable by default
5360  * all hcalls that were implemented before the hcall-enabling
5361  * facility was added.  Note this list should not include H_RTAS.
5362  */
5363 static unsigned int default_hcall_list[] = {
5364         H_REMOVE,
5365         H_ENTER,
5366         H_READ,
5367         H_PROTECT,
5368         H_BULK_REMOVE,
5369         H_GET_TCE,
5370         H_PUT_TCE,
5371         H_SET_DABR,
5372         H_SET_XDABR,
5373         H_CEDE,
5374         H_PROD,
5375         H_CONFER,
5376         H_REGISTER_VPA,
5377 #ifdef CONFIG_KVM_XICS
5378         H_EOI,
5379         H_CPPR,
5380         H_IPI,
5381         H_IPOLL,
5382         H_XIRR,
5383         H_XIRR_X,
5384 #endif
5385         0
5386 };
5387
5388 static void init_default_hcalls(void)
5389 {
5390         int i;
5391         unsigned int hcall;
5392
5393         for (i = 0; default_hcall_list[i]; ++i) {
5394                 hcall = default_hcall_list[i];
5395                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5396                 __set_bit(hcall / 4, default_enabled_hcalls);
5397         }
5398 }
5399
5400 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5401 {
5402         unsigned long lpcr;
5403         int radix;
5404         int err;
5405
5406         /* If not on a POWER9, reject it */
5407         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5408                 return -ENODEV;
5409
5410         /* If any unknown flags set, reject it */
5411         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5412                 return -EINVAL;
5413
5414         /* GR (guest radix) bit in process_table field must match */
5415         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5416         if (!!(cfg->process_table & PATB_GR) != radix)
5417                 return -EINVAL;
5418
5419         /* Process table size field must be reasonable, i.e. <= 24 */
5420         if ((cfg->process_table & PRTS_MASK) > 24)
5421                 return -EINVAL;
5422
5423         /* We can change a guest to/from radix now, if the host is radix */
5424         if (radix && !radix_enabled())
5425                 return -EINVAL;
5426
5427         /* If we're a nested hypervisor, we currently only support radix */
5428         if (kvmhv_on_pseries() && !radix)
5429                 return -EINVAL;
5430
5431         mutex_lock(&kvm->arch.mmu_setup_lock);
5432         if (radix != kvm_is_radix(kvm)) {
5433                 if (kvm->arch.mmu_ready) {
5434                         kvm->arch.mmu_ready = 0;
5435                         /* order mmu_ready vs. vcpus_running */
5436                         smp_mb();
5437                         if (atomic_read(&kvm->arch.vcpus_running)) {
5438                                 kvm->arch.mmu_ready = 1;
5439                                 err = -EBUSY;
5440                                 goto out_unlock;
5441                         }
5442                 }
5443                 if (radix)
5444                         err = kvmppc_switch_mmu_to_radix(kvm);
5445                 else
5446                         err = kvmppc_switch_mmu_to_hpt(kvm);
5447                 if (err)
5448                         goto out_unlock;
5449         }
5450
5451         kvm->arch.process_table = cfg->process_table;
5452         kvmppc_setup_partition_table(kvm);
5453
5454         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5455         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5456         err = 0;
5457
5458  out_unlock:
5459         mutex_unlock(&kvm->arch.mmu_setup_lock);
5460         return err;
5461 }
5462
5463 static int kvmhv_enable_nested(struct kvm *kvm)
5464 {
5465         if (!nested)
5466                 return -EPERM;
5467         if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5468                 return -ENODEV;
5469
5470         /* kvm == NULL means the caller is testing if the capability exists */
5471         if (kvm)
5472                 kvm->arch.nested_enable = true;
5473         return 0;
5474 }
5475
5476 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5477                                  int size)
5478 {
5479         int rc = -EINVAL;
5480
5481         if (kvmhv_vcpu_is_radix(vcpu)) {
5482                 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5483
5484                 if (rc > 0)
5485                         rc = -EINVAL;
5486         }
5487
5488         /* For now quadrants are the only way to access nested guest memory */
5489         if (rc && vcpu->arch.nested)
5490                 rc = -EAGAIN;
5491
5492         return rc;
5493 }
5494
5495 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5496                                 int size)
5497 {
5498         int rc = -EINVAL;
5499
5500         if (kvmhv_vcpu_is_radix(vcpu)) {
5501                 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5502
5503                 if (rc > 0)
5504                         rc = -EINVAL;
5505         }
5506
5507         /* For now quadrants are the only way to access nested guest memory */
5508         if (rc && vcpu->arch.nested)
5509                 rc = -EAGAIN;
5510
5511         return rc;
5512 }
5513
5514 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5515 {
5516         unpin_vpa(kvm, vpa);
5517         vpa->gpa = 0;
5518         vpa->pinned_addr = NULL;
5519         vpa->dirty = false;
5520         vpa->update_pending = 0;
5521 }
5522
5523 /*
5524  * Enable a guest to become a secure VM, or test whether
5525  * that could be enabled.
