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
  76 #include "book3s.h"
  77
  78 #define CREATE_TRACE_POINTS
  79 #include "trace_hv.h"
  80
  81 /* #define EXIT_DEBUG */
  82 /* #define EXIT_DEBUG_SIMPLE */
  83 /* #define EXIT_DEBUG_INT */
  84
  85 /* Used to indicate that a guest page fault needs to be handled */
  86 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  87 /* Used to indicate that a guest passthrough interrupt needs to be handled */
  88 #define RESUME_PASSTHROUGH      (RESUME_GUEST | RESUME_FLAG_ARCH2)
  89
  90 /* Used as a "null" value for timebase values */
  91 #define TB_NIL  (~(u64)0)
  92
  93 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
  94
  95 static int dynamic_mt_modes = 6;
  96 module_param(dynamic_mt_modes, int, 0644);
  97 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
  98 static int target_smt_mode;
  99 module_param(target_smt_mode, int, 0644);
 100 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
 101
 102 static bool indep_threads_mode = true;
 103 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
 104 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
 105
 106 static bool one_vm_per_core;
 107 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
 108 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)");
 109
 110 #ifdef CONFIG_KVM_XICS
 111 static struct kernel_param_ops module_param_ops = {
 112         .set = param_set_int,
 113         .get = param_get_int,
 114 };
 115
 116 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
 117 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
 118
 119 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
 120 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
 121 #endif
 122
 123 /* If set, guests are allowed to create and control nested guests */
 124 static bool nested = true;
 125 module_param(nested, bool, S_IRUGO | S_IWUSR);
 126 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
 127
 128 static inline bool nesting_enabled(struct kvm *kvm)
 129 {
 130         return kvm->arch.nested_enable && kvm_is_radix(kvm);
 131 }
 132
 133 /* If set, the threads on each CPU core have to be in the same MMU mode */
 134 static bool no_mixing_hpt_and_radix;
 135
 136 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
 137 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
 138
 139 /*
 140  * RWMR values for POWER8.  These control the rate at which PURR
 141  * and SPURR count and should be set according to the number of
 142  * online threads in the vcore being run.
 143  */
 144 #define RWMR_RPA_P8_1THREAD     0x164520C62609AECAUL
 145 #define RWMR_RPA_P8_2THREAD     0x7FFF2908450D8DA9UL
 146 #define RWMR_RPA_P8_3THREAD     0x164520C62609AECAUL
 147 #define RWMR_RPA_P8_4THREAD     0x199A421245058DA9UL
 148 #define RWMR_RPA_P8_5THREAD     0x164520C62609AECAUL
 149 #define RWMR_RPA_P8_6THREAD     0x164520C62609AECAUL
 150 #define RWMR_RPA_P8_7THREAD     0x164520C62609AECAUL
 151 #define RWMR_RPA_P8_8THREAD     0x164520C62609AECAUL
 152
 153 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
 154         RWMR_RPA_P8_1THREAD,
 155         RWMR_RPA_P8_1THREAD,
 156         RWMR_RPA_P8_2THREAD,
 157         RWMR_RPA_P8_3THREAD,
 158         RWMR_RPA_P8_4THREAD,
 159         RWMR_RPA_P8_5THREAD,
 160         RWMR_RPA_P8_6THREAD,
 161         RWMR_RPA_P8_7THREAD,
 162         RWMR_RPA_P8_8THREAD,
 163 };
 164
 165 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
 166                 int *ip)
 167 {
 168         int i = *ip;
 169         struct kvm_vcpu *vcpu;
 170
 171         while (++i < MAX_SMT_THREADS) {
 172                 vcpu = READ_ONCE(vc->runnable_threads[i]);
 173                 if (vcpu) {
 174                         *ip = i;
 175                         return vcpu;
 176                 }
 177         }
 178         return NULL;
 179 }
 180
 181 /* Used to traverse the list of runnable threads for a given vcore */
 182 #define for_each_runnable_thread(i, vcpu, vc) \
 183         for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
 184
 185 static bool kvmppc_ipi_thread(int cpu)
 186 {
 187         unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
 188
 189         /* If we're a nested hypervisor, fall back to ordinary IPIs for now */
 190         if (kvmhv_on_pseries())
 191                 return false;
 192
 193         /* On POWER9 we can use msgsnd to IPI any cpu */
 194         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
 195                 msg |= get_hard_smp_processor_id(cpu);
 196                 smp_mb();
 197                 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 198                 return true;
 199         }
 200
 201         /* On POWER8 for IPIs to threads in the same core, use msgsnd */
 202         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
 203                 preempt_disable();
 204                 if (cpu_first_thread_sibling(cpu) ==
 205                     cpu_first_thread_sibling(smp_processor_id())) {
 206                         msg |= cpu_thread_in_core(cpu);
 207                         smp_mb();
 208                         __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 209                         preempt_enable();
 210                         return true;
 211                 }
 212                 preempt_enable();
 213         }
 214
 215 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
 216         if (cpu >= 0 && cpu < nr_cpu_ids) {
 217                 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
 218                         xics_wake_cpu(cpu);
 219                         return true;
 220                 }
 221                 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
 222                 return true;
 223         }
 224 #endif
 225
 226         return false;
 227 }
 228
 229 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
 230 {
 231         int cpu;
 232         struct swait_queue_head *wqp;
 233
 234         wqp = kvm_arch_vcpu_wq(vcpu);
 235         if (swq_has_sleeper(wqp)) {
 236                 swake_up_one(wqp);
 237                 ++vcpu->stat.halt_wakeup;
 238         }
 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_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
 342 {
 343         /*
 344          * Check for illegal transactional state bit combination
 345          * and if we find it, force the TS field to a safe state.
 346          */
 347         if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
 348                 msr &= ~MSR_TS_MASK;
 349         vcpu->arch.shregs.msr = msr;
 350         kvmppc_end_cede(vcpu);
 351 }
 352
 353 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 354 {
 355         vcpu->arch.pvr = pvr;
 356 }
 357
 358 /* Dummy value used in computing PCR value below */
 359 #define PCR_ARCH_300    (PCR_ARCH_207 << 1)
 360
 361 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 362 {
 363         unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
 364         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 365
 366         /* We can (emulate) our own architecture version and anything older */
 367         if (cpu_has_feature(CPU_FTR_ARCH_300))
 368                 host_pcr_bit = PCR_ARCH_300;
 369         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
 370                 host_pcr_bit = PCR_ARCH_207;
 371         else if (cpu_has_feature(CPU_FTR_ARCH_206))
 372                 host_pcr_bit = PCR_ARCH_206;
 373         else
 374                 host_pcr_bit = PCR_ARCH_205;
 375
 376         /* Determine lowest PCR bit needed to run guest in given PVR level */
 377         guest_pcr_bit = host_pcr_bit;
 378         if (arch_compat) {
 379                 switch (arch_compat) {
 380                 case PVR_ARCH_205:
 381                         guest_pcr_bit = PCR_ARCH_205;
 382                         break;
 383                 case PVR_ARCH_206:
 384                 case PVR_ARCH_206p:
 385                         guest_pcr_bit = PCR_ARCH_206;
 386                         break;
 387                 case PVR_ARCH_207:
 388                         guest_pcr_bit = PCR_ARCH_207;
 389                         break;
 390                 case PVR_ARCH_300:
 391                         guest_pcr_bit = PCR_ARCH_300;
 392                         break;
 393                 default:
 394                         return -EINVAL;
 395                 }
 396         }
 397
 398         /* Check requested PCR bits don't exceed our capabilities */
 399         if (guest_pcr_bit > host_pcr_bit)
 400                 return -EINVAL;
 401
 402         spin_lock(&vc->lock);
 403         vc->arch_compat = arch_compat;
 404         /* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
 405         vc->pcr = host_pcr_bit - guest_pcr_bit;
 406         spin_unlock(&vc->lock);
 407
 408         return 0;
 409 }
 410
 411 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 412 {
 413         int r;
 414
 415         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 416         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 417                vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
 418         for (r = 0; r < 16; ++r)
 419                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 420                        r, kvmppc_get_gpr(vcpu, r),
 421                        r+16, kvmppc_get_gpr(vcpu, r+16));
 422         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 423                vcpu->arch.regs.ctr, vcpu->arch.regs.link);
 424         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 425                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 426         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 427                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 428         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 429                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 430         pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
 431                vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
 432         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 433         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 434                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 435         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 436         for (r = 0; r < vcpu->arch.slb_max; ++r)
 437                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 438                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 439         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 440                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 441                vcpu->arch.last_inst);
 442 }
 443
 444 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 445 {
 446         return kvm_get_vcpu_by_id(kvm, id);
 447 }
 448
 449 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 450 {
 451         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 452         vpa->yield_count = cpu_to_be32(1);
 453 }
 454
 455 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 456                    unsigned long addr, unsigned long len)
 457 {
 458         /* check address is cacheline aligned */
 459         if (addr & (L1_CACHE_BYTES - 1))
 460                 return -EINVAL;
 461         spin_lock(&vcpu->arch.vpa_update_lock);
 462         if (v->next_gpa != addr || v->len != len) {
 463                 v->next_gpa = addr;
 464                 v->len = addr ? len : 0;
 465                 v->update_pending = 1;
 466         }
 467         spin_unlock(&vcpu->arch.vpa_update_lock);
 468         return 0;
 469 }
 470
 471 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 472 struct reg_vpa {
 473         u32 dummy;
 474         union {
 475                 __be16 hword;
 476                 __be32 word;
 477         } length;
 478 };
 479
 480 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 481 {
 482         if (vpap->update_pending)
 483                 return vpap->next_gpa != 0;
 484         return vpap->pinned_addr != NULL;
 485 }
 486
 487 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 488                                        unsigned long flags,
 489                                        unsigned long vcpuid, unsigned long vpa)
 490 {
 491         struct kvm *kvm = vcpu->kvm;
 492         unsigned long len, nb;
 493         void *va;
 494         struct kvm_vcpu *tvcpu;
 495         int err;
 496         int subfunc;
 497         struct kvmppc_vpa *vpap;
 498
 499         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 500         if (!tvcpu)
 501                 return H_PARAMETER;
 502
 503         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 504         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 505             subfunc == H_VPA_REG_SLB) {
 506                 /* Registering new area - address must be cache-line aligned */
 507                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 508                         return H_PARAMETER;
 509
 510                 /* convert logical addr to kernel addr and read length */
 511                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 512                 if (va == NULL)
 513                         return H_PARAMETER;
 514                 if (subfunc == H_VPA_REG_VPA)
 515                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 516                 else
 517                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 518                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 519
 520                 /* Check length */
 521                 if (len > nb || len < sizeof(struct reg_vpa))
 522                         return H_PARAMETER;
 523         } else {
 524                 vpa = 0;
 525                 len = 0;
 526         }
 527
 528         err = H_PARAMETER;
 529         vpap = NULL;
 530         spin_lock(&tvcpu->arch.vpa_update_lock);
 531
 532         switch (subfunc) {
 533         case H_VPA_REG_VPA:             /* register VPA */
 534                 /*
 535                  * The size of our lppaca is 1kB because of the way we align
 536                  * it for the guest to avoid crossing a 4kB boundary. We only
 537                  * use 640 bytes of the structure though, so we should accept
 538                  * clients that set a size of 640.
 539                  */
 540                 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
 541                 if (len < sizeof(struct lppaca))
 542                         break;
 543                 vpap = &tvcpu->arch.vpa;
 544                 err = 0;
 545                 break;
 546
 547         case H_VPA_REG_DTL:             /* register DTL */
 548                 if (len < sizeof(struct dtl_entry))
 549                         break;
 550                 len -= len % sizeof(struct dtl_entry);
 551
 552                 /* Check that they have previously registered a VPA */
 553                 err = H_RESOURCE;
 554                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 555                         break;
 556
 557                 vpap = &tvcpu->arch.dtl;
 558                 err = 0;
 559                 break;
 560
 561         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 562                 /* Check that they have previously registered a VPA */
 563                 err = H_RESOURCE;
 564                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 565                         break;
 566
 567                 vpap = &tvcpu->arch.slb_shadow;
 568                 err = 0;
 569                 break;
 570
 571         case H_VPA_DEREG_VPA:           /* deregister VPA */
 572                 /* Check they don't still have a DTL or SLB buf registered */
 573                 err = H_RESOURCE;
 574                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 575                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 576                         break;
 577
 578                 vpap = &tvcpu->arch.vpa;
 579                 err = 0;
 580                 break;
 581
 582         case H_VPA_DEREG_DTL:           /* deregister DTL */
 583                 vpap = &tvcpu->arch.dtl;
 584                 err = 0;
 585                 break;
 586
 587         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 588                 vpap = &tvcpu->arch.slb_shadow;
 589                 err = 0;
 590                 break;
 591         }
 592
 593         if (vpap) {
 594                 vpap->next_gpa = vpa;
 595                 vpap->len = len;
 596                 vpap->update_pending = 1;
 597         }
 598
 599         spin_unlock(&tvcpu->arch.vpa_update_lock);
 600
 601         return err;
 602 }
 603
 604 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 605 {
 606         struct kvm *kvm = vcpu->kvm;
 607         void *va;
 608         unsigned long nb;
 609         unsigned long gpa;
 610
 611         /*
 612          * We need to pin the page pointed to by vpap->next_gpa,
 613          * but we can't call kvmppc_pin_guest_page under the lock
 614          * as it does get_user_pages() and down_read().  So we
 615          * have to drop the lock, pin the page, then get the lock
 616          * again and check that a new area didn't get registered
 617          * in the meantime.
 618          */
 619         for (;;) {
 620                 gpa = vpap->next_gpa;
 621                 spin_unlock(&vcpu->arch.vpa_update_lock);
 622                 va = NULL;
 623                 nb = 0;
 624                 if (gpa)
 625                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 626                 spin_lock(&vcpu->arch.vpa_update_lock);
 627                 if (gpa == vpap->next_gpa)
 628                         break;
 629                 /* sigh... unpin that one and try again */
 630                 if (va)
 631                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 632         }
 633
 634         vpap->update_pending = 0;
 635         if (va && nb < vpap->len) {
 636                 /*
 637                  * If it's now too short, it must be that userspace
 638                  * has changed the mappings underlying guest memory,
 639                  * so unregister the region.
 640                  */
 641                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 642                 va = NULL;
 643         }
 644         if (vpap->pinned_addr)
 645                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 646                                         vpap->dirty);
 647         vpap->gpa = gpa;
 648         vpap->pinned_addr = va;
 649         vpap->dirty = false;
 650         if (va)
 651                 vpap->pinned_end = va + vpap->len;
 652 }
 653
 654 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 655 {
 656         if (!(vcpu->arch.vpa.update_pending ||
 657               vcpu->arch.slb_shadow.update_pending ||
 658               vcpu->arch.dtl.update_pending))
 659                 return;
 660
 661         spin_lock(&vcpu->arch.vpa_update_lock);
 662         if (vcpu->arch.vpa.update_pending) {
 663                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 664                 if (vcpu->arch.vpa.pinned_addr)
 665                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 666         }
 667         if (vcpu->arch.dtl.update_pending) {
 668                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 669                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 670                 vcpu->arch.dtl_index = 0;
 671         }
 672         if (vcpu->arch.slb_shadow.update_pending)
 673                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 674         spin_unlock(&vcpu->arch.vpa_update_lock);
 675 }
 676
 677 /*
 678  * Return the accumulated stolen time for the vcore up until `now'.
 679  * The caller should hold the vcore lock.
 680  */
 681 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 682 {
 683         u64 p;
 684         unsigned long flags;
 685
 686         spin_lock_irqsave(&vc->stoltb_lock, flags);
 687         p = vc->stolen_tb;
 688         if (vc->vcore_state != VCORE_INACTIVE &&
 689             vc->preempt_tb != TB_NIL)
 690                 p += now - vc->preempt_tb;
 691         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 692         return p;
 693 }
 694
 695 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 696                                     struct kvmppc_vcore *vc)
 697 {
 698         struct dtl_entry *dt;
 699         struct lppaca *vpa;
 700         unsigned long stolen;
 701         unsigned long core_stolen;
 702         u64 now;
 703         unsigned long flags;
 704
 705         dt = vcpu->arch.dtl_ptr;
 706         vpa = vcpu->arch.vpa.pinned_addr;
 707         now = mftb();
 708         core_stolen = vcore_stolen_time(vc, now);
 709         stolen = core_stolen - vcpu->arch.stolen_logged;
 710         vcpu->arch.stolen_logged = core_stolen;
 711         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 712         stolen += vcpu->arch.busy_stolen;
 713         vcpu->arch.busy_stolen = 0;
 714         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 715         if (!dt || !vpa)
 716                 return;
 717         memset(dt, 0, sizeof(struct dtl_entry));
 718         dt->dispatch_reason = 7;
 719         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 720         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 721         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 722         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 723         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 724         ++dt;
 725         if (dt == vcpu->arch.dtl.pinned_end)
 726                 dt = vcpu->arch.dtl.pinned_addr;
 727         vcpu->arch.dtl_ptr = dt;
 728         /* order writing *dt vs. writing vpa->dtl_idx */
 729         smp_wmb();
 730         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 731         vcpu->arch.dtl.dirty = true;
 732 }
 733
 734 /* See if there is a doorbell interrupt pending for a vcpu */
 735 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
 736 {
 737         int thr;
 738         struct kvmppc_vcore *vc;
 739
 740         if (vcpu->arch.doorbell_request)
 741                 return true;
 742         /*
 743          * Ensure that the read of vcore->dpdes comes after the read
 744          * of vcpu->doorbell_request.  This barrier matches the
 745          * smp_wmb() in kvmppc_guest_entry_inject().
 746          */
 747         smp_rmb();
 748         vc = vcpu->arch.vcore;
 749         thr = vcpu->vcpu_id - vc->first_vcpuid;
 750         return !!(vc->dpdes & (1 << thr));
 751 }
 752
 753 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 754 {
 755         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 756                 return true;
 757         if ((!vcpu->arch.vcore->arch_compat) &&
 758             cpu_has_feature(CPU_FTR_ARCH_207S))
 759                 return true;
 760         return false;
 761 }
 762
 763 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 764                              unsigned long resource, unsigned long value1,
 765                              unsigned long value2)
 766 {
 767         switch (resource) {
 768         case H_SET_MODE_RESOURCE_SET_CIABR:
 769                 if (!kvmppc_power8_compatible(vcpu))
 770                         return H_P2;
 771                 if (value2)
 772                         return H_P4;
 773                 if (mflags)
 774                         return H_UNSUPPORTED_FLAG_START;
 775                 /* Guests can't breakpoint the hypervisor */
 776                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 777                         return H_P3;
 778                 vcpu->arch.ciabr  = value1;
 779                 return H_SUCCESS;
 780         case H_SET_MODE_RESOURCE_SET_DAWR:
 781                 if (!kvmppc_power8_compatible(vcpu))
 782                         return H_P2;
 783                 if (!ppc_breakpoint_available())
 784                         return H_P2;
 785                 if (mflags)
 786                         return H_UNSUPPORTED_FLAG_START;
 787                 if (value2 & DABRX_HYP)
 788                         return H_P4;
 789                 vcpu->arch.dawr  = value1;
 790                 vcpu->arch.dawrx = value2;
 791                 return H_SUCCESS;
 792         default:
 793                 return H_TOO_HARD;
 794         }
 795 }
 796
 797 /* Copy guest memory in place - must reside within a single memslot */
 798 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
 799                                   unsigned long len)
 800 {
 801         struct kvm_memory_slot *to_memslot = NULL;
 802         struct kvm_memory_slot *from_memslot = NULL;
 803         unsigned long to_addr, from_addr;
 804         int r;
 805
 806         /* Get HPA for from address */
 807         from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
 808         if (!from_memslot)
 809                 return -EFAULT;
 810         if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
 811                              << PAGE_SHIFT))
 812                 return -EINVAL;
 813         from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
 814         if (kvm_is_error_hva(from_addr))
 815                 return -EFAULT;
 816         from_addr |= (from & (PAGE_SIZE - 1));
 817
 818         /* Get HPA for to address */
 819         to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
 820         if (!to_memslot)
 821                 return -EFAULT;
 822         if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
 823                            << PAGE_SHIFT))
 824                 return -EINVAL;
 825         to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
 826         if (kvm_is_error_hva(to_addr))
 827                 return -EFAULT;
 828         to_addr |= (to & (PAGE_SIZE - 1));
 829
 830         /* Perform copy */
 831         r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
 832                              len);
 833         if (r)
 834                 return -EFAULT;
 835         mark_page_dirty(kvm, to >> PAGE_SHIFT);
 836         return 0;
 837 }
 838
 839 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
 840                                unsigned long dest, unsigned long src)
 841 {
 842         u64 pg_sz = SZ_4K;              /* 4K page size */
 843         u64 pg_mask = SZ_4K - 1;
 844         int ret;
 845
 846         /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
 847         if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
 848                       H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
 849                 return H_PARAMETER;
 850
 851         /* dest (and src if copy_page flag set) must be page aligned */
 852         if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
 853                 return H_PARAMETER;
 854
 855         /* zero and/or copy the page as determined by the flags */
 856         if (flags & H_COPY_PAGE) {
 857                 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
 858                 if (ret < 0)
 859                         return H_PARAMETER;
 860         } else if (flags & H_ZERO_PAGE) {
 861                 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
 862                 if (ret < 0)
 863                         return H_PARAMETER;
 864         }
 865
 866         /* We can ignore the remaining flags */
 867
 868         return H_SUCCESS;
 869 }
 870
 871 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 872 {
 873         struct kvmppc_vcore *vcore = target->arch.vcore;
 874
 875         /*
 876          * We expect to have been called by the real mode handler
 877          * (kvmppc_rm_h_confer()) which would have directly returned
 878          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 879          * have useful work to do and should not confer) so we don't
 880          * recheck that here.
