arch/powerpc/kvm/book3s_hv.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
   4  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
   5  *
   6  * Authors:
   7  *    Paul Mackerras <paulus@au1.ibm.com>
   8  *    Alexander Graf <agraf@suse.de>
   9  *    Kevin Wolf <mail@kevin-wolf.de>
  10  *
  11  * Description: KVM functions specific to running on Book 3S
  12  * processors in hypervisor mode (specifically POWER7 and later).
  13  *
  14  * This file is derived from arch/powerpc/kvm/book3s.c,
  15  * by Alexander Graf <agraf@suse.de>.
  16  */
  17
  18 #include <linux/kvm_host.h>
  19 #include <linux/kernel.h>
  20 #include <linux/err.h>
  21 #include <linux/slab.h>
  22 #include <linux/preempt.h>
  23 #include <linux/sched/signal.h>
  24 #include <linux/sched/stat.h>
  25 #include <linux/delay.h>
  26 #include <linux/export.h>
  27 #include <linux/fs.h>
  28 #include <linux/anon_inodes.h>
  29 #include <linux/cpu.h>
  30 #include <linux/cpumask.h>
  31 #include <linux/spinlock.h>
  32 #include <linux/page-flags.h>
  33 #include <linux/srcu.h>
  34 #include <linux/miscdevice.h>
  35 #include <linux/debugfs.h>
  36 #include <linux/gfp.h>
  37 #include <linux/vmalloc.h>
  38 #include <linux/highmem.h>
  39 #include <linux/hugetlb.h>
  40 #include <linux/kvm_irqfd.h>
  41 #include <linux/irqbypass.h>
  42 #include <linux/module.h>
  43 #include <linux/compiler.h>
  44 #include <linux/of.h>
  45
  46 #include <asm/ftrace.h>
  47 #include <asm/reg.h>
  48 #include <asm/ppc-opcode.h>
  49 #include <asm/asm-prototypes.h>
  50 #include <asm/archrandom.h>
  51 #include <asm/debug.h>
  52 #include <asm/disassemble.h>
  53 #include <asm/cputable.h>
  54 #include <asm/cacheflush.h>
  55 #include <linux/uaccess.h>
  56 #include <asm/io.h>
  57 #include <asm/kvm_ppc.h>
  58 #include <asm/kvm_book3s.h>
  59 #include <asm/mmu_context.h>
  60 #include <asm/lppaca.h>
  61 #include <asm/processor.h>
  62 #include <asm/cputhreads.h>
  63 #include <asm/page.h>
  64 #include <asm/hvcall.h>
  65 #include <asm/switch_to.h>
  66 #include <asm/smp.h>
  67 #include <asm/dbell.h>
  68 #include <asm/hmi.h>
  69 #include <asm/pnv-pci.h>
  70 #include <asm/mmu.h>
  71 #include <asm/opal.h>
  72 #include <asm/xics.h>
  73 #include <asm/xive.h>
  74 #include <asm/hw_breakpoint.h>
  75 #include <asm/kvm_host.h>
  76 #include <asm/kvm_book3s_uvmem.h>
  77 #include <asm/ultravisor.h>
  78
  79 #include "book3s.h"
  80
  81 #define CREATE_TRACE_POINTS
  82 #include "trace_hv.h"
  83
  84 /* #define EXIT_DEBUG */
  85 /* #define EXIT_DEBUG_SIMPLE */
  86 /* #define EXIT_DEBUG_INT */
  87
  88 /* Used to indicate that a guest page fault needs to be handled */
  89 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  90 /* Used to indicate that a guest passthrough interrupt needs to be handled */
  91 #define RESUME_PASSTHROUGH      (RESUME_GUEST | RESUME_FLAG_ARCH2)
  92
  93 /* Used as a "null" value for timebase values */
  94 #define TB_NIL  (~(u64)0)
  95
  96 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
  97
  98 static int dynamic_mt_modes = 6;
  99 module_param(dynamic_mt_modes, int, 0644);
 100 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
 101 static int target_smt_mode;
 102 module_param(target_smt_mode, int, 0644);
 103 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
 104
 105 static bool indep_threads_mode = true;
 106 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
 107 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
 108
 109 static bool one_vm_per_core;
 110 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
 111 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)");
 112
 113 #ifdef CONFIG_KVM_XICS
 114 static struct kernel_param_ops module_param_ops = {
 115         .set = param_set_int,
 116         .get = param_get_int,
 117 };
 118
 119 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
 120 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
 121
 122 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
 123 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
 124 #endif
 125
 126 /* If set, guests are allowed to create and control nested guests */
 127 static bool nested = true;
 128 module_param(nested, bool, S_IRUGO | S_IWUSR);
 129 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
 130
 131 static inline bool nesting_enabled(struct kvm *kvm)
 132 {
 133         return kvm->arch.nested_enable && kvm_is_radix(kvm);
 134 }
 135
 136 /* If set, the threads on each CPU core have to be in the same MMU mode */
 137 static bool no_mixing_hpt_and_radix;
 138
 139 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
 140
 141 /*
 142  * RWMR values for POWER8.  These control the rate at which PURR
 143  * and SPURR count and should be set according to the number of
 144  * online threads in the vcore being run.
 145  */
 146 #define RWMR_RPA_P8_1THREAD     0x164520C62609AECAUL
 147 #define RWMR_RPA_P8_2THREAD     0x7FFF2908450D8DA9UL
 148 #define RWMR_RPA_P8_3THREAD     0x164520C62609AECAUL
 149 #define RWMR_RPA_P8_4THREAD     0x199A421245058DA9UL
 150 #define RWMR_RPA_P8_5THREAD     0x164520C62609AECAUL
 151 #define RWMR_RPA_P8_6THREAD     0x164520C62609AECAUL
 152 #define RWMR_RPA_P8_7THREAD     0x164520C62609AECAUL
 153 #define RWMR_RPA_P8_8THREAD     0x164520C62609AECAUL
 154
 155 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
 156         RWMR_RPA_P8_1THREAD,
 157         RWMR_RPA_P8_1THREAD,
 158         RWMR_RPA_P8_2THREAD,
 159         RWMR_RPA_P8_3THREAD,
 160         RWMR_RPA_P8_4THREAD,
 161         RWMR_RPA_P8_5THREAD,
 162         RWMR_RPA_P8_6THREAD,
 163         RWMR_RPA_P8_7THREAD,
 164         RWMR_RPA_P8_8THREAD,
 165 };
 166
 167 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
 168                 int *ip)
 169 {
 170         int i = *ip;
 171         struct kvm_vcpu *vcpu;
 172
 173         while (++i < MAX_SMT_THREADS) {
 174                 vcpu = READ_ONCE(vc->runnable_threads[i]);
 175                 if (vcpu) {
 176                         *ip = i;
 177                         return vcpu;
 178                 }
 179         }
 180         return NULL;
 181 }
 182
 183 /* Used to traverse the list of runnable threads for a given vcore */
 184 #define for_each_runnable_thread(i, vcpu, vc) \
 185         for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
 186
 187 static bool kvmppc_ipi_thread(int cpu)
 188 {
 189         unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
 190
 191         /* If we're a nested hypervisor, fall back to ordinary IPIs for now */
 192         if (kvmhv_on_pseries())
 193                 return false;
 194
 195         /* On POWER9 we can use msgsnd to IPI any cpu */
 196         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
 197                 msg |= get_hard_smp_processor_id(cpu);
 198                 smp_mb();
 199                 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 200                 return true;
 201         }
 202
 203         /* On POWER8 for IPIs to threads in the same core, use msgsnd */
 204         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
 205                 preempt_disable();
 206                 if (cpu_first_thread_sibling(cpu) ==
 207                     cpu_first_thread_sibling(smp_processor_id())) {
 208                         msg |= cpu_thread_in_core(cpu);
 209                         smp_mb();
 210                         __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 211                         preempt_enable();
 212                         return true;
 213                 }
 214                 preempt_enable();
 215         }
 216
 217 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
 218         if (cpu >= 0 && cpu < nr_cpu_ids) {
 219                 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
 220                         xics_wake_cpu(cpu);
 221                         return true;
 222                 }
 223                 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
 224                 return true;
 225         }
 226 #endif
 227
 228         return false;
 229 }
 230
 231 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
 232 {
 233         int cpu;
 234         struct swait_queue_head *wqp;
 235
 236         wqp = kvm_arch_vcpu_wq(vcpu);
 237         if (swq_has_sleeper(wqp)) {
 238                 swake_up_one(wqp);
 239                 ++vcpu->stat.halt_wakeup;
 240         }
 241
 242         cpu = READ_ONCE(vcpu->arch.thread_cpu);
 243         if (cpu >= 0 && kvmppc_ipi_thread(cpu))
 244                 return;
 245
 246         /* CPU points to the first thread of the core */
 247         cpu = vcpu->cpu;
 248         if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
 249                 smp_send_reschedule(cpu);
 250 }
 251
 252 /*
 253  * We use the vcpu_load/put functions to measure stolen time.
 254  * Stolen time is counted as time when either the vcpu is able to
 255  * run as part of a virtual core, but the task running the vcore
 256  * is preempted or sleeping, or when the vcpu needs something done
 257  * in the kernel by the task running the vcpu, but that task is
 258  * preempted or sleeping.  Those two things have to be counted
 259  * separately, since one of the vcpu tasks will take on the job
 260  * of running the core, and the other vcpu tasks in the vcore will
 261  * sleep waiting for it to do that, but that sleep shouldn't count
 262  * as stolen time.
 263  *
 264  * Hence we accumulate stolen time when the vcpu can run as part of
 265  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 266  * needs its task to do other things in the kernel (for example,
 267  * service a page fault) in busy_stolen.  We don't accumulate
 268  * stolen time for a vcore when it is inactive, or for a vcpu
 269  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 270  * a misnomer; it means that the vcpu task is not executing in
 271  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 272  * the kernel.  We don't have any way of dividing up that time
 273  * between time that the vcpu is genuinely stopped, time that
 274  * the task is actively working on behalf of the vcpu, and time
 275  * that the task is preempted, so we don't count any of it as
 276  * stolen.
 277  *
 278  * Updates to busy_stolen are protected by arch.tbacct_lock;
 279  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
 280  * lock.  The stolen times are measured in units of timebase ticks.
 281  * (Note that the != TB_NIL checks below are purely defensive;
 282  * they should never fail.)
 283  */
 284
 285 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
 286 {
 287         unsigned long flags;
 288
 289         spin_lock_irqsave(&vc->stoltb_lock, flags);
 290         vc->preempt_tb = mftb();
 291         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 292 }
 293
 294 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
 295 {
 296         unsigned long flags;
 297
 298         spin_lock_irqsave(&vc->stoltb_lock, flags);
 299         if (vc->preempt_tb != TB_NIL) {
 300                 vc->stolen_tb += mftb() - vc->preempt_tb;
 301                 vc->preempt_tb = TB_NIL;
 302         }
 303         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 304 }
 305
 306 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 307 {
 308         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 309         unsigned long flags;
 310
 311         /*
 312          * We can test vc->runner without taking the vcore lock,
 313          * because only this task ever sets vc->runner to this
 314          * vcpu, and once it is set to this vcpu, only this task
 315          * ever sets it to NULL.
 316          */
 317         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 318                 kvmppc_core_end_stolen(vc);
 319
 320         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 321         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 322             vcpu->arch.busy_preempt != TB_NIL) {
 323                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 324                 vcpu->arch.busy_preempt = TB_NIL;
 325         }
 326         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 327 }
 328
 329 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 330 {
 331         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 332         unsigned long flags;
 333
 334         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 335                 kvmppc_core_start_stolen(vc);
 336
 337         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 338         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 339                 vcpu->arch.busy_preempt = mftb();
 340         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 341 }
 342
 343 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 344 {
 345         vcpu->arch.pvr = pvr;
 346 }
 347
 348 /* Dummy value used in computing PCR value below */
 349 #define PCR_ARCH_300    (PCR_ARCH_207 << 1)
 350
 351 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 352 {
 353         unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
 354         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 355
 356         /* We can (emulate) our own architecture version and anything older */
 357         if (cpu_has_feature(CPU_FTR_ARCH_300))
 358                 host_pcr_bit = PCR_ARCH_300;
 359         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
 360                 host_pcr_bit = PCR_ARCH_207;
 361         else if (cpu_has_feature(CPU_FTR_ARCH_206))
 362                 host_pcr_bit = PCR_ARCH_206;
 363         else
 364                 host_pcr_bit = PCR_ARCH_205;
 365
 366         /* Determine lowest PCR bit needed to run guest in given PVR level */
 367         guest_pcr_bit = host_pcr_bit;
 368         if (arch_compat) {
 369                 switch (arch_compat) {
 370                 case PVR_ARCH_205:
 371                         guest_pcr_bit = PCR_ARCH_205;
 372                         break;
 373                 case PVR_ARCH_206:
 374                 case PVR_ARCH_206p:
 375                         guest_pcr_bit = PCR_ARCH_206;
 376                         break;
 377                 case PVR_ARCH_207:
 378                         guest_pcr_bit = PCR_ARCH_207;
 379                         break;
 380                 case PVR_ARCH_300:
 381                         guest_pcr_bit = PCR_ARCH_300;
 382                         break;
 383                 default:
 384                         return -EINVAL;
 385                 }
 386         }
 387
 388         /* Check requested PCR bits don't exceed our capabilities */
 389         if (guest_pcr_bit > host_pcr_bit)
 390                 return -EINVAL;
 391
 392         spin_lock(&vc->lock);
 393         vc->arch_compat = arch_compat;
 394         /*
 395          * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
 396          * Also set all reserved PCR bits
 397          */
 398         vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
 399         spin_unlock(&vc->lock);
 400
 401         return 0;
 402 }
 403
 404 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 405 {
 406         int r;
 407
 408         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 409         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 410                vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
 411         for (r = 0; r < 16; ++r)
 412                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 413                        r, kvmppc_get_gpr(vcpu, r),
 414                        r+16, kvmppc_get_gpr(vcpu, r+16));
 415         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 416                vcpu->arch.regs.ctr, vcpu->arch.regs.link);
 417         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 418                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 419         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 420                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 421         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 422                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 423         pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
 424                vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
 425         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 426         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 427                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 428         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 429         for (r = 0; r < vcpu->arch.slb_max; ++r)
 430                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 431                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 432         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 433                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 434                vcpu->arch.last_inst);
 435 }
 436
 437 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 438 {
 439         return kvm_get_vcpu_by_id(kvm, id);
 440 }
 441
 442 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 443 {
 444         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 445         vpa->yield_count = cpu_to_be32(1);
 446 }
 447
 448 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 449                    unsigned long addr, unsigned long len)
 450 {
 451         /* check address is cacheline aligned */
 452         if (addr & (L1_CACHE_BYTES - 1))
 453                 return -EINVAL;
 454         spin_lock(&vcpu->arch.vpa_update_lock);
 455         if (v->next_gpa != addr || v->len != len) {
 456                 v->next_gpa = addr;
 457                 v->len = addr ? len : 0;
 458                 v->update_pending = 1;
 459         }
 460         spin_unlock(&vcpu->arch.vpa_update_lock);
 461         return 0;
 462 }
 463
 464 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 465 struct reg_vpa {
 466         u32 dummy;
 467         union {
 468                 __be16 hword;
 469                 __be32 word;
 470         } length;
 471 };
 472
 473 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 474 {
 475         if (vpap->update_pending)
 476                 return vpap->next_gpa != 0;
 477         return vpap->pinned_addr != NULL;
 478 }
 479
 480 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 481                                        unsigned long flags,
 482                                        unsigned long vcpuid, unsigned long vpa)
 483 {
 484         struct kvm *kvm = vcpu->kvm;
 485         unsigned long len, nb;
 486         void *va;
 487         struct kvm_vcpu *tvcpu;
 488         int err;
 489         int subfunc;
 490         struct kvmppc_vpa *vpap;
 491
 492         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 493         if (!tvcpu)
 494                 return H_PARAMETER;
 495
 496         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 497         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 498             subfunc == H_VPA_REG_SLB) {
 499                 /* Registering new area - address must be cache-line aligned */
 500                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 501                         return H_PARAMETER;
 502
 503                 /* convert logical addr to kernel addr and read length */
 504                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 505                 if (va == NULL)
 506                         return H_PARAMETER;
 507                 if (subfunc == H_VPA_REG_VPA)
 508                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 509                 else
 510                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 511                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 512
 513                 /* Check length */
 514                 if (len > nb || len < sizeof(struct reg_vpa))
 515                         return H_PARAMETER;
 516         } else {
 517                 vpa = 0;
 518                 len = 0;
 519         }
 520
 521         err = H_PARAMETER;
 522         vpap = NULL;
 523         spin_lock(&tvcpu->arch.vpa_update_lock);
 524
 525         switch (subfunc) {
 526         case H_VPA_REG_VPA:             /* register VPA */
 527                 /*
 528                  * The size of our lppaca is 1kB because of the way we align
 529                  * it for the guest to avoid crossing a 4kB boundary. We only
 530                  * use 640 bytes of the structure though, so we should accept
 531                  * clients that set a size of 640.
 532                  */
 533                 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
 534                 if (len < sizeof(struct lppaca))
 535                         break;
 536                 vpap = &tvcpu->arch.vpa;
 537                 err = 0;
 538                 break;
 539
 540         case H_VPA_REG_DTL:             /* register DTL */
 541                 if (len < sizeof(struct dtl_entry))
 542                         break;
 543                 len -= len % sizeof(struct dtl_entry);
 544
 545                 /* Check that they have previously registered a VPA */
 546                 err = H_RESOURCE;
 547                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 548                         break;
 549
 550                 vpap = &tvcpu->arch.dtl;
 551                 err = 0;
 552                 break;
 553
 554         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 555                 /* Check that they have previously registered a VPA */
 556                 err = H_RESOURCE;
 557                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 558                         break;
 559
 560                 vpap = &tvcpu->arch.slb_shadow;
 561                 err = 0;
 562                 break;
 563
 564         case H_VPA_DEREG_VPA:           /* deregister VPA */
 565                 /* Check they don't still have a DTL or SLB buf registered */
 566                 err = H_RESOURCE;
 567                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 568                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 569                         break;
 570
 571                 vpap = &tvcpu->arch.vpa;
 572                 err = 0;
 573                 break;
 574
 575         case H_VPA_DEREG_DTL:           /* deregister DTL */
 576                 vpap = &tvcpu->arch.dtl;
 577                 err = 0;
 578                 break;
 579
 580         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 581                 vpap = &tvcpu->arch.slb_shadow;
 582                 err = 0;
 583                 break;
 584         }
 585
 586         if (vpap) {
 587                 vpap->next_gpa = vpa;
 588                 vpap->len = len;
 589                 vpap->update_pending = 1;
 590         }
 591
 592         spin_unlock(&tvcpu->arch.vpa_update_lock);
 593
 594         return err;
 595 }
 596
 597 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 598 {
 599         struct kvm *kvm = vcpu->kvm;
 600         void *va;
 601         unsigned long nb;
 602         unsigned long gpa;
 603
 604         /*
 605          * We need to pin the page pointed to by vpap->next_gpa,
 606          * but we can't call kvmppc_pin_guest_page under the lock
 607          * as it does get_user_pages() and down_read().  So we
 608          * have to drop the lock, pin the page, then get the lock
 609          * again and check that a new area didn't get registered
 610          * in the meantime.
 611          */
 612         for (;;) {
 613                 gpa = vpap->next_gpa;
 614                 spin_unlock(&vcpu->arch.vpa_update_lock);
 615                 va = NULL;
 616                 nb = 0;
 617                 if (gpa)
 618                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 619                 spin_lock(&vcpu->arch.vpa_update_lock);
 620                 if (gpa == vpap->next_gpa)
 621                         break;
 622                 /* sigh... unpin that one and try again */
 623                 if (va)
 624                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 625         }
 626
 627         vpap->update_pending = 0;
 628         if (va && nb < vpap->len) {
 629                 /*
 630                  * If it's now too short, it must be that userspace
 631                  * has changed the mappings underlying guest memory,
 632                  * so unregister the region.