5526  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
5527  * tested (kvm == NULL) or enabled (kvm != NULL).
5528  */
5529 static int kvmhv_enable_svm(struct kvm *kvm)
5530 {
5531         if (!kvmppc_uvmem_available())
5532                 return -EINVAL;
5533         if (kvm)
5534                 kvm->arch.svm_enabled = 1;
5535         return 0;
5536 }
5537
5538 /*
5539  *  IOCTL handler to turn off secure mode of guest
5540  *
5541  * - Release all device pages
5542  * - Issue ucall to terminate the guest on the UV side
5543  * - Unpin the VPA pages.
5544  * - Reinit the partition scoped page tables
5545  */
5546 static int kvmhv_svm_off(struct kvm *kvm)
5547 {
5548         struct kvm_vcpu *vcpu;
5549         int mmu_was_ready;
5550         int srcu_idx;
5551         int ret = 0;
5552         int i;
5553
5554         if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5555                 return ret;
5556
5557         mutex_lock(&kvm->arch.mmu_setup_lock);
5558         mmu_was_ready = kvm->arch.mmu_ready;
5559         if (kvm->arch.mmu_ready) {
5560                 kvm->arch.mmu_ready = 0;
5561                 /* order mmu_ready vs. vcpus_running */
5562                 smp_mb();
5563                 if (atomic_read(&kvm->arch.vcpus_running)) {
5564                         kvm->arch.mmu_ready = 1;
5565                         ret = -EBUSY;
5566                         goto out;
5567                 }
5568         }
5569
5570         srcu_idx = srcu_read_lock(&kvm->srcu);
5571         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5572                 struct kvm_memory_slot *memslot;
5573                 struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5574
5575                 if (!slots)
5576                         continue;
5577
5578                 kvm_for_each_memslot(memslot, slots) {
5579                         kvmppc_uvmem_drop_pages(memslot, kvm, true);
5580                         uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5581                 }
5582         }
5583         srcu_read_unlock(&kvm->srcu, srcu_idx);
5584
5585         ret = uv_svm_terminate(kvm->arch.lpid);
5586         if (ret != U_SUCCESS) {
5587                 ret = -EINVAL;
5588                 goto out;
5589         }
5590
5591         /*
5592          * When secure guest is reset, all the guest pages are sent
5593          * to UV via UV_PAGE_IN before the non-boot vcpus get a
5594          * chance to run and unpin their VPA pages. Unpinning of all
5595          * VPA pages is done here explicitly so that VPA pages
5596          * can be migrated to the secure side.
5597          *
5598          * This is required to for the secure SMP guest to reboot
5599          * correctly.