 881          */
 882
 883         spin_lock(&vcore->lock);
 884         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 885             vcore->vcore_state != VCORE_INACTIVE &&
 886             vcore->runner)
 887                 target = vcore->runner;
 888         spin_unlock(&vcore->lock);
 889
 890         return kvm_vcpu_yield_to(target);
 891 }
 892
 893 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 894 {
 895         int yield_count = 0;
 896         struct lppaca *lppaca;
 897
 898         spin_lock(&vcpu->arch.vpa_update_lock);
 899         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 900         if (lppaca)
 901                 yield_count = be32_to_cpu(lppaca->yield_count);
 902         spin_unlock(&vcpu->arch.vpa_update_lock);
 903         return yield_count;
 904 }
 905
 906 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 907 {
 908         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 909         unsigned long target, ret = H_SUCCESS;
 910         int yield_count;
 911         struct kvm_vcpu *tvcpu;
 912         int idx, rc;
 913
 914         if (req <= MAX_HCALL_OPCODE &&
 915             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 916                 return RESUME_HOST;
 917
 918         switch (req) {
 919         case H_CEDE:
 920                 break;
 921         case H_PROD:
 922                 target = kvmppc_get_gpr(vcpu, 4);
 923                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 924                 if (!tvcpu) {
 925                         ret = H_PARAMETER;
 926                         break;
 927                 }
 928                 tvcpu->arch.prodded = 1;
 929                 smp_mb();
 930                 if (tvcpu->arch.ceded)
 931                         kvmppc_fast_vcpu_kick_hv(tvcpu);
 932                 break;
 933         case H_CONFER:
 934                 target = kvmppc_get_gpr(vcpu, 4);
 935                 if (target == -1)
 936                         break;
 937                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 938                 if (!tvcpu) {
 939                         ret = H_PARAMETER;
 940                         break;
 941                 }
 942                 yield_count = kvmppc_get_gpr(vcpu, 5);
 943                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 944                         break;
 945                 kvm_arch_vcpu_yield_to(tvcpu);
 946                 break;
 947         case H_REGISTER_VPA:
 948                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 949                                         kvmppc_get_gpr(vcpu, 5),
 950                                         kvmppc_get_gpr(vcpu, 6));
 951                 break;
 952         case H_RTAS:
 953                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 954                         return RESUME_HOST;
 955
 956                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 957                 rc = kvmppc_rtas_hcall(vcpu);
 958                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 959
 960                 if (rc == -ENOENT)
 961                         return RESUME_HOST;
 962                 else if (rc == 0)
 963                         break;
 964
 965                 /* Send the error out to userspace via KVM_RUN */
 966                 return rc;
 967         case H_LOGICAL_CI_LOAD:
 968                 ret = kvmppc_h_logical_ci_load(vcpu);
 969                 if (ret == H_TOO_HARD)
 970                         return RESUME_HOST;
 971                 break;
 972         case H_LOGICAL_CI_STORE:
 973                 ret = kvmppc_h_logical_ci_store(vcpu);
 974                 if (ret == H_TOO_HARD)
 975                         return RESUME_HOST;
 976                 break;
 977         case H_SET_MODE:
 978                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 979                                         kvmppc_get_gpr(vcpu, 5),
 980                                         kvmppc_get_gpr(vcpu, 6),
 981                                         kvmppc_get_gpr(vcpu, 7));
 982                 if (ret == H_TOO_HARD)
 983                         return RESUME_HOST;
 984                 break;
 985         case H_XIRR:
 986         case H_CPPR:
 987         case H_EOI:
 988         case H_IPI:
 989         case H_IPOLL:
 990         case H_XIRR_X:
 991                 if (kvmppc_xics_enabled(vcpu)) {
 992                         if (xics_on_xive()) {
 993                                 ret = H_NOT_AVAILABLE;
 994                                 return RESUME_GUEST;
 995                         }
 996                         ret = kvmppc_xics_hcall(vcpu, req);
 997                         break;
 998                 }
 999                 return RESUME_HOST;
1000         case H_SET_DABR:
1001                 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1002                 break;
1003         case H_SET_XDABR:
1004                 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1005                                                 kvmppc_get_gpr(vcpu, 5));
1006                 break;
1007 #ifdef CONFIG_SPAPR_TCE_IOMMU
1008         case H_GET_TCE:
1009                 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1010                                                 kvmppc_get_gpr(vcpu, 5));
1011                 if (ret == H_TOO_HARD)
1012                         return RESUME_HOST;
1013                 break;
1014         case H_PUT_TCE:
1015                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1016                                                 kvmppc_get_gpr(vcpu, 5),
1017                                                 kvmppc_get_gpr(vcpu, 6));
1018                 if (ret == H_TOO_HARD)
1019                         return RESUME_HOST;
1020                 break;
1021         case H_PUT_TCE_INDIRECT:
1022                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1023                                                 kvmppc_get_gpr(vcpu, 5),
1024                                                 kvmppc_get_gpr(vcpu, 6),
1025                                                 kvmppc_get_gpr(vcpu, 7));
1026                 if (ret == H_TOO_HARD)
1027                         return RESUME_HOST;
1028                 break;
1029         case H_STUFF_TCE:
1030                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1031                                                 kvmppc_get_gpr(vcpu, 5),
1032                                                 kvmppc_get_gpr(vcpu, 6),
1033                                                 kvmppc_get_gpr(vcpu, 7));
1034                 if (ret == H_TOO_HARD)
1035                         return RESUME_HOST;
1036                 break;
1037 #endif
1038         case H_RANDOM:
1039                 if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1040                         ret = H_HARDWARE;
1041                 break;
1042
1043         case H_SET_PARTITION_TABLE:
1044                 ret = H_FUNCTION;
1045                 if (nesting_enabled(vcpu->kvm))
1046                         ret = kvmhv_set_partition_table(vcpu);
1047                 break;
1048         case H_ENTER_NESTED:
1049                 ret = H_FUNCTION;
1050                 if (!nesting_enabled(vcpu->kvm))
1051                         break;
1052                 ret = kvmhv_enter_nested_guest(vcpu);
1053                 if (ret == H_INTERRUPT) {
1054                         kvmppc_set_gpr(vcpu, 3, 0);
1055                         vcpu->arch.hcall_needed = 0;
1056                         return -EINTR;
1057                 } else if (ret == H_TOO_HARD) {
1058                         kvmppc_set_gpr(vcpu, 3, 0);
1059                         vcpu->arch.hcall_needed = 0;
1060                         return RESUME_HOST;
1061                 }
1062                 break;
1063         case H_TLB_INVALIDATE:
1064                 ret = H_FUNCTION;
1065                 if (nesting_enabled(vcpu->kvm))
1066                         ret = kvmhv_do_nested_tlbie(vcpu);
1067                 break;
1068         case H_COPY_TOFROM_GUEST:
1069                 ret = H_FUNCTION;
1070                 if (nesting_enabled(vcpu->kvm))
1071                         ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1072                 break;
1073         case H_PAGE_INIT:
1074                 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1075                                          kvmppc_get_gpr(vcpu, 5),
1076                                          kvmppc_get_gpr(vcpu, 6));
1077                 break;
1078         default:
1079                 return RESUME_HOST;
1080         }
1081         kvmppc_set_gpr(vcpu, 3, ret);
1082         vcpu->arch.hcall_needed = 0;
1083         return RESUME_GUEST;
1084 }
1085
1086 /*
1087  * Handle H_CEDE in the nested virtualization case where we haven't
1088  * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1089  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1090  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1091  */
1092 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1093 {
1094         vcpu->arch.shregs.msr |= MSR_EE;
1095         vcpu->arch.ceded = 1;
1096         smp_mb();
1097         if (vcpu->arch.prodded) {
1098                 vcpu->arch.prodded = 0;
1099                 smp_mb();
1100                 vcpu->arch.ceded = 0;
1101         }
1102 }
1103
1104 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1105 {
1106         switch (cmd) {
1107         case H_CEDE:
1108         case H_PROD:
1109         case H_CONFER:
1110         case H_REGISTER_VPA:
1111         case H_SET_MODE:
1112         case H_LOGICAL_CI_LOAD:
1113         case H_LOGICAL_CI_STORE:
1114 #ifdef CONFIG_KVM_XICS
1115         case H_XIRR:
1116         case H_CPPR:
1117         case H_EOI:
1118         case H_IPI:
1119         case H_IPOLL:
1120         case H_XIRR_X:
1121 #endif
1122         case H_PAGE_INIT:
1123                 return 1;
1124         }
1125
1126         /* See if it's in the real-mode table */
1127         return kvmppc_hcall_impl_hv_realmode(cmd);
1128 }
1129
1130 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
1131                                         struct kvm_vcpu *vcpu)
1132 {
1133         u32 last_inst;
1134
1135         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1136                                         EMULATE_DONE) {
1137                 /*
1138                  * Fetch failed, so return to guest and
1139                  * try executing it again.
1140                  */
1141                 return RESUME_GUEST;
1142         }
1143
1144         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1145                 run->exit_reason = KVM_EXIT_DEBUG;
1146                 run->debug.arch.address = kvmppc_get_pc(vcpu);
1147                 return RESUME_HOST;
1148         } else {
1149                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1150                 return RESUME_GUEST;
1151         }
1152 }
1153
1154 static void do_nothing(void *x)
1155 {
1156 }
1157
1158 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1159 {
1160         int thr, cpu, pcpu, nthreads;
1161         struct kvm_vcpu *v;
1162         unsigned long dpdes;
1163
1164         nthreads = vcpu->kvm->arch.emul_smt_mode;
1165         dpdes = 0;
1166         cpu = vcpu->vcpu_id & ~(nthreads - 1);
1167         for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1168                 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1169                 if (!v)
1170                         continue;
1171                 /*
1172                  * If the vcpu is currently running on a physical cpu thread,
1173                  * interrupt it in order to pull it out of the guest briefly,
1174                  * which will update its vcore->dpdes value.
1175                  */
1176                 pcpu = READ_ONCE(v->cpu);
1177                 if (pcpu >= 0)
1178                         smp_call_function_single(pcpu, do_nothing, NULL, 1);
1179                 if (kvmppc_doorbell_pending(v))
1180                         dpdes |= 1 << thr;
1181         }
1182         return dpdes;
1183 }
1184
1185 /*
1186  * On POWER9, emulate doorbell-related instructions in order to
1187  * give the guest the illusion of running on a multi-threaded core.
1188  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1189  * and mfspr DPDES.
1190  */
1191 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1192 {
1193         u32 inst, rb, thr;
1194         unsigned long arg;
1195         struct kvm *kvm = vcpu->kvm;
1196         struct kvm_vcpu *tvcpu;
1197
1198         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1199                 return RESUME_GUEST;
1200         if (get_op(inst) != 31)
1201                 return EMULATE_FAIL;
1202         rb = get_rb(inst);
1203         thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1204         switch (get_xop(inst)) {
1205         case OP_31_XOP_MSGSNDP:
1206                 arg = kvmppc_get_gpr(vcpu, rb);
1207                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1208                         break;
1209                 arg &= 0x3f;
1210                 if (arg >= kvm->arch.emul_smt_mode)
1211                         break;
1212                 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1213                 if (!tvcpu)
1214                         break;
1215                 if (!tvcpu->arch.doorbell_request) {
1216                         tvcpu->arch.doorbell_request = 1;
1217                         kvmppc_fast_vcpu_kick_hv(tvcpu);
1218                 }
1219                 break;
1220         case OP_31_XOP_MSGCLRP:
1221                 arg = kvmppc_get_gpr(vcpu, rb);
1222                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1223                         break;
1224                 vcpu->arch.vcore->dpdes = 0;
1225                 vcpu->arch.doorbell_request = 0;
1226                 break;
1227         case OP_31_XOP_MFSPR:
1228                 switch (get_sprn(inst)) {
1229                 case SPRN_TIR:
1230                         arg = thr;
1231                         break;
1232                 case SPRN_DPDES:
1233                         arg = kvmppc_read_dpdes(vcpu);
1234                         break;
1235                 default:
1236                         return EMULATE_FAIL;
1237                 }
1238                 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1239                 break;
1240         default:
1241                 return EMULATE_FAIL;
1242         }
1243         kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1244         return RESUME_GUEST;
1245 }
1246
1247 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1248                                  struct task_struct *tsk)
1249 {
1250         int r = RESUME_HOST;
1251
1252         vcpu->stat.sum_exits++;
1253
1254         /*
1255          * This can happen if an interrupt occurs in the last stages
1256          * of guest entry or the first stages of guest exit (i.e. after
1257          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1258          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1259          * That can happen due to a bug, or due to a machine check
1260          * occurring at just the wrong time.
1261          */
1262         if (vcpu->arch.shregs.msr & MSR_HV) {
1263                 printk(KERN_EMERG "KVM trap in HV mode!\n");
1264                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1265                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1266                         vcpu->arch.shregs.msr);
1267                 kvmppc_dump_regs(vcpu);
1268                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1269                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1270                 return RESUME_HOST;
1271         }
1272         run->exit_reason = KVM_EXIT_UNKNOWN;
1273         run->ready_for_interrupt_injection = 1;
1274         switch (vcpu->arch.trap) {
1275         /* We're good on these - the host merely wanted to get our attention */
1276         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1277                 vcpu->stat.dec_exits++;
1278                 r = RESUME_GUEST;
1279                 break;
1280         case BOOK3S_INTERRUPT_EXTERNAL:
1281         case BOOK3S_INTERRUPT_H_DOORBELL:
1282         case BOOK3S_INTERRUPT_H_VIRT:
1283                 vcpu->stat.ext_intr_exits++;
1284                 r = RESUME_GUEST;
1285                 break;
1286         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1287         case BOOK3S_INTERRUPT_HMI:
1288         case BOOK3S_INTERRUPT_PERFMON:
1289         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1290                 r = RESUME_GUEST;
1291                 break;
1292         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1293                 /* Print the MCE event to host console. */
1294                 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1295
1296                 /*
1297                  * If the guest can do FWNMI, exit to userspace so it can
1298                  * deliver a FWNMI to the guest.
1299                  * Otherwise we synthesize a machine check for the guest
1300                  * so that it knows that the machine check occurred.
1301                  */
1302                 if (!vcpu->kvm->arch.fwnmi_enabled) {
1303                         ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1304                         kvmppc_core_queue_machine_check(vcpu, flags);
1305                         r = RESUME_GUEST;
1306                         break;
1307                 }
1308
1309                 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1310                 run->exit_reason = KVM_EXIT_NMI;
1311                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1312                 /* Clear out the old NMI status from run->flags */
1313                 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1314                 /* Now set the NMI status */
1315                 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1316                         run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1317                 else
1318                         run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1319
1320                 r = RESUME_HOST;
1321                 break;
1322         case BOOK3S_INTERRUPT_PROGRAM:
1323         {
1324                 ulong flags;
1325                 /*
1326                  * Normally program interrupts are delivered directly
1327                  * to the guest by the hardware, but we can get here
1328                  * as a result of a hypervisor emulation interrupt
1329                  * (e40) getting turned into a 700 by BML RTAS.
1330                  */
1331                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1332                 kvmppc_core_queue_program(vcpu, flags);
1333                 r = RESUME_GUEST;
1334                 break;
1335         }
1336         case BOOK3S_INTERRUPT_SYSCALL:
1337         {
1338                 /* hcall - punt to userspace */
1339                 int i;
1340
1341                 /* hypercall with MSR_PR has already been handled in rmode,
1342                  * and never reaches here.
1343                  */
1344
1345                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1346                 for (i = 0; i < 9; ++i)
1347                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1348                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1349                 vcpu->arch.hcall_needed = 1;
1350                 r = RESUME_HOST;
1351                 break;
1352         }
1353         /*
1354          * We get these next two if the guest accesses a page which it thinks
1355          * it has mapped but which is not actually present, either because
1356          * it is for an emulated I/O device or because the corresonding
1357          * host page has been paged out.  Any other HDSI/HISI interrupts
1358          * have been handled already.
1359          */
1360         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1361                 r = RESUME_PAGE_FAULT;
1362                 break;
1363         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1364                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1365                 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1366                         DSISR_SRR1_MATCH_64S;
1367                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1368                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1369                 r = RESUME_PAGE_FAULT;
1370                 break;
1371         /*
1372          * This occurs if the guest executes an illegal instruction.
1373          * If the guest debug is disabled, generate a program interrupt
1374          * to the guest. If guest debug is enabled, we need to check
1375          * whether the instruction is a software breakpoint instruction.
1376          * Accordingly return to Guest or Host.
1377          */
1378         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1379                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1380                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1381                                 swab32(vcpu->arch.emul_inst) :
1382                                 vcpu->arch.emul_inst;
1383                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1384                         r = kvmppc_emulate_debug_inst(run, vcpu);
1385                 } else {
1386                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1387                         r = RESUME_GUEST;
1388                 }
1389                 break;
1390         /*
1391          * This occurs if the guest (kernel or userspace), does something that
1392          * is prohibited by HFSCR.
1393          * On POWER9, this could be a doorbell instruction that we need
1394          * to emulate.
1395          * Otherwise, we just generate a program interrupt to the guest.
1396          */
1397         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1398                 r = EMULATE_FAIL;
1399                 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1400                     cpu_has_feature(CPU_FTR_ARCH_300))
1401                         r = kvmppc_emulate_doorbell_instr(vcpu);
1402                 if (r == EMULATE_FAIL) {
1403                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1404                         r = RESUME_GUEST;
1405                 }
1406                 break;
1407
1408 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1409         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1410                 /*
1411                  * This occurs for various TM-related instructions that
1412                  * we need to emulate on POWER9 DD2.2.  We have already
1413                  * handled the cases where the guest was in real-suspend
1414                  * mode and was transitioning to transactional state.
1415                  */
1416                 r = kvmhv_p9_tm_emulation(vcpu);
1417                 break;
1418 #endif
1419
1420         case BOOK3S_INTERRUPT_HV_RM_HARD:
1421                 r = RESUME_PASSTHROUGH;
1422                 break;
1423         default:
1424                 kvmppc_dump_regs(vcpu);
1425                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1426                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1427                         vcpu->arch.shregs.msr);
1428                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1429                 r = RESUME_HOST;
1430                 break;
1431         }
1432
1433         return r;
1434 }
1435
1436 static int kvmppc_handle_nested_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1437 {
1438         int r;
1439         int srcu_idx;
1440
1441         vcpu->stat.sum_exits++;
1442
1443         /*
1444          * This can happen if an interrupt occurs in the last stages
1445          * of guest entry or the first stages of guest exit (i.e. after
1446          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1447          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1448          * That can happen due to a bug, or due to a machine check
1449          * occurring at just the wrong time.
1450          */
1451         if (vcpu->arch.shregs.msr & MSR_HV) {
1452                 pr_emerg("KVM trap in HV mode while nested!\n");
1453                 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1454                          vcpu->arch.trap, kvmppc_get_pc(vcpu),
1455                          vcpu->arch.shregs.msr);
1456                 kvmppc_dump_regs(vcpu);
1457                 return RESUME_HOST;
1458         }
1459         switch (vcpu->arch.trap) {
1460         /* We're good on these - the host merely wanted to get our attention */
1461         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1462                 vcpu->stat.dec_exits++;
1463                 r = RESUME_GUEST;
1464                 break;
1465         case BOOK3S_INTERRUPT_EXTERNAL:
1466                 vcpu->stat.ext_intr_exits++;
1467                 r = RESUME_HOST;
1468                 break;
1469         case BOOK3S_INTERRUPT_H_DOORBELL:
1470         case BOOK3S_INTERRUPT_H_VIRT:
1471                 vcpu->stat.ext_intr_exits++;
1472                 r = RESUME_GUEST;
1473                 break;
1474         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1475         case BOOK3S_INTERRUPT_HMI:
1476         case BOOK3S_INTERRUPT_PERFMON:
1477         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1478                 r = RESUME_GUEST;
1479                 break;
1480         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1481                 /* Pass the machine check to the L1 guest */
1482                 r = RESUME_HOST;
1483                 /* Print the MCE event to host console. */
1484                 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1485                 break;
1486         /*
1487          * We get these next two if the guest accesses a page which it thinks
1488          * it has mapped but which is not actually present, either because
1489          * it is for an emulated I/O device or because the corresonding
1490          * host page has been paged out.
1491          */
1492         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1493                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1494                 r = kvmhv_nested_page_fault(run, vcpu);
1495                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1496                 break;
1497         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1498                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1499                 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1500                                          DSISR_SRR1_MATCH_64S;
1501                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1502                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1503                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1504                 r = kvmhv_nested_page_fault(run, vcpu);
1505                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1506                 break;
1507
1508 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1509         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1510                 /*
1511                  * This occurs for various TM-related instructions that
1512                  * we need to emulate on POWER9 DD2.2.  We have already
1513                  * handled the cases where the guest was in real-suspend
1514                  * mode and was transitioning to transactional state.