 633                  */
 634                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 635                 va = NULL;
 636         }
 637         if (vpap->pinned_addr)
 638                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 639                                         vpap->dirty);
 640         vpap->gpa = gpa;
 641         vpap->pinned_addr = va;
 642         vpap->dirty = false;
 643         if (va)
 644                 vpap->pinned_end = va + vpap->len;
 645 }
 646
 647 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 648 {
 649         if (!(vcpu->arch.vpa.update_pending ||
 650               vcpu->arch.slb_shadow.update_pending ||
 651               vcpu->arch.dtl.update_pending))
 652                 return;
 653
 654         spin_lock(&vcpu->arch.vpa_update_lock);
 655         if (vcpu->arch.vpa.update_pending) {
 656                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 657                 if (vcpu->arch.vpa.pinned_addr)
 658                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 659         }
 660         if (vcpu->arch.dtl.update_pending) {
 661                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 662                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 663                 vcpu->arch.dtl_index = 0;
 664         }
 665         if (vcpu->arch.slb_shadow.update_pending)
 666                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 667         spin_unlock(&vcpu->arch.vpa_update_lock);
 668 }
 669
 670 /*
 671  * Return the accumulated stolen time for the vcore up until `now'.
 672  * The caller should hold the vcore lock.
 673  */
 674 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 675 {
 676         u64 p;
 677         unsigned long flags;
 678
 679         spin_lock_irqsave(&vc->stoltb_lock, flags);
 680         p = vc->stolen_tb;
 681         if (vc->vcore_state != VCORE_INACTIVE &&
 682             vc->preempt_tb != TB_NIL)
 683                 p += now - vc->preempt_tb;
 684         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 685         return p;
 686 }
 687
 688 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 689                                     struct kvmppc_vcore *vc)
 690 {
 691         struct dtl_entry *dt;
 692         struct lppaca *vpa;
 693         unsigned long stolen;
 694         unsigned long core_stolen;
 695         u64 now;
 696         unsigned long flags;
 697
 698         dt = vcpu->arch.dtl_ptr;
 699         vpa = vcpu->arch.vpa.pinned_addr;
 700         now = mftb();
 701         core_stolen = vcore_stolen_time(vc, now);
 702         stolen = core_stolen - vcpu->arch.stolen_logged;
 703         vcpu->arch.stolen_logged = core_stolen;
 704         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 705         stolen += vcpu->arch.busy_stolen;
 706         vcpu->arch.busy_stolen = 0;
 707         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 708         if (!dt || !vpa)
 709                 return;
 710         memset(dt, 0, sizeof(struct dtl_entry));
 711         dt->dispatch_reason = 7;
 712         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 713         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 714         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 715         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 716         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 717         ++dt;
 718         if (dt == vcpu->arch.dtl.pinned_end)
 719                 dt = vcpu->arch.dtl.pinned_addr;
 720         vcpu->arch.dtl_ptr = dt;
 721         /* order writing *dt vs. writing vpa->dtl_idx */
 722         smp_wmb();
 723         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 724         vcpu->arch.dtl.dirty = true;
 725 }
 726
 727 /* See if there is a doorbell interrupt pending for a vcpu */
 728 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
 729 {
 730         int thr;
 731         struct kvmppc_vcore *vc;
 732
 733         if (vcpu->arch.doorbell_request)
 734                 return true;
 735         /*
 736          * Ensure that the read of vcore->dpdes comes after the read
 737          * of vcpu->doorbell_request.  This barrier matches the
 738          * smp_wmb() in kvmppc_guest_entry_inject().
 739          */
 740         smp_rmb();
 741         vc = vcpu->arch.vcore;
 742         thr = vcpu->vcpu_id - vc->first_vcpuid;
 743         return !!(vc->dpdes & (1 << thr));
 744 }
 745
 746 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 747 {
 748         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 749                 return true;
 750         if ((!vcpu->arch.vcore->arch_compat) &&
 751             cpu_has_feature(CPU_FTR_ARCH_207S))
 752                 return true;
 753         return false;
 754 }
 755
 756 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 757                              unsigned long resource, unsigned long value1,
 758                              unsigned long value2)
 759 {
 760         switch (resource) {
 761         case H_SET_MODE_RESOURCE_SET_CIABR:
 762                 if (!kvmppc_power8_compatible(vcpu))
 763                         return H_P2;
 764                 if (value2)
 765                         return H_P4;
 766                 if (mflags)
 767                         return H_UNSUPPORTED_FLAG_START;
 768                 /* Guests can't breakpoint the hypervisor */
 769                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 770                         return H_P3;
 771                 vcpu->arch.ciabr  = value1;
 772                 return H_SUCCESS;
 773         case H_SET_MODE_RESOURCE_SET_DAWR:
 774                 if (!kvmppc_power8_compatible(vcpu))
 775                         return H_P2;
 776                 if (!ppc_breakpoint_available())
 777                         return H_P2;
 778                 if (mflags)
 779                         return H_UNSUPPORTED_FLAG_START;
 780                 if (value2 & DABRX_HYP)
 781                         return H_P4;
 782                 vcpu->arch.dawr  = value1;
 783                 vcpu->arch.dawrx = value2;
 784                 return H_SUCCESS;
 785         case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
 786                 /* KVM does not support mflags=2 (AIL=2) */
 787                 if (mflags != 0 && mflags != 3)
 788                         return H_UNSUPPORTED_FLAG_START;
 789                 return H_TOO_HARD;
 790         default:
 791                 return H_TOO_HARD;
 792         }
 793 }
 794
 795 /* Copy guest memory in place - must reside within a single memslot */
 796 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
 797                                   unsigned long len)
 798 {
 799         struct kvm_memory_slot *to_memslot = NULL;
 800         struct kvm_memory_slot *from_memslot = NULL;
 801         unsigned long to_addr, from_addr;
 802         int r;
 803
 804         /* Get HPA for from address */
 805         from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
 806         if (!from_memslot)
 807                 return -EFAULT;
 808         if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
 809                              << PAGE_SHIFT))
 810                 return -EINVAL;
 811         from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
 812         if (kvm_is_error_hva(from_addr))
 813                 return -EFAULT;
 814         from_addr |= (from & (PAGE_SIZE - 1));
 815
 816         /* Get HPA for to address */
 817         to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
 818         if (!to_memslot)
 819                 return -EFAULT;
 820         if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
 821                            << PAGE_SHIFT))
 822                 return -EINVAL;
 823         to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
 824         if (kvm_is_error_hva(to_addr))
 825                 return -EFAULT;
 826         to_addr |= (to & (PAGE_SIZE - 1));
 827
 828         /* Perform copy */
 829         r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
 830                              len);
 831         if (r)
 832                 return -EFAULT;
 833         mark_page_dirty(kvm, to >> PAGE_SHIFT);
 834         return 0;
 835 }
 836
 837 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
 838                                unsigned long dest, unsigned long src)
 839 {
 840         u64 pg_sz = SZ_4K;              /* 4K page size */
 841         u64 pg_mask = SZ_4K - 1;
 842         int ret;
 843
 844         /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
 845         if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
 846                       H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
 847                 return H_PARAMETER;
 848
 849         /* dest (and src if copy_page flag set) must be page aligned */
 850         if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
 851                 return H_PARAMETER;
 852
 853         /* zero and/or copy the page as determined by the flags */
 854         if (flags & H_COPY_PAGE) {
 855                 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
 856                 if (ret < 0)
 857                         return H_PARAMETER;
 858         } else if (flags & H_ZERO_PAGE) {
 859                 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
 860                 if (ret < 0)
 861                         return H_PARAMETER;
 862         }
 863
 864         /* We can ignore the remaining flags */
 865
 866         return H_SUCCESS;
 867 }
 868
 869 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 870 {
 871         struct kvmppc_vcore *vcore = target->arch.vcore;
 872
 873         /*
 874          * We expect to have been called by the real mode handler
 875          * (kvmppc_rm_h_confer()) which would have directly returned
 876          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 877          * have useful work to do and should not confer) so we don't
 878          * recheck that here.
 879          */
 880
 881         spin_lock(&vcore->lock);
 882         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 883             vcore->vcore_state != VCORE_INACTIVE &&
 884             vcore->runner)
 885                 target = vcore->runner;
 886         spin_unlock(&vcore->lock);
 887
 888         return kvm_vcpu_yield_to(target);
 889 }
 890
 891 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 892 {
 893         int yield_count = 0;
 894         struct lppaca *lppaca;
 895
 896         spin_lock(&vcpu->arch.vpa_update_lock);
 897         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 898         if (lppaca)
 899                 yield_count = be32_to_cpu(lppaca->yield_count);
 900         spin_unlock(&vcpu->arch.vpa_update_lock);
 901         return yield_count;
 902 }
 903
 904 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 905 {
 906         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 907         unsigned long target, ret = H_SUCCESS;
 908         int yield_count;
 909         struct kvm_vcpu *tvcpu;
 910         int idx, rc;
 911
 912         if (req <= MAX_HCALL_OPCODE &&
 913             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 914                 return RESUME_HOST;
 915
 916         switch (req) {
 917         case H_CEDE:
 918                 break;
 919         case H_PROD:
 920                 target = kvmppc_get_gpr(vcpu, 4);
 921                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 922                 if (!tvcpu) {
 923                         ret = H_PARAMETER;
 924                         break;
 925                 }
 926                 tvcpu->arch.prodded = 1;
 927                 smp_mb();
 928                 if (tvcpu->arch.ceded)
 929                         kvmppc_fast_vcpu_kick_hv(tvcpu);
 930                 break;
 931         case H_CONFER:
 932                 target = kvmppc_get_gpr(vcpu, 4);
 933                 if (target == -1)
 934                         break;
 935                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 936                 if (!tvcpu) {
 937                         ret = H_PARAMETER;
 938                         break;
 939                 }
 940                 yield_count = kvmppc_get_gpr(vcpu, 5);
 941                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 942                         break;
 943                 kvm_arch_vcpu_yield_to(tvcpu);
 944                 break;
 945         case H_REGISTER_VPA:
 946                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 947                                         kvmppc_get_gpr(vcpu, 5),
 948                                         kvmppc_get_gpr(vcpu, 6));
 949                 break;
 950         case H_RTAS:
 951                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 952                         return RESUME_HOST;
 953
 954                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 955                 rc = kvmppc_rtas_hcall(vcpu);
 956                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 957
 958                 if (rc == -ENOENT)
 959                         return RESUME_HOST;
 960                 else if (rc == 0)
 961                         break;
 962
 963                 /* Send the error out to userspace via KVM_RUN */
 964                 return rc;
 965         case H_LOGICAL_CI_LOAD:
 966                 ret = kvmppc_h_logical_ci_load(vcpu);
 967                 if (ret == H_TOO_HARD)
 968                         return RESUME_HOST;
 969                 break;
 970         case H_LOGICAL_CI_STORE:
 971                 ret = kvmppc_h_logical_ci_store(vcpu);
 972                 if (ret == H_TOO_HARD)
 973                         return RESUME_HOST;
 974                 break;
 975         case H_SET_MODE:
 976                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 977                                         kvmppc_get_gpr(vcpu, 5),
 978                                         kvmppc_get_gpr(vcpu, 6),
 979                                         kvmppc_get_gpr(vcpu, 7));
 980                 if (ret == H_TOO_HARD)
 981                         return RESUME_HOST;
 982                 break;
 983         case H_XIRR:
 984         case H_CPPR:
 985         case H_EOI:
 986         case H_IPI:
 987         case H_IPOLL:
 988         case H_XIRR_X:
 989                 if (kvmppc_xics_enabled(vcpu)) {
 990                         if (xics_on_xive()) {
 991                                 ret = H_NOT_AVAILABLE;
 992                                 return RESUME_GUEST;
 993                         }
 994                         ret = kvmppc_xics_hcall(vcpu, req);
 995                         break;
 996                 }
 997                 return RESUME_HOST;
 998         case H_SET_DABR:
 999                 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1000                 break;
1001         case H_SET_XDABR:
1002                 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1003                                                 kvmppc_get_gpr(vcpu, 5));
1004                 break;
1005 #ifdef CONFIG_SPAPR_TCE_IOMMU
1006         case H_GET_TCE:
1007                 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1008                                                 kvmppc_get_gpr(vcpu, 5));
1009                 if (ret == H_TOO_HARD)
1010                         return RESUME_HOST;
1011                 break;
1012         case H_PUT_TCE:
1013                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1014                                                 kvmppc_get_gpr(vcpu, 5),
1015                                                 kvmppc_get_gpr(vcpu, 6));
1016                 if (ret == H_TOO_HARD)
1017                         return RESUME_HOST;
1018                 break;
1019         case H_PUT_TCE_INDIRECT:
1020                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1021                                                 kvmppc_get_gpr(vcpu, 5),
1022                                                 kvmppc_get_gpr(vcpu, 6),
1023                                                 kvmppc_get_gpr(vcpu, 7));
1024                 if (ret == H_TOO_HARD)
1025                         return RESUME_HOST;
1026                 break;
1027         case H_STUFF_TCE:
1028                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1029                                                 kvmppc_get_gpr(vcpu, 5),
1030                                                 kvmppc_get_gpr(vcpu, 6),
1031                                                 kvmppc_get_gpr(vcpu, 7));
1032                 if (ret == H_TOO_HARD)
1033                         return RESUME_HOST;
1034                 break;
1035 #endif
1036         case H_RANDOM:
1037                 if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1038                         ret = H_HARDWARE;
1039                 break;
1040
1041         case H_SET_PARTITION_TABLE:
1042                 ret = H_FUNCTION;
1043                 if (nesting_enabled(vcpu->kvm))
1044                         ret = kvmhv_set_partition_table(vcpu);
1045                 break;
1046         case H_ENTER_NESTED:
1047                 ret = H_FUNCTION;
1048                 if (!nesting_enabled(vcpu->kvm))
1049                         break;
1050                 ret = kvmhv_enter_nested_guest(vcpu);
1051                 if (ret == H_INTERRUPT) {
1052                         kvmppc_set_gpr(vcpu, 3, 0);
1053                         vcpu->arch.hcall_needed = 0;
1054                         return -EINTR;
1055                 } else if (ret == H_TOO_HARD) {
1056                         kvmppc_set_gpr(vcpu, 3, 0);
1057                         vcpu->arch.hcall_needed = 0;
1058                         return RESUME_HOST;
1059                 }
1060                 break;
1061         case H_TLB_INVALIDATE:
1062                 ret = H_FUNCTION;
1063                 if (nesting_enabled(vcpu->kvm))
1064                         ret = kvmhv_do_nested_tlbie(vcpu);
1065                 break;
1066         case H_COPY_TOFROM_GUEST:
1067                 ret = H_FUNCTION;
1068                 if (nesting_enabled(vcpu->kvm))
1069                         ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1070                 break;
1071         case H_PAGE_INIT:
1072                 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1073                                          kvmppc_get_gpr(vcpu, 5),
1074                                          kvmppc_get_gpr(vcpu, 6));
1075                 break;
1076         case H_SVM_PAGE_IN:
1077                 ret = kvmppc_h_svm_page_in(vcpu->kvm,
1078                                            kvmppc_get_gpr(vcpu, 4),
1079                                            kvmppc_get_gpr(vcpu, 5),
1080                                            kvmppc_get_gpr(vcpu, 6));
1081                 break;
1082         case H_SVM_PAGE_OUT:
1083                 ret = kvmppc_h_svm_page_out(vcpu->kvm,
1084                                             kvmppc_get_gpr(vcpu, 4),
1085                                             kvmppc_get_gpr(vcpu, 5),
1086                                             kvmppc_get_gpr(vcpu, 6));
1087                 break;
1088         case H_SVM_INIT_START:
1089                 ret = kvmppc_h_svm_init_start(vcpu->kvm);
1090                 break;
1091         case H_SVM_INIT_DONE:
1092                 ret = kvmppc_h_svm_init_done(vcpu->kvm);
1093                 break;
1094         case H_SVM_INIT_ABORT:
1095                 ret = kvmppc_h_svm_init_abort(vcpu->kvm);
1096                 break;
1097
1098         default:
1099                 return RESUME_HOST;
1100         }
1101         kvmppc_set_gpr(vcpu, 3, ret);
1102         vcpu->arch.hcall_needed = 0;
1103         return RESUME_GUEST;
1104 }
1105
1106 /*
1107  * Handle H_CEDE in the nested virtualization case where we haven't
1108  * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1109  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1110  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1111  */
1112 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1113 {
1114         vcpu->arch.shregs.msr |= MSR_EE;
1115         vcpu->arch.ceded = 1;
1116         smp_mb();
1117         if (vcpu->arch.prodded) {
1118                 vcpu->arch.prodded = 0;
1119                 smp_mb();
1120                 vcpu->arch.ceded = 0;
1121         }
1122 }
1123
1124 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1125 {
1126         switch (cmd) {
1127         case H_CEDE:
1128         case H_PROD:
1129         case H_CONFER:
1130         case H_REGISTER_VPA:
1131         case H_SET_MODE:
1132         case H_LOGICAL_CI_LOAD:
1133         case H_LOGICAL_CI_STORE:
1134 #ifdef CONFIG_KVM_XICS
1135         case H_XIRR:
1136         case H_CPPR:
1137         case H_EOI:
1138         case H_IPI:
1139         case H_IPOLL:
1140         case H_XIRR_X:
1141 #endif
1142         case H_PAGE_INIT:
1143                 return 1;
1144         }
1145
1146         /* See if it's in the real-mode table */
1147         return kvmppc_hcall_impl_hv_realmode(cmd);
1148 }
1149
1150 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
1151                                         struct kvm_vcpu *vcpu)
1152 {
1153         u32 last_inst;
1154
1155         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1156                                         EMULATE_DONE) {
1157                 /*
1158                  * Fetch failed, so return to guest and
1159                  * try executing it again.
1160                  */
1161                 return RESUME_GUEST;
1162         }
1163
1164         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1165                 run->exit_reason = KVM_EXIT_DEBUG;
1166                 run->debug.arch.address = kvmppc_get_pc(vcpu);
1167                 return RESUME_HOST;
1168         } else {
1169                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1170                 return RESUME_GUEST;
1171         }
1172 }
1173
1174 static void do_nothing(void *x)
1175 {
1176 }
1177
1178 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1179 {
1180         int thr, cpu, pcpu, nthreads;
1181         struct kvm_vcpu *v;
1182         unsigned long dpdes;
1183
1184         nthreads = vcpu->kvm->arch.emul_smt_mode;
1185         dpdes = 0;
1186         cpu = vcpu->vcpu_id & ~(nthreads - 1);
1187         for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1188                 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1189                 if (!v)
1190                         continue;
1191                 /*
1192                  * If the vcpu is currently running on a physical cpu thread,
1193                  * interrupt it in order to pull it out of the guest briefly,
1194                  * which will update its vcore->dpdes value.
1195                  */
1196                 pcpu = READ_ONCE(v->cpu);
1197                 if (pcpu >= 0)
1198                         smp_call_function_single(pcpu, do_nothing, NULL, 1);
1199                 if (kvmppc_doorbell_pending(v))
1200                         dpdes |= 1 << thr;
1201         }
1202         return dpdes;
1203 }
1204
1205 /*
1206  * On POWER9, emulate doorbell-related instructions in order to
1207  * give the guest the illusion of running on a multi-threaded core.
1208  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1209  * and mfspr DPDES.
1210  */
1211 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1212 {
1213         u32 inst, rb, thr;
1214         unsigned long arg;
1215         struct kvm *kvm = vcpu->kvm;
1216         struct kvm_vcpu *tvcpu;
1217
1218         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1219                 return RESUME_GUEST;
1220         if (get_op(inst) != 31)
1221                 return EMULATE_FAIL;
1222         rb = get_rb(inst);
1223         thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1224         switch (get_xop(inst)) {
1225         case OP_31_XOP_MSGSNDP:
1226                 arg = kvmppc_get_gpr(vcpu, rb);
1227                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1228                         break;
1229                 arg &= 0x3f;
1230                 if (arg >= kvm->arch.emul_smt_mode)
1231                         break;
1232                 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1233                 if (!tvcpu)
1234                         break;
1235                 if (!tvcpu->arch.doorbell_request) {
1236                         tvcpu->arch.doorbell_request = 1;
1237                         kvmppc_fast_vcpu_kick_hv(tvcpu);
1238                 }
1239                 break;
1240         case OP_31_XOP_MSGCLRP:
1241                 arg = kvmppc_get_gpr(vcpu, rb);
1242                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1243                         break;
1244                 vcpu->arch.vcore->dpdes = 0;
1245                 vcpu->arch.doorbell_request = 0;
1246                 break;
1247         case OP_31_XOP_MFSPR:
1248                 switch (get_sprn(inst)) {
1249                 case SPRN_TIR:
1250                         arg = thr;
1251                         break;
1252                 case SPRN_DPDES:
1253                         arg = kvmppc_read_dpdes(vcpu);
1254                         break;
1255                 default:
1256                         return EMULATE_FAIL;
1257                 }
1258                 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1259                 break;
1260         default:
1261                 return EMULATE_FAIL;
1262         }
1263         kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1264         return RESUME_GUEST;
1265 }
1266
1267 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1268                                  struct task_struct *tsk)
1269 {
1270         int r = RESUME_HOST;
1271
1272         vcpu->stat.sum_exits++;
1273
1274         /*
1275          * This can happen if an interrupt occurs in the last stages
1276          * of guest entry or the first stages of guest exit (i.e. after
1277          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1278          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1279          * That can happen due to a bug, or due to a machine check
1280          * occurring at just the wrong time.