5600          */
5601         kvm_for_each_vcpu(i, vcpu, kvm) {
5602                 spin_lock(&vcpu->arch.vpa_update_lock);
5603                 unpin_vpa_reset(kvm, &vcpu->arch.dtl);
5604                 unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
5605                 unpin_vpa_reset(kvm, &vcpu->arch.vpa);
5606                 spin_unlock(&vcpu->arch.vpa_update_lock);
5607         }
5608
5609         kvmppc_setup_partition_table(kvm);
5610         kvm->arch.secure_guest = 0;
5611         kvm->arch.mmu_ready = mmu_was_ready;
5612 out:
5613         mutex_unlock(&kvm->arch.mmu_setup_lock);
5614         return ret;
5615 }
5616
5617 static int kvmhv_enable_dawr1(struct kvm *kvm)
5618 {
5619         if (!cpu_has_feature(CPU_FTR_DAWR1))
5620                 return -ENODEV;
5621
5622         /* kvm == NULL means the caller is testing if the capability exists */
5623         if (kvm)
5624                 kvm->arch.dawr1_enabled = true;
5625         return 0;
5626 }
5627
5628 static bool kvmppc_hash_v3_possible(void)
5629 {
5630         if (radix_enabled() && no_mixing_hpt_and_radix)
5631                 return false;
5632
5633         return cpu_has_feature(CPU_FTR_ARCH_300) &&
5634                 cpu_has_feature(CPU_FTR_HVMODE);
5635 }
5636
5637 static struct kvmppc_ops kvm_ops_hv = {
5638         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5639         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5640         .get_one_reg = kvmppc_get_one_reg_hv,
5641         .set_one_reg = kvmppc_set_one_reg_hv,
5642         .vcpu_load   = kvmppc_core_vcpu_load_hv,
5643         .vcpu_put    = kvmppc_core_vcpu_put_hv,
5644         .inject_interrupt = kvmppc_inject_interrupt_hv,
5645         .set_msr     = kvmppc_set_msr_hv,
5646         .vcpu_run    = kvmppc_vcpu_run_hv,
5647         .vcpu_create = kvmppc_core_vcpu_create_hv,
5648         .vcpu_free   = kvmppc_core_vcpu_free_hv,
5649         .check_requests = kvmppc_core_check_requests_hv,
5650         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5651         .flush_memslot  = kvmppc_core_flush_memslot_hv,
5652         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5653         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5654         .unmap_hva_range = kvm_unmap_hva_range_hv,
5655         .age_hva  = kvm_age_hva_hv,
5656         .test_age_hva = kvm_test_age_hva_hv,
5657         .set_spte_hva = kvm_set_spte_hva_hv,
5658         .free_memslot = kvmppc_core_free_memslot_hv,
5659         .init_vm =  kvmppc_core_init_vm_hv,
5660         .destroy_vm = kvmppc_core_destroy_vm_hv,
5661         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5662         .emulate_op = kvmppc_core_emulate_op_hv,
5663         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5664         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5665         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5666         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5667         .hcall_implemented = kvmppc_hcall_impl_hv,
5668 #ifdef CONFIG_KVM_XICS
5669         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5670         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5671 #endif
5672         .configure_mmu = kvmhv_configure_mmu,
5673         .get_rmmu_info = kvmhv_get_rmmu_info,
5674         .set_smt_mode = kvmhv_set_smt_mode,
5675         .enable_nested = kvmhv_enable_nested,
5676         .load_from_eaddr = kvmhv_load_from_eaddr,
5677         .store_to_eaddr = kvmhv_store_to_eaddr,
5678         .enable_svm = kvmhv_enable_svm,
5679         .svm_off = kvmhv_svm_off,
5680         .enable_dawr1 = kvmhv_enable_dawr1,
5681         .hash_v3_possible = kvmppc_hash_v3_possible,
5682 };
5683
5684 static int kvm_init_subcore_bitmap(void)
5685 {
5686         int i, j;
5687         int nr_cores = cpu_nr_cores();
5688         struct sibling_subcore_state *sibling_subcore_state;
5689
5690         for (i = 0; i < nr_cores; i++) {
5691                 int first_cpu = i * threads_per_core;
5692                 int node = cpu_to_node(first_cpu);