1515                  */
1516                 r = kvmhv_p9_tm_emulation(vcpu);
1517                 break;
1518 #endif
1519
1520         case BOOK3S_INTERRUPT_HV_RM_HARD:
1521                 vcpu->arch.trap = 0;
1522                 r = RESUME_GUEST;
1523                 if (!xics_on_xive())
1524                         kvmppc_xics_rm_complete(vcpu, 0);
1525                 break;
1526         default:
1527                 r = RESUME_HOST;
1528                 break;
1529         }
1530
1531         return r;
1532 }
1533
1534 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1535                                             struct kvm_sregs *sregs)
1536 {
1537         int i;
1538
1539         memset(sregs, 0, sizeof(struct kvm_sregs));
1540         sregs->pvr = vcpu->arch.pvr;
1541         for (i = 0; i < vcpu->arch.slb_max; i++) {
1542                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1543                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1544         }
1545
1546         return 0;
1547 }
1548
1549 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1550                                             struct kvm_sregs *sregs)
1551 {
1552         int i, j;
1553
1554         /* Only accept the same PVR as the host's, since we can't spoof it */
1555         if (sregs->pvr != vcpu->arch.pvr)
1556                 return -EINVAL;
1557
1558         j = 0;
1559         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1560                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1561                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1562                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1563                         ++j;
1564                 }
1565         }
1566         vcpu->arch.slb_max = j;
1567
1568         return 0;
1569 }
1570
1571 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1572                 bool preserve_top32)
1573 {
1574         struct kvm *kvm = vcpu->kvm;
1575         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1576         u64 mask;
1577
1578         spin_lock(&vc->lock);
1579         /*
1580          * If ILE (interrupt little-endian) has changed, update the
1581          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1582          */
1583         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1584                 struct kvm_vcpu *vcpu;
1585                 int i;
1586
1587                 kvm_for_each_vcpu(i, vcpu, kvm) {
1588                         if (vcpu->arch.vcore != vc)
1589                                 continue;
1590                         if (new_lpcr & LPCR_ILE)
1591                                 vcpu->arch.intr_msr |= MSR_LE;
1592                         else
1593                                 vcpu->arch.intr_msr &= ~MSR_LE;
1594                 }
1595         }
1596
1597         /*
1598          * Userspace can only modify DPFD (default prefetch depth),
1599          * ILE (interrupt little-endian) and TC (translation control).
1600          * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1601          */
1602         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1603         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1604                 mask |= LPCR_AIL;
1605         /*
1606          * On POWER9, allow userspace to enable large decrementer for the
1607          * guest, whether or not the host has it enabled.
1608          */
1609         if (cpu_has_feature(CPU_FTR_ARCH_300))
1610                 mask |= LPCR_LD;
1611
1612         /* Broken 32-bit version of LPCR must not clear top bits */
1613         if (preserve_top32)
1614                 mask &= 0xFFFFFFFF;
1615         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1616         spin_unlock(&vc->lock);
1617 }
1618
1619 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1620                                  union kvmppc_one_reg *val)
1621 {
1622         int r = 0;
1623         long int i;
1624
1625         switch (id) {
1626         case KVM_REG_PPC_DEBUG_INST:
1627                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1628                 break;
1629         case KVM_REG_PPC_HIOR:
1630                 *val = get_reg_val(id, 0);
1631                 break;
1632         case KVM_REG_PPC_DABR:
1633                 *val = get_reg_val(id, vcpu->arch.dabr);
1634                 break;
1635         case KVM_REG_PPC_DABRX:
1636                 *val = get_reg_val(id, vcpu->arch.dabrx);
1637                 break;
1638         case KVM_REG_PPC_DSCR:
1639                 *val = get_reg_val(id, vcpu->arch.dscr);
1640                 break;
1641         case KVM_REG_PPC_PURR:
1642                 *val = get_reg_val(id, vcpu->arch.purr);
1643                 break;
1644         case KVM_REG_PPC_SPURR:
1645                 *val = get_reg_val(id, vcpu->arch.spurr);
1646                 break;
1647         case KVM_REG_PPC_AMR:
1648                 *val = get_reg_val(id, vcpu->arch.amr);
1649                 break;
1650         case KVM_REG_PPC_UAMOR:
1651                 *val = get_reg_val(id, vcpu->arch.uamor);
1652                 break;
1653         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1654                 i = id - KVM_REG_PPC_MMCR0;
1655                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1656                 break;
1657         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1658                 i = id - KVM_REG_PPC_PMC1;
1659                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1660                 break;
1661         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1662                 i = id - KVM_REG_PPC_SPMC1;
1663                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1664                 break;
1665         case KVM_REG_PPC_SIAR:
1666                 *val = get_reg_val(id, vcpu->arch.siar);
1667                 break;
1668         case KVM_REG_PPC_SDAR:
1669                 *val = get_reg_val(id, vcpu->arch.sdar);
1670                 break;
1671         case KVM_REG_PPC_SIER:
1672                 *val = get_reg_val(id, vcpu->arch.sier);
1673                 break;
1674         case KVM_REG_PPC_IAMR:
1675                 *val = get_reg_val(id, vcpu->arch.iamr);
1676                 break;
1677         case KVM_REG_PPC_PSPB:
1678                 *val = get_reg_val(id, vcpu->arch.pspb);
1679                 break;
1680         case KVM_REG_PPC_DPDES:
1681                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1682                 break;
1683         case KVM_REG_PPC_VTB:
1684                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1685                 break;
1686         case KVM_REG_PPC_DAWR:
1687                 *val = get_reg_val(id, vcpu->arch.dawr);
1688                 break;
1689         case KVM_REG_PPC_DAWRX:
1690                 *val = get_reg_val(id, vcpu->arch.dawrx);
1691                 break;
1692         case KVM_REG_PPC_CIABR:
1693                 *val = get_reg_val(id, vcpu->arch.ciabr);
1694                 break;
1695         case KVM_REG_PPC_CSIGR:
1696                 *val = get_reg_val(id, vcpu->arch.csigr);
1697                 break;
1698         case KVM_REG_PPC_TACR:
1699                 *val = get_reg_val(id, vcpu->arch.tacr);
1700                 break;
1701         case KVM_REG_PPC_TCSCR:
1702                 *val = get_reg_val(id, vcpu->arch.tcscr);
1703                 break;
1704         case KVM_REG_PPC_PID:
1705                 *val = get_reg_val(id, vcpu->arch.pid);
1706                 break;
1707         case KVM_REG_PPC_ACOP:
1708                 *val = get_reg_val(id, vcpu->arch.acop);
1709                 break;
1710         case KVM_REG_PPC_WORT:
1711                 *val = get_reg_val(id, vcpu->arch.wort);
1712                 break;
1713         case KVM_REG_PPC_TIDR:
1714                 *val = get_reg_val(id, vcpu->arch.tid);
1715                 break;
1716         case KVM_REG_PPC_PSSCR:
1717                 *val = get_reg_val(id, vcpu->arch.psscr);
1718                 break;
1719         case KVM_REG_PPC_VPA_ADDR:
1720                 spin_lock(&vcpu->arch.vpa_update_lock);
1721                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1722                 spin_unlock(&vcpu->arch.vpa_update_lock);
1723                 break;
1724         case KVM_REG_PPC_VPA_SLB:
1725                 spin_lock(&vcpu->arch.vpa_update_lock);
1726                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1727                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1728                 spin_unlock(&vcpu->arch.vpa_update_lock);
1729                 break;
1730         case KVM_REG_PPC_VPA_DTL:
1731                 spin_lock(&vcpu->arch.vpa_update_lock);
1732                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1733                 val->vpaval.length = vcpu->arch.dtl.len;
1734                 spin_unlock(&vcpu->arch.vpa_update_lock);
1735                 break;
1736         case KVM_REG_PPC_TB_OFFSET:
1737                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1738                 break;
1739         case KVM_REG_PPC_LPCR:
1740         case KVM_REG_PPC_LPCR_64:
1741                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1742                 break;
1743         case KVM_REG_PPC_PPR:
1744                 *val = get_reg_val(id, vcpu->arch.ppr);
1745                 break;
1746 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1747         case KVM_REG_PPC_TFHAR:
1748                 *val = get_reg_val(id, vcpu->arch.tfhar);
1749                 break;
1750         case KVM_REG_PPC_TFIAR:
1751                 *val = get_reg_val(id, vcpu->arch.tfiar);
1752                 break;
1753         case KVM_REG_PPC_TEXASR:
1754                 *val = get_reg_val(id, vcpu->arch.texasr);
1755                 break;
1756         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1757                 i = id - KVM_REG_PPC_TM_GPR0;
1758                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1759                 break;
1760         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1761         {
1762                 int j;
1763                 i = id - KVM_REG_PPC_TM_VSR0;
1764                 if (i < 32)
1765                         for (j = 0; j < TS_FPRWIDTH; j++)
1766                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1767                 else {
1768                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1769                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1770                         else
1771                                 r = -ENXIO;
1772                 }
1773                 break;
1774         }
1775         case KVM_REG_PPC_TM_CR:
1776                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1777                 break;
1778         case KVM_REG_PPC_TM_XER:
1779                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1780                 break;
1781         case KVM_REG_PPC_TM_LR:
1782                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1783                 break;
1784         case KVM_REG_PPC_TM_CTR:
1785                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1786                 break;
1787         case KVM_REG_PPC_TM_FPSCR:
1788                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1789                 break;
1790         case KVM_REG_PPC_TM_AMR:
1791                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1792                 break;
1793         case KVM_REG_PPC_TM_PPR:
1794                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1795                 break;
1796         case KVM_REG_PPC_TM_VRSAVE:
1797                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1798                 break;
1799         case KVM_REG_PPC_TM_VSCR:
1800                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1801                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1802                 else
1803                         r = -ENXIO;
1804                 break;
1805         case KVM_REG_PPC_TM_DSCR:
1806                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1807                 break;
1808         case KVM_REG_PPC_TM_TAR:
1809                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1810                 break;
1811 #endif
1812         case KVM_REG_PPC_ARCH_COMPAT:
1813                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1814                 break;
1815         case KVM_REG_PPC_DEC_EXPIRY:
1816                 *val = get_reg_val(id, vcpu->arch.dec_expires +
1817                                    vcpu->arch.vcore->tb_offset);
1818                 break;
1819         case KVM_REG_PPC_ONLINE:
1820                 *val = get_reg_val(id, vcpu->arch.online);
1821                 break;
1822         case KVM_REG_PPC_PTCR:
1823                 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1824                 break;
1825         default:
1826                 r = -EINVAL;
1827                 break;
1828         }
1829
1830         return r;
1831 }
1832
1833 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1834                                  union kvmppc_one_reg *val)
1835 {
1836         int r = 0;
1837         long int i;
1838         unsigned long addr, len;
1839
1840         switch (id) {
1841         case KVM_REG_PPC_HIOR:
1842                 /* Only allow this to be set to zero */
1843                 if (set_reg_val(id, *val))
1844                         r = -EINVAL;
1845                 break;
1846         case KVM_REG_PPC_DABR:
1847                 vcpu->arch.dabr = set_reg_val(id, *val);
1848                 break;
1849         case KVM_REG_PPC_DABRX:
1850                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1851                 break;
1852         case KVM_REG_PPC_DSCR:
1853                 vcpu->arch.dscr = set_reg_val(id, *val);
1854                 break;
1855         case KVM_REG_PPC_PURR:
1856                 vcpu->arch.purr = set_reg_val(id, *val);
1857                 break;
1858         case KVM_REG_PPC_SPURR:
1859                 vcpu->arch.spurr = set_reg_val(id, *val);
1860                 break;
1861         case KVM_REG_PPC_AMR:
1862                 vcpu->arch.amr = set_reg_val(id, *val);
1863                 break;
1864         case KVM_REG_PPC_UAMOR:
1865                 vcpu->arch.uamor = set_reg_val(id, *val);
1866                 break;
1867         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1868                 i = id - KVM_REG_PPC_MMCR0;
1869                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1870                 break;
1871         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1872                 i = id - KVM_REG_PPC_PMC1;
1873                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1874                 break;
1875         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1876                 i = id - KVM_REG_PPC_SPMC1;
1877                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1878                 break;
1879         case KVM_REG_PPC_SIAR:
1880                 vcpu->arch.siar = set_reg_val(id, *val);
1881                 break;
1882         case KVM_REG_PPC_SDAR:
1883                 vcpu->arch.sdar = set_reg_val(id, *val);
1884                 break;
1885         case KVM_REG_PPC_SIER:
1886                 vcpu->arch.sier = set_reg_val(id, *val);
1887                 break;
1888         case KVM_REG_PPC_IAMR:
1889                 vcpu->arch.iamr = set_reg_val(id, *val);
1890                 break;
1891         case KVM_REG_PPC_PSPB:
1892                 vcpu->arch.pspb = set_reg_val(id, *val);
1893                 break;
1894         case KVM_REG_PPC_DPDES:
1895                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1896                 break;
1897         case KVM_REG_PPC_VTB:
1898                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1899                 break;
1900         case KVM_REG_PPC_DAWR:
1901                 vcpu->arch.dawr = set_reg_val(id, *val);
1902                 break;
1903         case KVM_REG_PPC_DAWRX:
1904                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1905                 break;
1906         case KVM_REG_PPC_CIABR:
1907                 vcpu->arch.ciabr = set_reg_val(id, *val);
1908                 /* Don't allow setting breakpoints in hypervisor code */
1909                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1910                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1911                 break;
1912         case KVM_REG_PPC_CSIGR:
1913                 vcpu->arch.csigr = set_reg_val(id, *val);
1914                 break;
1915         case KVM_REG_PPC_TACR:
1916                 vcpu->arch.tacr = set_reg_val(id, *val);
1917                 break;
1918         case KVM_REG_PPC_TCSCR:
1919                 vcpu->arch.tcscr = set_reg_val(id, *val);
1920                 break;
1921         case KVM_REG_PPC_PID:
1922                 vcpu->arch.pid = set_reg_val(id, *val);
1923                 break;
1924         case KVM_REG_PPC_ACOP:
1925                 vcpu->arch.acop = set_reg_val(id, *val);
1926                 break;
1927         case KVM_REG_PPC_WORT:
1928                 vcpu->arch.wort = set_reg_val(id, *val);
1929                 break;
1930         case KVM_REG_PPC_TIDR:
1931                 vcpu->arch.tid = set_reg_val(id, *val);
1932                 break;
1933         case KVM_REG_PPC_PSSCR:
1934                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1935                 break;
1936         case KVM_REG_PPC_VPA_ADDR:
1937                 addr = set_reg_val(id, *val);
1938                 r = -EINVAL;
1939                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1940                               vcpu->arch.dtl.next_gpa))
1941                         break;
1942                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1943                 break;
1944         case KVM_REG_PPC_VPA_SLB:
1945                 addr = val->vpaval.addr;
1946                 len = val->vpaval.length;
1947                 r = -EINVAL;
1948                 if (addr && !vcpu->arch.vpa.next_gpa)
1949                         break;
1950                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1951                 break;
1952         case KVM_REG_PPC_VPA_DTL:
1953                 addr = val->vpaval.addr;
1954                 len = val->vpaval.length;
1955                 r = -EINVAL;
1956                 if (addr && (len < sizeof(struct dtl_entry) ||
1957                              !vcpu->arch.vpa.next_gpa))
1958                         break;
1959                 len -= len % sizeof(struct dtl_entry);
1960                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1961                 break;
1962         case KVM_REG_PPC_TB_OFFSET:
1963                 /* round up to multiple of 2^24 */
1964                 vcpu->arch.vcore->tb_offset =
1965                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1966                 break;
1967         case KVM_REG_PPC_LPCR:
1968                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1969                 break;
1970         case KVM_REG_PPC_LPCR_64:
1971                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1972                 break;
1973         case KVM_REG_PPC_PPR:
1974                 vcpu->arch.ppr = set_reg_val(id, *val);
1975                 break;
1976 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1977         case KVM_REG_PPC_TFHAR:
1978                 vcpu->arch.tfhar = set_reg_val(id, *val);
1979                 break;
1980         case KVM_REG_PPC_TFIAR:
1981                 vcpu->arch.tfiar = set_reg_val(id, *val);
1982                 break;
1983         case KVM_REG_PPC_TEXASR:
1984                 vcpu->arch.texasr = set_reg_val(id, *val);
1985                 break;
1986         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1987                 i = id - KVM_REG_PPC_TM_GPR0;
1988                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1989                 break;
1990         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1991         {
1992                 int j;
1993                 i = id - KVM_REG_PPC_TM_VSR0;
1994                 if (i < 32)
1995                         for (j = 0; j < TS_FPRWIDTH; j++)
1996                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1997                 else
1998                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1999                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2000                         else
2001                                 r = -ENXIO;
2002                 break;
2003         }
2004         case KVM_REG_PPC_TM_CR:
2005                 vcpu->arch.cr_tm = set_reg_val(id, *val);
2006                 break;
2007         case KVM_REG_PPC_TM_XER:
2008                 vcpu->arch.xer_tm = set_reg_val(id, *val);
2009                 break;
2010         case KVM_REG_PPC_TM_LR:
2011                 vcpu->arch.lr_tm = set_reg_val(id, *val);
2012                 break;
2013         case KVM_REG_PPC_TM_CTR:
2014                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2015                 break;
2016         case KVM_REG_PPC_TM_FPSCR:
2017                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2018                 break;
2019         case KVM_REG_PPC_TM_AMR:
2020                 vcpu->arch.amr_tm = set_reg_val(id, *val);
2021                 break;
2022         case KVM_REG_PPC_TM_PPR:
2023                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2024                 break;
2025         case KVM_REG_PPC_TM_VRSAVE:
2026                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2027                 break;
2028         case KVM_REG_PPC_TM_VSCR:
2029                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2030                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2031                 else
2032                         r = - ENXIO;
2033                 break;
2034         case KVM_REG_PPC_TM_DSCR:
2035                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2036                 break;
2037         case KVM_REG_PPC_TM_TAR:
2038                 vcpu->arch.tar_tm = set_reg_val(id, *val);
2039                 break;
2040 #endif
2041         case KVM_REG_PPC_ARCH_COMPAT:
2042                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2043                 break;
2044         case KVM_REG_PPC_DEC_EXPIRY:
2045                 vcpu->arch.dec_expires = set_reg_val(id, *val) -
2046                         vcpu->arch.vcore->tb_offset;
2047                 break;
2048         case KVM_REG_PPC_ONLINE:
2049                 i = set_reg_val(id, *val);
2050                 if (i && !vcpu->arch.online)
2051                         atomic_inc(&vcpu->arch.vcore->online_count);
2052                 else if (!i && vcpu->arch.online)
2053                         atomic_dec(&vcpu->arch.vcore->online_count);
2054                 vcpu->arch.online = i;
2055                 break;
2056         case KVM_REG_PPC_PTCR:
2057                 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2058                 break;
2059         default:
2060                 r = -EINVAL;
2061                 break;
2062         }
2063
2064         return r;
2065 }
2066
2067 /*
2068  * On POWER9, threads are independent and can be in different partitions.
2069  * Therefore we consider each thread to be a subcore.
2070  * There is a restriction that all threads have to be in the same
2071  * MMU mode (radix or HPT), unfortunately, but since we only support
2072  * HPT guests on a HPT host so far, that isn't an impediment yet.