1281          */
1282         if (vcpu->arch.shregs.msr & MSR_HV) {
1283                 printk(KERN_EMERG "KVM trap in HV mode!\n");
1284                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1285                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1286                         vcpu->arch.shregs.msr);
1287                 kvmppc_dump_regs(vcpu);
1288                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1289                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1290                 return RESUME_HOST;
1291         }
1292         run->exit_reason = KVM_EXIT_UNKNOWN;
1293         run->ready_for_interrupt_injection = 1;
1294         switch (vcpu->arch.trap) {
1295         /* We're good on these - the host merely wanted to get our attention */
1296         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1297                 vcpu->stat.dec_exits++;
1298                 r = RESUME_GUEST;
1299                 break;
1300         case BOOK3S_INTERRUPT_EXTERNAL:
1301         case BOOK3S_INTERRUPT_H_DOORBELL:
1302         case BOOK3S_INTERRUPT_H_VIRT:
1303                 vcpu->stat.ext_intr_exits++;
1304                 r = RESUME_GUEST;
1305                 break;
1306         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1307         case BOOK3S_INTERRUPT_HMI:
1308         case BOOK3S_INTERRUPT_PERFMON:
1309         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1310                 r = RESUME_GUEST;
1311                 break;
1312         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1313                 /* Print the MCE event to host console. */
1314                 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1315
1316                 /*
1317                  * If the guest can do FWNMI, exit to userspace so it can
1318                  * deliver a FWNMI to the guest.
1319                  * Otherwise we synthesize a machine check for the guest
1320                  * so that it knows that the machine check occurred.
1321                  */
1322                 if (!vcpu->kvm->arch.fwnmi_enabled) {
1323                         ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1324                         kvmppc_core_queue_machine_check(vcpu, flags);
1325                         r = RESUME_GUEST;
1326                         break;
1327                 }
1328
1329                 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1330                 run->exit_reason = KVM_EXIT_NMI;
1331                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1332                 /* Clear out the old NMI status from run->flags */
1333                 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1334                 /* Now set the NMI status */
1335                 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1336                         run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1337                 else
1338                         run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1339
1340                 r = RESUME_HOST;
1341                 break;
1342         case BOOK3S_INTERRUPT_PROGRAM:
1343         {
1344                 ulong flags;
1345                 /*
1346                  * Normally program interrupts are delivered directly
1347                  * to the guest by the hardware, but we can get here
1348                  * as a result of a hypervisor emulation interrupt
1349                  * (e40) getting turned into a 700 by BML RTAS.
1350                  */
1351                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1352                 kvmppc_core_queue_program(vcpu, flags);
1353                 r = RESUME_GUEST;
1354                 break;
1355         }
1356         case BOOK3S_INTERRUPT_SYSCALL:
1357         {
1358                 /* hcall - punt to userspace */
1359                 int i;
1360
1361                 /* hypercall with MSR_PR has already been handled in rmode,
1362                  * and never reaches here.
1363                  */
1364
1365                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1366                 for (i = 0; i < 9; ++i)
1367                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1368                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1369                 vcpu->arch.hcall_needed = 1;
1370                 r = RESUME_HOST;
1371                 break;
1372         }
1373         /*
1374          * We get these next two if the guest accesses a page which it thinks
1375          * it has mapped but which is not actually present, either because
1376          * it is for an emulated I/O device or because the corresonding
1377          * host page has been paged out.  Any other HDSI/HISI interrupts
1378          * have been handled already.
1379          */
1380         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1381                 r = RESUME_PAGE_FAULT;
1382                 break;
1383         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1384                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1385                 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1386                         DSISR_SRR1_MATCH_64S;
1387                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1388                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1389                 r = RESUME_PAGE_FAULT;
1390                 break;
1391         /*
1392          * This occurs if the guest executes an illegal instruction.
1393          * If the guest debug is disabled, generate a program interrupt
1394          * to the guest. If guest debug is enabled, we need to check
1395          * whether the instruction is a software breakpoint instruction.
1396          * Accordingly return to Guest or Host.
1397          */
1398         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1399                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1400                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1401                                 swab32(vcpu->arch.emul_inst) :
1402                                 vcpu->arch.emul_inst;
1403                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1404                         r = kvmppc_emulate_debug_inst(run, vcpu);
1405                 } else {
1406                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1407                         r = RESUME_GUEST;
1408                 }
1409                 break;
1410         /*
1411          * This occurs if the guest (kernel or userspace), does something that
1412          * is prohibited by HFSCR.
1413          * On POWER9, this could be a doorbell instruction that we need
1414          * to emulate.
1415          * Otherwise, we just generate a program interrupt to the guest.
1416          */
1417         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1418                 r = EMULATE_FAIL;
1419                 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1420                     cpu_has_feature(CPU_FTR_ARCH_300))
1421                         r = kvmppc_emulate_doorbell_instr(vcpu);
1422                 if (r == EMULATE_FAIL) {
1423                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1424                         r = RESUME_GUEST;
1425                 }
1426                 break;
1427
1428 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1429         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1430                 /*
1431                  * This occurs for various TM-related instructions that
1432                  * we need to emulate on POWER9 DD2.2.  We have already
1433                  * handled the cases where the guest was in real-suspend
1434                  * mode and was transitioning to transactional state.
1435                  */
1436                 r = kvmhv_p9_tm_emulation(vcpu);
1437                 break;
1438 #endif
1439
1440         case BOOK3S_INTERRUPT_HV_RM_HARD:
1441                 r = RESUME_PASSTHROUGH;
1442                 break;
1443         default:
1444                 kvmppc_dump_regs(vcpu);
1445                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1446                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1447                         vcpu->arch.shregs.msr);
1448                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1449                 r = RESUME_HOST;
1450                 break;
1451         }
1452
1453         return r;
1454 }
1455
1456 static int kvmppc_handle_nested_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1457 {
1458         int r;
1459         int srcu_idx;
1460
1461         vcpu->stat.sum_exits++;
1462
1463         /*
1464          * This can happen if an interrupt occurs in the last stages
1465          * of guest entry or the first stages of guest exit (i.e. after
1466          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1467          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1468          * That can happen due to a bug, or due to a machine check
1469          * occurring at just the wrong time.
1470          */
1471         if (vcpu->arch.shregs.msr & MSR_HV) {
1472                 pr_emerg("KVM trap in HV mode while nested!\n");
1473                 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1474                          vcpu->arch.trap, kvmppc_get_pc(vcpu),
1475                          vcpu->arch.shregs.msr);
1476                 kvmppc_dump_regs(vcpu);
1477                 return RESUME_HOST;
1478         }
1479         switch (vcpu->arch.trap) {
1480         /* We're good on these - the host merely wanted to get our attention */
1481         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1482                 vcpu->stat.dec_exits++;
1483                 r = RESUME_GUEST;
1484                 break;
1485         case BOOK3S_INTERRUPT_EXTERNAL:
1486                 vcpu->stat.ext_intr_exits++;
1487                 r = RESUME_HOST;
1488                 break;
1489         case BOOK3S_INTERRUPT_H_DOORBELL:
1490         case BOOK3S_INTERRUPT_H_VIRT:
1491                 vcpu->stat.ext_intr_exits++;
1492                 r = RESUME_GUEST;
1493                 break;
1494         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1495         case BOOK3S_INTERRUPT_HMI:
1496         case BOOK3S_INTERRUPT_PERFMON:
1497         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1498                 r = RESUME_GUEST;
1499                 break;
1500         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1501                 /* Pass the machine check to the L1 guest */
1502                 r = RESUME_HOST;
1503                 /* Print the MCE event to host console. */
1504                 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1505                 break;
1506         /*
1507          * We get these next two if the guest accesses a page which it thinks
1508          * it has mapped but which is not actually present, either because
1509          * it is for an emulated I/O device or because the corresonding
1510          * host page has been paged out.
1511          */
1512         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1513                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1514                 r = kvmhv_nested_page_fault(run, vcpu);
1515                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1516                 break;
1517         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1518                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1519                 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1520                                          DSISR_SRR1_MATCH_64S;
1521                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1522                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1523                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1524                 r = kvmhv_nested_page_fault(run, vcpu);
1525                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1526                 break;
1527
1528 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1529         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1530                 /*
1531                  * This occurs for various TM-related instructions that
1532                  * we need to emulate on POWER9 DD2.2.  We have already
1533                  * handled the cases where the guest was in real-suspend
1534                  * mode and was transitioning to transactional state.
1535                  */
1536                 r = kvmhv_p9_tm_emulation(vcpu);
1537                 break;
1538 #endif
1539
1540         case BOOK3S_INTERRUPT_HV_RM_HARD:
1541                 vcpu->arch.trap = 0;
1542                 r = RESUME_GUEST;
1543                 if (!xics_on_xive())
1544                         kvmppc_xics_rm_complete(vcpu, 0);
1545                 break;
1546         default:
1547                 r = RESUME_HOST;
1548                 break;
1549         }
1550
1551         return r;
1552 }
1553
1554 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1555                                             struct kvm_sregs *sregs)
1556 {
1557         int i;
1558
1559         memset(sregs, 0, sizeof(struct kvm_sregs));
1560         sregs->pvr = vcpu->arch.pvr;
1561         for (i = 0; i < vcpu->arch.slb_max; i++) {
1562                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1563                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1564         }
1565
1566         return 0;
1567 }
1568
1569 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1570                                             struct kvm_sregs *sregs)
1571 {
1572         int i, j;
1573
1574         /* Only accept the same PVR as the host's, since we can't spoof it */
1575         if (sregs->pvr != vcpu->arch.pvr)
1576                 return -EINVAL;
1577
1578         j = 0;
1579         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1580                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1581                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1582                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1583                         ++j;
1584                 }
1585         }
1586         vcpu->arch.slb_max = j;
1587
1588         return 0;
1589 }
1590
1591 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1592                 bool preserve_top32)
1593 {
1594         struct kvm *kvm = vcpu->kvm;
1595         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1596         u64 mask;
1597
1598         spin_lock(&vc->lock);
1599         /*
1600          * If ILE (interrupt little-endian) has changed, update the
1601          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1602          */
1603         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1604                 struct kvm_vcpu *vcpu;
1605                 int i;
1606
1607                 kvm_for_each_vcpu(i, vcpu, kvm) {
1608                         if (vcpu->arch.vcore != vc)
1609                                 continue;
1610                         if (new_lpcr & LPCR_ILE)
1611                                 vcpu->arch.intr_msr |= MSR_LE;
1612                         else
1613                                 vcpu->arch.intr_msr &= ~MSR_LE;
1614                 }
1615         }
1616
1617         /*
1618          * Userspace can only modify DPFD (default prefetch depth),
1619          * ILE (interrupt little-endian) and TC (translation control).
1620          * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1621          */
1622         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1623         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1624                 mask |= LPCR_AIL;
1625         /*
1626          * On POWER9, allow userspace to enable large decrementer for the
1627          * guest, whether or not the host has it enabled.
1628          */
1629         if (cpu_has_feature(CPU_FTR_ARCH_300))
1630                 mask |= LPCR_LD;
1631
1632         /* Broken 32-bit version of LPCR must not clear top bits */
1633         if (preserve_top32)
1634                 mask &= 0xFFFFFFFF;
1635         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1636         spin_unlock(&vc->lock);
1637 }
1638
1639 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1640                                  union kvmppc_one_reg *val)
1641 {
1642         int r = 0;
1643         long int i;
1644
1645         switch (id) {
1646         case KVM_REG_PPC_DEBUG_INST:
1647                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1648                 break;
1649         case KVM_REG_PPC_HIOR:
1650                 *val = get_reg_val(id, 0);
1651                 break;
1652         case KVM_REG_PPC_DABR:
1653                 *val = get_reg_val(id, vcpu->arch.dabr);
1654                 break;
1655         case KVM_REG_PPC_DABRX:
1656                 *val = get_reg_val(id, vcpu->arch.dabrx);
1657                 break;
1658         case KVM_REG_PPC_DSCR:
1659                 *val = get_reg_val(id, vcpu->arch.dscr);
1660                 break;
1661         case KVM_REG_PPC_PURR:
1662                 *val = get_reg_val(id, vcpu->arch.purr);
1663                 break;
1664         case KVM_REG_PPC_SPURR:
1665                 *val = get_reg_val(id, vcpu->arch.spurr);
1666                 break;
1667         case KVM_REG_PPC_AMR:
1668                 *val = get_reg_val(id, vcpu->arch.amr);
1669                 break;
1670         case KVM_REG_PPC_UAMOR:
1671                 *val = get_reg_val(id, vcpu->arch.uamor);
1672                 break;
1673         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1674                 i = id - KVM_REG_PPC_MMCR0;
1675                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1676                 break;
1677         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1678                 i = id - KVM_REG_PPC_PMC1;
1679                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1680                 break;
1681         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1682                 i = id - KVM_REG_PPC_SPMC1;
1683                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1684                 break;
1685         case KVM_REG_PPC_SIAR:
1686                 *val = get_reg_val(id, vcpu->arch.siar);
1687                 break;
1688         case KVM_REG_PPC_SDAR:
1689                 *val = get_reg_val(id, vcpu->arch.sdar);
1690                 break;
1691         case KVM_REG_PPC_SIER:
1692                 *val = get_reg_val(id, vcpu->arch.sier);
1693                 break;
1694         case KVM_REG_PPC_IAMR:
1695                 *val = get_reg_val(id, vcpu->arch.iamr);
1696                 break;
1697         case KVM_REG_PPC_PSPB:
1698                 *val = get_reg_val(id, vcpu->arch.pspb);
1699                 break;
1700         case KVM_REG_PPC_DPDES:
1701                 /*
1702                  * On POWER9, where we are emulating msgsndp etc.,
1703                  * we return 1 bit for each vcpu, which can come from
1704                  * either vcore->dpdes or doorbell_request.
1705                  * On POWER8, doorbell_request is 0.
1706                  */
1707                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes |
1708                                    vcpu->arch.doorbell_request);
1709                 break;
1710         case KVM_REG_PPC_VTB:
1711                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1712                 break;
1713         case KVM_REG_PPC_DAWR:
1714                 *val = get_reg_val(id, vcpu->arch.dawr);
1715                 break;
1716         case KVM_REG_PPC_DAWRX:
1717                 *val = get_reg_val(id, vcpu->arch.dawrx);
1718                 break;
1719         case KVM_REG_PPC_CIABR:
1720                 *val = get_reg_val(id, vcpu->arch.ciabr);
1721                 break;
1722         case KVM_REG_PPC_CSIGR:
1723                 *val = get_reg_val(id, vcpu->arch.csigr);
1724                 break;
1725         case KVM_REG_PPC_TACR:
1726                 *val = get_reg_val(id, vcpu->arch.tacr);
1727                 break;
1728         case KVM_REG_PPC_TCSCR:
1729                 *val = get_reg_val(id, vcpu->arch.tcscr);
1730                 break;
1731         case KVM_REG_PPC_PID:
1732                 *val = get_reg_val(id, vcpu->arch.pid);
1733                 break;
1734         case KVM_REG_PPC_ACOP:
1735                 *val = get_reg_val(id, vcpu->arch.acop);
1736                 break;
1737         case KVM_REG_PPC_WORT:
1738                 *val = get_reg_val(id, vcpu->arch.wort);
1739                 break;
1740         case KVM_REG_PPC_TIDR:
1741                 *val = get_reg_val(id, vcpu->arch.tid);
1742                 break;
1743         case KVM_REG_PPC_PSSCR:
1744                 *val = get_reg_val(id, vcpu->arch.psscr);
1745                 break;
1746         case KVM_REG_PPC_VPA_ADDR:
1747                 spin_lock(&vcpu->arch.vpa_update_lock);
1748                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1749                 spin_unlock(&vcpu->arch.vpa_update_lock);
1750                 break;
1751         case KVM_REG_PPC_VPA_SLB:
1752                 spin_lock(&vcpu->arch.vpa_update_lock);
1753                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1754                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1755                 spin_unlock(&vcpu->arch.vpa_update_lock);
1756                 break;
1757         case KVM_REG_PPC_VPA_DTL:
1758                 spin_lock(&vcpu->arch.vpa_update_lock);
1759                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1760                 val->vpaval.length = vcpu->arch.dtl.len;
1761                 spin_unlock(&vcpu->arch.vpa_update_lock);
1762                 break;
1763         case KVM_REG_PPC_TB_OFFSET:
1764                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1765                 break;
1766         case KVM_REG_PPC_LPCR:
1767         case KVM_REG_PPC_LPCR_64:
1768                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1769                 break;
1770         case KVM_REG_PPC_PPR:
1771                 *val = get_reg_val(id, vcpu->arch.ppr);
1772                 break;
1773 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1774         case KVM_REG_PPC_TFHAR:
1775                 *val = get_reg_val(id, vcpu->arch.tfhar);
1776                 break;
1777         case KVM_REG_PPC_TFIAR:
1778                 *val = get_reg_val(id, vcpu->arch.tfiar);
1779                 break;
1780         case KVM_REG_PPC_TEXASR:
1781                 *val = get_reg_val(id, vcpu->arch.texasr);
1782                 break;
1783         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1784                 i = id - KVM_REG_PPC_TM_GPR0;
1785                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1786                 break;
1787         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1788         {
1789                 int j;
1790                 i = id - KVM_REG_PPC_TM_VSR0;
1791                 if (i < 32)
1792                         for (j = 0; j < TS_FPRWIDTH; j++)
1793                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1794                 else {
1795                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1796                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1797                         else
1798                                 r = -ENXIO;
1799                 }
1800                 break;
1801         }
1802         case KVM_REG_PPC_TM_CR:
1803                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1804                 break;
1805         case KVM_REG_PPC_TM_XER:
1806                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1807                 break;
1808         case KVM_REG_PPC_TM_LR:
1809                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1810                 break;
1811         case KVM_REG_PPC_TM_CTR:
1812                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1813                 break;
1814         case KVM_REG_PPC_TM_FPSCR:
1815                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1816                 break;
1817         case KVM_REG_PPC_TM_AMR:
1818                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1819                 break;
1820         case KVM_REG_PPC_TM_PPR:
1821                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1822                 break;
1823         case KVM_REG_PPC_TM_VRSAVE:
1824                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1825                 break;
1826         case KVM_REG_PPC_TM_VSCR:
1827                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1828                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1829                 else
1830                         r = -ENXIO;
1831                 break;
1832         case KVM_REG_PPC_TM_DSCR:
1833                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1834                 break;
1835         case KVM_REG_PPC_TM_TAR:
1836                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1837                 break;
1838 #endif
1839         case KVM_REG_PPC_ARCH_COMPAT:
1840                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1841                 break;
1842         case KVM_REG_PPC_DEC_EXPIRY:
1843                 *val = get_reg_val(id, vcpu->arch.dec_expires +
1844                                    vcpu->arch.vcore->tb_offset);
1845                 break;
1846         case KVM_REG_PPC_ONLINE:
1847                 *val = get_reg_val(id, vcpu->arch.online);
1848                 break;
1849         case KVM_REG_PPC_PTCR:
1850                 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1851                 break;
1852         default:
1853                 r = -EINVAL;
1854                 break;