5693
5694                 /* Ignore if it is already allocated. */
5695                 if (paca_ptrs[first_cpu]->sibling_subcore_state)
5696                         continue;
5697
5698                 sibling_subcore_state =
5699                         kzalloc_node(sizeof(struct sibling_subcore_state),
5700                                                         GFP_KERNEL, node);
5701                 if (!sibling_subcore_state)
5702                         return -ENOMEM;
5703
5704
5705                 for (j = 0; j < threads_per_core; j++) {
5706                         int cpu = first_cpu + j;
5707
5708                         paca_ptrs[cpu]->sibling_subcore_state =
5709                                                 sibling_subcore_state;
5710                 }
5711         }
5712         return 0;
5713 }
5714
5715 static int kvmppc_radix_possible(void)
5716 {
5717         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5718 }
5719
5720 static int kvmppc_book3s_init_hv(void)
5721 {
5722         int r;
5723
5724         if (!tlbie_capable) {
5725                 pr_err("KVM-HV: Host does not support TLBIE\n");
5726                 return -ENODEV;
5727         }
5728
5729         /*
5730          * FIXME!! Do we need to check on all cpus ?
5731          */
5732         r = kvmppc_core_check_processor_compat_hv();
5733         if (r < 0)
5734                 return -ENODEV;
5735
5736         r = kvmhv_nested_init();
5737         if (r)
5738                 return r;
5739
5740         r = kvm_init_subcore_bitmap();
5741         if (r)
5742                 return r;
5743
5744         /*
5745          * We need a way of accessing the XICS interrupt controller,
5746          * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5747          * indirectly, via OPAL.
5748          */
5749 #ifdef CONFIG_SMP
5750         if (!xics_on_xive() && !kvmhv_on_pseries() &&
5751             !local_paca->kvm_hstate.xics_phys) {
5752                 struct device_node *np;
5753
5754                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5755                 if (!np) {
5756                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5757                         return -ENODEV;
5758                 }
5759                 /* presence of intc confirmed - node can be dropped again */
5760                 of_node_put(np);
5761         }
5762 #endif
5763
5764         kvm_ops_hv.owner = THIS_MODULE;
5765         kvmppc_hv_ops = &kvm_ops_hv;
5766
5767         init_default_hcalls();
5768
5769         init_vcore_lists();
5770
5771         r = kvmppc_mmu_hv_init();
5772         if (r)
5773                 return r;
5774
5775         if (kvmppc_radix_possible())
5776                 r = kvmppc_radix_init();
5777
5778         /*
5779          * POWER9 chips before version 2.02 can't have some threads in
5780          * HPT mode and some in radix mode on the same core.
5781          */
5782         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5783                 unsigned int pvr = mfspr(SPRN_PVR);
5784                 if ((pvr >> 16) == PVR_POWER9 &&
5785                     (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5786                      ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5787                         no_mixing_hpt_and_radix = true;
5788         }
5789
5790         r = kvmppc_uvmem_init();
5791         if (r < 0)
5792                 pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
5793
5794         return r;
5795 }
5796
5797 static void kvmppc_book3s_exit_hv(void)
5798 {
5799         kvmppc_uvmem_free();
5800         kvmppc_free_host_rm_ops();
5801         if (kvmppc_radix_possible())
5802                 kvmppc_radix_exit();
5803         kvmppc_hv_ops = NULL;
5804         kvmhv_nested_exit();
5805 }
5806
5807 module_init(kvmppc_book3s_init_hv);
5808 module_exit(kvmppc_book3s_exit_hv);
5809 MODULE_LICENSE("GPL");
5810 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5811 MODULE_ALIAS("devname:kvm");