2073  */
2074 static int threads_per_vcore(struct kvm *kvm)
2075 {
2076         if (kvm->arch.threads_indep)
2077                 return 1;
2078         return threads_per_subcore;
2079 }
2080
2081 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2082 {
2083         struct kvmppc_vcore *vcore;
2084
2085         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2086
2087         if (vcore == NULL)
2088                 return NULL;
2089
2090         spin_lock_init(&vcore->lock);
2091         spin_lock_init(&vcore->stoltb_lock);
2092         init_swait_queue_head(&vcore->wq);
2093         vcore->preempt_tb = TB_NIL;
2094         vcore->lpcr = kvm->arch.lpcr;
2095         vcore->first_vcpuid = id;
2096         vcore->kvm = kvm;
2097         INIT_LIST_HEAD(&vcore->preempt_list);
2098
2099         return vcore;
2100 }
2101
2102 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2103 static struct debugfs_timings_element {
2104         const char *name;
2105         size_t offset;
2106 } timings[] = {
2107         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
2108         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
2109         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
2110         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
2111         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
2112 };
2113
2114 #define N_TIMINGS       (ARRAY_SIZE(timings))
2115
2116 struct debugfs_timings_state {
2117         struct kvm_vcpu *vcpu;
2118         unsigned int    buflen;
2119         char            buf[N_TIMINGS * 100];
2120 };
2121
2122 static int debugfs_timings_open(struct inode *inode, struct file *file)
2123 {
2124         struct kvm_vcpu *vcpu = inode->i_private;
2125         struct debugfs_timings_state *p;
2126
2127         p = kzalloc(sizeof(*p), GFP_KERNEL);
2128         if (!p)
2129                 return -ENOMEM;
2130
2131         kvm_get_kvm(vcpu->kvm);
2132         p->vcpu = vcpu;
2133         file->private_data = p;
2134
2135         return nonseekable_open(inode, file);
2136 }
2137
2138 static int debugfs_timings_release(struct inode *inode, struct file *file)
2139 {
2140         struct debugfs_timings_state *p = file->private_data;
2141
2142         kvm_put_kvm(p->vcpu->kvm);
2143         kfree(p);
2144         return 0;
2145 }
2146
2147 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2148                                     size_t len, loff_t *ppos)
2149 {
2150         struct debugfs_timings_state *p = file->private_data;
2151         struct kvm_vcpu *vcpu = p->vcpu;
2152         char *s, *buf_end;
2153         struct kvmhv_tb_accumulator tb;
2154         u64 count;
2155         loff_t pos;
2156         ssize_t n;
2157         int i, loops;
2158         bool ok;
2159
2160         if (!p->buflen) {
2161                 s = p->buf;
2162                 buf_end = s + sizeof(p->buf);
2163                 for (i = 0; i < N_TIMINGS; ++i) {
2164                         struct kvmhv_tb_accumulator *acc;
2165
2166                         acc = (struct kvmhv_tb_accumulator *)
2167                                 ((unsigned long)vcpu + timings[i].offset);
2168                         ok = false;
2169                         for (loops = 0; loops < 1000; ++loops) {
2170                                 count = acc->seqcount;
2171                                 if (!(count & 1)) {
2172                                         smp_rmb();
2173                                         tb = *acc;
2174                                         smp_rmb();
2175                                         if (count == acc->seqcount) {
2176                                                 ok = true;
2177                                                 break;
2178                                         }
2179                                 }
2180                                 udelay(1);
2181                         }
2182                         if (!ok)
2183                                 snprintf(s, buf_end - s, "%s: stuck\n",
2184                                         timings[i].name);
2185                         else
2186                                 snprintf(s, buf_end - s,
2187                                         "%s: %llu %llu %llu %llu\n",
2188                                         timings[i].name, count / 2,
2189                                         tb_to_ns(tb.tb_total),
2190                                         tb_to_ns(tb.tb_min),
2191                                         tb_to_ns(tb.tb_max));
2192                         s += strlen(s);
2193                 }
2194                 p->buflen = s - p->buf;
2195         }
2196
2197         pos = *ppos;
2198         if (pos >= p->buflen)
2199                 return 0;
2200         if (len > p->buflen - pos)
2201                 len = p->buflen - pos;
2202         n = copy_to_user(buf, p->buf + pos, len);
2203         if (n) {
2204                 if (n == len)
2205                         return -EFAULT;
2206                 len -= n;
2207         }
2208         *ppos = pos + len;
2209         return len;
2210 }
2211
2212 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2213                                      size_t len, loff_t *ppos)
2214 {
2215         return -EACCES;
2216 }
2217
2218 static const struct file_operations debugfs_timings_ops = {
2219         .owner   = THIS_MODULE,
2220         .open    = debugfs_timings_open,
2221         .release = debugfs_timings_release,
2222         .read    = debugfs_timings_read,
2223         .write   = debugfs_timings_write,
2224         .llseek  = generic_file_llseek,
2225 };
2226
2227 /* Create a debugfs directory for the vcpu */
2228 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2229 {
2230         char buf[16];
2231         struct kvm *kvm = vcpu->kvm;
2232
2233         snprintf(buf, sizeof(buf), "vcpu%u", id);
2234         if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
2235                 return;
2236         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2237         if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
2238                 return;
2239         vcpu->arch.debugfs_timings =
2240                 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
2241                                     vcpu, &debugfs_timings_ops);
2242 }
2243
2244 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2245 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2246 {
2247 }
2248 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2249
2250 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
2251                                                    unsigned int id)
2252 {
2253         struct kvm_vcpu *vcpu;
2254         int err;
2255         int core;
2256         struct kvmppc_vcore *vcore;
2257
2258         err = -ENOMEM;
2259         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2260         if (!vcpu)
2261                 goto out;
2262
2263         err = kvm_vcpu_init(vcpu, kvm, id);
2264         if (err)
2265                 goto free_vcpu;
2266
2267         vcpu->arch.shared = &vcpu->arch.shregs;
2268 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2269         /*
2270          * The shared struct is never shared on HV,
2271          * so we can always use host endianness
2272          */
2273 #ifdef __BIG_ENDIAN__
2274         vcpu->arch.shared_big_endian = true;
2275 #else
2276         vcpu->arch.shared_big_endian = false;
2277 #endif
2278 #endif
2279         vcpu->arch.mmcr[0] = MMCR0_FC;
2280         vcpu->arch.ctrl = CTRL_RUNLATCH;
2281         /* default to host PVR, since we can't spoof it */
2282         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2283         spin_lock_init(&vcpu->arch.vpa_update_lock);
2284         spin_lock_init(&vcpu->arch.tbacct_lock);
2285         vcpu->arch.busy_preempt = TB_NIL;
2286         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2287
2288         /*
2289          * Set the default HFSCR for the guest from the host value.
2290          * This value is only used on POWER9.
2291          * On POWER9, we want to virtualize the doorbell facility, so we
2292          * don't set the HFSCR_MSGP bit, and that causes those instructions
2293          * to trap and then we emulate them.
2294          */
2295         vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2296                 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2297         if (cpu_has_feature(CPU_FTR_HVMODE)) {
2298                 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2299                 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2300                         vcpu->arch.hfscr |= HFSCR_TM;
2301         }
2302         if (cpu_has_feature(CPU_FTR_TM_COMP))
2303                 vcpu->arch.hfscr |= HFSCR_TM;
2304
2305         kvmppc_mmu_book3s_hv_init(vcpu);
2306
2307         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2308
2309         init_waitqueue_head(&vcpu->arch.cpu_run);
2310
2311         mutex_lock(&kvm->lock);
2312         vcore = NULL;
2313         err = -EINVAL;
2314         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2315                 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2316                         pr_devel("KVM: VCPU ID too high\n");
2317                         core = KVM_MAX_VCORES;
2318                 } else {
2319                         BUG_ON(kvm->arch.smt_mode != 1);
2320                         core = kvmppc_pack_vcpu_id(kvm, id);
2321                 }
2322         } else {
2323                 core = id / kvm->arch.smt_mode;
2324         }
2325         if (core < KVM_MAX_VCORES) {
2326                 vcore = kvm->arch.vcores[core];
2327                 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2328                         pr_devel("KVM: collision on id %u", id);
2329                         vcore = NULL;
2330                 } else if (!vcore) {
2331                         /*
2332                          * Take mmu_setup_lock for mutual exclusion
2333                          * with kvmppc_update_lpcr().
2334                          */
2335                         err = -ENOMEM;
2336                         vcore = kvmppc_vcore_create(kvm,
2337                                         id & ~(kvm->arch.smt_mode - 1));
2338                         mutex_lock(&kvm->arch.mmu_setup_lock);
2339                         kvm->arch.vcores[core] = vcore;
2340                         kvm->arch.online_vcores++;
2341                         mutex_unlock(&kvm->arch.mmu_setup_lock);
2342                 }
2343         }
2344         mutex_unlock(&kvm->lock);
2345
2346         if (!vcore)
2347                 goto free_vcpu;
2348
2349         spin_lock(&vcore->lock);
2350         ++vcore->num_threads;
2351         spin_unlock(&vcore->lock);
2352         vcpu->arch.vcore = vcore;
2353         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2354         vcpu->arch.thread_cpu = -1;
2355         vcpu->arch.prev_cpu = -1;
2356
2357         vcpu->arch.cpu_type = KVM_CPU_3S_64;
2358         kvmppc_sanity_check(vcpu);
2359
2360         debugfs_vcpu_init(vcpu, id);
2361
2362         return vcpu;
2363
2364 free_vcpu:
2365         kmem_cache_free(kvm_vcpu_cache, vcpu);
2366 out:
2367         return ERR_PTR(err);
2368 }
2369
2370 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2371                               unsigned long flags)
2372 {
2373         int err;
2374         int esmt = 0;
2375
2376         if (flags)
2377                 return -EINVAL;
2378         if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2379                 return -EINVAL;
2380         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2381                 /*
2382                  * On POWER8 (or POWER7), the threading mode is "strict",
2383                  * so we pack smt_mode vcpus per vcore.
2384                  */
2385                 if (smt_mode > threads_per_subcore)
2386                         return -EINVAL;
2387         } else {
2388                 /*
2389                  * On POWER9, the threading mode is "loose",
2390                  * so each vcpu gets its own vcore.
2391                  */
2392                 esmt = smt_mode;
2393                 smt_mode = 1;
2394         }
2395         mutex_lock(&kvm->lock);
2396         err = -EBUSY;
2397         if (!kvm->arch.online_vcores) {
2398                 kvm->arch.smt_mode = smt_mode;
2399                 kvm->arch.emul_smt_mode = esmt;
2400                 err = 0;
2401         }
2402         mutex_unlock(&kvm->lock);
2403
2404         return err;
2405 }
2406
2407 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2408 {
2409         if (vpa->pinned_addr)
2410                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2411                                         vpa->dirty);
2412 }
2413
2414 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2415 {
2416         spin_lock(&vcpu->arch.vpa_update_lock);
2417         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2418         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2419         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2420         spin_unlock(&vcpu->arch.vpa_update_lock);
2421         kvm_vcpu_uninit(vcpu);
2422         kmem_cache_free(kvm_vcpu_cache, vcpu);
2423 }
2424
2425 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2426 {
2427         /* Indicate we want to get back into the guest */
2428         return 1;
2429 }
2430
2431 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2432 {
2433         unsigned long dec_nsec, now;
2434
2435         now = get_tb();
2436         if (now > vcpu->arch.dec_expires) {
2437                 /* decrementer has already gone negative */
2438                 kvmppc_core_queue_dec(vcpu);
2439                 kvmppc_core_prepare_to_enter(vcpu);
2440                 return;
2441         }
2442         dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2443         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2444         vcpu->arch.timer_running = 1;
2445 }
2446
2447 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2448 {
2449         vcpu->arch.ceded = 0;
2450         if (vcpu->arch.timer_running) {
2451                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2452                 vcpu->arch.timer_running = 0;
2453         }
2454 }
2455
2456 extern int __kvmppc_vcore_entry(void);
2457
2458 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2459                                    struct kvm_vcpu *vcpu)
2460 {
2461         u64 now;
2462
2463         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2464                 return;
2465         spin_lock_irq(&vcpu->arch.tbacct_lock);
2466         now = mftb();
2467         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2468                 vcpu->arch.stolen_logged;
2469         vcpu->arch.busy_preempt = now;
2470         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2471         spin_unlock_irq(&vcpu->arch.tbacct_lock);
2472         --vc->n_runnable;
2473         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2474 }
2475
2476 static int kvmppc_grab_hwthread(int cpu)
2477 {
2478         struct paca_struct *tpaca;
2479         long timeout = 10000;
2480
2481         tpaca = paca_ptrs[cpu];
2482
2483         /* Ensure the thread won't go into the kernel if it wakes */
2484         tpaca->kvm_hstate.kvm_vcpu = NULL;
2485         tpaca->kvm_hstate.kvm_vcore = NULL;
2486         tpaca->kvm_hstate.napping = 0;
2487         smp_wmb();
2488         tpaca->kvm_hstate.hwthread_req = 1;
2489
2490         /*
2491          * If the thread is already executing in the kernel (e.g. handling
2492          * a stray interrupt), wait for it to get back to nap mode.
2493          * The smp_mb() is to ensure that our setting of hwthread_req
2494          * is visible before we look at hwthread_state, so if this
2495          * races with the code at system_reset_pSeries and the thread
2496          * misses our setting of hwthread_req, we are sure to see its
2497          * setting of hwthread_state, and vice versa.
2498          */
2499         smp_mb();
2500         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2501                 if (--timeout <= 0) {
2502                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
2503                         return -EBUSY;
2504                 }
2505                 udelay(1);
2506         }
2507         return 0;
2508 }
2509
2510 static void kvmppc_release_hwthread(int cpu)
2511 {
2512         struct paca_struct *tpaca;
2513
2514         tpaca = paca_ptrs[cpu];
2515         tpaca->kvm_hstate.hwthread_req = 0;
2516         tpaca->kvm_hstate.kvm_vcpu = NULL;
2517         tpaca->kvm_hstate.kvm_vcore = NULL;
2518         tpaca->kvm_hstate.kvm_split_mode = NULL;
2519 }
2520
2521 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2522 {
2523         struct kvm_nested_guest *nested = vcpu->arch.nested;
2524         cpumask_t *cpu_in_guest;
2525         int i;
2526
2527         cpu = cpu_first_thread_sibling(cpu);
2528         if (nested) {
2529                 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2530                 cpu_in_guest = &nested->cpu_in_guest;
2531         } else {
2532                 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2533                 cpu_in_guest = &kvm->arch.cpu_in_guest;
2534         }
2535         /*
2536          * Make sure setting of bit in need_tlb_flush precedes
2537          * testing of cpu_in_guest bits.  The matching barrier on
2538          * the other side is the first smp_mb() in kvmppc_run_core().
2539          */
2540         smp_mb();
2541         for (i = 0; i < threads_per_core; ++i)
2542                 if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2543                         smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2544 }
2545
2546 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2547 {
2548         struct kvm_nested_guest *nested = vcpu->arch.nested;
2549         struct kvm *kvm = vcpu->kvm;
2550         int prev_cpu;
2551
2552         if (!cpu_has_feature(CPU_FTR_HVMODE))
2553                 return;
2554
2555         if (nested)
2556                 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2557         else
2558                 prev_cpu = vcpu->arch.prev_cpu;
2559
2560         /*
2561          * With radix, the guest can do TLB invalidations itself,
2562          * and it could choose to use the local form (tlbiel) if
2563          * it is invalidating a translation that has only ever been
2564          * used on one vcpu.  However, that doesn't mean it has
2565          * only ever been used on one physical cpu, since vcpus
2566          * can move around between pcpus.  To cope with this, when
2567          * a vcpu moves from one pcpu to another, we need to tell
2568          * any vcpus running on the same core as this vcpu previously
2569          * ran to flush the TLB.  The TLB is shared between threads,
2570          * so we use a single bit in .need_tlb_flush for all 4 threads.
2571          */
2572         if (prev_cpu != pcpu) {
2573                 if (prev_cpu >= 0 &&
2574                     cpu_first_thread_sibling(prev_cpu) !=
2575                     cpu_first_thread_sibling(pcpu))
2576                         radix_flush_cpu(kvm, prev_cpu, vcpu);
2577                 if (nested)
2578                         nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2579                 else
2580                         vcpu->arch.prev_cpu = pcpu;
2581         }
2582 }
2583
2584 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2585 {
2586         int cpu;
2587         struct paca_struct *tpaca;
2588         struct kvm *kvm = vc->kvm;
2589
2590         cpu = vc->pcpu;
2591         if (vcpu) {
2592                 if (vcpu->arch.timer_running) {
2593                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2594                         vcpu->arch.timer_running = 0;
2595                 }
2596                 cpu += vcpu->arch.ptid;
2597                 vcpu->cpu = vc->pcpu;
2598                 vcpu->arch.thread_cpu = cpu;
2599                 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2600         }
2601         tpaca = paca_ptrs[cpu];
2602         tpaca->kvm_hstate.kvm_vcpu = vcpu;
2603         tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2604         tpaca->kvm_hstate.fake_suspend = 0;
2605         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2606         smp_wmb();
2607         tpaca->kvm_hstate.kvm_vcore = vc;
2608         if (cpu != smp_processor_id())
2609                 kvmppc_ipi_thread(cpu);
2610 }
2611
2612 static void kvmppc_wait_for_nap(int n_threads)
2613 {
2614         int cpu = smp_processor_id();
2615         int i, loops;
2616
2617         if (n_threads <= 1)
2618                 return;
2619         for (loops = 0; loops < 1000000; ++loops) {
2620                 /*
2621                  * Check if all threads are finished.
2622                  * We set the vcore pointer when starting a thread
2623                  * and the thread clears it when finished, so we look
2624                  * for any threads that still have a non-NULL vcore ptr.
2625                  */
2626                 for (i = 1; i < n_threads; ++i)
2627                         if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2628                                 break;
2629                 if (i == n_threads) {
2630                         HMT_medium();
2631                         return;
2632                 }
2633                 HMT_low();
2634         }
2635         HMT_medium();
2636         for (i = 1; i < n_threads; ++i)
2637                 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2638                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2639 }
2640
2641 /*
2642  * Check that we are on thread 0 and that any other threads in
2643  * this core are off-line.  Then grab the threads so they can't
2644  * enter the kernel.
2645  */
2646 static int on_primary_thread(void)
2647 {
2648         int cpu = smp_processor_id();
2649         int thr;
2650
2651         /* Are we on a primary subcore? */
2652         if (cpu_thread_in_subcore(cpu))
2653                 return 0;
2654
2655         thr = 0;
2656         while (++thr < threads_per_subcore)
2657                 if (cpu_online(cpu + thr))
2658                         return 0;
2659
2660         /* Grab all hw threads so they can't go into the kernel */
2661         for (thr = 1; thr < threads_per_subcore; ++thr) {
2662                 if (kvmppc_grab_hwthread(cpu + thr)) {
2663                         /* Couldn't grab one; let the others go */
2664                         do {
2665                                 kvmppc_release_hwthread(cpu + thr);
2666                         } while (--thr > 0);
2667                         return 0;
2668                 }
2669         }
2670         return 1;
2671 }
2672
2673 /*
2674  * A list of virtual cores for each physical CPU.
2675  * These are vcores that could run but their runner VCPU tasks are
2676  * (or may be) preempted.
2677  */
2678 struct preempted_vcore_list {
2679         struct list_head        list;
2680         spinlock_t              lock;
2681 };
2682
2683 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2684
2685 static void init_vcore_lists(void)
2686 {
2687         int cpu;
2688
2689         for_each_possible_cpu(cpu) {
2690                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2691                 spin_lock_init(&lp->lock);
2692                 INIT_LIST_HEAD(&lp->list);
2693         }
2694 }
2695
2696 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2697 {
2698         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2699
2700         vc->vcore_state = VCORE_PREEMPT;
2701         vc->pcpu = smp_processor_id();
2702         if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2703                 spin_lock(&lp->lock);
2704                 list_add_tail(&vc->preempt_list, &lp->list);
2705                 spin_unlock(&lp->lock);
2706         }
2707
2708         /* Start accumulating stolen time */
2709         kvmppc_core_start_stolen(vc);
2710 }
2711
2712 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2713 {
2714         struct preempted_vcore_list *lp;
2715
2716         kvmppc_core_end_stolen(vc);
2717         if (!list_empty(&vc->preempt_list)) {
2718                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2719                 spin_lock(&lp->lock);
2720                 list_del_init(&vc->preempt_list);
2721                 spin_unlock(&lp->lock);
2722         }
2723         vc->vcore_state = VCORE_INACTIVE;
2724 }
2725
2726 /*
2727  * This stores information about the virtual cores currently
2728  * assigned to a physical core.
2729  */
2730 struct core_info {
2731         int             n_subcores;
2732         int             max_subcore_threads;
2733         int             total_threads;
2734         int             subcore_threads[MAX_SUBCORES];
2735         struct kvmppc_vcore *vc[MAX_SUBCORES];
2736 };
2737
2738 /*
2739  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2740  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2741  */
2742 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2743
2744 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2745 {
2746         memset(cip, 0, sizeof(*cip));
2747         cip->n_subcores = 1;
2748         cip->max_subcore_threads = vc->num_threads;
2749         cip->total_threads = vc->num_threads;
2750         cip->subcore_threads[0] = vc->num_threads;
2751         cip->vc[0] = vc;
2752 }
2753
2754 static bool subcore_config_ok(int n_subcores, int n_threads)
2755 {
2756         /*
2757          * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2758          * split-core mode, with one thread per subcore.
2759          */
2760         if (cpu_has_feature(CPU_FTR_ARCH_300))
2761                 return n_subcores <= 4 && n_threads == 1;
2762
2763         /* On POWER8, can only dynamically split if unsplit to begin with */
2764         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2765                 return false;
2766         if (n_subcores > MAX_SUBCORES)
2767                 return false;
2768         if (n_subcores > 1) {
2769                 if (!(dynamic_mt_modes & 2))
2770                         n_subcores = 4;
2771                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2772                         return false;
2773         }
2774
2775         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2776 }
2777
2778 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2779 {
2780         vc->entry_exit_map = 0;
2781         vc->in_guest = 0;
2782         vc->napping_threads = 0;
2783         vc->conferring_threads = 0;
2784         vc->tb_offset_applied = 0;
2785 }
2786
2787 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2788 {
2789         int n_threads = vc->num_threads;
2790         int sub;
2791
2792         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2793                 return false;
2794
2795         /* In one_vm_per_core mode, require all vcores to be from the same vm */
2796         if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2797                 return false;
2798
2799         /* Some POWER9 chips require all threads to be in the same MMU mode */
2800         if (no_mixing_hpt_and_radix &&
2801             kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2802                 return false;
2803
2804         if (n_threads < cip->max_subcore_threads)
2805                 n_threads = cip->max_subcore_threads;
2806         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2807                 return false;
2808         cip->max_subcore_threads = n_threads;
2809
2810         sub = cip->n_subcores;
2811         ++cip->n_subcores;
2812         cip->total_threads += vc->num_threads;
2813         cip->subcore_threads[sub] = vc->num_threads;
2814         cip->vc[sub] = vc;
2815         init_vcore_to_run(vc);
2816         list_del_init(&vc->preempt_list);
2817
2818         return true;
2819 }
2820
2821 /*
2822  * Work out whether it is possible to piggyback the execution of
2823  * vcore *pvc onto the execution of the other vcores described in *cip.