1855         }
1856
1857         return r;
1858 }
1859
1860 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1861                                  union kvmppc_one_reg *val)
1862 {
1863         int r = 0;
1864         long int i;
1865         unsigned long addr, len;
1866
1867         switch (id) {
1868         case KVM_REG_PPC_HIOR:
1869                 /* Only allow this to be set to zero */
1870                 if (set_reg_val(id, *val))
1871                         r = -EINVAL;
1872                 break;
1873         case KVM_REG_PPC_DABR:
1874                 vcpu->arch.dabr = set_reg_val(id, *val);
1875                 break;
1876         case KVM_REG_PPC_DABRX:
1877                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1878                 break;
1879         case KVM_REG_PPC_DSCR:
1880                 vcpu->arch.dscr = set_reg_val(id, *val);
1881                 break;
1882         case KVM_REG_PPC_PURR:
1883                 vcpu->arch.purr = set_reg_val(id, *val);
1884                 break;
1885         case KVM_REG_PPC_SPURR:
1886                 vcpu->arch.spurr = set_reg_val(id, *val);
1887                 break;
1888         case KVM_REG_PPC_AMR:
1889                 vcpu->arch.amr = set_reg_val(id, *val);
1890                 break;
1891         case KVM_REG_PPC_UAMOR:
1892                 vcpu->arch.uamor = set_reg_val(id, *val);
1893                 break;
1894         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1895                 i = id - KVM_REG_PPC_MMCR0;
1896                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1897                 break;
1898         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1899                 i = id - KVM_REG_PPC_PMC1;
1900                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1901                 break;
1902         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1903                 i = id - KVM_REG_PPC_SPMC1;
1904                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1905                 break;
1906         case KVM_REG_PPC_SIAR:
1907                 vcpu->arch.siar = set_reg_val(id, *val);
1908                 break;
1909         case KVM_REG_PPC_SDAR:
1910                 vcpu->arch.sdar = set_reg_val(id, *val);
1911                 break;
1912         case KVM_REG_PPC_SIER:
1913                 vcpu->arch.sier = set_reg_val(id, *val);
1914                 break;
1915         case KVM_REG_PPC_IAMR:
1916                 vcpu->arch.iamr = set_reg_val(id, *val);
1917                 break;
1918         case KVM_REG_PPC_PSPB:
1919                 vcpu->arch.pspb = set_reg_val(id, *val);
1920                 break;
1921         case KVM_REG_PPC_DPDES:
1922                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1923                 break;
1924         case KVM_REG_PPC_VTB:
1925                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1926                 break;
1927         case KVM_REG_PPC_DAWR:
1928                 vcpu->arch.dawr = set_reg_val(id, *val);
1929                 break;
1930         case KVM_REG_PPC_DAWRX:
1931                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1932                 break;
1933         case KVM_REG_PPC_CIABR:
1934                 vcpu->arch.ciabr = set_reg_val(id, *val);
1935                 /* Don't allow setting breakpoints in hypervisor code */
1936                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1937                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1938                 break;
1939         case KVM_REG_PPC_CSIGR:
1940                 vcpu->arch.csigr = set_reg_val(id, *val);
1941                 break;
1942         case KVM_REG_PPC_TACR:
1943                 vcpu->arch.tacr = set_reg_val(id, *val);
1944                 break;
1945         case KVM_REG_PPC_TCSCR:
1946                 vcpu->arch.tcscr = set_reg_val(id, *val);
1947                 break;
1948         case KVM_REG_PPC_PID:
1949                 vcpu->arch.pid = set_reg_val(id, *val);
1950                 break;
1951         case KVM_REG_PPC_ACOP:
1952                 vcpu->arch.acop = set_reg_val(id, *val);
1953                 break;
1954         case KVM_REG_PPC_WORT:
1955                 vcpu->arch.wort = set_reg_val(id, *val);
1956                 break;
1957         case KVM_REG_PPC_TIDR:
1958                 vcpu->arch.tid = set_reg_val(id, *val);
1959                 break;
1960         case KVM_REG_PPC_PSSCR:
1961                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1962                 break;
1963         case KVM_REG_PPC_VPA_ADDR:
1964                 addr = set_reg_val(id, *val);
1965                 r = -EINVAL;
1966                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1967                               vcpu->arch.dtl.next_gpa))
1968                         break;
1969                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1970                 break;
1971         case KVM_REG_PPC_VPA_SLB:
1972                 addr = val->vpaval.addr;
1973                 len = val->vpaval.length;
1974                 r = -EINVAL;
1975                 if (addr && !vcpu->arch.vpa.next_gpa)
1976                         break;
1977                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1978                 break;
1979         case KVM_REG_PPC_VPA_DTL:
1980                 addr = val->vpaval.addr;
1981                 len = val->vpaval.length;
1982                 r = -EINVAL;
1983                 if (addr && (len < sizeof(struct dtl_entry) ||
1984                              !vcpu->arch.vpa.next_gpa))
1985                         break;
1986                 len -= len % sizeof(struct dtl_entry);
1987                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1988                 break;
1989         case KVM_REG_PPC_TB_OFFSET:
1990                 /* round up to multiple of 2^24 */
1991                 vcpu->arch.vcore->tb_offset =
1992                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1993                 break;
1994         case KVM_REG_PPC_LPCR:
1995                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1996                 break;
1997         case KVM_REG_PPC_LPCR_64:
1998                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1999                 break;
2000         case KVM_REG_PPC_PPR:
2001                 vcpu->arch.ppr = set_reg_val(id, *val);
2002                 break;
2003 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2004         case KVM_REG_PPC_TFHAR:
2005                 vcpu->arch.tfhar = set_reg_val(id, *val);
2006                 break;
2007         case KVM_REG_PPC_TFIAR:
2008                 vcpu->arch.tfiar = set_reg_val(id, *val);
2009                 break;
2010         case KVM_REG_PPC_TEXASR:
2011                 vcpu->arch.texasr = set_reg_val(id, *val);
2012                 break;
2013         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2014                 i = id - KVM_REG_PPC_TM_GPR0;
2015                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2016                 break;
2017         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2018         {
2019                 int j;
2020                 i = id - KVM_REG_PPC_TM_VSR0;
2021                 if (i < 32)
2022                         for (j = 0; j < TS_FPRWIDTH; j++)
2023                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2024                 else
2025                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
2026                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2027                         else
2028                                 r = -ENXIO;
2029                 break;
2030         }
2031         case KVM_REG_PPC_TM_CR:
2032                 vcpu->arch.cr_tm = set_reg_val(id, *val);
2033                 break;
2034         case KVM_REG_PPC_TM_XER:
2035                 vcpu->arch.xer_tm = set_reg_val(id, *val);
2036                 break;
2037         case KVM_REG_PPC_TM_LR:
2038                 vcpu->arch.lr_tm = set_reg_val(id, *val);
2039                 break;
2040         case KVM_REG_PPC_TM_CTR:
2041                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2042                 break;
2043         case KVM_REG_PPC_TM_FPSCR:
2044                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2045                 break;
2046         case KVM_REG_PPC_TM_AMR:
2047                 vcpu->arch.amr_tm = set_reg_val(id, *val);
2048                 break;
2049         case KVM_REG_PPC_TM_PPR:
2050                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2051                 break;
2052         case KVM_REG_PPC_TM_VRSAVE:
2053                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2054                 break;
2055         case KVM_REG_PPC_TM_VSCR:
2056                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2057                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2058                 else
2059                         r = - ENXIO;
2060                 break;
2061         case KVM_REG_PPC_TM_DSCR:
2062                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2063                 break;
2064         case KVM_REG_PPC_TM_TAR:
2065                 vcpu->arch.tar_tm = set_reg_val(id, *val);
2066                 break;
2067 #endif
2068         case KVM_REG_PPC_ARCH_COMPAT:
2069                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2070                 break;
2071         case KVM_REG_PPC_DEC_EXPIRY:
2072                 vcpu->arch.dec_expires = set_reg_val(id, *val) -
2073                         vcpu->arch.vcore->tb_offset;
2074                 break;
2075         case KVM_REG_PPC_ONLINE:
2076                 i = set_reg_val(id, *val);
2077                 if (i && !vcpu->arch.online)
2078                         atomic_inc(&vcpu->arch.vcore->online_count);
2079                 else if (!i && vcpu->arch.online)
2080                         atomic_dec(&vcpu->arch.vcore->online_count);
2081                 vcpu->arch.online = i;
2082                 break;
2083         case KVM_REG_PPC_PTCR:
2084                 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2085                 break;
2086         default:
2087                 r = -EINVAL;
2088                 break;
2089         }
2090
2091         return r;
2092 }
2093
2094 /*
2095  * On POWER9, threads are independent and can be in different partitions.
2096  * Therefore we consider each thread to be a subcore.
2097  * There is a restriction that all threads have to be in the same
2098  * MMU mode (radix or HPT), unfortunately, but since we only support
2099  * HPT guests on a HPT host so far, that isn't an impediment yet.
2100  */
2101 static int threads_per_vcore(struct kvm *kvm)
2102 {
2103         if (kvm->arch.threads_indep)
2104                 return 1;
2105         return threads_per_subcore;
2106 }
2107
2108 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2109 {
2110         struct kvmppc_vcore *vcore;
2111
2112         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2113
2114         if (vcore == NULL)
2115                 return NULL;
2116
2117         spin_lock_init(&vcore->lock);
2118         spin_lock_init(&vcore->stoltb_lock);
2119         init_swait_queue_head(&vcore->wq);
2120         vcore->preempt_tb = TB_NIL;
2121         vcore->lpcr = kvm->arch.lpcr;
2122         vcore->first_vcpuid = id;
2123         vcore->kvm = kvm;
2124         INIT_LIST_HEAD(&vcore->preempt_list);
2125
2126         return vcore;
2127 }
2128
2129 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2130 static struct debugfs_timings_element {
2131         const char *name;
2132         size_t offset;
2133 } timings[] = {
2134         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
2135         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
2136         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
2137         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
2138         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
2139 };
2140
2141 #define N_TIMINGS       (ARRAY_SIZE(timings))
2142
2143 struct debugfs_timings_state {
2144         struct kvm_vcpu *vcpu;
2145         unsigned int    buflen;
2146         char            buf[N_TIMINGS * 100];
2147 };
2148
2149 static int debugfs_timings_open(struct inode *inode, struct file *file)
2150 {
2151         struct kvm_vcpu *vcpu = inode->i_private;
2152         struct debugfs_timings_state *p;
2153
2154         p = kzalloc(sizeof(*p), GFP_KERNEL);
2155         if (!p)
2156                 return -ENOMEM;
2157
2158         kvm_get_kvm(vcpu->kvm);
2159         p->vcpu = vcpu;
2160         file->private_data = p;
2161
2162         return nonseekable_open(inode, file);
2163 }
2164
2165 static int debugfs_timings_release(struct inode *inode, struct file *file)
2166 {
2167         struct debugfs_timings_state *p = file->private_data;
2168
2169         kvm_put_kvm(p->vcpu->kvm);
2170         kfree(p);
2171         return 0;
2172 }
2173
2174 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2175                                     size_t len, loff_t *ppos)
2176 {
2177         struct debugfs_timings_state *p = file->private_data;
2178         struct kvm_vcpu *vcpu = p->vcpu;
2179         char *s, *buf_end;
2180         struct kvmhv_tb_accumulator tb;
2181         u64 count;
2182         loff_t pos;
2183         ssize_t n;
2184         int i, loops;
2185         bool ok;
2186
2187         if (!p->buflen) {
2188                 s = p->buf;
2189                 buf_end = s + sizeof(p->buf);
2190                 for (i = 0; i < N_TIMINGS; ++i) {
2191                         struct kvmhv_tb_accumulator *acc;
2192
2193                         acc = (struct kvmhv_tb_accumulator *)
2194                                 ((unsigned long)vcpu + timings[i].offset);
2195                         ok = false;
2196                         for (loops = 0; loops < 1000; ++loops) {
2197                                 count = acc->seqcount;
2198                                 if (!(count & 1)) {
2199                                         smp_rmb();
2200                                         tb = *acc;
2201                                         smp_rmb();
2202                                         if (count == acc->seqcount) {
2203                                                 ok = true;
2204                                                 break;
2205                                         }
2206                                 }
2207                                 udelay(1);
2208                         }
2209                         if (!ok)
2210                                 snprintf(s, buf_end - s, "%s: stuck\n",
2211                                         timings[i].name);
2212                         else
2213                                 snprintf(s, buf_end - s,
2214                                         "%s: %llu %llu %llu %llu\n",
2215                                         timings[i].name, count / 2,
2216                                         tb_to_ns(tb.tb_total),
2217                                         tb_to_ns(tb.tb_min),
2218                                         tb_to_ns(tb.tb_max));
2219                         s += strlen(s);
2220                 }
2221                 p->buflen = s - p->buf;
2222         }
2223
2224         pos = *ppos;
2225         if (pos >= p->buflen)
2226                 return 0;
2227         if (len > p->buflen - pos)
2228                 len = p->buflen - pos;
2229         n = copy_to_user(buf, p->buf + pos, len);
2230         if (n) {
2231                 if (n == len)
2232                         return -EFAULT;
2233                 len -= n;
2234         }
2235         *ppos = pos + len;
2236         return len;
2237 }
2238
2239 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2240                                      size_t len, loff_t *ppos)
2241 {
2242         return -EACCES;
2243 }
2244
2245 static const struct file_operations debugfs_timings_ops = {
2246         .owner   = THIS_MODULE,
2247         .open    = debugfs_timings_open,
2248         .release = debugfs_timings_release,
2249         .read    = debugfs_timings_read,
2250         .write   = debugfs_timings_write,
2251         .llseek  = generic_file_llseek,
2252 };
2253
2254 /* Create a debugfs directory for the vcpu */
2255 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2256 {
2257         char buf[16];
2258         struct kvm *kvm = vcpu->kvm;
2259
2260         snprintf(buf, sizeof(buf), "vcpu%u", id);
2261         if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
2262                 return;
2263         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2264         if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
2265                 return;
2266         vcpu->arch.debugfs_timings =
2267                 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
2268                                     vcpu, &debugfs_timings_ops);
2269 }
2270
2271 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2272 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2273 {
2274 }
2275 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2276
2277 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2278 {
2279         int err;
2280         int core;
2281         struct kvmppc_vcore *vcore;
2282         struct kvm *kvm;
2283         unsigned int id;
2284
2285         kvm = vcpu->kvm;
2286         id = vcpu->vcpu_id;
2287
2288         vcpu->arch.shared = &vcpu->arch.shregs;
2289 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2290         /*
2291          * The shared struct is never shared on HV,
2292          * so we can always use host endianness
2293          */
2294 #ifdef __BIG_ENDIAN__
2295         vcpu->arch.shared_big_endian = true;
2296 #else
2297         vcpu->arch.shared_big_endian = false;
2298 #endif
2299 #endif
2300         vcpu->arch.mmcr[0] = MMCR0_FC;
2301         vcpu->arch.ctrl = CTRL_RUNLATCH;
2302         /* default to host PVR, since we can't spoof it */
2303         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2304         spin_lock_init(&vcpu->arch.vpa_update_lock);
2305         spin_lock_init(&vcpu->arch.tbacct_lock);
2306         vcpu->arch.busy_preempt = TB_NIL;
2307         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2308
2309         /*
2310          * Set the default HFSCR for the guest from the host value.
2311          * This value is only used on POWER9.
2312          * On POWER9, we want to virtualize the doorbell facility, so we
2313          * don't set the HFSCR_MSGP bit, and that causes those instructions
2314          * to trap and then we emulate them.
2315          */
2316         vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2317                 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2318         if (cpu_has_feature(CPU_FTR_HVMODE)) {
2319                 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2320                 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2321                         vcpu->arch.hfscr |= HFSCR_TM;
2322         }
2323         if (cpu_has_feature(CPU_FTR_TM_COMP))
2324                 vcpu->arch.hfscr |= HFSCR_TM;
2325
2326         kvmppc_mmu_book3s_hv_init(vcpu);
2327
2328         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2329
2330         init_waitqueue_head(&vcpu->arch.cpu_run);
2331
2332         mutex_lock(&kvm->lock);
2333         vcore = NULL;
2334         err = -EINVAL;
2335         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2336                 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2337                         pr_devel("KVM: VCPU ID too high\n");
2338                         core = KVM_MAX_VCORES;
2339                 } else {
2340                         BUG_ON(kvm->arch.smt_mode != 1);
2341                         core = kvmppc_pack_vcpu_id(kvm, id);
2342                 }
2343         } else {
2344                 core = id / kvm->arch.smt_mode;
2345         }
2346         if (core < KVM_MAX_VCORES) {
2347                 vcore = kvm->arch.vcores[core];
2348                 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2349                         pr_devel("KVM: collision on id %u", id);
2350                         vcore = NULL;
2351                 } else if (!vcore) {
2352                         /*
2353                          * Take mmu_setup_lock for mutual exclusion
2354                          * with kvmppc_update_lpcr().
2355                          */
2356                         err = -ENOMEM;
2357                         vcore = kvmppc_vcore_create(kvm,
2358                                         id & ~(kvm->arch.smt_mode - 1));
2359                         mutex_lock(&kvm->arch.mmu_setup_lock);
2360                         kvm->arch.vcores[core] = vcore;
2361                         kvm->arch.online_vcores++;
2362                         mutex_unlock(&kvm->arch.mmu_setup_lock);
2363                 }
2364         }
2365         mutex_unlock(&kvm->lock);
2366
2367         if (!vcore)
2368                 return err;
2369
2370         spin_lock(&vcore->lock);
2371         ++vcore->num_threads;
2372         spin_unlock(&vcore->lock);
2373         vcpu->arch.vcore = vcore;
2374         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2375         vcpu->arch.thread_cpu = -1;
2376         vcpu->arch.prev_cpu = -1;
2377
2378         vcpu->arch.cpu_type = KVM_CPU_3S_64;
2379         kvmppc_sanity_check(vcpu);
2380
2381         debugfs_vcpu_init(vcpu, id);
2382
2383         return 0;
2384 }
2385
2386 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2387                               unsigned long flags)
2388 {
2389         int err;
2390         int esmt = 0;
2391
2392         if (flags)
2393                 return -EINVAL;
2394         if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2395                 return -EINVAL;
2396         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2397                 /*
2398                  * On POWER8 (or POWER7), the threading mode is "strict",
2399                  * so we pack smt_mode vcpus per vcore.
2400                  */
2401                 if (smt_mode > threads_per_subcore)
2402                         return -EINVAL;
2403         } else {
2404                 /*
2405                  * On POWER9, the threading mode is "loose",
2406                  * so each vcpu gets its own vcore.
2407                  */
2408                 esmt = smt_mode;
2409                 smt_mode = 1;
2410         }
2411         mutex_lock(&kvm->lock);
2412         err = -EBUSY;
2413         if (!kvm->arch.online_vcores) {
2414                 kvm->arch.smt_mode = smt_mode;
2415                 kvm->arch.emul_smt_mode = esmt;
2416                 err = 0;
2417         }
2418         mutex_unlock(&kvm->lock);
2419
2420         return err;
2421 }
2422
2423 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2424 {
2425         if (vpa->pinned_addr)
2426                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2427                                         vpa->dirty);
2428 }
2429
2430 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2431 {
2432         spin_lock(&vcpu->arch.vpa_update_lock);
2433         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2434         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2435         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2436         spin_unlock(&vcpu->arch.vpa_update_lock);
2437 }
2438
2439 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2440 {
2441         /* Indicate we want to get back into the guest */
2442         return 1;
2443 }
2444
2445 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2446 {
2447         unsigned long dec_nsec, now;
2448
2449         now = get_tb();
2450         if (now > vcpu->arch.dec_expires) {
2451                 /* decrementer has already gone negative */
2452                 kvmppc_core_queue_dec(vcpu);
2453                 kvmppc_core_prepare_to_enter(vcpu);
2454                 return;
2455         }
2456         dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2457         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2458         vcpu->arch.timer_running = 1;
2459 }
2460
2461 extern int __kvmppc_vcore_entry(void);
2462
2463 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2464                                    struct kvm_vcpu *vcpu)
2465 {
2466         u64 now;
2467
2468         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2469                 return;
2470         spin_lock_irq(&vcpu->arch.tbacct_lock);
2471         now = mftb();
2472         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2473                 vcpu->arch.stolen_logged;
2474         vcpu->arch.busy_preempt = now;
2475         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2476         spin_unlock_irq(&vcpu->arch.tbacct_lock);
2477         --vc->n_runnable;
2478         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2479 }
2480
2481 static int kvmppc_grab_hwthread(int cpu)
2482 {
2483         struct paca_struct *tpaca;
2484         long timeout = 10000;
2485
2486         tpaca = paca_ptrs[cpu];
2487
2488         /* Ensure the thread won't go into the kernel if it wakes */
2489         tpaca->kvm_hstate.kvm_vcpu = NULL;
2490         tpaca->kvm_hstate.kvm_vcore = NULL;
2491         tpaca->kvm_hstate.napping = 0;
2492         smp_wmb();
2493         tpaca->kvm_hstate.hwthread_req = 1;
2494
2495         /*
2496          * If the thread is already executing in the kernel (e.g. handling
2497          * a stray interrupt), wait for it to get back to nap mode.
2498          * The smp_mb() is to ensure that our setting of hwthread_req
2499          * is visible before we look at hwthread_state, so if this
2500          * races with the code at system_reset_pSeries and the thread
2501          * misses our setting of hwthread_req, we are sure to see its
2502          * setting of hwthread_state, and vice versa.
