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