2824  */
2825 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2826                           int target_threads)
2827 {
2828         if (cip->total_threads + pvc->num_threads > target_threads)
2829                 return false;
2830
2831         return can_dynamic_split(pvc, cip);
2832 }
2833
2834 static void prepare_threads(struct kvmppc_vcore *vc)
2835 {
2836         int i;
2837         struct kvm_vcpu *vcpu;
2838
2839         for_each_runnable_thread(i, vcpu, vc) {
2840                 if (signal_pending(vcpu->arch.run_task))
2841                         vcpu->arch.ret = -EINTR;
2842                 else if (vcpu->arch.vpa.update_pending ||
2843                          vcpu->arch.slb_shadow.update_pending ||
2844                          vcpu->arch.dtl.update_pending)
2845                         vcpu->arch.ret = RESUME_GUEST;
2846                 else
2847                         continue;
2848                 kvmppc_remove_runnable(vc, vcpu);
2849                 wake_up(&vcpu->arch.cpu_run);
2850         }
2851 }
2852
2853 static void collect_piggybacks(struct core_info *cip, int target_threads)
2854 {
2855         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2856         struct kvmppc_vcore *pvc, *vcnext;
2857
2858         spin_lock(&lp->lock);
2859         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2860                 if (!spin_trylock(&pvc->lock))
2861                         continue;
2862                 prepare_threads(pvc);
2863                 if (!pvc->n_runnable) {
2864                         list_del_init(&pvc->preempt_list);
2865                         if (pvc->runner == NULL) {
2866                                 pvc->vcore_state = VCORE_INACTIVE;
2867                                 kvmppc_core_end_stolen(pvc);
2868                         }
2869                         spin_unlock(&pvc->lock);
2870                         continue;
2871                 }
2872                 if (!can_piggyback(pvc, cip, target_threads)) {
2873                         spin_unlock(&pvc->lock);
2874                         continue;
2875                 }
2876                 kvmppc_core_end_stolen(pvc);
2877                 pvc->vcore_state = VCORE_PIGGYBACK;
2878                 if (cip->total_threads >= target_threads)
2879                         break;
2880         }
2881         spin_unlock(&lp->lock);
2882 }
2883
2884 static bool recheck_signals(struct core_info *cip)
2885 {
2886         int sub, i;
2887         struct kvm_vcpu *vcpu;
2888
2889         for (sub = 0; sub < cip->n_subcores; ++sub)
2890                 for_each_runnable_thread(i, vcpu, cip->vc[sub])
2891                         if (signal_pending(vcpu->arch.run_task))
2892                                 return true;
2893         return false;
2894 }
2895
2896 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2897 {
2898         int still_running = 0, i;
2899         u64 now;
2900         long ret;
2901         struct kvm_vcpu *vcpu;
2902
2903         spin_lock(&vc->lock);
2904         now = get_tb();
2905         for_each_runnable_thread(i, vcpu, vc) {
2906                 /*
2907                  * It's safe to unlock the vcore in the loop here, because
2908                  * for_each_runnable_thread() is safe against removal of
2909                  * the vcpu, and the vcore state is VCORE_EXITING here,
2910                  * so any vcpus becoming runnable will have their arch.trap
2911                  * set to zero and can't actually run in the guest.
2912                  */
2913                 spin_unlock(&vc->lock);
2914                 /* cancel pending dec exception if dec is positive */
2915                 if (now < vcpu->arch.dec_expires &&
2916                     kvmppc_core_pending_dec(vcpu))
2917                         kvmppc_core_dequeue_dec(vcpu);
2918
2919                 trace_kvm_guest_exit(vcpu);
2920
2921                 ret = RESUME_GUEST;
2922                 if (vcpu->arch.trap)
2923                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2924                                                     vcpu->arch.run_task);
2925
2926                 vcpu->arch.ret = ret;
2927                 vcpu->arch.trap = 0;
2928
2929                 spin_lock(&vc->lock);
2930                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2931                         if (vcpu->arch.pending_exceptions)
2932                                 kvmppc_core_prepare_to_enter(vcpu);
2933                         if (vcpu->arch.ceded)
2934                                 kvmppc_set_timer(vcpu);
2935                         else
2936                                 ++still_running;
2937                 } else {
2938                         kvmppc_remove_runnable(vc, vcpu);
2939                         wake_up(&vcpu->arch.cpu_run);
2940                 }
2941         }
2942         if (!is_master) {
2943                 if (still_running > 0) {
2944                         kvmppc_vcore_preempt(vc);
2945                 } else if (vc->runner) {
2946                         vc->vcore_state = VCORE_PREEMPT;
2947                         kvmppc_core_start_stolen(vc);
2948                 } else {
2949                         vc->vcore_state = VCORE_INACTIVE;
2950                 }
2951                 if (vc->n_runnable > 0 && vc->runner == NULL) {
2952                         /* make sure there's a candidate runner awake */
2953                         i = -1;
2954                         vcpu = next_runnable_thread(vc, &i);
2955                         wake_up(&vcpu->arch.cpu_run);
2956                 }
2957         }
2958         spin_unlock(&vc->lock);
2959 }
2960
2961 /*
2962  * Clear core from the list of active host cores as we are about to
2963  * enter the guest. Only do this if it is the primary thread of the
2964  * core (not if a subcore) that is entering the guest.
2965  */
2966 static inline int kvmppc_clear_host_core(unsigned int cpu)
2967 {
2968         int core;
2969
2970         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2971                 return 0;
2972         /*
2973          * Memory barrier can be omitted here as we will do a smp_wmb()
2974          * later in kvmppc_start_thread and we need ensure that state is
2975          * visible to other CPUs only after we enter guest.
2976          */
2977         core = cpu >> threads_shift;
2978         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2979         return 0;
2980 }
2981
2982 /*
2983  * Advertise this core as an active host core since we exited the guest
2984  * Only need to do this if it is the primary thread of the core that is
2985  * exiting.
2986  */
2987 static inline int kvmppc_set_host_core(unsigned int cpu)
2988 {
2989         int core;
2990
2991         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2992                 return 0;
2993
2994         /*
2995          * Memory barrier can be omitted here because we do a spin_unlock
2996          * immediately after this which provides the memory barrier.
2997          */
2998         core = cpu >> threads_shift;
2999         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3000         return 0;
3001 }
3002
3003 static void set_irq_happened(int trap)
3004 {
3005         switch (trap) {
3006         case BOOK3S_INTERRUPT_EXTERNAL:
3007                 local_paca->irq_happened |= PACA_IRQ_EE;
3008                 break;
3009         case BOOK3S_INTERRUPT_H_DOORBELL:
3010                 local_paca->irq_happened |= PACA_IRQ_DBELL;
3011                 break;
3012         case BOOK3S_INTERRUPT_HMI:
3013                 local_paca->irq_happened |= PACA_IRQ_HMI;
3014                 break;
3015         case BOOK3S_INTERRUPT_SYSTEM_RESET:
3016                 replay_system_reset();
3017                 break;
3018         }
3019 }
3020
3021 /*
3022  * Run a set of guest threads on a physical core.
3023  * Called with vc->lock held.
3024  */
3025 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3026 {
3027         struct kvm_vcpu *vcpu;
3028         int i;
3029         int srcu_idx;
3030         struct core_info core_info;
3031         struct kvmppc_vcore *pvc;
3032         struct kvm_split_mode split_info, *sip;
3033         int split, subcore_size, active;
3034         int sub;
3035         bool thr0_done;
3036         unsigned long cmd_bit, stat_bit;
3037         int pcpu, thr;
3038         int target_threads;
3039         int controlled_threads;
3040         int trap;
3041         bool is_power8;
3042         bool hpt_on_radix;
3043
3044         /*
3045          * Remove from the list any threads that have a signal pending
3046          * or need a VPA update done
3047          */
3048         prepare_threads(vc);
3049
3050         /* if the runner is no longer runnable, let the caller pick a new one */
3051         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3052                 return;
3053
3054         /*
3055          * Initialize *vc.
3056          */
3057         init_vcore_to_run(vc);
3058         vc->preempt_tb = TB_NIL;
3059
3060         /*
3061          * Number of threads that we will be controlling: the same as
3062          * the number of threads per subcore, except on POWER9,
3063          * where it's 1 because the threads are (mostly) independent.
3064          */
3065         controlled_threads = threads_per_vcore(vc->kvm);
3066
3067         /*
3068          * Make sure we are running on primary threads, and that secondary
3069          * threads are offline.  Also check if the number of threads in this
3070          * guest are greater than the current system threads per guest.
3071          * On POWER9, we need to be not in independent-threads mode if
3072          * this is a HPT guest on a radix host machine where the
3073          * CPU threads may not be in different MMU modes.
3074          */
3075         hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3076                 !kvm_is_radix(vc->kvm);
3077         if (((controlled_threads > 1) &&
3078              ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3079             (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3080                 for_each_runnable_thread(i, vcpu, vc) {
3081                         vcpu->arch.ret = -EBUSY;
3082                         kvmppc_remove_runnable(vc, vcpu);
3083                         wake_up(&vcpu->arch.cpu_run);
3084                 }
3085                 goto out;
3086         }
3087
3088         /*
3089          * See if we could run any other vcores on the physical core
3090          * along with this one.
3091          */
3092         init_core_info(&core_info, vc);
3093         pcpu = smp_processor_id();
3094         target_threads = controlled_threads;
3095         if (target_smt_mode && target_smt_mode < target_threads)
3096                 target_threads = target_smt_mode;
3097         if (vc->num_threads < target_threads)
3098                 collect_piggybacks(&core_info, target_threads);
3099
3100         /*
3101          * On radix, arrange for TLB flushing if necessary.
3102          * This has to be done before disabling interrupts since
3103          * it uses smp_call_function().
3104          */
3105         pcpu = smp_processor_id();
3106         if (kvm_is_radix(vc->kvm)) {
3107                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3108                         for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3109                                 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3110         }
3111
3112         /*
3113          * Hard-disable interrupts, and check resched flag and signals.
3114          * If we need to reschedule or deliver a signal, clean up
3115          * and return without going into the guest(s).
3116          * If the mmu_ready flag has been cleared, don't go into the
3117          * guest because that means a HPT resize operation is in progress.
3118          */
3119         local_irq_disable();
3120         hard_irq_disable();
3121         if (lazy_irq_pending() || need_resched() ||
3122             recheck_signals(&core_info) || !vc->kvm->arch.mmu_ready) {
3123                 local_irq_enable();
3124                 vc->vcore_state = VCORE_INACTIVE;
3125                 /* Unlock all except the primary vcore */
3126                 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3127                         pvc = core_info.vc[sub];
3128                         /* Put back on to the preempted vcores list */
3129                         kvmppc_vcore_preempt(pvc);
3130                         spin_unlock(&pvc->lock);
3131                 }
3132                 for (i = 0; i < controlled_threads; ++i)
3133                         kvmppc_release_hwthread(pcpu + i);
3134                 return;
3135         }
3136
3137         kvmppc_clear_host_core(pcpu);
3138
3139         /* Decide on micro-threading (split-core) mode */
3140         subcore_size = threads_per_subcore;
3141         cmd_bit = stat_bit = 0;
3142         split = core_info.n_subcores;
3143         sip = NULL;
3144         is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3145                 && !cpu_has_feature(CPU_FTR_ARCH_300);
3146
3147         if (split > 1 || hpt_on_radix) {
3148                 sip = &split_info;
3149                 memset(&split_info, 0, sizeof(split_info));
3150                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3151                         split_info.vc[sub] = core_info.vc[sub];
3152
3153                 if (is_power8) {
3154                         if (split == 2 && (dynamic_mt_modes & 2)) {
3155                                 cmd_bit = HID0_POWER8_1TO2LPAR;
3156                                 stat_bit = HID0_POWER8_2LPARMODE;
3157                         } else {
3158                                 split = 4;
3159                                 cmd_bit = HID0_POWER8_1TO4LPAR;
3160                                 stat_bit = HID0_POWER8_4LPARMODE;
3161                         }
3162                         subcore_size = MAX_SMT_THREADS / split;
3163                         split_info.rpr = mfspr(SPRN_RPR);
3164                         split_info.pmmar = mfspr(SPRN_PMMAR);
3165                         split_info.ldbar = mfspr(SPRN_LDBAR);
3166                         split_info.subcore_size = subcore_size;
3167                 } else {
3168                         split_info.subcore_size = 1;
3169                         if (hpt_on_radix) {
3170                                 /* Use the split_info for LPCR/LPIDR changes */
3171                                 split_info.lpcr_req = vc->lpcr;
3172                                 split_info.lpidr_req = vc->kvm->arch.lpid;
3173                                 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3174                                 split_info.do_set = 1;
3175                         }
3176                 }
3177
3178                 /* order writes to split_info before kvm_split_mode pointer */
3179                 smp_wmb();
3180         }
3181
3182         for (thr = 0; thr < controlled_threads; ++thr) {
3183                 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3184
3185                 paca->kvm_hstate.tid = thr;
3186                 paca->kvm_hstate.napping = 0;
3187                 paca->kvm_hstate.kvm_split_mode = sip;
3188         }
3189
3190         /* Initiate micro-threading (split-core) on POWER8 if required */
3191         if (cmd_bit) {
3192                 unsigned long hid0 = mfspr(SPRN_HID0);
3193
3194                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3195                 mb();
3196                 mtspr(SPRN_HID0, hid0);
3197                 isync();
3198                 for (;;) {
3199                         hid0 = mfspr(SPRN_HID0);
3200                         if (hid0 & stat_bit)
3201                                 break;
3202                         cpu_relax();
3203                 }
3204         }
3205
3206         /*
3207          * On POWER8, set RWMR register.
3208          * Since it only affects PURR and SPURR, it doesn't affect
3209          * the host, so we don't save/restore the host value.
3210          */
3211         if (is_power8) {
3212                 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3213                 int n_online = atomic_read(&vc->online_count);
3214
3215                 /*
3216                  * Use the 8-thread value if we're doing split-core
3217                  * or if the vcore's online count looks bogus.
3218                  */
3219                 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3220                     n_online >= 1 && n_online <= MAX_SMT_THREADS)
3221                         rwmr_val = p8_rwmr_values[n_online];
3222                 mtspr(SPRN_RWMR, rwmr_val);
3223         }
3224
3225         /* Start all the threads */
3226         active = 0;
3227         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3228                 thr = is_power8 ? subcore_thread_map[sub] : sub;
3229                 thr0_done = false;
3230                 active |= 1 << thr;
3231                 pvc = core_info.vc[sub];
3232                 pvc->pcpu = pcpu + thr;
3233                 for_each_runnable_thread(i, vcpu, pvc) {
3234                         kvmppc_start_thread(vcpu, pvc);
3235                         kvmppc_create_dtl_entry(vcpu, pvc);
3236                         trace_kvm_guest_enter(vcpu);
3237                         if (!vcpu->arch.ptid)
3238                                 thr0_done = true;
3239                         active |= 1 << (thr + vcpu->arch.ptid);
3240                 }
3241                 /*
3242                  * We need to start the first thread of each subcore
3243                  * even if it doesn't have a vcpu.
3244                  */
3245                 if (!thr0_done)
3246                         kvmppc_start_thread(NULL, pvc);
3247         }
3248
3249         /*
3250          * Ensure that split_info.do_nap is set after setting
3251          * the vcore pointer in the PACA of the secondaries.
3252          */
3253         smp_mb();
3254
3255         /*
3256          * When doing micro-threading, poke the inactive threads as well.
3257          * This gets them to the nap instruction after kvm_do_nap,
3258          * which reduces the time taken to unsplit later.
3259          * For POWER9 HPT guest on radix host, we need all the secondary
3260          * threads woken up so they can do the LPCR/LPIDR change.
3261          */
3262         if (cmd_bit || hpt_on_radix) {
3263                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
3264                 for (thr = 1; thr < threads_per_subcore; ++thr)
3265                         if (!(active & (1 << thr)))
3266                                 kvmppc_ipi_thread(pcpu + thr);
3267         }
3268
3269         vc->vcore_state = VCORE_RUNNING;
3270         preempt_disable();
3271
3272         trace_kvmppc_run_core(vc, 0);
3273
3274         for (sub = 0; sub < core_info.n_subcores; ++sub)
3275                 spin_unlock(&core_info.vc[sub]->lock);
3276
3277         guest_enter_irqoff();
3278
3279         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3280
3281         this_cpu_disable_ftrace();
3282
3283         /*
3284          * Interrupts will be enabled once we get into the guest,
3285          * so tell lockdep that we're about to enable interrupts.
3286          */
3287         trace_hardirqs_on();
3288
3289         trap = __kvmppc_vcore_entry();
3290
3291         trace_hardirqs_off();
3292
3293         this_cpu_enable_ftrace();
3294
3295         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3296
3297         set_irq_happened(trap);
3298
3299         spin_lock(&vc->lock);
3300         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3301         vc->vcore_state = VCORE_EXITING;
3302
3303         /* wait for secondary threads to finish writing their state to memory */
3304         kvmppc_wait_for_nap(controlled_threads);
3305
3306         /* Return to whole-core mode if we split the core earlier */
3307         if (cmd_bit) {
3308                 unsigned long hid0 = mfspr(SPRN_HID0);
3309                 unsigned long loops = 0;
3310
3311                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3312                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3313                 mb();
3314                 mtspr(SPRN_HID0, hid0);
3315                 isync();
3316                 for (;;) {
3317                         hid0 = mfspr(SPRN_HID0);
3318                         if (!(hid0 & stat_bit))
3319                                 break;
3320                         cpu_relax();
3321                         ++loops;
3322                 }
3323         } else if (hpt_on_radix) {
3324                 /* Wait for all threads to have seen final sync */
3325                 for (thr = 1; thr < controlled_threads; ++thr) {
3326                         struct paca_struct *paca = paca_ptrs[pcpu + thr];
3327
3328                         while (paca->kvm_hstate.kvm_split_mode) {
3329                                 HMT_low();
3330                                 barrier();
3331                         }
3332                         HMT_medium();
3333                 }
3334         }
3335         split_info.do_nap = 0;
3336
3337         kvmppc_set_host_core(pcpu);
3338
3339         local_irq_enable();
3340         guest_exit();
3341
3342         /* Let secondaries go back to the offline loop */
3343         for (i = 0; i < controlled_threads; ++i) {
3344                 kvmppc_release_hwthread(pcpu + i);
3345                 if (sip && sip->napped[i])
3346                         kvmppc_ipi_thread(pcpu + i);
3347                 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3348         }
3349
3350         spin_unlock(&vc->lock);
3351
3352         /* make sure updates to secondary vcpu structs are visible now */
3353         smp_mb();
3354
3355         preempt_enable();
3356
3357         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3358                 pvc = core_info.vc[sub];
3359                 post_guest_process(pvc, pvc == vc);
3360         }
3361
3362         spin_lock(&vc->lock);
3363
3364  out:
3365         vc->vcore_state = VCORE_INACTIVE;
3366         trace_kvmppc_run_core(vc, 1);
3367 }
3368
3369 /*
3370  * Load up hypervisor-mode registers on P9.
3371  */
3372 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3373                                      unsigned long lpcr)
3374 {
3375         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3376         s64 hdec;
3377         u64 tb, purr, spurr;
3378         int trap;
3379         unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3380         unsigned long host_ciabr = mfspr(SPRN_CIABR);
3381         unsigned long host_dawr = mfspr(SPRN_DAWR);
3382         unsigned long host_dawrx = mfspr(SPRN_DAWRX);
3383         unsigned long host_psscr = mfspr(SPRN_PSSCR);
3384         unsigned long host_pidr = mfspr(SPRN_PID);
3385
3386         hdec = time_limit - mftb();
3387         if (hdec < 0)
3388                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3389         mtspr(SPRN_HDEC, hdec);
3390
3391         if (vc->tb_offset) {
3392                 u64 new_tb = mftb() + vc->tb_offset;
3393                 mtspr(SPRN_TBU40, new_tb);
3394                 tb = mftb();
3395                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3396                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3397                 vc->tb_offset_applied = vc->tb_offset;
3398         }
3399
3400         if (vc->pcr)
3401                 mtspr(SPRN_PCR, vc->pcr);
3402         mtspr(SPRN_DPDES, vc->dpdes);
3403         mtspr(SPRN_VTB, vc->vtb);
3404
3405         local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3406         local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3407         mtspr(SPRN_PURR, vcpu->arch.purr);
3408         mtspr(SPRN_SPURR, vcpu->arch.spurr);
3409
3410         if (dawr_enabled()) {
3411                 mtspr(SPRN_DAWR, vcpu->arch.dawr);
3412                 mtspr(SPRN_DAWRX, vcpu->arch.dawrx);
3413         }
3414         mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3415         mtspr(SPRN_IC, vcpu->arch.ic);
3416         mtspr(SPRN_PID, vcpu->arch.pid);
3417
3418         mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3419               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3420
3421         mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3422
3423         mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3424         mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3425         mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3426         mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3427
3428         mtspr(SPRN_AMOR, ~0UL);
3429
3430         mtspr(SPRN_LPCR, lpcr);
3431         isync();
3432
3433         kvmppc_xive_push_vcpu(vcpu);
3434
3435         mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3436         mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3437
3438         trap = __kvmhv_vcpu_entry_p9(vcpu);
3439
3440         /* Advance host PURR/SPURR by the amount used by guest */
3441         purr = mfspr(SPRN_PURR);
3442         spurr = mfspr(SPRN_SPURR);
3443         mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3444               purr - vcpu->arch.purr);
3445         mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3446               spurr - vcpu->arch.spurr);
3447         vcpu->arch.purr = purr;
3448         vcpu->arch.spurr = spurr;
3449
3450         vcpu->arch.ic = mfspr(SPRN_IC);
3451         vcpu->arch.pid = mfspr(SPRN_PID);
3452         vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3453
3454         vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3455         vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3456         vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3457         vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3458
3459         /* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3460         mtspr(SPRN_PSSCR, host_psscr |
3461               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3462         mtspr(SPRN_HFSCR, host_hfscr);
3463         mtspr(SPRN_CIABR, host_ciabr);
3464         mtspr(SPRN_DAWR, host_dawr);
3465         mtspr(SPRN_DAWRX, host_dawrx);
3466         mtspr(SPRN_PID, host_pidr);
3467
3468         /*
3469          * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3470          * case we interrupted the guest between a tlbie and a ptesync.
3471          */
3472         asm volatile("eieio; tlbsync; ptesync");
3473
3474         mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid);    /* restore host LPID */
3475         isync();
3476
3477         vc->dpdes = mfspr(SPRN_DPDES);
3478         vc->vtb = mfspr(SPRN_VTB);
3479         mtspr(SPRN_DPDES, 0);
3480         if (vc->pcr)
3481                 mtspr(SPRN_PCR, 0);
3482
3483         if (vc->tb_offset_applied) {
3484                 u64 new_tb = mftb() - vc->tb_offset_applied;
3485                 mtspr(SPRN_TBU40, new_tb);
3486                 tb = mftb();
3487                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3488                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3489                 vc->tb_offset_applied = 0;
3490         }
3491
3492         mtspr(SPRN_HDEC, 0x7fffffff);
3493         mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3494
3495         return trap;
3496 }
3497
3498 /*
3499  * Virtual-mode guest entry for POWER9 and later when the host and
3500  * guest are both using the radix MMU.  The LPIDR has already been set.