2503          */
2504         smp_mb();
2505         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2506                 if (--timeout <= 0) {
2507                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
2508                         return -EBUSY;
2509                 }
2510                 udelay(1);
2511         }
2512         return 0;
2513 }
2514
2515 static void kvmppc_release_hwthread(int cpu)
2516 {
2517         struct paca_struct *tpaca;
2518
2519         tpaca = paca_ptrs[cpu];
2520         tpaca->kvm_hstate.hwthread_req = 0;
2521         tpaca->kvm_hstate.kvm_vcpu = NULL;
2522         tpaca->kvm_hstate.kvm_vcore = NULL;
2523         tpaca->kvm_hstate.kvm_split_mode = NULL;
2524 }
2525
2526 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2527 {
2528         struct kvm_nested_guest *nested = vcpu->arch.nested;
2529         cpumask_t *cpu_in_guest;
2530         int i;
2531
2532         cpu = cpu_first_thread_sibling(cpu);
2533         if (nested) {
2534                 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2535                 cpu_in_guest = &nested->cpu_in_guest;
2536         } else {
2537                 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2538                 cpu_in_guest = &kvm->arch.cpu_in_guest;
2539         }
2540         /*
2541          * Make sure setting of bit in need_tlb_flush precedes
2542          * testing of cpu_in_guest bits.  The matching barrier on
2543          * the other side is the first smp_mb() in kvmppc_run_core().
2544          */
2545         smp_mb();
2546         for (i = 0; i < threads_per_core; ++i)
2547                 if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2548                         smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2549 }
2550
2551 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2552 {
2553         struct kvm_nested_guest *nested = vcpu->arch.nested;
2554         struct kvm *kvm = vcpu->kvm;
2555         int prev_cpu;
2556
2557         if (!cpu_has_feature(CPU_FTR_HVMODE))
2558                 return;
2559
2560         if (nested)
2561                 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2562         else
2563                 prev_cpu = vcpu->arch.prev_cpu;
2564
2565         /*
2566          * With radix, the guest can do TLB invalidations itself,
2567          * and it could choose to use the local form (tlbiel) if
2568          * it is invalidating a translation that has only ever been
2569          * used on one vcpu.  However, that doesn't mean it has
2570          * only ever been used on one physical cpu, since vcpus
2571          * can move around between pcpus.  To cope with this, when
2572          * a vcpu moves from one pcpu to another, we need to tell
2573          * any vcpus running on the same core as this vcpu previously
2574          * ran to flush the TLB.  The TLB is shared between threads,
2575          * so we use a single bit in .need_tlb_flush for all 4 threads.
2576          */
2577         if (prev_cpu != pcpu) {
2578                 if (prev_cpu >= 0 &&
2579                     cpu_first_thread_sibling(prev_cpu) !=
2580                     cpu_first_thread_sibling(pcpu))
2581                         radix_flush_cpu(kvm, prev_cpu, vcpu);
2582                 if (nested)
2583                         nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2584                 else
2585                         vcpu->arch.prev_cpu = pcpu;
2586         }
2587 }
2588
2589 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2590 {
2591         int cpu;
2592         struct paca_struct *tpaca;
2593         struct kvm *kvm = vc->kvm;
2594
2595         cpu = vc->pcpu;
2596         if (vcpu) {
2597                 if (vcpu->arch.timer_running) {
2598                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2599                         vcpu->arch.timer_running = 0;
2600                 }
2601                 cpu += vcpu->arch.ptid;
2602                 vcpu->cpu = vc->pcpu;
2603                 vcpu->arch.thread_cpu = cpu;
2604                 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2605         }
2606         tpaca = paca_ptrs[cpu];
2607         tpaca->kvm_hstate.kvm_vcpu = vcpu;
2608         tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2609         tpaca->kvm_hstate.fake_suspend = 0;
2610         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2611         smp_wmb();
2612         tpaca->kvm_hstate.kvm_vcore = vc;
2613         if (cpu != smp_processor_id())
2614                 kvmppc_ipi_thread(cpu);
2615 }
2616
2617 static void kvmppc_wait_for_nap(int n_threads)
2618 {
2619         int cpu = smp_processor_id();
2620         int i, loops;
2621
2622         if (n_threads <= 1)
2623                 return;
2624         for (loops = 0; loops < 1000000; ++loops) {
2625                 /*
2626                  * Check if all threads are finished.
2627                  * We set the vcore pointer when starting a thread
2628                  * and the thread clears it when finished, so we look
2629                  * for any threads that still have a non-NULL vcore ptr.
2630                  */
2631                 for (i = 1; i < n_threads; ++i)
2632                         if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2633                                 break;
2634                 if (i == n_threads) {
2635                         HMT_medium();
2636                         return;
2637                 }
2638                 HMT_low();
2639         }
2640         HMT_medium();
2641         for (i = 1; i < n_threads; ++i)
2642                 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2643                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2644 }
2645
2646 /*
2647  * Check that we are on thread 0 and that any other threads in
2648  * this core are off-line.  Then grab the threads so they can't
2649  * enter the kernel.
2650  */
2651 static int on_primary_thread(void)
2652 {
2653         int cpu = smp_processor_id();
2654         int thr;
2655
2656         /* Are we on a primary subcore? */
2657         if (cpu_thread_in_subcore(cpu))
2658                 return 0;
2659
2660         thr = 0;
2661         while (++thr < threads_per_subcore)
2662                 if (cpu_online(cpu + thr))
2663                         return 0;
2664
2665         /* Grab all hw threads so they can't go into the kernel */
2666         for (thr = 1; thr < threads_per_subcore; ++thr) {
2667                 if (kvmppc_grab_hwthread(cpu + thr)) {
2668                         /* Couldn't grab one; let the others go */
2669                         do {
2670                                 kvmppc_release_hwthread(cpu + thr);
2671                         } while (--thr > 0);
2672                         return 0;
2673                 }
2674         }
2675         return 1;
2676 }
2677
2678 /*
2679  * A list of virtual cores for each physical CPU.
2680  * These are vcores that could run but their runner VCPU tasks are
2681  * (or may be) preempted.
2682  */
2683 struct preempted_vcore_list {
2684         struct list_head        list;
2685         spinlock_t              lock;
2686 };
2687
2688 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2689
2690 static void init_vcore_lists(void)
2691 {
2692         int cpu;
2693
2694         for_each_possible_cpu(cpu) {
2695                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2696                 spin_lock_init(&lp->lock);
2697                 INIT_LIST_HEAD(&lp->list);
2698         }
2699 }
2700
2701 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2702 {
2703         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2704
2705         vc->vcore_state = VCORE_PREEMPT;
2706         vc->pcpu = smp_processor_id();
2707         if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2708                 spin_lock(&lp->lock);
2709                 list_add_tail(&vc->preempt_list, &lp->list);
2710                 spin_unlock(&lp->lock);
2711         }
2712
2713         /* Start accumulating stolen time */
2714         kvmppc_core_start_stolen(vc);
2715 }
2716
2717 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2718 {
2719         struct preempted_vcore_list *lp;
2720
2721         kvmppc_core_end_stolen(vc);
2722         if (!list_empty(&vc->preempt_list)) {
2723                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2724                 spin_lock(&lp->lock);
2725                 list_del_init(&vc->preempt_list);
2726                 spin_unlock(&lp->lock);
2727         }
2728         vc->vcore_state = VCORE_INACTIVE;
2729 }
2730
2731 /*
2732  * This stores information about the virtual cores currently
2733  * assigned to a physical core.
2734  */
2735 struct core_info {
2736         int             n_subcores;
2737         int             max_subcore_threads;
2738         int             total_threads;
2739         int             subcore_threads[MAX_SUBCORES];
2740         struct kvmppc_vcore *vc[MAX_SUBCORES];
2741 };
2742
2743 /*
2744  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2745  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2746  */
2747 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2748
2749 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2750 {
2751         memset(cip, 0, sizeof(*cip));
2752         cip->n_subcores = 1;
2753         cip->max_subcore_threads = vc->num_threads;
2754         cip->total_threads = vc->num_threads;
2755         cip->subcore_threads[0] = vc->num_threads;
2756         cip->vc[0] = vc;
2757 }
2758
2759 static bool subcore_config_ok(int n_subcores, int n_threads)
2760 {
2761         /*
2762          * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2763          * split-core mode, with one thread per subcore.
2764          */
2765         if (cpu_has_feature(CPU_FTR_ARCH_300))
2766                 return n_subcores <= 4 && n_threads == 1;
2767
2768         /* On POWER8, can only dynamically split if unsplit to begin with */
2769         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2770                 return false;
2771         if (n_subcores > MAX_SUBCORES)
2772                 return false;
2773         if (n_subcores > 1) {
2774                 if (!(dynamic_mt_modes & 2))
2775                         n_subcores = 4;
2776                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2777                         return false;
2778         }
2779
2780         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2781 }
2782
2783 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2784 {
2785         vc->entry_exit_map = 0;
2786         vc->in_guest = 0;
2787         vc->napping_threads = 0;
2788         vc->conferring_threads = 0;
2789         vc->tb_offset_applied = 0;
2790 }
2791
2792 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2793 {
2794         int n_threads = vc->num_threads;
2795         int sub;
2796
2797         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2798                 return false;
2799
2800         /* In one_vm_per_core mode, require all vcores to be from the same vm */
2801         if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2802                 return false;
2803
2804         /* Some POWER9 chips require all threads to be in the same MMU mode */
2805         if (no_mixing_hpt_and_radix &&
2806             kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2807                 return false;
2808
2809         if (n_threads < cip->max_subcore_threads)
2810                 n_threads = cip->max_subcore_threads;
2811         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2812                 return false;
2813         cip->max_subcore_threads = n_threads;
2814
2815         sub = cip->n_subcores;
2816         ++cip->n_subcores;
2817         cip->total_threads += vc->num_threads;
2818         cip->subcore_threads[sub] = vc->num_threads;
2819         cip->vc[sub] = vc;
2820         init_vcore_to_run(vc);
2821         list_del_init(&vc->preempt_list);
2822
2823         return true;
2824 }
2825
2826 /*
2827  * Work out whether it is possible to piggyback the execution of
2828  * vcore *pvc onto the execution of the other vcores described in *cip.
2829  */
2830 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2831                           int target_threads)
2832 {
2833         if (cip->total_threads + pvc->num_threads > target_threads)
2834                 return false;
2835
2836         return can_dynamic_split(pvc, cip);
2837 }
2838
2839 static void prepare_threads(struct kvmppc_vcore *vc)
2840 {
2841         int i;
2842         struct kvm_vcpu *vcpu;
2843
2844         for_each_runnable_thread(i, vcpu, vc) {
2845                 if (signal_pending(vcpu->arch.run_task))
2846                         vcpu->arch.ret = -EINTR;
2847                 else if (vcpu->arch.vpa.update_pending ||
2848                          vcpu->arch.slb_shadow.update_pending ||
2849                          vcpu->arch.dtl.update_pending)
2850                         vcpu->arch.ret = RESUME_GUEST;
2851                 else
2852                         continue;
2853                 kvmppc_remove_runnable(vc, vcpu);
2854                 wake_up(&vcpu->arch.cpu_run);
2855         }
2856 }
2857
2858 static void collect_piggybacks(struct core_info *cip, int target_threads)
2859 {
2860         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2861         struct kvmppc_vcore *pvc, *vcnext;
2862
2863         spin_lock(&lp->lock);
2864         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2865                 if (!spin_trylock(&pvc->lock))
2866                         continue;
2867                 prepare_threads(pvc);
2868                 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
2869                         list_del_init(&pvc->preempt_list);
2870                         if (pvc->runner == NULL) {
2871                                 pvc->vcore_state = VCORE_INACTIVE;
2872                                 kvmppc_core_end_stolen(pvc);
2873                         }
2874                         spin_unlock(&pvc->lock);
2875                         continue;
2876                 }
2877                 if (!can_piggyback(pvc, cip, target_threads)) {
2878                         spin_unlock(&pvc->lock);
2879                         continue;
2880                 }
2881                 kvmppc_core_end_stolen(pvc);
2882                 pvc->vcore_state = VCORE_PIGGYBACK;
2883                 if (cip->total_threads >= target_threads)
2884                         break;
2885         }
2886         spin_unlock(&lp->lock);
2887 }
2888
2889 static bool recheck_signals_and_mmu(struct core_info *cip)
2890 {
2891         int sub, i;
2892         struct kvm_vcpu *vcpu;
2893         struct kvmppc_vcore *vc;
2894
2895         for (sub = 0; sub < cip->n_subcores; ++sub) {
2896                 vc = cip->vc[sub];
2897                 if (!vc->kvm->arch.mmu_ready)
2898                         return true;
2899                 for_each_runnable_thread(i, vcpu, vc)
2900                         if (signal_pending(vcpu->arch.run_task))
2901                                 return true;
2902         }
2903         return false;
2904 }
2905
2906 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2907 {
2908         int still_running = 0, i;
2909         u64 now;
2910         long ret;
2911         struct kvm_vcpu *vcpu;
2912
2913         spin_lock(&vc->lock);
2914         now = get_tb();
2915         for_each_runnable_thread(i, vcpu, vc) {
2916                 /*
2917                  * It's safe to unlock the vcore in the loop here, because
2918                  * for_each_runnable_thread() is safe against removal of
2919                  * the vcpu, and the vcore state is VCORE_EXITING here,
2920                  * so any vcpus becoming runnable will have their arch.trap
2921                  * set to zero and can't actually run in the guest.
2922                  */
2923                 spin_unlock(&vc->lock);
2924                 /* cancel pending dec exception if dec is positive */
2925                 if (now < vcpu->arch.dec_expires &&
2926                     kvmppc_core_pending_dec(vcpu))
2927                         kvmppc_core_dequeue_dec(vcpu);
2928
2929                 trace_kvm_guest_exit(vcpu);
2930
2931                 ret = RESUME_GUEST;
2932                 if (vcpu->arch.trap)
2933                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2934                                                     vcpu->arch.run_task);
2935
2936                 vcpu->arch.ret = ret;
2937                 vcpu->arch.trap = 0;
2938
2939                 spin_lock(&vc->lock);
2940                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2941                         if (vcpu->arch.pending_exceptions)
2942                                 kvmppc_core_prepare_to_enter(vcpu);
2943                         if (vcpu->arch.ceded)
2944                                 kvmppc_set_timer(vcpu);
2945                         else
2946                                 ++still_running;
2947                 } else {
2948                         kvmppc_remove_runnable(vc, vcpu);
2949                         wake_up(&vcpu->arch.cpu_run);
2950                 }
2951         }
2952         if (!is_master) {
2953                 if (still_running > 0) {
2954                         kvmppc_vcore_preempt(vc);
2955                 } else if (vc->runner) {
2956                         vc->vcore_state = VCORE_PREEMPT;
2957                         kvmppc_core_start_stolen(vc);
2958                 } else {
2959                         vc->vcore_state = VCORE_INACTIVE;
2960                 }
2961                 if (vc->n_runnable > 0 && vc->runner == NULL) {
2962                         /* make sure there's a candidate runner awake */
2963                         i = -1;
2964                         vcpu = next_runnable_thread(vc, &i);
2965                         wake_up(&vcpu->arch.cpu_run);
2966                 }
2967         }
2968         spin_unlock(&vc->lock);
2969 }
2970
2971 /*
2972  * Clear core from the list of active host cores as we are about to
2973  * enter the guest. Only do this if it is the primary thread of the
2974  * core (not if a subcore) that is entering the guest.
2975  */
2976 static inline int kvmppc_clear_host_core(unsigned int cpu)
2977 {
2978         int core;
2979
2980         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2981                 return 0;
2982         /*
2983          * Memory barrier can be omitted here as we will do a smp_wmb()
2984          * later in kvmppc_start_thread and we need ensure that state is
2985          * visible to other CPUs only after we enter guest.
2986          */
2987         core = cpu >> threads_shift;
2988         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2989         return 0;
2990 }
2991
2992 /*
2993  * Advertise this core as an active host core since we exited the guest
2994  * Only need to do this if it is the primary thread of the core that is
2995  * exiting.
2996  */
2997 static inline int kvmppc_set_host_core(unsigned int cpu)
2998 {
2999         int core;
3000
3001         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3002                 return 0;
3003
3004         /*
3005          * Memory barrier can be omitted here because we do a spin_unlock
3006          * immediately after this which provides the memory barrier.
3007          */
3008         core = cpu >> threads_shift;
3009         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3010         return 0;
3011 }
3012
3013 static void set_irq_happened(int trap)
3014 {
3015         switch (trap) {
3016         case BOOK3S_INTERRUPT_EXTERNAL:
3017                 local_paca->irq_happened |= PACA_IRQ_EE;
3018                 break;
3019         case BOOK3S_INTERRUPT_H_DOORBELL:
3020                 local_paca->irq_happened |= PACA_IRQ_DBELL;
3021                 break;
3022         case BOOK3S_INTERRUPT_HMI:
3023                 local_paca->irq_happened |= PACA_IRQ_HMI;
3024                 break;
3025         case BOOK3S_INTERRUPT_SYSTEM_RESET:
3026                 replay_system_reset();
3027                 break;
3028         }
3029 }
3030
3031 /*
3032  * Run a set of guest threads on a physical core.
3033  * Called with vc->lock held.
3034  */
3035 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3036 {
3037         struct kvm_vcpu *vcpu;
3038         int i;
3039         int srcu_idx;
3040         struct core_info core_info;
3041         struct kvmppc_vcore *pvc;
3042         struct kvm_split_mode split_info, *sip;
3043         int split, subcore_size, active;
3044         int sub;
3045         bool thr0_done;
3046         unsigned long cmd_bit, stat_bit;
3047         int pcpu, thr;
3048         int target_threads;
3049         int controlled_threads;
3050         int trap;
3051         bool is_power8;
3052         bool hpt_on_radix;
3053
3054         /*
3055          * Remove from the list any threads that have a signal pending
3056          * or need a VPA update done
3057          */
3058         prepare_threads(vc);
3059
3060         /* if the runner is no longer runnable, let the caller pick a new one */
3061         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3062                 return;
3063
3064         /*
3065          * Initialize *vc.
3066          */
3067         init_vcore_to_run(vc);
3068         vc->preempt_tb = TB_NIL;
3069
3070         /*
3071          * Number of threads that we will be controlling: the same as
3072          * the number of threads per subcore, except on POWER9,
3073          * where it's 1 because the threads are (mostly) independent.
3074          */
3075         controlled_threads = threads_per_vcore(vc->kvm);
3076
3077         /*
3078          * Make sure we are running on primary threads, and that secondary
3079          * threads are offline.  Also check if the number of threads in this
3080          * guest are greater than the current system threads per guest.
3081          * On POWER9, we need to be not in independent-threads mode if
3082          * this is a HPT guest on a radix host machine where the
3083          * CPU threads may not be in different MMU modes.
3084          */
3085         hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3086                 !kvm_is_radix(vc->kvm);
3087         if (((controlled_threads > 1) &&
3088              ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3089             (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3090                 for_each_runnable_thread(i, vcpu, vc) {
3091                         vcpu->arch.ret = -EBUSY;
3092                         kvmppc_remove_runnable(vc, vcpu);
3093                         wake_up(&vcpu->arch.cpu_run);
3094                 }
3095                 goto out;
3096         }
3097
3098         /*
3099          * See if we could run any other vcores on the physical core
3100          * along with this one.
3101          */
3102         init_core_info(&core_info, vc);
3103         pcpu = smp_processor_id();
3104         target_threads = controlled_threads;
3105         if (target_smt_mode && target_smt_mode < target_threads)
3106                 target_threads = target_smt_mode;
3107         if (vc->num_threads < target_threads)
3108                 collect_piggybacks(&core_info, target_threads);
3109
3110         /*
3111          * On radix, arrange for TLB flushing if necessary.
3112          * This has to be done before disabling interrupts since
3113          * it uses smp_call_function().
3114          */
3115         pcpu = smp_processor_id();
3116         if (kvm_is_radix(vc->kvm)) {
3117                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3118                         for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3119                                 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3120         }
3121
3122         /*
3123          * Hard-disable interrupts, and check resched flag and signals.
3124          * If we need to reschedule or deliver a signal, clean up
3125          * and return without going into the guest(s).
3126          * If the mmu_ready flag has been cleared, don't go into the
3127          * guest because that means a HPT resize operation is in progress.