3501  */
3502 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3503                          unsigned long lpcr)
3504 {
3505         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3506         unsigned long host_dscr = mfspr(SPRN_DSCR);
3507         unsigned long host_tidr = mfspr(SPRN_TIDR);
3508         unsigned long host_iamr = mfspr(SPRN_IAMR);
3509         unsigned long host_amr = mfspr(SPRN_AMR);
3510         s64 dec;
3511         u64 tb;
3512         int trap, save_pmu;
3513
3514         dec = mfspr(SPRN_DEC);
3515         tb = mftb();
3516         if (dec < 512)
3517                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3518         local_paca->kvm_hstate.dec_expires = dec + tb;
3519         if (local_paca->kvm_hstate.dec_expires < time_limit)
3520                 time_limit = local_paca->kvm_hstate.dec_expires;
3521
3522         vcpu->arch.ceded = 0;
3523
3524         kvmhv_save_host_pmu();          /* saves it to PACA kvm_hstate */
3525
3526         kvmppc_subcore_enter_guest();
3527
3528         vc->entry_exit_map = 1;
3529         vc->in_guest = 1;
3530
3531         if (vcpu->arch.vpa.pinned_addr) {
3532                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3533                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3534                 lp->yield_count = cpu_to_be32(yield_count);
3535                 vcpu->arch.vpa.dirty = 1;
3536         }
3537
3538         if (cpu_has_feature(CPU_FTR_TM) ||
3539             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3540                 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3541
3542         kvmhv_load_guest_pmu(vcpu);
3543
3544         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3545         load_fp_state(&vcpu->arch.fp);
3546 #ifdef CONFIG_ALTIVEC
3547         load_vr_state(&vcpu->arch.vr);
3548 #endif
3549         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3550
3551         mtspr(SPRN_DSCR, vcpu->arch.dscr);
3552         mtspr(SPRN_IAMR, vcpu->arch.iamr);
3553         mtspr(SPRN_PSPB, vcpu->arch.pspb);
3554         mtspr(SPRN_FSCR, vcpu->arch.fscr);
3555         mtspr(SPRN_TAR, vcpu->arch.tar);
3556         mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3557         mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3558         mtspr(SPRN_BESCR, vcpu->arch.bescr);
3559         mtspr(SPRN_WORT, vcpu->arch.wort);
3560         mtspr(SPRN_TIDR, vcpu->arch.tid);
3561         mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3562         mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3563         mtspr(SPRN_AMR, vcpu->arch.amr);
3564         mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3565
3566         if (!(vcpu->arch.ctrl & 1))
3567                 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3568
3569         mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3570
3571         if (kvmhv_on_pseries()) {
3572                 /* call our hypervisor to load up HV regs and go */
3573                 struct hv_guest_state hvregs;
3574
3575                 kvmhv_save_hv_regs(vcpu, &hvregs);
3576                 hvregs.lpcr = lpcr;
3577                 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3578                 hvregs.version = HV_GUEST_STATE_VERSION;
3579                 if (vcpu->arch.nested) {
3580                         hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3581                         hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3582                 } else {
3583                         hvregs.lpid = vcpu->kvm->arch.lpid;
3584                         hvregs.vcpu_token = vcpu->vcpu_id;
3585                 }
3586                 hvregs.hdec_expiry = time_limit;
3587                 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3588                                           __pa(&vcpu->arch.regs));
3589                 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3590                 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3591                 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3592                 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3593
3594                 /* H_CEDE has to be handled now, not later */
3595                 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3596                     kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3597                         kvmppc_nested_cede(vcpu);
3598                         trap = 0;
3599                 }
3600         } else {
3601                 trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3602         }
3603
3604         vcpu->arch.slb_max = 0;
3605         dec = mfspr(SPRN_DEC);
3606         tb = mftb();
3607         vcpu->arch.dec_expires = dec + tb;
3608         vcpu->cpu = -1;
3609         vcpu->arch.thread_cpu = -1;
3610         vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3611
3612         vcpu->arch.iamr = mfspr(SPRN_IAMR);
3613         vcpu->arch.pspb = mfspr(SPRN_PSPB);
3614         vcpu->arch.fscr = mfspr(SPRN_FSCR);
3615         vcpu->arch.tar = mfspr(SPRN_TAR);
3616         vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3617         vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3618         vcpu->arch.bescr = mfspr(SPRN_BESCR);
3619         vcpu->arch.wort = mfspr(SPRN_WORT);
3620         vcpu->arch.tid = mfspr(SPRN_TIDR);
3621         vcpu->arch.amr = mfspr(SPRN_AMR);
3622         vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3623         vcpu->arch.dscr = mfspr(SPRN_DSCR);
3624
3625         mtspr(SPRN_PSPB, 0);
3626         mtspr(SPRN_WORT, 0);
3627         mtspr(SPRN_UAMOR, 0);
3628         mtspr(SPRN_DSCR, host_dscr);
3629         mtspr(SPRN_TIDR, host_tidr);
3630         mtspr(SPRN_IAMR, host_iamr);
3631         mtspr(SPRN_PSPB, 0);
3632
3633         if (host_amr != vcpu->arch.amr)
3634                 mtspr(SPRN_AMR, host_amr);
3635
3636         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3637         store_fp_state(&vcpu->arch.fp);
3638 #ifdef CONFIG_ALTIVEC
3639         store_vr_state(&vcpu->arch.vr);
3640 #endif
3641         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3642
3643         if (cpu_has_feature(CPU_FTR_TM) ||
3644             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3645                 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3646
3647         save_pmu = 1;
3648         if (vcpu->arch.vpa.pinned_addr) {
3649                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3650                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3651                 lp->yield_count = cpu_to_be32(yield_count);
3652                 vcpu->arch.vpa.dirty = 1;
3653                 save_pmu = lp->pmcregs_in_use;
3654         }
3655
3656         kvmhv_save_guest_pmu(vcpu, save_pmu);
3657
3658         vc->entry_exit_map = 0x101;
3659         vc->in_guest = 0;
3660
3661         mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3662         mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3663
3664         kvmhv_load_host_pmu();
3665
3666         kvmppc_subcore_exit_guest();
3667
3668         return trap;
3669 }
3670
3671 /*
3672  * Wait for some other vcpu thread to execute us, and
3673  * wake us up when we need to handle something in the host.
3674  */
3675 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3676                                  struct kvm_vcpu *vcpu, int wait_state)
3677 {
3678         DEFINE_WAIT(wait);
3679
3680         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3681         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3682                 spin_unlock(&vc->lock);
3683                 schedule();
3684                 spin_lock(&vc->lock);
3685         }
3686         finish_wait(&vcpu->arch.cpu_run, &wait);
3687 }
3688
3689 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3690 {
3691         if (!halt_poll_ns_grow)
3692                 return;
3693
3694         vc->halt_poll_ns *= halt_poll_ns_grow;
3695         if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3696                 vc->halt_poll_ns = halt_poll_ns_grow_start;
3697 }
3698
3699 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3700 {
3701         if (halt_poll_ns_shrink == 0)
3702                 vc->halt_poll_ns = 0;
3703         else
3704                 vc->halt_poll_ns /= halt_poll_ns_shrink;
3705 }
3706
3707 #ifdef CONFIG_KVM_XICS
3708 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3709 {
3710         if (!xics_on_xive())
3711                 return false;
3712         return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3713                 vcpu->arch.xive_saved_state.cppr;
3714 }
3715 #else
3716 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3717 {
3718         return false;
3719 }
3720 #endif /* CONFIG_KVM_XICS */
3721
3722 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3723 {
3724         if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3725             kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3726                 return true;
3727
3728         return false;
3729 }
3730
3731 /*
3732  * Check to see if any of the runnable vcpus on the vcore have pending
3733  * exceptions or are no longer ceded
3734  */
3735 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3736 {
3737         struct kvm_vcpu *vcpu;
3738         int i;
3739
3740         for_each_runnable_thread(i, vcpu, vc) {
3741                 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3742                         return 1;
3743         }
3744
3745         return 0;
3746 }
3747
3748 /*
3749  * All the vcpus in this vcore are idle, so wait for a decrementer
3750  * or external interrupt to one of the vcpus.  vc->lock is held.
3751  */
3752 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3753 {
3754         ktime_t cur, start_poll, start_wait;
3755         int do_sleep = 1;
3756         u64 block_ns;
3757         DECLARE_SWAITQUEUE(wait);
3758
3759         /* Poll for pending exceptions and ceded state */
3760         cur = start_poll = ktime_get();
3761         if (vc->halt_poll_ns) {
3762                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3763                 ++vc->runner->stat.halt_attempted_poll;
3764
3765                 vc->vcore_state = VCORE_POLLING;
3766                 spin_unlock(&vc->lock);
3767
3768                 do {
3769                         if (kvmppc_vcore_check_block(vc)) {
3770                                 do_sleep = 0;
3771                                 break;
3772                         }
3773                         cur = ktime_get();
3774                 } while (single_task_running() && ktime_before(cur, stop));
3775
3776                 spin_lock(&vc->lock);
3777                 vc->vcore_state = VCORE_INACTIVE;
3778
3779                 if (!do_sleep) {
3780                         ++vc->runner->stat.halt_successful_poll;
3781                         goto out;
3782                 }
3783         }
3784
3785         prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3786
3787         if (kvmppc_vcore_check_block(vc)) {
3788                 finish_swait(&vc->wq, &wait);
3789                 do_sleep = 0;
3790                 /* If we polled, count this as a successful poll */
3791                 if (vc->halt_poll_ns)
3792                         ++vc->runner->stat.halt_successful_poll;
3793                 goto out;
3794         }
3795
3796         start_wait = ktime_get();
3797
3798         vc->vcore_state = VCORE_SLEEPING;
3799         trace_kvmppc_vcore_blocked(vc, 0);
3800         spin_unlock(&vc->lock);
3801         schedule();
3802         finish_swait(&vc->wq, &wait);
3803         spin_lock(&vc->lock);
3804         vc->vcore_state = VCORE_INACTIVE;
3805         trace_kvmppc_vcore_blocked(vc, 1);
3806         ++vc->runner->stat.halt_successful_wait;
3807
3808         cur = ktime_get();
3809
3810 out:
3811         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3812
3813         /* Attribute wait time */
3814         if (do_sleep) {
3815                 vc->runner->stat.halt_wait_ns +=
3816                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
3817                 /* Attribute failed poll time */
3818                 if (vc->halt_poll_ns)
3819                         vc->runner->stat.halt_poll_fail_ns +=
3820                                 ktime_to_ns(start_wait) -
3821                                 ktime_to_ns(start_poll);
3822         } else {
3823                 /* Attribute successful poll time */
3824                 if (vc->halt_poll_ns)
3825                         vc->runner->stat.halt_poll_success_ns +=
3826                                 ktime_to_ns(cur) -
3827                                 ktime_to_ns(start_poll);
3828         }
3829
3830         /* Adjust poll time */
3831         if (halt_poll_ns) {
3832                 if (block_ns <= vc->halt_poll_ns)
3833                         ;
3834                 /* We slept and blocked for longer than the max halt time */
3835                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3836                         shrink_halt_poll_ns(vc);
3837                 /* We slept and our poll time is too small */
3838                 else if (vc->halt_poll_ns < halt_poll_ns &&
3839                                 block_ns < halt_poll_ns)
3840                         grow_halt_poll_ns(vc);
3841                 if (vc->halt_poll_ns > halt_poll_ns)
3842                         vc->halt_poll_ns = halt_poll_ns;
3843         } else
3844                 vc->halt_poll_ns = 0;
3845
3846         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3847 }
3848
3849 /*
3850  * This never fails for a radix guest, as none of the operations it does
3851  * for a radix guest can fail or have a way to report failure.
3852  * kvmhv_run_single_vcpu() relies on this fact.
3853  */
3854 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3855 {
3856         int r = 0;
3857         struct kvm *kvm = vcpu->kvm;
3858
3859         mutex_lock(&kvm->arch.mmu_setup_lock);
3860         if (!kvm->arch.mmu_ready) {
3861                 if (!kvm_is_radix(kvm))
3862                         r = kvmppc_hv_setup_htab_rma(vcpu);
3863                 if (!r) {
3864                         if (cpu_has_feature(CPU_FTR_ARCH_300))
3865                                 kvmppc_setup_partition_table(kvm);
3866                         kvm->arch.mmu_ready = 1;
3867                 }
3868         }
3869         mutex_unlock(&kvm->arch.mmu_setup_lock);
3870         return r;
3871 }
3872
3873 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3874 {
3875         int n_ceded, i, r;
3876         struct kvmppc_vcore *vc;
3877         struct kvm_vcpu *v;
3878
3879         trace_kvmppc_run_vcpu_enter(vcpu);
3880
3881         kvm_run->exit_reason = 0;
3882         vcpu->arch.ret = RESUME_GUEST;
3883         vcpu->arch.trap = 0;
3884         kvmppc_update_vpas(vcpu);
3885
3886         /*
3887          * Synchronize with other threads in this virtual core
3888          */
3889         vc = vcpu->arch.vcore;
3890         spin_lock(&vc->lock);
3891         vcpu->arch.ceded = 0;
3892         vcpu->arch.run_task = current;
3893         vcpu->arch.kvm_run = kvm_run;
3894         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3895         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3896         vcpu->arch.busy_preempt = TB_NIL;
3897         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3898         ++vc->n_runnable;
3899
3900         /*
3901          * This happens the first time this is called for a vcpu.
3902          * If the vcore is already running, we may be able to start
3903          * this thread straight away and have it join in.
3904          */
3905         if (!signal_pending(current)) {
3906                 if ((vc->vcore_state == VCORE_PIGGYBACK ||
3907                      vc->vcore_state == VCORE_RUNNING) &&
3908                            !VCORE_IS_EXITING(vc)) {
3909                         kvmppc_create_dtl_entry(vcpu, vc);
3910                         kvmppc_start_thread(vcpu, vc);
3911                         trace_kvm_guest_enter(vcpu);
3912                 } else if (vc->vcore_state == VCORE_SLEEPING) {
3913                         swake_up_one(&vc->wq);
3914                 }
3915
3916         }
3917
3918         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3919                !signal_pending(current)) {
3920                 /* See if the MMU is ready to go */
3921                 if (!vcpu->kvm->arch.mmu_ready) {
3922                         spin_unlock(&vc->lock);
3923                         r = kvmhv_setup_mmu(vcpu);
3924                         spin_lock(&vc->lock);
3925                         if (r) {
3926                                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3927                                 kvm_run->fail_entry.
3928                                         hardware_entry_failure_reason = 0;
3929                                 vcpu->arch.ret = r;
3930                                 break;
3931                         }
3932                 }
3933
3934                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3935                         kvmppc_vcore_end_preempt(vc);
3936
3937                 if (vc->vcore_state != VCORE_INACTIVE) {
3938                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3939                         continue;
3940                 }
3941                 for_each_runnable_thread(i, v, vc) {
3942                         kvmppc_core_prepare_to_enter(v);
3943                         if (signal_pending(v->arch.run_task)) {
3944                                 kvmppc_remove_runnable(vc, v);
3945                                 v->stat.signal_exits++;
3946                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3947                                 v->arch.ret = -EINTR;
3948                                 wake_up(&v->arch.cpu_run);
3949                         }
3950                 }
3951                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3952                         break;
3953                 n_ceded = 0;
3954                 for_each_runnable_thread(i, v, vc) {
3955                         if (!kvmppc_vcpu_woken(v))
3956                                 n_ceded += v->arch.ceded;
3957                         else
3958                                 v->arch.ceded = 0;
3959                 }
3960                 vc->runner = vcpu;
3961                 if (n_ceded == vc->n_runnable) {
3962                         kvmppc_vcore_blocked(vc);
3963                 } else if (need_resched()) {
3964                         kvmppc_vcore_preempt(vc);
3965                         /* Let something else run */
3966                         cond_resched_lock(&vc->lock);
3967                         if (vc->vcore_state == VCORE_PREEMPT)
3968                                 kvmppc_vcore_end_preempt(vc);
3969                 } else {
3970                         kvmppc_run_core(vc);
3971                 }
3972                 vc->runner = NULL;
3973         }
3974
3975         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3976                (vc->vcore_state == VCORE_RUNNING ||
3977                 vc->vcore_state == VCORE_EXITING ||
3978                 vc->vcore_state == VCORE_PIGGYBACK))
3979                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3980
3981         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3982                 kvmppc_vcore_end_preempt(vc);
3983
3984         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3985                 kvmppc_remove_runnable(vc, vcpu);
3986                 vcpu->stat.signal_exits++;
3987                 kvm_run->exit_reason = KVM_EXIT_INTR;
3988                 vcpu->arch.ret = -EINTR;
3989         }
3990
3991         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3992                 /* Wake up some vcpu to run the core */
3993                 i = -1;
3994                 v = next_runnable_thread(vc, &i);
3995                 wake_up(&v->arch.cpu_run);
3996         }
3997
3998         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3999         spin_unlock(&vc->lock);
4000         return vcpu->arch.ret;
4001 }
4002
4003 int kvmhv_run_single_vcpu(struct kvm_run *kvm_run,
4004                           struct kvm_vcpu *vcpu, u64 time_limit,
4005                           unsigned long lpcr)
4006 {
4007         int trap, r, pcpu;
4008         int srcu_idx, lpid;
4009         struct kvmppc_vcore *vc;
4010         struct kvm *kvm = vcpu->kvm;
4011         struct kvm_nested_guest *nested = vcpu->arch.nested;
4012
4013         trace_kvmppc_run_vcpu_enter(vcpu);
4014
4015         kvm_run->exit_reason = 0;
4016         vcpu->arch.ret = RESUME_GUEST;
4017         vcpu->arch.trap = 0;
4018
4019         vc = vcpu->arch.vcore;
4020         vcpu->arch.ceded = 0;
4021         vcpu->arch.run_task = current;
4022         vcpu->arch.kvm_run = kvm_run;
4023         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4024         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4025         vcpu->arch.busy_preempt = TB_NIL;
4026         vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4027         vc->runnable_threads[0] = vcpu;
4028         vc->n_runnable = 1;
4029         vc->runner = vcpu;
4030
4031         /* See if the MMU is ready to go */
4032         if (!kvm->arch.mmu_ready)
4033                 kvmhv_setup_mmu(vcpu);
4034
4035         if (need_resched())
4036                 cond_resched();
4037
4038         kvmppc_update_vpas(vcpu);
4039
4040         init_vcore_to_run(vc);
4041         vc->preempt_tb = TB_NIL;
4042
4043         preempt_disable();
4044         pcpu = smp_processor_id();
4045         vc->pcpu = pcpu;
4046         kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4047
4048         local_irq_disable();
4049         hard_irq_disable();
4050         if (signal_pending(current))
4051                 goto sigpend;
4052         if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4053                 goto out;
4054
4055         if (!nested) {
4056                 kvmppc_core_prepare_to_enter(vcpu);
4057                 if (vcpu->arch.doorbell_request) {
4058                         vc->dpdes = 1;
4059                         smp_wmb();
4060                         vcpu->arch.doorbell_request = 0;
4061                 }
4062                 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4063                              &vcpu->arch.pending_exceptions))
4064                         lpcr |= LPCR_MER;
4065         } else if (vcpu->arch.pending_exceptions ||
4066                    vcpu->arch.doorbell_request ||
4067                    xive_interrupt_pending(vcpu)) {
4068                 vcpu->arch.ret = RESUME_HOST;
4069                 goto out;
4070         }
4071
4072         kvmppc_clear_host_core(pcpu);
4073
4074         local_paca->kvm_hstate.tid = 0;
4075         local_paca->kvm_hstate.napping = 0;
4076         local_paca->kvm_hstate.kvm_split_mode = NULL;
4077         kvmppc_start_thread(vcpu, vc);
4078         kvmppc_create_dtl_entry(vcpu, vc);
4079         trace_kvm_guest_enter(vcpu);
4080
4081         vc->vcore_state = VCORE_RUNNING;
4082         trace_kvmppc_run_core(vc, 0);
4083
4084         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4085                 lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4086                 mtspr(SPRN_LPID, lpid);
4087                 isync();
4088                 kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4089         }
4090
4091         guest_enter_irqoff();
4092
4093         srcu_idx = srcu_read_lock(&kvm->srcu);
4094
4095         this_cpu_disable_ftrace();
4096
4097         /* Tell lockdep that we're about to enable interrupts */
4098         trace_hardirqs_on();
4099
4100         trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4101         vcpu->arch.trap = trap;
4102
4103         trace_hardirqs_off();
4104
4105         this_cpu_enable_ftrace();
4106
4107         srcu_read_unlock(&kvm->srcu, srcu_idx);
4108
4109         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4110                 mtspr(SPRN_LPID, kvm->arch.host_lpid);
4111                 isync();
4112         }
4113
4114         set_irq_happened(trap);
4115
4116         kvmppc_set_host_core(pcpu);
4117
4118         local_irq_enable();
4119         guest_exit();
4120
4121         cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4122
4123         preempt_enable();
4124
4125         /* cancel pending decrementer exception if DEC is now positive */
4126         if (get_tb() < vcpu->arch.dec_expires && kvmppc_core_pending_dec(vcpu))
4127                 kvmppc_core_dequeue_dec(vcpu);
4128
4129         trace_kvm_guest_exit(vcpu);
4130         r = RESUME_GUEST;
4131         if (trap) {
4132                 if (!nested)
4133                         r = kvmppc_handle_exit_hv(kvm_run, vcpu, current);
4134                 else
4135                         r = kvmppc_handle_nested_exit(kvm_run, vcpu);
4136         }
4137         vcpu->arch.ret = r;
4138
4139         if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4140             !kvmppc_vcpu_woken(vcpu)) {
4141                 kvmppc_set_timer(vcpu);
4142                 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4143                         if (signal_pending(current)) {
4144                                 vcpu->stat.signal_exits++;
4145                                 kvm_run->exit_reason = KVM_EXIT_INTR;
4146                                 vcpu->arch.ret = -EINTR;
4147                                 break;
4148                         }
4149                         spin_lock(&vc->lock);
4150                         kvmppc_vcore_blocked(vc);
4151                         spin_unlock(&vc->lock);
4152                 }
4153         }
4154         vcpu->arch.ceded = 0;
4155
4156         vc->vcore_state = VCORE_INACTIVE;
4157         trace_kvmppc_run_core(vc, 1);
4158
4159  done:
4160         kvmppc_remove_runnable(vc, vcpu);
4161         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4162
4163         return vcpu->arch.ret;
4164
4165  sigpend:
4166         vcpu->stat.signal_exits++;
4167         kvm_run->exit_reason = KVM_EXIT_INTR;
4168         vcpu->arch.ret = -EINTR;
4169  out:
4170         local_irq_enable();
4171         preempt_enable();
4172         goto done;
4173 }
4174
4175 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
4176 {
4177         int r;
4178         int srcu_idx;
4179         unsigned long ebb_regs[3] = {}; /* shut up GCC */
4180         unsigned long user_tar = 0;
4181         unsigned int user_vrsave;
4182         struct kvm *kvm;
4183
4184         if (!vcpu->arch.sane) {
4185                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4186                 return -EINVAL;
4187         }
4188
4189         /*
4190          * Don't allow entry with a suspended transaction, because
4191          * the guest entry/exit code will lose it.