3128          */
3129         local_irq_disable();
3130         hard_irq_disable();
3131         if (lazy_irq_pending() || need_resched() ||
3132             recheck_signals_and_mmu(&core_info)) {
3133                 local_irq_enable();
3134                 vc->vcore_state = VCORE_INACTIVE;
3135                 /* Unlock all except the primary vcore */
3136                 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3137                         pvc = core_info.vc[sub];
3138                         /* Put back on to the preempted vcores list */
3139                         kvmppc_vcore_preempt(pvc);
3140                         spin_unlock(&pvc->lock);
3141                 }
3142                 for (i = 0; i < controlled_threads; ++i)
3143                         kvmppc_release_hwthread(pcpu + i);
3144                 return;
3145         }
3146
3147         kvmppc_clear_host_core(pcpu);
3148
3149         /* Decide on micro-threading (split-core) mode */
3150         subcore_size = threads_per_subcore;
3151         cmd_bit = stat_bit = 0;
3152         split = core_info.n_subcores;
3153         sip = NULL;
3154         is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3155                 && !cpu_has_feature(CPU_FTR_ARCH_300);
3156
3157         if (split > 1 || hpt_on_radix) {
3158                 sip = &split_info;
3159                 memset(&split_info, 0, sizeof(split_info));
3160                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3161                         split_info.vc[sub] = core_info.vc[sub];
3162
3163                 if (is_power8) {
3164                         if (split == 2 && (dynamic_mt_modes & 2)) {
3165                                 cmd_bit = HID0_POWER8_1TO2LPAR;
3166                                 stat_bit = HID0_POWER8_2LPARMODE;
3167                         } else {
3168                                 split = 4;
3169                                 cmd_bit = HID0_POWER8_1TO4LPAR;
3170                                 stat_bit = HID0_POWER8_4LPARMODE;
3171                         }
3172                         subcore_size = MAX_SMT_THREADS / split;
3173                         split_info.rpr = mfspr(SPRN_RPR);
3174                         split_info.pmmar = mfspr(SPRN_PMMAR);
3175                         split_info.ldbar = mfspr(SPRN_LDBAR);
3176                         split_info.subcore_size = subcore_size;
3177                 } else {
3178                         split_info.subcore_size = 1;
3179                         if (hpt_on_radix) {
3180                                 /* Use the split_info for LPCR/LPIDR changes */
3181                                 split_info.lpcr_req = vc->lpcr;
3182                                 split_info.lpidr_req = vc->kvm->arch.lpid;
3183                                 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3184                                 split_info.do_set = 1;
3185                         }
3186                 }
3187
3188                 /* order writes to split_info before kvm_split_mode pointer */
3189                 smp_wmb();
3190         }
3191
3192         for (thr = 0; thr < controlled_threads; ++thr) {
3193                 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3194
3195                 paca->kvm_hstate.tid = thr;
3196                 paca->kvm_hstate.napping = 0;
3197                 paca->kvm_hstate.kvm_split_mode = sip;
3198         }
3199
3200         /* Initiate micro-threading (split-core) on POWER8 if required */
3201         if (cmd_bit) {
3202                 unsigned long hid0 = mfspr(SPRN_HID0);
3203
3204                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3205                 mb();
3206                 mtspr(SPRN_HID0, hid0);
3207                 isync();
3208                 for (;;) {
3209                         hid0 = mfspr(SPRN_HID0);
3210                         if (hid0 & stat_bit)
3211                                 break;
3212                         cpu_relax();
3213                 }
3214         }
3215
3216         /*
3217          * On POWER8, set RWMR register.
3218          * Since it only affects PURR and SPURR, it doesn't affect
3219          * the host, so we don't save/restore the host value.
3220          */
3221         if (is_power8) {
3222                 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3223                 int n_online = atomic_read(&vc->online_count);
3224
3225                 /*
3226                  * Use the 8-thread value if we're doing split-core
3227                  * or if the vcore's online count looks bogus.
3228                  */
3229                 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3230                     n_online >= 1 && n_online <= MAX_SMT_THREADS)
3231                         rwmr_val = p8_rwmr_values[n_online];
3232                 mtspr(SPRN_RWMR, rwmr_val);
3233         }
3234
3235         /* Start all the threads */
3236         active = 0;
3237         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3238                 thr = is_power8 ? subcore_thread_map[sub] : sub;
3239                 thr0_done = false;
3240                 active |= 1 << thr;
3241                 pvc = core_info.vc[sub];
3242                 pvc->pcpu = pcpu + thr;
3243                 for_each_runnable_thread(i, vcpu, pvc) {
3244                         kvmppc_start_thread(vcpu, pvc);
3245                         kvmppc_create_dtl_entry(vcpu, pvc);
3246                         trace_kvm_guest_enter(vcpu);
3247                         if (!vcpu->arch.ptid)
3248                                 thr0_done = true;
3249                         active |= 1 << (thr + vcpu->arch.ptid);
3250                 }
3251                 /*
3252                  * We need to start the first thread of each subcore
3253                  * even if it doesn't have a vcpu.
3254                  */
3255                 if (!thr0_done)
3256                         kvmppc_start_thread(NULL, pvc);
3257         }
3258
3259         /*
3260          * Ensure that split_info.do_nap is set after setting
3261          * the vcore pointer in the PACA of the secondaries.
3262          */
3263         smp_mb();
3264
3265         /*
3266          * When doing micro-threading, poke the inactive threads as well.
3267          * This gets them to the nap instruction after kvm_do_nap,
3268          * which reduces the time taken to unsplit later.
3269          * For POWER9 HPT guest on radix host, we need all the secondary
3270          * threads woken up so they can do the LPCR/LPIDR change.
3271          */
3272         if (cmd_bit || hpt_on_radix) {
3273                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
3274                 for (thr = 1; thr < threads_per_subcore; ++thr)
3275                         if (!(active & (1 << thr)))
3276                                 kvmppc_ipi_thread(pcpu + thr);
3277         }
3278
3279         vc->vcore_state = VCORE_RUNNING;
3280         preempt_disable();
3281
3282         trace_kvmppc_run_core(vc, 0);
3283
3284         for (sub = 0; sub < core_info.n_subcores; ++sub)
3285                 spin_unlock(&core_info.vc[sub]->lock);
3286
3287         guest_enter_irqoff();
3288
3289         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3290
3291         this_cpu_disable_ftrace();
3292
3293         /*
3294          * Interrupts will be enabled once we get into the guest,
3295          * so tell lockdep that we're about to enable interrupts.
3296          */
3297         trace_hardirqs_on();
3298
3299         trap = __kvmppc_vcore_entry();
3300
3301         trace_hardirqs_off();
3302
3303         this_cpu_enable_ftrace();
3304
3305         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3306
3307         set_irq_happened(trap);
3308
3309         spin_lock(&vc->lock);
3310         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3311         vc->vcore_state = VCORE_EXITING;
3312
3313         /* wait for secondary threads to finish writing their state to memory */
3314         kvmppc_wait_for_nap(controlled_threads);
3315
3316         /* Return to whole-core mode if we split the core earlier */
3317         if (cmd_bit) {
3318                 unsigned long hid0 = mfspr(SPRN_HID0);
3319                 unsigned long loops = 0;
3320
3321                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3322                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3323                 mb();
3324                 mtspr(SPRN_HID0, hid0);
3325                 isync();
3326                 for (;;) {
3327                         hid0 = mfspr(SPRN_HID0);
3328                         if (!(hid0 & stat_bit))
3329                                 break;
3330                         cpu_relax();
3331                         ++loops;
3332                 }
3333         } else if (hpt_on_radix) {
3334                 /* Wait for all threads to have seen final sync */
3335                 for (thr = 1; thr < controlled_threads; ++thr) {
3336                         struct paca_struct *paca = paca_ptrs[pcpu + thr];
3337
3338                         while (paca->kvm_hstate.kvm_split_mode) {
3339                                 HMT_low();
3340                                 barrier();
3341                         }
3342                         HMT_medium();
3343                 }
3344         }
3345         split_info.do_nap = 0;
3346
3347         kvmppc_set_host_core(pcpu);
3348
3349         local_irq_enable();
3350         guest_exit();
3351
3352         /* Let secondaries go back to the offline loop */
3353         for (i = 0; i < controlled_threads; ++i) {
3354                 kvmppc_release_hwthread(pcpu + i);
3355                 if (sip && sip->napped[i])
3356                         kvmppc_ipi_thread(pcpu + i);
3357                 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3358         }
3359
3360         spin_unlock(&vc->lock);
3361
3362         /* make sure updates to secondary vcpu structs are visible now */
3363         smp_mb();
3364
3365         preempt_enable();
3366
3367         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3368                 pvc = core_info.vc[sub];
3369                 post_guest_process(pvc, pvc == vc);
3370         }
3371
3372         spin_lock(&vc->lock);
3373
3374  out:
3375         vc->vcore_state = VCORE_INACTIVE;
3376         trace_kvmppc_run_core(vc, 1);
3377 }
3378
3379 /*
3380  * Load up hypervisor-mode registers on P9.
3381  */
3382 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3383                                      unsigned long lpcr)
3384 {
3385         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3386         s64 hdec;
3387         u64 tb, purr, spurr;
3388         int trap;
3389         unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3390         unsigned long host_ciabr = mfspr(SPRN_CIABR);
3391         unsigned long host_dawr = mfspr(SPRN_DAWR);
3392         unsigned long host_dawrx = mfspr(SPRN_DAWRX);
3393         unsigned long host_psscr = mfspr(SPRN_PSSCR);
3394         unsigned long host_pidr = mfspr(SPRN_PID);
3395
3396         hdec = time_limit - mftb();
3397         if (hdec < 0)
3398                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3399         mtspr(SPRN_HDEC, hdec);
3400
3401         if (vc->tb_offset) {
3402                 u64 new_tb = mftb() + vc->tb_offset;
3403                 mtspr(SPRN_TBU40, new_tb);
3404                 tb = mftb();
3405                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3406                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3407                 vc->tb_offset_applied = vc->tb_offset;
3408         }
3409
3410         if (vc->pcr)
3411                 mtspr(SPRN_PCR, vc->pcr | PCR_MASK);
3412         mtspr(SPRN_DPDES, vc->dpdes);
3413         mtspr(SPRN_VTB, vc->vtb);
3414
3415         local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3416         local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3417         mtspr(SPRN_PURR, vcpu->arch.purr);
3418         mtspr(SPRN_SPURR, vcpu->arch.spurr);
3419
3420         if (dawr_enabled()) {
3421                 mtspr(SPRN_DAWR, vcpu->arch.dawr);
3422                 mtspr(SPRN_DAWRX, vcpu->arch.dawrx);
3423         }
3424         mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3425         mtspr(SPRN_IC, vcpu->arch.ic);
3426         mtspr(SPRN_PID, vcpu->arch.pid);
3427
3428         mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3429               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3430
3431         mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3432
3433         mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3434         mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3435         mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3436         mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3437
3438         mtspr(SPRN_AMOR, ~0UL);
3439
3440         mtspr(SPRN_LPCR, lpcr);
3441         isync();
3442
3443         kvmppc_xive_push_vcpu(vcpu);
3444
3445         mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3446         mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3447
3448         trap = __kvmhv_vcpu_entry_p9(vcpu);
3449
3450         /* Advance host PURR/SPURR by the amount used by guest */
3451         purr = mfspr(SPRN_PURR);
3452         spurr = mfspr(SPRN_SPURR);
3453         mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3454               purr - vcpu->arch.purr);
3455         mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3456               spurr - vcpu->arch.spurr);
3457         vcpu->arch.purr = purr;
3458         vcpu->arch.spurr = spurr;
3459
3460         vcpu->arch.ic = mfspr(SPRN_IC);
3461         vcpu->arch.pid = mfspr(SPRN_PID);
3462         vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3463
3464         vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3465         vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3466         vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3467         vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3468
3469         /* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3470         mtspr(SPRN_PSSCR, host_psscr |
3471               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3472         mtspr(SPRN_HFSCR, host_hfscr);
3473         mtspr(SPRN_CIABR, host_ciabr);
3474         mtspr(SPRN_DAWR, host_dawr);
3475         mtspr(SPRN_DAWRX, host_dawrx);
3476         mtspr(SPRN_PID, host_pidr);
3477
3478         /*
3479          * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3480          * case we interrupted the guest between a tlbie and a ptesync.
3481          */
3482         asm volatile("eieio; tlbsync; ptesync");
3483
3484         mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid);    /* restore host LPID */
3485         isync();
3486
3487         vc->dpdes = mfspr(SPRN_DPDES);
3488         vc->vtb = mfspr(SPRN_VTB);
3489         mtspr(SPRN_DPDES, 0);
3490         if (vc->pcr)
3491                 mtspr(SPRN_PCR, PCR_MASK);
3492
3493         if (vc->tb_offset_applied) {
3494                 u64 new_tb = mftb() - vc->tb_offset_applied;
3495                 mtspr(SPRN_TBU40, new_tb);
3496                 tb = mftb();
3497                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3498                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3499                 vc->tb_offset_applied = 0;
3500         }
3501
3502         mtspr(SPRN_HDEC, 0x7fffffff);
3503         mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3504
3505         return trap;
3506 }
3507
3508 /*
3509  * Virtual-mode guest entry for POWER9 and later when the host and
3510  * guest are both using the radix MMU.  The LPIDR has already been set.
3511  */
3512 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3513                          unsigned long lpcr)
3514 {
3515         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3516         unsigned long host_dscr = mfspr(SPRN_DSCR);
3517         unsigned long host_tidr = mfspr(SPRN_TIDR);
3518         unsigned long host_iamr = mfspr(SPRN_IAMR);
3519         unsigned long host_amr = mfspr(SPRN_AMR);
3520         s64 dec;
3521         u64 tb;
3522         int trap, save_pmu;
3523
3524         dec = mfspr(SPRN_DEC);
3525         tb = mftb();
3526         if (dec < 512)
3527                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3528         local_paca->kvm_hstate.dec_expires = dec + tb;
3529         if (local_paca->kvm_hstate.dec_expires < time_limit)
3530                 time_limit = local_paca->kvm_hstate.dec_expires;
3531
3532         vcpu->arch.ceded = 0;
3533
3534         kvmhv_save_host_pmu();          /* saves it to PACA kvm_hstate */
3535
3536         kvmppc_subcore_enter_guest();
3537
3538         vc->entry_exit_map = 1;
3539         vc->in_guest = 1;
3540
3541         if (vcpu->arch.vpa.pinned_addr) {
3542                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3543                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3544                 lp->yield_count = cpu_to_be32(yield_count);
3545                 vcpu->arch.vpa.dirty = 1;
3546         }
3547
3548         if (cpu_has_feature(CPU_FTR_TM) ||
3549             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3550                 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3551
3552         kvmhv_load_guest_pmu(vcpu);
3553
3554         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3555         load_fp_state(&vcpu->arch.fp);
3556 #ifdef CONFIG_ALTIVEC
3557         load_vr_state(&vcpu->arch.vr);
3558 #endif
3559         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3560
3561         mtspr(SPRN_DSCR, vcpu->arch.dscr);
3562         mtspr(SPRN_IAMR, vcpu->arch.iamr);
3563         mtspr(SPRN_PSPB, vcpu->arch.pspb);
3564         mtspr(SPRN_FSCR, vcpu->arch.fscr);
3565         mtspr(SPRN_TAR, vcpu->arch.tar);
3566         mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3567         mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3568         mtspr(SPRN_BESCR, vcpu->arch.bescr);
3569         mtspr(SPRN_WORT, vcpu->arch.wort);
3570         mtspr(SPRN_TIDR, vcpu->arch.tid);
3571         mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3572         mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3573         mtspr(SPRN_AMR, vcpu->arch.amr);
3574         mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3575
3576         if (!(vcpu->arch.ctrl & 1))
3577                 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3578
3579         mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3580
3581         if (kvmhv_on_pseries()) {
3582                 /*
3583                  * We need to save and restore the guest visible part of the
3584                  * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3585                  * doesn't do this for us. Note only required if pseries since
3586                  * this is done in kvmhv_load_hv_regs_and_go() below otherwise.
3587                  */
3588                 unsigned long host_psscr;
3589                 /* call our hypervisor to load up HV regs and go */
3590                 struct hv_guest_state hvregs;
3591
3592                 host_psscr = mfspr(SPRN_PSSCR_PR);
3593                 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3594                 kvmhv_save_hv_regs(vcpu, &hvregs);
3595                 hvregs.lpcr = lpcr;
3596                 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3597                 hvregs.version = HV_GUEST_STATE_VERSION;
3598                 if (vcpu->arch.nested) {
3599                         hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3600                         hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3601                 } else {
3602                         hvregs.lpid = vcpu->kvm->arch.lpid;
3603                         hvregs.vcpu_token = vcpu->vcpu_id;
3604                 }
3605                 hvregs.hdec_expiry = time_limit;
3606                 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3607                                           __pa(&vcpu->arch.regs));
3608                 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3609                 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3610                 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3611                 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3612                 vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3613                 mtspr(SPRN_PSSCR_PR, host_psscr);
3614
3615                 /* H_CEDE has to be handled now, not later */
3616                 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3617                     kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3618                         kvmppc_nested_cede(vcpu);
3619                         trap = 0;
3620                 }
3621         } else {
3622                 trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3623         }
3624
3625         vcpu->arch.slb_max = 0;
3626         dec = mfspr(SPRN_DEC);
3627         if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
3628                 dec = (s32) dec;
3629         tb = mftb();
3630         vcpu->arch.dec_expires = dec + tb;
3631         vcpu->cpu = -1;
3632         vcpu->arch.thread_cpu = -1;
3633         vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3634
3635         vcpu->arch.iamr = mfspr(SPRN_IAMR);
3636         vcpu->arch.pspb = mfspr(SPRN_PSPB);
3637         vcpu->arch.fscr = mfspr(SPRN_FSCR);
3638         vcpu->arch.tar = mfspr(SPRN_TAR);
3639         vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3640         vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3641         vcpu->arch.bescr = mfspr(SPRN_BESCR);
3642         vcpu->arch.wort = mfspr(SPRN_WORT);
3643         vcpu->arch.tid = mfspr(SPRN_TIDR);
3644         vcpu->arch.amr = mfspr(SPRN_AMR);
3645         vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3646         vcpu->arch.dscr = mfspr(SPRN_DSCR);
3647
3648         mtspr(SPRN_PSPB, 0);
3649         mtspr(SPRN_WORT, 0);
3650         mtspr(SPRN_UAMOR, 0);
3651         mtspr(SPRN_DSCR, host_dscr);
3652         mtspr(SPRN_TIDR, host_tidr);
3653         mtspr(SPRN_IAMR, host_iamr);
3654         mtspr(SPRN_PSPB, 0);
3655
3656         if (host_amr != vcpu->arch.amr)
3657                 mtspr(SPRN_AMR, host_amr);
3658
3659         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3660         store_fp_state(&vcpu->arch.fp);
3661 #ifdef CONFIG_ALTIVEC
3662         store_vr_state(&vcpu->arch.vr);
3663 #endif
3664         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3665
3666         if (cpu_has_feature(CPU_FTR_TM) ||
3667             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3668                 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3669
3670         save_pmu = 1;
3671         if (vcpu->arch.vpa.pinned_addr) {
3672                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3673                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3674                 lp->yield_count = cpu_to_be32(yield_count);
3675                 vcpu->arch.vpa.dirty = 1;
3676                 save_pmu = lp->pmcregs_in_use;
3677         }
3678         /* Must save pmu if this guest is capable of running nested guests */
3679         save_pmu |= nesting_enabled(vcpu->kvm);
3680
3681         kvmhv_save_guest_pmu(vcpu, save_pmu);
3682
3683         vc->entry_exit_map = 0x101;
3684         vc->in_guest = 0;
3685
3686         mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3687         mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3688
3689         kvmhv_load_host_pmu();
3690
3691         kvmppc_subcore_exit_guest();
3692
3693         return trap;
3694 }
3695
3696 /*
3697  * Wait for some other vcpu thread to execute us, and
3698  * wake us up when we need to handle something in the host.
3699  */
3700 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3701                                  struct kvm_vcpu *vcpu, int wait_state)
3702 {
3703         DEFINE_WAIT(wait);
3704
3705         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3706         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3707                 spin_unlock(&vc->lock);
3708                 schedule();
3709                 spin_lock(&vc->lock);
3710         }
3711         finish_wait(&vcpu->arch.cpu_run, &wait);
3712 }
3713
3714 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3715 {
3716         if (!halt_poll_ns_grow)
3717                 return;
3718
3719         vc->halt_poll_ns *= halt_poll_ns_grow;
3720         if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3721                 vc->halt_poll_ns = halt_poll_ns_grow_start;
3722 }
3723
3724 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3725 {
3726         if (halt_poll_ns_shrink == 0)
3727                 vc->halt_poll_ns = 0;
3728         else
3729                 vc->halt_poll_ns /= halt_poll_ns_shrink;
3730 }
3731
3732 #ifdef CONFIG_KVM_XICS
3733 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3734 {
3735         if (!xics_on_xive())
3736                 return false;
3737         return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3738                 vcpu->arch.xive_saved_state.cppr;
3739 }
3740 #else
3741 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3742 {
3743         return false;
3744 }
3745 #endif /* CONFIG_KVM_XICS */
3746
3747 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3748 {
3749         if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3750             kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3751                 return true;
3752
3753         return false;
3754 }
3755
3756 /*
3757  * Check to see if any of the runnable vcpus on the vcore have pending
3758  * exceptions or are no longer ceded
3759  */
3760 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3761 {
3762         struct kvm_vcpu *vcpu;
3763         int i;
3764
3765         for_each_runnable_thread(i, vcpu, vc) {
3766                 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3767                         return 1;
3768         }
3769
3770         return 0;
3771 }
3772
3773 /*
3774  * All the vcpus in this vcore are idle, so wait for a decrementer
3775  * or external interrupt to one of the vcpus.  vc->lock is held.
3776  */
3777 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3778 {
3779         ktime_t cur, start_poll, start_wait;
3780         int do_sleep = 1;
3781         u64 block_ns;
3782         DECLARE_SWAITQUEUE(wait);
3783
3784         /* Poll for pending exceptions and ceded state */
3785         cur = start_poll = ktime_get();
3786         if (vc->halt_poll_ns) {
3787                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3788                 ++vc->runner->stat.halt_attempted_poll;
3789
3790                 vc->vcore_state = VCORE_POLLING;
3791                 spin_unlock(&vc->lock);
3792
3793                 do {
3794                         if (kvmppc_vcore_check_block(vc)) {
3795                                 do_sleep = 0;
3796                                 break;
3797                         }
3798                         cur = ktime_get();
3799                 } while (single_task_running() && ktime_before(cur, stop));
3800
3801                 spin_lock(&vc->lock);
3802                 vc->vcore_state = VCORE_INACTIVE;
3803
3804                 if (!do_sleep) {
3805                         ++vc->runner->stat.halt_successful_poll;
3806                         goto out;
3807                 }
3808         }
3809
3810         prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3811
3812         if (kvmppc_vcore_check_block(vc)) {
3813                 finish_swait(&vc->wq, &wait);
3814                 do_sleep = 0;
3815                 /* If we polled, count this as a successful poll */
3816                 if (vc->halt_poll_ns)
3817                         ++vc->runner->stat.halt_successful_poll;
3818                 goto out;
3819         }
3820
3821         start_wait = ktime_get();
3822
3823         vc->vcore_state = VCORE_SLEEPING;
3824         trace_kvmppc_vcore_blocked(vc, 0);
3825         spin_unlock(&vc->lock);
3826         schedule();
3827         finish_swait(&vc->wq, &wait);
3828         spin_lock(&vc->lock);
3829         vc->vcore_state = VCORE_INACTIVE;
3830         trace_kvmppc_vcore_blocked(vc, 1);
3831         ++vc->runner->stat.halt_successful_wait;
3832
3833         cur = ktime_get();
3834
3835 out:
3836         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3837
3838         /* Attribute wait time */
3839         if (do_sleep) {
3840                 vc->runner->stat.halt_wait_ns +=
3841                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
3842                 /* Attribute failed poll time */
3843                 if (vc->halt_poll_ns)
3844                         vc->runner->stat.halt_poll_fail_ns +=
3845                                 ktime_to_ns(start_wait) -
3846                                 ktime_to_ns(start_poll);
3847         } else {
3848                 /* Attribute successful poll time */
3849                 if (vc->halt_poll_ns)
3850                         vc->runner->stat.halt_poll_success_ns +=
3851                                 ktime_to_ns(cur) -
3852                                 ktime_to_ns(start_poll);
3853         }
3854
3855         /* Adjust poll time */
3856         if (halt_poll_ns) {
3857                 if (block_ns <= vc->halt_poll_ns)
3858                         ;
3859                 /* We slept and blocked for longer than the max halt time */
3860                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3861                         shrink_halt_poll_ns(vc);
3862                 /* We slept and our poll time is too small */
3863                 else if (vc->halt_poll_ns < halt_poll_ns &&
3864                                 block_ns < halt_poll_ns)
3865                         grow_halt_poll_ns(vc);
3866                 if (vc->halt_poll_ns > halt_poll_ns)
3867                         vc->halt_poll_ns = halt_poll_ns;
3868         } else
3869                 vc->halt_poll_ns = 0;
3870
3871         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3872 }
3873
3874 /*
3875  * This never fails for a radix guest, as none of the operations it does
3876  * for a radix guest can fail or have a way to report failure.
3877  * kvmhv_run_single_vcpu() relies on this fact.
3878  */
3879 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3880 {
3881         int r = 0;
3882         struct kvm *kvm = vcpu->kvm;
3883
3884         mutex_lock(&kvm->arch.mmu_setup_lock);
3885         if (!kvm->arch.mmu_ready) {
3886                 if (!kvm_is_radix(kvm))
3887                         r = kvmppc_hv_setup_htab_rma(vcpu);
3888                 if (!r) {
3889                         if (cpu_has_feature(CPU_FTR_ARCH_300))
3890                                 kvmppc_setup_partition_table(kvm);
3891                         kvm->arch.mmu_ready = 1;
3892                 }
3893         }
3894         mutex_unlock(&kvm->arch.mmu_setup_lock);
3895         return r;
3896 }
3897
3898 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3899 {
3900         int n_ceded, i, r;
3901         struct kvmppc_vcore *vc;
3902         struct kvm_vcpu *v;
3903
3904         trace_kvmppc_run_vcpu_enter(vcpu);
3905
3906         kvm_run->exit_reason = 0;
3907         vcpu->arch.ret = RESUME_GUEST;
3908         vcpu->arch.trap = 0;
3909         kvmppc_update_vpas(vcpu);
3910
3911         /*
3912          * Synchronize with other threads in this virtual core
3913          */
3914         vc = vcpu->arch.vcore;
3915         spin_lock(&vc->lock);
3916         vcpu->arch.ceded = 0;
3917         vcpu->arch.run_task = current;
3918         vcpu->arch.kvm_run = kvm_run;
3919         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3920         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3921         vcpu->arch.busy_preempt = TB_NIL;
3922         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3923         ++vc->n_runnable;
3924
3925         /*
3926          * This happens the first time this is called for a vcpu.
3927          * If the vcore is already running, we may be able to start
3928          * this thread straight away and have it join in.
3929          */
3930         if (!signal_pending(current)) {
3931                 if ((vc->vcore_state == VCORE_PIGGYBACK ||
3932                      vc->vcore_state == VCORE_RUNNING) &&
3933                            !VCORE_IS_EXITING(vc)) {
3934                         kvmppc_create_dtl_entry(vcpu, vc);
3935                         kvmppc_start_thread(vcpu, vc);
3936                         trace_kvm_guest_enter(vcpu);
3937                 } else if (vc->vcore_state == VCORE_SLEEPING) {
3938                         swake_up_one(&vc->wq);
3939                 }
3940
3941         }
3942
3943         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3944                !signal_pending(current)) {
3945                 /* See if the MMU is ready to go */
3946                 if (!vcpu->kvm->arch.mmu_ready) {
3947                         spin_unlock(&vc->lock);
3948                         r = kvmhv_setup_mmu(vcpu);
3949                         spin_lock(&vc->lock);
3950                         if (r) {
3951                                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3952                                 kvm_run->fail_entry.
3953                                         hardware_entry_failure_reason = 0;
3954                                 vcpu->arch.ret = r;
3955                                 break;
3956                         }
3957                 }
3958
3959                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3960                         kvmppc_vcore_end_preempt(vc);
3961
3962                 if (vc->vcore_state != VCORE_INACTIVE) {
3963                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3964                         continue;
3965                 }
3966                 for_each_runnable_thread(i, v, vc) {
3967                         kvmppc_core_prepare_to_enter(v);
3968                         if (signal_pending(v->arch.run_task)) {
3969                                 kvmppc_remove_runnable(vc, v);
3970                                 v->stat.signal_exits++;
3971                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3972                                 v->arch.ret = -EINTR;
3973                                 wake_up(&v->arch.cpu_run);
3974                         }
3975                 }
3976                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3977                         break;
3978                 n_ceded = 0;
3979                 for_each_runnable_thread(i, v, vc) {
3980                         if (!kvmppc_vcpu_woken(v))
3981                                 n_ceded += v->arch.ceded;
3982                         else
3983                                 v->arch.ceded = 0;
3984                 }
3985                 vc->runner = vcpu;
3986                 if (n_ceded == vc->n_runnable) {
3987                         kvmppc_vcore_blocked(vc);
3988                 } else if (need_resched()) {
3989                         kvmppc_vcore_preempt(vc);
3990                         /* Let something else run */
3991                         cond_resched_lock(&vc->lock);
3992                         if (vc->vcore_state == VCORE_PREEMPT)
3993                                 kvmppc_vcore_end_preempt(vc);
3994                 } else {
3995                         kvmppc_run_core(vc);
3996                 }
3997                 vc->runner = NULL;
3998         }
3999
4000         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4001                (vc->vcore_state == VCORE_RUNNING ||
4002                 vc->vcore_state == VCORE_EXITING ||
4003                 vc->vcore_state == VCORE_PIGGYBACK))
4004                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4005
4006         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4007                 kvmppc_vcore_end_preempt(vc);
4008
4009         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4010                 kvmppc_remove_runnable(vc, vcpu);
4011                 vcpu->stat.signal_exits++;
4012                 kvm_run->exit_reason = KVM_EXIT_INTR;
4013                 vcpu->arch.ret = -EINTR;
4014         }
4015
4016         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4017                 /* Wake up some vcpu to run the core */
4018                 i = -1;
4019                 v = next_runnable_thread(vc, &i);
4020                 wake_up(&v->arch.cpu_run);
4021         }
4022
4023         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4024         spin_unlock(&vc->lock);
4025         return vcpu->arch.ret;
4026 }
4027
4028 int kvmhv_run_single_vcpu(struct kvm_run *kvm_run,
4029                           struct kvm_vcpu *vcpu, u64 time_limit,
4030                           unsigned long lpcr)
4031 {
4032         int trap, r, pcpu;
4033         int srcu_idx, lpid;
4034         struct kvmppc_vcore *vc;
4035         struct kvm *kvm = vcpu->kvm;
4036         struct kvm_nested_guest *nested = vcpu->arch.nested;
4037
4038         trace_kvmppc_run_vcpu_enter(vcpu);
4039
4040         kvm_run->exit_reason = 0;
4041         vcpu->arch.ret = RESUME_GUEST;
4042         vcpu->arch.trap = 0;
4043
4044         vc = vcpu->arch.vcore;
4045         vcpu->arch.ceded = 0;
4046         vcpu->arch.run_task = current;
4047         vcpu->arch.kvm_run = kvm_run;
4048         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4049         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4050         vcpu->arch.busy_preempt = TB_NIL;
4051         vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4052         vc->runnable_threads[0] = vcpu;
4053         vc->n_runnable = 1;
4054         vc->runner = vcpu;
4055
4056         /* See if the MMU is ready to go */
4057         if (!kvm->arch.mmu_ready)
4058                 kvmhv_setup_mmu(vcpu);
4059
4060         if (need_resched())
4061                 cond_resched();
4062
4063         kvmppc_update_vpas(vcpu);
4064
4065         init_vcore_to_run(vc);
4066         vc->preempt_tb = TB_NIL;
4067
4068         preempt_disable();
4069         pcpu = smp_processor_id();
4070         vc->pcpu = pcpu;
4071         kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4072
4073         local_irq_disable();
4074         hard_irq_disable();
4075         if (signal_pending(current))
4076                 goto sigpend;
4077         if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4078                 goto out;
4079
4080         if (!nested) {
4081                 kvmppc_core_prepare_to_enter(vcpu);
4082                 if (vcpu->arch.doorbell_request) {
4083                         vc->dpdes = 1;
4084                         smp_wmb();
4085                         vcpu->arch.doorbell_request = 0;
4086                 }
4087                 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4088                              &vcpu->arch.pending_exceptions))
4089                         lpcr |= LPCR_MER;
4090         } else if (vcpu->arch.pending_exceptions ||
4091                    vcpu->arch.doorbell_request ||
4092                    xive_interrupt_pending(vcpu)) {
4093                 vcpu->arch.ret = RESUME_HOST;
4094                 goto out;
4095         }
4096
4097         kvmppc_clear_host_core(pcpu);
4098
4099         local_paca->kvm_hstate.tid = 0;
4100         local_paca->kvm_hstate.napping = 0;
4101         local_paca->kvm_hstate.kvm_split_mode = NULL;
4102         kvmppc_start_thread(vcpu, vc);
4103         kvmppc_create_dtl_entry(vcpu, vc);
4104         trace_kvm_guest_enter(vcpu);
4105
4106         vc->vcore_state = VCORE_RUNNING;
4107         trace_kvmppc_run_core(vc, 0);
4108
4109         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4110                 lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4111                 mtspr(SPRN_LPID, lpid);
4112                 isync();
4113                 kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4114         }
4115
4116         guest_enter_irqoff();
4117
4118         srcu_idx = srcu_read_lock(&kvm->srcu);
4119
4120         this_cpu_disable_ftrace();
4121
4122         /* Tell lockdep that we're about to enable interrupts */
4123         trace_hardirqs_on();
4124
4125         trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4126         vcpu->arch.trap = trap;
4127
4128         trace_hardirqs_off();
4129
4130         this_cpu_enable_ftrace();
4131
4132         srcu_read_unlock(&kvm->srcu, srcu_idx);
4133
4134         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4135                 mtspr(SPRN_LPID, kvm->arch.host_lpid);
4136                 isync();
4137         }
4138
4139         set_irq_happened(trap);
4140
4141         kvmppc_set_host_core(pcpu);
4142
4143         local_irq_enable();
4144         guest_exit();
4145
4146         cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4147
4148         preempt_enable();
4149
4150         /*
4151          * cancel pending decrementer exception if DEC is now positive, or if
4152          * entering a nested guest in which case the decrementer is now owned
4153          * by L2 and the L1 decrementer is provided in hdec_expires
4154          */
4155         if (kvmppc_core_pending_dec(vcpu) &&
4156                         ((get_tb() < vcpu->arch.dec_expires) ||
4157                          (trap == BOOK3S_INTERRUPT_SYSCALL &&
4158                           kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4159                 kvmppc_core_dequeue_dec(vcpu);
4160
4161         trace_kvm_guest_exit(vcpu);
4162         r = RESUME_GUEST;
4163         if (trap) {
4164                 if (!nested)
4165                         r = kvmppc_handle_exit_hv(kvm_run, vcpu, current);
4166                 else
4167                         r = kvmppc_handle_nested_exit(kvm_run, vcpu);
4168         }
4169         vcpu->arch.ret = r;
4170
4171         if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4172             !kvmppc_vcpu_woken(vcpu)) {
4173                 kvmppc_set_timer(vcpu);
4174                 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4175                         if (signal_pending(current)) {
4176                                 vcpu->stat.signal_exits++;
4177                                 kvm_run->exit_reason = KVM_EXIT_INTR;
4178                                 vcpu->arch.ret = -EINTR;
4179                                 break;
4180                         }
4181                         spin_lock(&vc->lock);
4182                         kvmppc_vcore_blocked(vc);
4183                         spin_unlock(&vc->lock);
4184                 }
4185         }
4186         vcpu->arch.ceded = 0;
4187
4188         vc->vcore_state = VCORE_INACTIVE;
4189         trace_kvmppc_run_core(vc, 1);
4190
4191  done:
4192         kvmppc_remove_runnable(vc, vcpu);
4193         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4194
4195         return vcpu->arch.ret;
4196
4197  sigpend:
4198         vcpu->stat.signal_exits++;
4199         kvm_run->exit_reason = KVM_EXIT_INTR;
4200         vcpu->arch.ret = -EINTR;
4201  out:
4202         local_irq_enable();
4203         preempt_enable();
4204         goto done;
4205 }
4206
4207 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
4208 {
4209         int r;
4210         int srcu_idx;
4211         unsigned long ebb_regs[3] = {}; /* shut up GCC */
4212         unsigned long user_tar = 0;
4213         unsigned int user_vrsave;
4214         struct kvm *kvm;
4215
4216         if (!vcpu->arch.sane) {
4217                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4218                 return -EINVAL;
4219         }
4220
4221         /*
4222          * Don't allow entry with a suspended transaction, because
4223          * the guest entry/exit code will lose it.
4224          * If the guest has TM enabled, save away their TM-related SPRs
4225          * (they will get restored by the TM unavailable interrupt).
4226          */
4227 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4228         if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4229             (current->thread.regs->msr & MSR_TM)) {
4230                 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4231                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4232                         run->fail_entry.hardware_entry_failure_reason = 0;
4233                         return -EINVAL;
4234                 }
4235                 /* Enable TM so we can read the TM SPRs */
4236                 mtmsr(mfmsr() | MSR_TM);
4237                 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4238                 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4239                 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4240                 current->thread.regs->msr &= ~MSR_TM;
4241         }
4242 #endif
4243
4244         /*
4245          * Force online to 1 for the sake of old userspace which doesn't
4246          * set it.
4247          */
4248         if (!vcpu->arch.online) {
4249                 atomic_inc(&vcpu->arch.vcore->online_count);
4250                 vcpu->arch.online = 1;
4251         }
4252
4253         kvmppc_core_prepare_to_enter(vcpu);
4254
4255         /* No need to go into the guest when all we'll do is come back out */
4256         if (signal_pending(current)) {
4257                 run->exit_reason = KVM_EXIT_INTR;
4258                 return -EINTR;
4259         }
4260
4261         kvm = vcpu->kvm;
4262         atomic_inc(&kvm->arch.vcpus_running);
4263         /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4264         smp_mb();
4265
4266         flush_all_to_thread(current);
4267
4268         /* Save userspace EBB and other register values */
4269         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4270                 ebb_regs[0] = mfspr(SPRN_EBBHR);
4271                 ebb_regs[1] = mfspr(SPRN_EBBRR);
4272                 ebb_regs[2] = mfspr(SPRN_BESCR);
4273                 user_tar = mfspr(SPRN_TAR);
4274         }
4275         user_vrsave = mfspr(SPRN_VRSAVE);
4276
4277         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
4278         vcpu->arch.pgdir = kvm->mm->pgd;
4279         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4280
4281         do {
4282                 /*
4283                  * The early POWER9 chips that can't mix radix and HPT threads
4284                  * on the same core also need the workaround for the problem
4285                  * where the TLB would prefetch entries in the guest exit path
4286                  * for radix guests using the guest PIDR value and LPID 0.
4287                  * The workaround is in the old path (kvmppc_run_vcpu())
4288                  * but not the new path (kvmhv_run_single_vcpu()).
4289                  */
4290                 if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4291                     !no_mixing_hpt_and_radix)
4292                         r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0,
4293                                                   vcpu->arch.vcore->lpcr);
4294                 else
4295                         r = kvmppc_run_vcpu(run, vcpu);
4296
4297                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4298                     !(vcpu->arch.shregs.msr & MSR_PR)) {
4299                         trace_kvm_hcall_enter(vcpu);
4300                         r = kvmppc_pseries_do_hcall(vcpu);
4301                         trace_kvm_hcall_exit(vcpu, r);
4302                         kvmppc_core_prepare_to_enter(vcpu);
4303                 } else if (r == RESUME_PAGE_FAULT) {
4304                         srcu_idx = srcu_read_lock(&kvm->srcu);
4305                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
4306                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4307                         srcu_read_unlock(&kvm->srcu, srcu_idx);
4308                 } else if (r == RESUME_PASSTHROUGH) {
4309                         if (WARN_ON(xics_on_xive()))
4310                                 r = H_SUCCESS;
4311                         else
4312                                 r = kvmppc_xics_rm_complete(vcpu, 0);
4313                 }
4314         } while (is_kvmppc_resume_guest(r));
4315
4316         /* Restore userspace EBB and other register values */
4317         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4318                 mtspr(SPRN_EBBHR, ebb_regs[0]);
4319                 mtspr(SPRN_EBBRR, ebb_regs[1]);
4320                 mtspr(SPRN_BESCR, ebb_regs[2]);
4321                 mtspr(SPRN_TAR, user_tar);
4322                 mtspr(SPRN_FSCR, current->thread.fscr);
4323         }
4324         mtspr(SPRN_VRSAVE, user_vrsave);
4325
4326         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4327         atomic_dec(&kvm->arch.vcpus_running);
4328         return r;
4329 }
4330
4331 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4332                                      int shift, int sllp)
4333 {
4334         (*sps)->page_shift = shift;
4335         (*sps)->slb_enc = sllp;
4336         (*sps)->enc[0].page_shift = shift;
4337         (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4338         /*
4339          * Add 16MB MPSS support (may get filtered out by userspace)
4340          */
4341         if (shift != 24) {
4342                 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4343                 if (penc != -1) {
4344                         (*sps)->enc[1].page_shift = 24;
4345                         (*sps)->enc[1].pte_enc = penc;
4346                 }
4347         }
4348         (*sps)++;
4349 }
4350
4351 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4352                                          struct kvm_ppc_smmu_info *info)
4353 {
4354         struct kvm_ppc_one_seg_page_size *sps;
4355
4356         /*
4357          * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4358          * POWER7 doesn't support keys for instruction accesses,
4359          * POWER8 and POWER9 do.