4192          * If the guest has TM enabled, save away their TM-related SPRs
4193          * (they will get restored by the TM unavailable interrupt).
4194          */
4195 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4196         if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4197             (current->thread.regs->msr & MSR_TM)) {
4198                 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4199                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4200                         run->fail_entry.hardware_entry_failure_reason = 0;
4201                         return -EINVAL;
4202                 }
4203                 /* Enable TM so we can read the TM SPRs */
4204                 mtmsr(mfmsr() | MSR_TM);
4205                 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4206                 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4207                 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4208                 current->thread.regs->msr &= ~MSR_TM;
4209         }
4210 #endif
4211
4212         /*
4213          * Force online to 1 for the sake of old userspace which doesn't
4214          * set it.
4215          */
4216         if (!vcpu->arch.online) {
4217                 atomic_inc(&vcpu->arch.vcore->online_count);
4218                 vcpu->arch.online = 1;
4219         }
4220
4221         kvmppc_core_prepare_to_enter(vcpu);
4222
4223         /* No need to go into the guest when all we'll do is come back out */
4224         if (signal_pending(current)) {
4225                 run->exit_reason = KVM_EXIT_INTR;
4226                 return -EINTR;
4227         }
4228
4229         kvm = vcpu->kvm;
4230         atomic_inc(&kvm->arch.vcpus_running);
4231         /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4232         smp_mb();
4233
4234         flush_all_to_thread(current);
4235
4236         /* Save userspace EBB and other register values */
4237         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4238                 ebb_regs[0] = mfspr(SPRN_EBBHR);
4239                 ebb_regs[1] = mfspr(SPRN_EBBRR);
4240                 ebb_regs[2] = mfspr(SPRN_BESCR);
4241                 user_tar = mfspr(SPRN_TAR);
4242         }
4243         user_vrsave = mfspr(SPRN_VRSAVE);
4244
4245         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
4246         vcpu->arch.pgdir = current->mm->pgd;
4247         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4248
4249         do {
4250                 /*
4251                  * The early POWER9 chips that can't mix radix and HPT threads
4252                  * on the same core also need the workaround for the problem
4253                  * where the TLB would prefetch entries in the guest exit path
4254                  * for radix guests using the guest PIDR value and LPID 0.
4255                  * The workaround is in the old path (kvmppc_run_vcpu())
4256                  * but not the new path (kvmhv_run_single_vcpu()).
4257                  */
4258                 if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4259                     !no_mixing_hpt_and_radix)
4260                         r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0,
4261                                                   vcpu->arch.vcore->lpcr);
4262                 else
4263                         r = kvmppc_run_vcpu(run, vcpu);
4264
4265                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4266                     !(vcpu->arch.shregs.msr & MSR_PR)) {
4267                         trace_kvm_hcall_enter(vcpu);
4268                         r = kvmppc_pseries_do_hcall(vcpu);
4269                         trace_kvm_hcall_exit(vcpu, r);
4270                         kvmppc_core_prepare_to_enter(vcpu);
4271                 } else if (r == RESUME_PAGE_FAULT) {
4272                         srcu_idx = srcu_read_lock(&kvm->srcu);
4273                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
4274                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4275                         srcu_read_unlock(&kvm->srcu, srcu_idx);
4276                 } else if (r == RESUME_PASSTHROUGH) {
4277                         if (WARN_ON(xics_on_xive()))
4278                                 r = H_SUCCESS;
4279                         else
4280                                 r = kvmppc_xics_rm_complete(vcpu, 0);
4281                 }
4282         } while (is_kvmppc_resume_guest(r));
4283
4284         /* Restore userspace EBB and other register values */
4285         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4286                 mtspr(SPRN_EBBHR, ebb_regs[0]);
4287                 mtspr(SPRN_EBBRR, ebb_regs[1]);
4288                 mtspr(SPRN_BESCR, ebb_regs[2]);
4289                 mtspr(SPRN_TAR, user_tar);
4290                 mtspr(SPRN_FSCR, current->thread.fscr);
4291         }
4292         mtspr(SPRN_VRSAVE, user_vrsave);
4293
4294         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4295         atomic_dec(&kvm->arch.vcpus_running);
4296         return r;
4297 }
4298
4299 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4300                                      int shift, int sllp)
4301 {
4302         (*sps)->page_shift = shift;
4303         (*sps)->slb_enc = sllp;
4304         (*sps)->enc[0].page_shift = shift;
4305         (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4306         /*
4307          * Add 16MB MPSS support (may get filtered out by userspace)
4308          */
4309         if (shift != 24) {
4310                 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4311                 if (penc != -1) {
4312                         (*sps)->enc[1].page_shift = 24;
4313                         (*sps)->enc[1].pte_enc = penc;
4314                 }
4315         }
4316         (*sps)++;
4317 }
4318
4319 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4320                                          struct kvm_ppc_smmu_info *info)
4321 {
4322         struct kvm_ppc_one_seg_page_size *sps;
4323
4324         /*
4325          * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4326          * POWER7 doesn't support keys for instruction accesses,
4327          * POWER8 and POWER9 do.
4328          */
4329         info->data_keys = 32;
4330         info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4331
4332         /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4333         info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4334         info->slb_size = 32;
4335
4336         /* We only support these sizes for now, and no muti-size segments */
4337         sps = &info->sps[0];
4338         kvmppc_add_seg_page_size(&sps, 12, 0);
4339         kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4340         kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4341
4342         /* If running as a nested hypervisor, we don't support HPT guests */
4343         if (kvmhv_on_pseries())
4344                 info->flags |= KVM_PPC_NO_HASH;
4345
4346         return 0;
4347 }
4348
4349 /*
4350  * Get (and clear) the dirty memory log for a memory slot.
4351  */
4352 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4353                                          struct kvm_dirty_log *log)
4354 {
4355         struct kvm_memslots *slots;
4356         struct kvm_memory_slot *memslot;
4357         int i, r;
4358         unsigned long n;
4359         unsigned long *buf, *p;
4360         struct kvm_vcpu *vcpu;
4361
4362         mutex_lock(&kvm->slots_lock);
4363
4364         r = -EINVAL;
4365         if (log->slot >= KVM_USER_MEM_SLOTS)
4366                 goto out;
4367
4368         slots = kvm_memslots(kvm);
4369         memslot = id_to_memslot(slots, log->slot);
4370         r = -ENOENT;
4371         if (!memslot->dirty_bitmap)
4372                 goto out;
4373
4374         /*
4375          * Use second half of bitmap area because both HPT and radix
4376          * accumulate bits in the first half.
4377          */
4378         n = kvm_dirty_bitmap_bytes(memslot);
4379         buf = memslot->dirty_bitmap + n / sizeof(long);
4380         memset(buf, 0, n);
4381
4382         if (kvm_is_radix(kvm))
4383                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4384         else
4385                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4386         if (r)
4387                 goto out;
4388
4389         /*
4390          * We accumulate dirty bits in the first half of the
4391          * memslot's dirty_bitmap area, for when pages are paged
4392          * out or modified by the host directly.  Pick up these
4393          * bits and add them to the map.
4394          */
4395         p = memslot->dirty_bitmap;
4396         for (i = 0; i < n / sizeof(long); ++i)
4397                 buf[i] |= xchg(&p[i], 0);
4398
4399         /* Harvest dirty bits from VPA and DTL updates */
4400         /* Note: we never modify the SLB shadow buffer areas */
4401         kvm_for_each_vcpu(i, vcpu, kvm) {
4402                 spin_lock(&vcpu->arch.vpa_update_lock);
4403                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4404                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4405                 spin_unlock(&vcpu->arch.vpa_update_lock);
4406         }
4407
4408         r = -EFAULT;
4409         if (copy_to_user(log->dirty_bitmap, buf, n))
4410                 goto out;
4411
4412         r = 0;
4413 out:
4414         mutex_unlock(&kvm->slots_lock);
4415         return r;
4416 }
4417
4418 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
4419                                         struct kvm_memory_slot *dont)
4420 {
4421         if (!dont || free->arch.rmap != dont->arch.rmap) {
4422                 vfree(free->arch.rmap);
4423                 free->arch.rmap = NULL;
4424         }
4425 }
4426
4427 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
4428                                          unsigned long npages)
4429 {
4430         slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
4431         if (!slot->arch.rmap)
4432                 return -ENOMEM;
4433
4434         return 0;
4435 }
4436
4437 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4438                                         struct kvm_memory_slot *memslot,
4439                                         const struct kvm_userspace_memory_region *mem)
4440 {
4441         return 0;
4442 }
4443
4444 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4445                                 const struct kvm_userspace_memory_region *mem,
4446                                 const struct kvm_memory_slot *old,
4447                                 const struct kvm_memory_slot *new,
4448                                 enum kvm_mr_change change)
4449 {
4450         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4451
4452         /*
4453          * If we are making a new memslot, it might make
4454          * some address that was previously cached as emulated
4455          * MMIO be no longer emulated MMIO, so invalidate
4456          * all the caches of emulated MMIO translations.
4457          */
4458         if (npages)
4459                 atomic64_inc(&kvm->arch.mmio_update);
4460
4461         /*
4462          * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4463          * have already called kvm_arch_flush_shadow_memslot() to
4464          * flush shadow mappings.  For KVM_MR_CREATE we have no
4465          * previous mappings.  So the only case to handle is
4466          * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4467          * has been changed.
4468          * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4469          * to get rid of any THP PTEs in the partition-scoped page tables
4470          * so we can track dirtiness at the page level; we flush when
4471          * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4472          * using THP PTEs.
4473          */
4474         if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4475             ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4476                 kvmppc_radix_flush_memslot(kvm, old);
4477 }
4478
4479 /*
4480  * Update LPCR values in kvm->arch and in vcores.
4481  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4482  * of kvm->arch.lpcr update).
4483  */
4484 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4485 {
4486         long int i;
4487         u32 cores_done = 0;
4488
4489         if ((kvm->arch.lpcr & mask) == lpcr)
4490                 return;
4491
4492         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4493
4494         for (i = 0; i < KVM_MAX_VCORES; ++i) {
4495                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4496                 if (!vc)
4497                         continue;
4498                 spin_lock(&vc->lock);
4499                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4500                 spin_unlock(&vc->lock);
4501                 if (++cores_done >= kvm->arch.online_vcores)
4502                         break;
4503         }
4504 }
4505
4506 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
4507 {
4508         return;
4509 }
4510
4511 void kvmppc_setup_partition_table(struct kvm *kvm)
4512 {
4513         unsigned long dw0, dw1;
4514
4515         if (!kvm_is_radix(kvm)) {
4516                 /* PS field - page size for VRMA */
4517                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4518                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4519                 /* HTABSIZE and HTABORG fields */
4520                 dw0 |= kvm->arch.sdr1;
4521
4522                 /* Second dword as set by userspace */
4523                 dw1 = kvm->arch.process_table;
4524         } else {
4525                 dw0 = PATB_HR | radix__get_tree_size() |
4526                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4527                 dw1 = PATB_GR | kvm->arch.process_table;
4528         }
4529         kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4530 }
4531
4532 /*
4533  * Set up HPT (hashed page table) and RMA (real-mode area).
4534  * Must be called with kvm->arch.mmu_setup_lock held.
4535  */
4536 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4537 {
4538         int err = 0;
4539         struct kvm *kvm = vcpu->kvm;
4540         unsigned long hva;
4541         struct kvm_memory_slot *memslot;
4542         struct vm_area_struct *vma;
4543         unsigned long lpcr = 0, senc;
4544         unsigned long psize, porder;
4545         int srcu_idx;
4546
4547         /* Allocate hashed page table (if not done already) and reset it */
4548         if (!kvm->arch.hpt.virt) {
4549                 int order = KVM_DEFAULT_HPT_ORDER;
4550                 struct kvm_hpt_info info;
4551
4552                 err = kvmppc_allocate_hpt(&info, order);
4553                 /* If we get here, it means userspace didn't specify a
4554                  * size explicitly.  So, try successively smaller
4555                  * sizes if the default failed. */
4556                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4557                         err  = kvmppc_allocate_hpt(&info, order);
4558
4559                 if (err < 0) {
4560                         pr_err("KVM: Couldn't alloc HPT\n");
4561                         goto out;
4562                 }
4563
4564                 kvmppc_set_hpt(kvm, &info);
4565         }
4566
4567         /* Look up the memslot for guest physical address 0 */
4568         srcu_idx = srcu_read_lock(&kvm->srcu);
4569         memslot = gfn_to_memslot(kvm, 0);
4570
4571         /* We must have some memory at 0 by now */
4572         err = -EINVAL;
4573         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4574                 goto out_srcu;
4575
4576         /* Look up the VMA for the start of this memory slot */
4577         hva = memslot->userspace_addr;
4578         down_read(&current->mm->mmap_sem);
4579         vma = find_vma(current->mm, hva);
4580         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4581                 goto up_out;
4582
4583         psize = vma_kernel_pagesize(vma);
4584
4585         up_read(&current->mm->mmap_sem);
4586
4587         /* We can handle 4k, 64k or 16M pages in the VRMA */
4588         if (psize >= 0x1000000)
4589                 psize = 0x1000000;
4590         else if (psize >= 0x10000)
4591                 psize = 0x10000;
4592         else
4593                 psize = 0x1000;
4594         porder = __ilog2(psize);
4595
4596         senc = slb_pgsize_encoding(psize);
4597         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4598                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4599         /* Create HPTEs in the hash page table for the VRMA */
4600         kvmppc_map_vrma(vcpu, memslot, porder);
4601
4602         /* Update VRMASD field in the LPCR */
4603         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4604                 /* the -4 is to account for senc values starting at 0x10 */
4605                 lpcr = senc << (LPCR_VRMASD_SH - 4);
4606                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4607         }
4608
4609         /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4610         smp_wmb();
4611         err = 0;
4612  out_srcu:
4613         srcu_read_unlock(&kvm->srcu, srcu_idx);
4614  out:
4615         return err;
4616
4617  up_out:
4618         up_read(&current->mm->mmap_sem);
4619         goto out_srcu;
4620 }
4621
4622 /*
4623  * Must be called with kvm->arch.mmu_setup_lock held and
4624  * mmu_ready = 0 and no vcpus running.
4625  */
4626 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4627 {
4628         if (nesting_enabled(kvm))
4629                 kvmhv_release_all_nested(kvm);
4630         kvmppc_rmap_reset(kvm);
4631         kvm->arch.process_table = 0;
4632         /* Mutual exclusion with kvm_unmap_hva_range etc. */
4633         spin_lock(&kvm->mmu_lock);
4634         kvm->arch.radix = 0;
4635         spin_unlock(&kvm->mmu_lock);
4636         kvmppc_free_radix(kvm);
4637         kvmppc_update_lpcr(kvm, LPCR_VPM1,
4638                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4639         return 0;
4640 }
4641
4642 /*
4643  * Must be called with kvm->arch.mmu_setup_lock held and
4644  * mmu_ready = 0 and no vcpus running.
4645  */
4646 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4647 {
4648         int err;
4649
4650         err = kvmppc_init_vm_radix(kvm);
4651         if (err)
4652                 return err;
4653         kvmppc_rmap_reset(kvm);
4654         /* Mutual exclusion with kvm_unmap_hva_range etc. */
4655         spin_lock(&kvm->mmu_lock);
4656         kvm->arch.radix = 1;
4657         spin_unlock(&kvm->mmu_lock);
4658         kvmppc_free_hpt(&kvm->arch.hpt);
4659         kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4660                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4661         return 0;
4662 }
4663
4664 #ifdef CONFIG_KVM_XICS
4665 /*
4666  * Allocate a per-core structure for managing state about which cores are
4667  * running in the host versus the guest and for exchanging data between
4668  * real mode KVM and CPU running in the host.
4669  * This is only done for the first VM.
4670  * The allocated structure stays even if all VMs have stopped.
4671  * It is only freed when the kvm-hv module is unloaded.
4672  * It's OK for this routine to fail, we just don't support host
4673  * core operations like redirecting H_IPI wakeups.
4674  */
4675 void kvmppc_alloc_host_rm_ops(void)
4676 {
4677         struct kvmppc_host_rm_ops *ops;
4678         unsigned long l_ops;
4679         int cpu, core;
4680         int size;
4681
4682         /* Not the first time here ? */
4683         if (kvmppc_host_rm_ops_hv != NULL)
4684                 return;
4685
4686         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4687         if (!ops)
4688                 return;
4689
4690         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4691         ops->rm_core = kzalloc(size, GFP_KERNEL);
4692
4693         if (!ops->rm_core) {
4694                 kfree(ops);
4695                 return;
4696         }
4697
4698         cpus_read_lock();
4699
4700         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4701                 if (!cpu_online(cpu))
4702                         continue;
4703
4704                 core = cpu >> threads_shift;
4705                 ops->rm_core[core].rm_state.in_host = 1;
4706         }
4707
4708         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4709
4710         /*
4711          * Make the contents of the kvmppc_host_rm_ops structure visible
4712          * to other CPUs before we assign it to the global variable.
4713          * Do an atomic assignment (no locks used here), but if someone
4714          * beats us to it, just free our copy and return.
4715          */
4716         smp_wmb();
4717         l_ops = (unsigned long) ops;
4718
4719         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4720                 cpus_read_unlock();
4721                 kfree(ops->rm_core);
4722                 kfree(ops);
4723                 return;
4724         }
4725
4726         cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4727                                              "ppc/kvm_book3s:prepare",
4728                                              kvmppc_set_host_core,
4729                                              kvmppc_clear_host_core);
4730         cpus_read_unlock();
4731 }
4732
4733 void kvmppc_free_host_rm_ops(void)
4734 {
4735         if (kvmppc_host_rm_ops_hv) {
4736                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4737                 kfree(kvmppc_host_rm_ops_hv->rm_core);
4738                 kfree(kvmppc_host_rm_ops_hv);
4739                 kvmppc_host_rm_ops_hv = NULL;
4740         }
4741 }
4742 #endif
4743
4744 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4745 {
4746         unsigned long lpcr, lpid;
4747         char buf[32];
4748         int ret;
4749
4750         mutex_init(&kvm->arch.mmu_setup_lock);
4751
4752         /* Allocate the guest's logical partition ID */
4753
4754         lpid = kvmppc_alloc_lpid();
4755         if ((long)lpid < 0)
4756                 return -ENOMEM;
4757         kvm->arch.lpid = lpid;
4758
4759         kvmppc_alloc_host_rm_ops();
4760
4761         kvmhv_vm_nested_init(kvm);
4762
4763         /*
4764          * Since we don't flush the TLB when tearing down a VM,
4765          * and this lpid might have previously been used,
4766          * make sure we flush on each core before running the new VM.
4767          * On POWER9, the tlbie in mmu_partition_table_set_entry()
4768          * does this flush for us.
4769          */
4770         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4771                 cpumask_setall(&kvm->arch.need_tlb_flush);
4772
4773         /* Start out with the default set of hcalls enabled */
4774         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4775                sizeof(kvm->arch.enabled_hcalls));
4776
4777         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4778                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4779
4780         /* Init LPCR for virtual RMA mode */
4781         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4782                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
4783                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4784                 lpcr &= LPCR_PECE | LPCR_LPES;
4785         } else {
4786                 lpcr = 0;
4787         }
4788         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4789                 LPCR_VPM0 | LPCR_VPM1;
4790         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4791                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4792         /* On POWER8 turn on online bit to enable PURR/SPURR */
4793         if (cpu_has_feature(CPU_FTR_ARCH_207S))
4794                 lpcr |= LPCR_ONL;
4795         /*
4796          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4797          * Set HVICE bit to enable hypervisor virtualization interrupts.
4798          * Set HEIC to prevent OS interrupts to go to hypervisor (should
4799          * be unnecessary but better safe than sorry in case we re-enable
4800          * EE in HV mode with this LPCR still set)
4801          */
4802         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4803                 lpcr &= ~LPCR_VPM0;
4804                 lpcr |= LPCR_HVICE | LPCR_HEIC;
4805
4806                 /*
4807                  * If xive is enabled, we route 0x500 interrupts directly
4808                  * to the guest.
4809                  */
4810                 if (xics_on_xive())
4811                         lpcr |= LPCR_LPES;
4812         }
4813
4814         /*
4815          * If the host uses radix, the guest starts out as radix.
4816          */
4817         if (radix_enabled()) {
4818                 kvm->arch.radix = 1;
4819                 kvm->arch.mmu_ready = 1;
4820                 lpcr &= ~LPCR_VPM1;
4821                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4822                 ret = kvmppc_init_vm_radix(kvm);
4823                 if (ret) {
4824                         kvmppc_free_lpid(kvm->arch.lpid);
4825                         return ret;
4826                 }
4827                 kvmppc_setup_partition_table(kvm);
4828         }
4829
4830         kvm->arch.lpcr = lpcr;
4831
4832         /* Initialization for future HPT resizes */
4833         kvm->arch.resize_hpt = NULL;
4834
4835         /*
4836          * Work out how many sets the TLB has, for the use of
4837          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4838          */
4839         if (radix_enabled())
4840                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
4841         else if (cpu_has_feature(CPU_FTR_ARCH_300))
4842                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
4843         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4844                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
4845         else
4846                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
4847
4848         /*
4849          * Track that we now have a HV mode VM active. This blocks secondary
4850          * CPU threads from coming online.
4851          * On POWER9, we only need to do this if the "indep_threads_mode"
4852          * module parameter has been set to N.
4853          */
4854         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4855                 if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4856                         pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4857                         kvm->arch.threads_indep = true;
4858                 } else {
4859                         kvm->arch.threads_indep = indep_threads_mode;
4860                 }
4861         }
4862         if (!kvm->arch.threads_indep)
4863                 kvm_hv_vm_activated();
4864
4865         /*
4866          * Initialize smt_mode depending on processor.
4867          * POWER8 and earlier have to use "strict" threading, where
4868          * all vCPUs in a vcore have to run on the same (sub)core,
4869          * whereas on POWER9 the threads can each run a different
4870          * guest.
4871          */
4872         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4873                 kvm->arch.smt_mode = threads_per_subcore;
4874         else
4875                 kvm->arch.smt_mode = 1;
4876         kvm->arch.emul_smt_mode = 1;
4877
4878         /*
4879          * Create a debugfs directory for the VM
4880          */
4881         snprintf(buf, sizeof(buf), "vm%d", current->pid);
4882         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4883         kvmppc_mmu_debugfs_init(kvm);
4884         if (radix_enabled())
4885                 kvmhv_radix_debugfs_init(kvm);
4886
4887         return 0;
4888 }
4889
4890 static void kvmppc_free_vcores(struct kvm *kvm)
4891 {
4892         long int i;
4893
4894         for (i = 0; i < KVM_MAX_VCORES; ++i)
4895                 kfree(kvm->arch.vcores[i]);
4896         kvm->arch.online_vcores = 0;
4897 }
4898
4899 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4900 {
4901         debugfs_remove_recursive(kvm->arch.debugfs_dir);
4902
4903         if (!kvm->arch.threads_indep)
4904                 kvm_hv_vm_deactivated();
4905
4906         kvmppc_free_vcores(kvm);
4907
4908
4909         if (kvm_is_radix(kvm))
4910                 kvmppc_free_radix(kvm);
4911         else
4912                 kvmppc_free_hpt(&kvm->arch.hpt);
4913
4914         /* Perform global invalidation and return lpid to the pool */
4915         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4916                 if (nesting_enabled(kvm))
4917                         kvmhv_release_all_nested(kvm);
4918                 kvm->arch.process_table = 0;
4919                 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
4920         }
4921         kvmppc_free_lpid(kvm->arch.lpid);
4922
4923         kvmppc_free_pimap(kvm);
4924 }
4925
4926 /* We don't need to emulate any privileged instructions or dcbz */
4927 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4928                                      unsigned int inst, int *advance)
4929 {
4930         return EMULATE_FAIL;
4931 }
4932
4933 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4934                                         ulong spr_val)
4935 {
4936         return EMULATE_FAIL;
4937 }
4938
4939 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4940                                         ulong *spr_val)
4941 {
4942         return EMULATE_FAIL;
4943 }
4944
4945 static int kvmppc_core_check_processor_compat_hv(void)
4946 {
4947         if (cpu_has_feature(CPU_FTR_HVMODE) &&
4948             cpu_has_feature(CPU_FTR_ARCH_206))
4949                 return 0;
4950
4951         /* POWER9 in radix mode is capable of being a nested hypervisor. */
4952         if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
4953                 return 0;
4954
4955         return -EIO;
4956 }
4957
4958 #ifdef CONFIG_KVM_XICS
4959
4960 void kvmppc_free_pimap(struct kvm *kvm)
4961 {
4962         kfree(kvm->arch.pimap);
4963 }
4964
4965 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
4966 {
4967         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
4968 }
4969
4970 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4971 {
4972         struct irq_desc *desc;
4973         struct kvmppc_irq_map *irq_map;
4974         struct kvmppc_passthru_irqmap *pimap;
4975         struct irq_chip *chip;
4976         int i, rc = 0;
4977
4978         if (!kvm_irq_bypass)
4979                 return 1;
4980
4981         desc = irq_to_desc(host_irq);
4982         if (!desc)
4983                 return -EIO;
4984
4985         mutex_lock(&kvm->lock);
4986
4987         pimap = kvm->arch.pimap;
4988         if (pimap == NULL) {
4989                 /* First call, allocate structure to hold IRQ map */
4990                 pimap = kvmppc_alloc_pimap();
4991                 if (pimap == NULL) {
4992                         mutex_unlock(&kvm->lock);
4993                         return -ENOMEM;
4994                 }
4995                 kvm->arch.pimap = pimap;
4996         }
4997
4998         /*
4999          * For now, we only support interrupts for which the EOI operation
5000          * is an OPAL call followed by a write to XIRR, since that's
5001          * what our real-mode EOI code does, or a XIVE interrupt
5002          */
5003         chip = irq_data_get_irq_chip(&desc->irq_data);
5004         if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5005                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5006                         host_irq, guest_gsi);
5007                 mutex_unlock(&kvm->lock);
5008                 return -ENOENT;
5009         }
5010
5011         /*
5012          * See if we already have an entry for this guest IRQ number.
5013          * If it's mapped to a hardware IRQ number, that's an error,
5014          * otherwise re-use this entry.
5015          */
5016         for (i = 0; i < pimap->n_mapped; i++) {
5017                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5018                         if (pimap->mapped[i].r_hwirq) {
5019                                 mutex_unlock(&kvm->lock);
5020                                 return -EINVAL;
5021                         }
5022                         break;
5023                 }
5024         }
5025
5026         if (i == KVMPPC_PIRQ_MAPPED) {
5027                 mutex_unlock(&kvm->lock);
5028                 return -EAGAIN;         /* table is full */
5029         }
5030
5031         irq_map = &pimap->mapped[i];
5032
5033         irq_map->v_hwirq = guest_gsi;
5034         irq_map->desc = desc;
5035
5036         /*
5037          * Order the above two stores before the next to serialize with
5038          * the KVM real mode handler.
5039          */
5040         smp_wmb();
5041         irq_map->r_hwirq = desc->irq_data.hwirq;
5042
5043         if (i == pimap->n_mapped)
5044                 pimap->n_mapped++;
5045
5046         if (xics_on_xive())
5047                 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5048         else
5049                 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5050         if (rc)
5051                 irq_map->r_hwirq = 0;
5052
5053         mutex_unlock(&kvm->lock);
5054
5055         return 0;
5056 }
5057
5058 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5059 {
5060         struct irq_desc *desc;
5061         struct kvmppc_passthru_irqmap *pimap;
5062         int i, rc = 0;
5063
5064         if (!kvm_irq_bypass)
5065                 return 0;
5066
5067         desc = irq_to_desc(host_irq);
5068         if (!desc)
5069                 return -EIO;
5070
5071         mutex_lock(&kvm->lock);
5072         if (!kvm->arch.pimap)
5073                 goto unlock;
5074
5075         pimap = kvm->arch.pimap;
5076
5077         for (i = 0; i < pimap->n_mapped; i++) {
5078                 if (guest_gsi == pimap->mapped[i].v_hwirq)
5079                         break;
5080         }
5081
5082         if (i == pimap->n_mapped) {
5083                 mutex_unlock(&kvm->lock);
5084                 return -ENODEV;
5085         }
5086
5087         if (xics_on_xive())
5088                 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5089         else
5090                 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5091
5092         /* invalidate the entry (what do do on error from the above ?) */
5093         pimap->mapped[i].r_hwirq = 0;
5094
5095         /*
5096          * We don't free this structure even when the count goes to
5097          * zero. The structure is freed when we destroy the VM.
5098          */
5099  unlock:
5100         mutex_unlock(&kvm->lock);
5101         return rc;
5102 }
5103
5104 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5105                                              struct irq_bypass_producer *prod)
5106 {
5107         int ret = 0;
5108         struct kvm_kernel_irqfd *irqfd =
5109                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5110
5111         irqfd->producer = prod;
5112
5113         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5114         if (ret)
5115                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5116                         prod->irq, irqfd->gsi, ret);
5117
5118         return ret;
5119 }
5120
5121 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5122                                               struct irq_bypass_producer *prod)
5123 {
5124         int ret;
5125         struct kvm_kernel_irqfd *irqfd =
5126                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5127
5128         irqfd->producer = NULL;
5129
5130         /*
5131          * When producer of consumer is unregistered, we change back to
5132          * default external interrupt handling mode - KVM real mode
5133          * will switch back to host.
5134          */
5135         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5136         if (ret)
5137                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5138                         prod->irq, irqfd->gsi, ret);
5139 }
5140 #endif
5141
5142 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5143                                  unsigned int ioctl, unsigned long arg)
5144 {
5145         struct kvm *kvm __maybe_unused = filp->private_data;
5146         void __user *argp = (void __user *)arg;
5147         long r;
5148
5149         switch (ioctl) {
5150
5151         case KVM_PPC_ALLOCATE_HTAB: {
5152                 u32 htab_order;
5153
5154                 r = -EFAULT;
5155                 if (get_user(htab_order, (u32 __user *)argp))
5156                         break;
5157                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5158                 if (r)
5159                         break;
5160                 r = 0;
5161                 break;
5162         }
5163
5164         case KVM_PPC_GET_HTAB_FD: {
5165                 struct kvm_get_htab_fd ghf;
5166
5167                 r = -EFAULT;
5168                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5169                         break;
5170                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5171                 break;
5172         }
5173
5174         case KVM_PPC_RESIZE_HPT_PREPARE: {
5175                 struct kvm_ppc_resize_hpt rhpt;
5176
5177                 r = -EFAULT;
5178                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5179                         break;
5180
5181                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5182                 break;
5183         }
5184
5185         case KVM_PPC_RESIZE_HPT_COMMIT: {
5186                 struct kvm_ppc_resize_hpt rhpt;
5187
5188                 r = -EFAULT;
5189                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5190                         break;
5191
5192                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5193                 break;
5194         }
5195
5196         default:
5197                 r = -ENOTTY;
5198         }
5199
5200         return r;
5201 }
5202
5203 /*
5204  * List of hcall numbers to enable by default.
5205  * For compatibility with old userspace, we enable by default
5206  * all hcalls that were implemented before the hcall-enabling
5207  * facility was added.  Note this list should not include H_RTAS.
5208  */
5209 static unsigned int default_hcall_list[] = {
5210         H_REMOVE,
5211         H_ENTER,
5212         H_READ,
5213         H_PROTECT,
5214         H_BULK_REMOVE,
5215         H_GET_TCE,
5216         H_PUT_TCE,
5217         H_SET_DABR,
5218         H_SET_XDABR,
5219         H_CEDE,
5220         H_PROD,
5221         H_CONFER,
5222         H_REGISTER_VPA,
5223 #ifdef CONFIG_KVM_XICS
5224         H_EOI,
5225         H_CPPR,
5226         H_IPI,
5227         H_IPOLL,
5228         H_XIRR,
5229         H_XIRR_X,
5230 #endif
5231         0
5232 };
5233
5234 static void init_default_hcalls(void)
5235 {
5236         int i;
5237         unsigned int hcall;
5238
5239         for (i = 0; default_hcall_list[i]; ++i) {
5240                 hcall = default_hcall_list[i];
5241                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5242                 __set_bit(hcall / 4, default_enabled_hcalls);
5243         }
5244 }
5245
5246 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5247 {
5248         unsigned long lpcr;
5249         int radix;
5250         int err;
5251
5252         /* If not on a POWER9, reject it */
5253         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5254                 return -ENODEV;
5255
5256         /* If any unknown flags set, reject it */
5257         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5258                 return -EINVAL;
5259
5260         /* GR (guest radix) bit in process_table field must match */
5261         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5262         if (!!(cfg->process_table & PATB_GR) != radix)
5263                 return -EINVAL;
5264
5265         /* Process table size field must be reasonable, i.e. <= 24 */
5266         if ((cfg->process_table & PRTS_MASK) > 24)
5267                 return -EINVAL;
5268
5269         /* We can change a guest to/from radix now, if the host is radix */
5270         if (radix && !radix_enabled())
5271                 return -EINVAL;
5272
5273         /* If we're a nested hypervisor, we currently only support radix */
5274         if (kvmhv_on_pseries() && !radix)
5275                 return -EINVAL;
5276
5277         mutex_lock(&kvm->arch.mmu_setup_lock);
5278         if (radix != kvm_is_radix(kvm)) {
5279                 if (kvm->arch.mmu_ready) {
5280                         kvm->arch.mmu_ready = 0;
5281                         /* order mmu_ready vs. vcpus_running */
5282                         smp_mb();
5283                         if (atomic_read(&kvm->arch.vcpus_running)) {
5284                                 kvm->arch.mmu_ready = 1;
5285                                 err = -EBUSY;
5286                                 goto out_unlock;
5287                         }
5288                 }
5289                 if (radix)
5290                         err = kvmppc_switch_mmu_to_radix(kvm);
5291                 else
5292                         err = kvmppc_switch_mmu_to_hpt(kvm);
5293                 if (err)
5294                         goto out_unlock;
5295         }
5296
5297         kvm->arch.process_table = cfg->process_table;
5298         kvmppc_setup_partition_table(kvm);
5299
5300         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5301         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5302         err = 0;
5303
5304  out_unlock:
5305         mutex_unlock(&kvm->arch.mmu_setup_lock);
5306         return err;
5307 }
5308
5309 static int kvmhv_enable_nested(struct kvm *kvm)
5310 {
5311         if (!nested)
5312                 return -EPERM;
5313         if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5314                 return -ENODEV;
5315
5316         /* kvm == NULL means the caller is testing if the capability exists */
5317         if (kvm)
5318                 kvm->arch.nested_enable = true;
5319         return 0;
5320 }
5321
5322 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5323                                  int size)
5324 {
5325         int rc = -EINVAL;
5326
5327         if (kvmhv_vcpu_is_radix(vcpu)) {
5328                 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5329
5330                 if (rc > 0)
5331                         rc = -EINVAL;
5332         }
5333
5334         /* For now quadrants are the only way to access nested guest memory */
5335         if (rc && vcpu->arch.nested)
5336                 rc = -EAGAIN;
5337
5338         return rc;
5339 }
5340
5341 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5342                                 int size)
5343 {
5344         int rc = -EINVAL;
5345
5346         if (kvmhv_vcpu_is_radix(vcpu)) {
5347                 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5348
5349                 if (rc > 0)
5350                         rc = -EINVAL;
5351         }
5352
5353         /* For now quadrants are the only way to access nested guest memory */
5354         if (rc && vcpu->arch.nested)
5355                 rc = -EAGAIN;
5356
5357         return rc;
5358 }
5359
5360 static struct kvmppc_ops kvm_ops_hv = {
5361         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5362         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5363         .get_one_reg = kvmppc_get_one_reg_hv,
5364         .set_one_reg = kvmppc_set_one_reg_hv,
5365         .vcpu_load   = kvmppc_core_vcpu_load_hv,
5366         .vcpu_put    = kvmppc_core_vcpu_put_hv,
5367         .set_msr     = kvmppc_set_msr_hv,
5368         .vcpu_run    = kvmppc_vcpu_run_hv,
5369         .vcpu_create = kvmppc_core_vcpu_create_hv,
5370         .vcpu_free   = kvmppc_core_vcpu_free_hv,
5371         .check_requests = kvmppc_core_check_requests_hv,
5372         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5373         .flush_memslot  = kvmppc_core_flush_memslot_hv,
5374         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5375         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5376         .unmap_hva_range = kvm_unmap_hva_range_hv,
5377         .age_hva  = kvm_age_hva_hv,
5378         .test_age_hva = kvm_test_age_hva_hv,
5379         .set_spte_hva = kvm_set_spte_hva_hv,
5380         .mmu_destroy  = kvmppc_mmu_destroy_hv,
5381         .free_memslot = kvmppc_core_free_memslot_hv,
5382         .create_memslot = kvmppc_core_create_memslot_hv,
5383         .init_vm =  kvmppc_core_init_vm_hv,
5384         .destroy_vm = kvmppc_core_destroy_vm_hv,
5385         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5386         .emulate_op = kvmppc_core_emulate_op_hv,
5387         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5388         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5389         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5390         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5391         .hcall_implemented = kvmppc_hcall_impl_hv,
5392 #ifdef CONFIG_KVM_XICS
5393         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5394         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5395 #endif
5396         .configure_mmu = kvmhv_configure_mmu,
5397         .get_rmmu_info = kvmhv_get_rmmu_info,
5398         .set_smt_mode = kvmhv_set_smt_mode,
5399         .enable_nested = kvmhv_enable_nested,
5400         .load_from_eaddr = kvmhv_load_from_eaddr,
5401         .store_to_eaddr = kvmhv_store_to_eaddr,
5402 };
5403
5404 static int kvm_init_subcore_bitmap(void)
5405 {
5406         int i, j;
5407         int nr_cores = cpu_nr_cores();
5408         struct sibling_subcore_state *sibling_subcore_state;
5409
5410         for (i = 0; i < nr_cores; i++) {
5411                 int first_cpu = i * threads_per_core;
5412                 int node = cpu_to_node(first_cpu);
5413
5414                 /* Ignore if it is already allocated. */
5415                 if (paca_ptrs[first_cpu]->sibling_subcore_state)
5416                         continue;
5417
5418                 sibling_subcore_state =
5419                         kzalloc_node(sizeof(struct sibling_subcore_state),
5420                                                         GFP_KERNEL, node);
5421                 if (!sibling_subcore_state)
5422                         return -ENOMEM;
5423
5424
5425                 for (j = 0; j < threads_per_core; j++) {
5426                         int cpu = first_cpu + j;
5427
5428                         paca_ptrs[cpu]->sibling_subcore_state =
5429                                                 sibling_subcore_state;
5430                 }
5431         }
5432         return 0;
5433 }
5434
5435 static int kvmppc_radix_possible(void)
5436 {
5437         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5438 }
5439
5440 static int kvmppc_book3s_init_hv(void)
5441 {
5442         int r;
5443         /*
5444          * FIXME!! Do we need to check on all cpus ?
5445          */
5446         r = kvmppc_core_check_processor_compat_hv();
5447         if (r < 0)
5448                 return -ENODEV;
5449
5450         r = kvmhv_nested_init();
5451         if (r)
5452                 return r;
5453
5454         r = kvm_init_subcore_bitmap();
5455         if (r)
5456                 return r;
5457
5458         /*
5459          * We need a way of accessing the XICS interrupt controller,
5460          * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5461          * indirectly, via OPAL.
5462          */
5463 #ifdef CONFIG_SMP
5464         if (!xics_on_xive() && !kvmhv_on_pseries() &&
5465             !local_paca->kvm_hstate.xics_phys) {
5466                 struct device_node *np;
5467
5468                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5469                 if (!np) {
5470                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5471                         return -ENODEV;
5472                 }
5473                 /* presence of intc confirmed - node can be dropped again */
5474                 of_node_put(np);
5475         }
5476 #endif
5477
5478         kvm_ops_hv.owner = THIS_MODULE;
5479         kvmppc_hv_ops = &kvm_ops_hv;
5480
5481         init_default_hcalls();
5482
5483         init_vcore_lists();
5484
5485         r = kvmppc_mmu_hv_init();
5486         if (r)
5487                 return r;
5488
5489         if (kvmppc_radix_possible())
5490                 r = kvmppc_radix_init();
5491
5492         /*
5493          * POWER9 chips before version 2.02 can't have some threads in
5494          * HPT mode and some in radix mode on the same core.
5495          */
5496         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5497                 unsigned int pvr = mfspr(SPRN_PVR);
5498                 if ((pvr >> 16) == PVR_POWER9 &&
5499                     (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5500                      ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5501                         no_mixing_hpt_and_radix = true;
5502         }
5503
5504         return r;
5505 }
5506
5507 static void kvmppc_book3s_exit_hv(void)
5508 {
5509         kvmppc_free_host_rm_ops();
5510         if (kvmppc_radix_possible())
5511                 kvmppc_radix_exit();
5512         kvmppc_hv_ops = NULL;
5513         kvmhv_nested_exit();
5514 }
5515
5516 module_init(kvmppc_book3s_init_hv);
5517 module_exit(kvmppc_book3s_exit_hv);
5518 MODULE_LICENSE("GPL");
5519 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5520 MODULE_ALIAS("devname:kvm");