4360          */
4361         info->data_keys = 32;
4362         info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4363
4364         /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4365         info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4366         info->slb_size = 32;
4367
4368         /* We only support these sizes for now, and no muti-size segments */
4369         sps = &info->sps[0];
4370         kvmppc_add_seg_page_size(&sps, 12, 0);
4371         kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4372         kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4373
4374         /* If running as a nested hypervisor, we don't support HPT guests */
4375         if (kvmhv_on_pseries())
4376                 info->flags |= KVM_PPC_NO_HASH;
4377
4378         return 0;
4379 }
4380
4381 /*
4382  * Get (and clear) the dirty memory log for a memory slot.
4383  */
4384 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4385                                          struct kvm_dirty_log *log)
4386 {
4387         struct kvm_memslots *slots;
4388         struct kvm_memory_slot *memslot;
4389         int i, r;
4390         unsigned long n;
4391         unsigned long *buf, *p;
4392         struct kvm_vcpu *vcpu;
4393
4394         mutex_lock(&kvm->slots_lock);
4395
4396         r = -EINVAL;
4397         if (log->slot >= KVM_USER_MEM_SLOTS)
4398                 goto out;
4399
4400         slots = kvm_memslots(kvm);
4401         memslot = id_to_memslot(slots, log->slot);
4402         r = -ENOENT;
4403         if (!memslot->dirty_bitmap)
4404                 goto out;
4405
4406         /*
4407          * Use second half of bitmap area because both HPT and radix
4408          * accumulate bits in the first half.
4409          */
4410         n = kvm_dirty_bitmap_bytes(memslot);
4411         buf = memslot->dirty_bitmap + n / sizeof(long);
4412         memset(buf, 0, n);
4413
4414         if (kvm_is_radix(kvm))
4415                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4416         else
4417                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4418         if (r)
4419                 goto out;
4420
4421         /*
4422          * We accumulate dirty bits in the first half of the
4423          * memslot's dirty_bitmap area, for when pages are paged
4424          * out or modified by the host directly.  Pick up these
4425          * bits and add them to the map.
4426          */
4427         p = memslot->dirty_bitmap;
4428         for (i = 0; i < n / sizeof(long); ++i)
4429                 buf[i] |= xchg(&p[i], 0);
4430
4431         /* Harvest dirty bits from VPA and DTL updates */
4432         /* Note: we never modify the SLB shadow buffer areas */
4433         kvm_for_each_vcpu(i, vcpu, kvm) {
4434                 spin_lock(&vcpu->arch.vpa_update_lock);
4435                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4436                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4437                 spin_unlock(&vcpu->arch.vpa_update_lock);
4438         }
4439
4440         r = -EFAULT;
4441         if (copy_to_user(log->dirty_bitmap, buf, n))
4442                 goto out;
4443
4444         r = 0;
4445 out:
4446         mutex_unlock(&kvm->slots_lock);
4447         return r;
4448 }
4449
4450 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
4451                                         struct kvm_memory_slot *dont)
4452 {
4453         if (!dont || free->arch.rmap != dont->arch.rmap) {
4454                 vfree(free->arch.rmap);
4455                 free->arch.rmap = NULL;
4456         }
4457 }
4458
4459 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
4460                                          unsigned long npages)
4461 {
4462         slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
4463         if (!slot->arch.rmap)
4464                 return -ENOMEM;
4465
4466         return 0;
4467 }
4468
4469 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4470                                         struct kvm_memory_slot *memslot,
4471                                         const struct kvm_userspace_memory_region *mem)
4472 {
4473         return 0;
4474 }
4475
4476 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4477                                 const struct kvm_userspace_memory_region *mem,
4478                                 const struct kvm_memory_slot *old,
4479                                 const struct kvm_memory_slot *new,
4480                                 enum kvm_mr_change change)
4481 {
4482         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4483
4484         /*
4485          * If we are making a new memslot, it might make
4486          * some address that was previously cached as emulated
4487          * MMIO be no longer emulated MMIO, so invalidate
4488          * all the caches of emulated MMIO translations.
4489          */
4490         if (npages)
4491                 atomic64_inc(&kvm->arch.mmio_update);
4492
4493         /*
4494          * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4495          * have already called kvm_arch_flush_shadow_memslot() to
4496          * flush shadow mappings.  For KVM_MR_CREATE we have no
4497          * previous mappings.  So the only case to handle is
4498          * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4499          * has been changed.
4500          * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4501          * to get rid of any THP PTEs in the partition-scoped page tables
4502          * so we can track dirtiness at the page level; we flush when
4503          * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4504          * using THP PTEs.
4505          */
4506         if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4507             ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4508                 kvmppc_radix_flush_memslot(kvm, old);
4509         /*
4510          * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4511          */
4512         if (!kvm->arch.secure_guest)
4513                 return;
4514
4515         switch (change) {
4516         case KVM_MR_CREATE:
4517                 if (kvmppc_uvmem_slot_init(kvm, new))
4518                         return;
4519                 uv_register_mem_slot(kvm->arch.lpid,
4520                                      new->base_gfn << PAGE_SHIFT,
4521                                      new->npages * PAGE_SIZE,
4522                                      0, new->id);
4523                 break;
4524         case KVM_MR_DELETE:
4525                 uv_unregister_mem_slot(kvm->arch.lpid, old->id);
4526                 kvmppc_uvmem_slot_free(kvm, old);
4527                 break;
4528         default:
4529                 /* TODO: Handle KVM_MR_MOVE */
4530                 break;
4531         }
4532 }
4533
4534 /*
4535  * Update LPCR values in kvm->arch and in vcores.
4536  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4537  * of kvm->arch.lpcr update).
4538  */
4539 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4540 {
4541         long int i;
4542         u32 cores_done = 0;
4543
4544         if ((kvm->arch.lpcr & mask) == lpcr)
4545                 return;
4546
4547         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4548
4549         for (i = 0; i < KVM_MAX_VCORES; ++i) {
4550                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4551                 if (!vc)
4552                         continue;
4553                 spin_lock(&vc->lock);
4554                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4555                 spin_unlock(&vc->lock);
4556                 if (++cores_done >= kvm->arch.online_vcores)
4557                         break;
4558         }
4559 }
4560
4561 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
4562 {
4563         return;
4564 }
4565
4566 void kvmppc_setup_partition_table(struct kvm *kvm)
4567 {
4568         unsigned long dw0, dw1;
4569
4570         if (!kvm_is_radix(kvm)) {
4571                 /* PS field - page size for VRMA */
4572                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4573                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4574                 /* HTABSIZE and HTABORG fields */
4575                 dw0 |= kvm->arch.sdr1;
4576
4577                 /* Second dword as set by userspace */
4578                 dw1 = kvm->arch.process_table;
4579         } else {
4580                 dw0 = PATB_HR | radix__get_tree_size() |
4581                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4582                 dw1 = PATB_GR | kvm->arch.process_table;
4583         }
4584         kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4585 }
4586
4587 /*
4588  * Set up HPT (hashed page table) and RMA (real-mode area).
4589  * Must be called with kvm->arch.mmu_setup_lock held.
4590  */
4591 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4592 {
4593         int err = 0;
4594         struct kvm *kvm = vcpu->kvm;
4595         unsigned long hva;
4596         struct kvm_memory_slot *memslot;
4597         struct vm_area_struct *vma;
4598         unsigned long lpcr = 0, senc;
4599         unsigned long psize, porder;
4600         int srcu_idx;
4601
4602         /* Allocate hashed page table (if not done already) and reset it */
4603         if (!kvm->arch.hpt.virt) {
4604                 int order = KVM_DEFAULT_HPT_ORDER;
4605                 struct kvm_hpt_info info;
4606
4607                 err = kvmppc_allocate_hpt(&info, order);
4608                 /* If we get here, it means userspace didn't specify a
4609                  * size explicitly.  So, try successively smaller
4610                  * sizes if the default failed. */
4611                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4612                         err  = kvmppc_allocate_hpt(&info, order);
4613
4614                 if (err < 0) {
4615                         pr_err("KVM: Couldn't alloc HPT\n");
4616                         goto out;
4617                 }
4618
4619                 kvmppc_set_hpt(kvm, &info);
4620         }
4621
4622         /* Look up the memslot for guest physical address 0 */
4623         srcu_idx = srcu_read_lock(&kvm->srcu);
4624         memslot = gfn_to_memslot(kvm, 0);
4625
4626         /* We must have some memory at 0 by now */
4627         err = -EINVAL;
4628         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4629                 goto out_srcu;
4630
4631         /* Look up the VMA for the start of this memory slot */
4632         hva = memslot->userspace_addr;
4633         down_read(&kvm->mm->mmap_sem);
4634         vma = find_vma(kvm->mm, hva);
4635         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4636                 goto up_out;
4637
4638         psize = vma_kernel_pagesize(vma);
4639
4640         up_read(&kvm->mm->mmap_sem);
4641
4642         /* We can handle 4k, 64k or 16M pages in the VRMA */
4643         if (psize >= 0x1000000)
4644                 psize = 0x1000000;
4645         else if (psize >= 0x10000)
4646                 psize = 0x10000;
4647         else
4648                 psize = 0x1000;
4649         porder = __ilog2(psize);
4650
4651         senc = slb_pgsize_encoding(psize);
4652         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4653                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4654         /* Create HPTEs in the hash page table for the VRMA */
4655         kvmppc_map_vrma(vcpu, memslot, porder);
4656
4657         /* Update VRMASD field in the LPCR */
4658         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4659                 /* the -4 is to account for senc values starting at 0x10 */
4660                 lpcr = senc << (LPCR_VRMASD_SH - 4);
4661                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4662         }
4663
4664         /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4665         smp_wmb();
4666         err = 0;
4667  out_srcu:
4668         srcu_read_unlock(&kvm->srcu, srcu_idx);
4669  out:
4670         return err;
4671
4672  up_out:
4673         up_read(&kvm->mm->mmap_sem);
4674         goto out_srcu;
4675 }
4676
4677 /*
4678  * Must be called with kvm->arch.mmu_setup_lock held and
4679  * mmu_ready = 0 and no vcpus running.
4680  */
4681 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4682 {
4683         if (nesting_enabled(kvm))
4684                 kvmhv_release_all_nested(kvm);
4685         kvmppc_rmap_reset(kvm);
4686         kvm->arch.process_table = 0;
4687         /* Mutual exclusion with kvm_unmap_hva_range etc. */
4688         spin_lock(&kvm->mmu_lock);
4689         kvm->arch.radix = 0;
4690         spin_unlock(&kvm->mmu_lock);
4691         kvmppc_free_radix(kvm);
4692         kvmppc_update_lpcr(kvm, LPCR_VPM1,
4693                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4694         return 0;
4695 }
4696
4697 /*
4698  * Must be called with kvm->arch.mmu_setup_lock held and
4699  * mmu_ready = 0 and no vcpus running.
4700  */
4701 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4702 {
4703         int err;
4704
4705         err = kvmppc_init_vm_radix(kvm);
4706         if (err)
4707                 return err;
4708         kvmppc_rmap_reset(kvm);
4709         /* Mutual exclusion with kvm_unmap_hva_range etc. */
4710         spin_lock(&kvm->mmu_lock);
4711         kvm->arch.radix = 1;
4712         spin_unlock(&kvm->mmu_lock);
4713         kvmppc_free_hpt(&kvm->arch.hpt);
4714         kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4715                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4716         return 0;
4717 }
4718
4719 #ifdef CONFIG_KVM_XICS
4720 /*
4721  * Allocate a per-core structure for managing state about which cores are
4722  * running in the host versus the guest and for exchanging data between
4723  * real mode KVM and CPU running in the host.
4724  * This is only done for the first VM.
4725  * The allocated structure stays even if all VMs have stopped.
4726  * It is only freed when the kvm-hv module is unloaded.
4727  * It's OK for this routine to fail, we just don't support host
4728  * core operations like redirecting H_IPI wakeups.
4729  */
4730 void kvmppc_alloc_host_rm_ops(void)
4731 {
4732         struct kvmppc_host_rm_ops *ops;
4733         unsigned long l_ops;
4734         int cpu, core;
4735         int size;
4736
4737         /* Not the first time here ? */
4738         if (kvmppc_host_rm_ops_hv != NULL)
4739                 return;
4740
4741         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4742         if (!ops)
4743                 return;
4744
4745         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4746         ops->rm_core = kzalloc(size, GFP_KERNEL);
4747
4748         if (!ops->rm_core) {
4749                 kfree(ops);
4750                 return;
4751         }
4752
4753         cpus_read_lock();
4754
4755         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4756                 if (!cpu_online(cpu))
4757                         continue;
4758
4759                 core = cpu >> threads_shift;
4760                 ops->rm_core[core].rm_state.in_host = 1;
4761         }
4762
4763         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4764
4765         /*
4766          * Make the contents of the kvmppc_host_rm_ops structure visible
4767          * to other CPUs before we assign it to the global variable.
4768          * Do an atomic assignment (no locks used here), but if someone
4769          * beats us to it, just free our copy and return.
4770          */
4771         smp_wmb();
4772         l_ops = (unsigned long) ops;
4773
4774         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4775                 cpus_read_unlock();
4776                 kfree(ops->rm_core);
4777                 kfree(ops);
4778                 return;
4779         }
4780
4781         cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4782                                              "ppc/kvm_book3s:prepare",
4783                                              kvmppc_set_host_core,
4784                                              kvmppc_clear_host_core);
4785         cpus_read_unlock();
4786 }
4787
4788 void kvmppc_free_host_rm_ops(void)
4789 {
4790         if (kvmppc_host_rm_ops_hv) {
4791                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4792                 kfree(kvmppc_host_rm_ops_hv->rm_core);
4793                 kfree(kvmppc_host_rm_ops_hv);
4794                 kvmppc_host_rm_ops_hv = NULL;
4795         }
4796 }
4797 #endif
4798
4799 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4800 {
4801         unsigned long lpcr, lpid;
4802         char buf[32];
4803         int ret;
4804
4805         mutex_init(&kvm->arch.uvmem_lock);
4806         INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
4807         mutex_init(&kvm->arch.mmu_setup_lock);
4808
4809         /* Allocate the guest's logical partition ID */
4810
4811         lpid = kvmppc_alloc_lpid();
4812         if ((long)lpid < 0)
4813                 return -ENOMEM;
4814         kvm->arch.lpid = lpid;
4815
4816         kvmppc_alloc_host_rm_ops();
4817
4818         kvmhv_vm_nested_init(kvm);
4819
4820         /*
4821          * Since we don't flush the TLB when tearing down a VM,
4822          * and this lpid might have previously been used,
4823          * make sure we flush on each core before running the new VM.
4824          * On POWER9, the tlbie in mmu_partition_table_set_entry()
4825          * does this flush for us.
4826          */
4827         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4828                 cpumask_setall(&kvm->arch.need_tlb_flush);
4829
4830         /* Start out with the default set of hcalls enabled */
4831         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4832                sizeof(kvm->arch.enabled_hcalls));
4833
4834         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4835                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4836
4837         /* Init LPCR for virtual RMA mode */
4838         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4839                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
4840                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4841                 lpcr &= LPCR_PECE | LPCR_LPES;
4842         } else {
4843                 lpcr = 0;
4844         }
4845         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4846                 LPCR_VPM0 | LPCR_VPM1;
4847         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4848                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4849         /* On POWER8 turn on online bit to enable PURR/SPURR */
4850         if (cpu_has_feature(CPU_FTR_ARCH_207S))
4851                 lpcr |= LPCR_ONL;
4852         /*
4853          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4854          * Set HVICE bit to enable hypervisor virtualization interrupts.
4855          * Set HEIC to prevent OS interrupts to go to hypervisor (should
4856          * be unnecessary but better safe than sorry in case we re-enable
4857          * EE in HV mode with this LPCR still set)
4858          */
4859         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4860                 lpcr &= ~LPCR_VPM0;
4861                 lpcr |= LPCR_HVICE | LPCR_HEIC;
4862
4863                 /*
4864                  * If xive is enabled, we route 0x500 interrupts directly
4865                  * to the guest.
4866                  */
4867                 if (xics_on_xive())
4868                         lpcr |= LPCR_LPES;
4869         }
4870
4871         /*
4872          * If the host uses radix, the guest starts out as radix.
4873          */
4874         if (radix_enabled()) {
4875                 kvm->arch.radix = 1;
4876                 kvm->arch.mmu_ready = 1;
4877                 lpcr &= ~LPCR_VPM1;
4878                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4879                 ret = kvmppc_init_vm_radix(kvm);
4880                 if (ret) {
4881                         kvmppc_free_lpid(kvm->arch.lpid);
4882                         return ret;
4883                 }
4884                 kvmppc_setup_partition_table(kvm);
4885         }
4886
4887         kvm->arch.lpcr = lpcr;
4888
4889         /* Initialization for future HPT resizes */
4890         kvm->arch.resize_hpt = NULL;
4891
4892         /*
4893          * Work out how many sets the TLB has, for the use of
4894          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4895          */
4896         if (radix_enabled())
4897                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
4898         else if (cpu_has_feature(CPU_FTR_ARCH_300))
4899                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
4900         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4901                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
4902         else
4903                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
4904
4905         /*
4906          * Track that we now have a HV mode VM active. This blocks secondary
4907          * CPU threads from coming online.
4908          * On POWER9, we only need to do this if the "indep_threads_mode"
4909          * module parameter has been set to N.
4910          */
4911         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4912                 if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4913                         pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4914                         kvm->arch.threads_indep = true;
4915                 } else {
4916                         kvm->arch.threads_indep = indep_threads_mode;
4917                 }
4918         }
4919         if (!kvm->arch.threads_indep)
4920                 kvm_hv_vm_activated();
4921
4922         /*
4923          * Initialize smt_mode depending on processor.
4924          * POWER8 and earlier have to use "strict" threading, where
4925          * all vCPUs in a vcore have to run on the same (sub)core,
4926          * whereas on POWER9 the threads can each run a different
4927          * guest.
4928          */
4929         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4930                 kvm->arch.smt_mode = threads_per_subcore;
4931         else
4932                 kvm->arch.smt_mode = 1;
4933         kvm->arch.emul_smt_mode = 1;
4934
4935         /*
4936          * Create a debugfs directory for the VM
4937          */
4938         snprintf(buf, sizeof(buf), "vm%d", current->pid);
4939         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4940         kvmppc_mmu_debugfs_init(kvm);
4941         if (radix_enabled())
4942                 kvmhv_radix_debugfs_init(kvm);
4943
4944         return 0;
4945 }
4946
4947 static void kvmppc_free_vcores(struct kvm *kvm)
4948 {
4949         long int i;
4950
4951         for (i = 0; i < KVM_MAX_VCORES; ++i)
4952                 kfree(kvm->arch.vcores[i]);
4953         kvm->arch.online_vcores = 0;
4954 }
4955
4956 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4957 {
4958         debugfs_remove_recursive(kvm->arch.debugfs_dir);
4959
4960         if (!kvm->arch.threads_indep)
4961                 kvm_hv_vm_deactivated();
4962
4963         kvmppc_free_vcores(kvm);
4964
4965
4966         if (kvm_is_radix(kvm))
4967                 kvmppc_free_radix(kvm);
4968         else
4969                 kvmppc_free_hpt(&kvm->arch.hpt);
4970
4971         /* Perform global invalidation and return lpid to the pool */
4972         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4973                 if (nesting_enabled(kvm))
4974                         kvmhv_release_all_nested(kvm);
4975                 kvm->arch.process_table = 0;
4976                 if (kvm->arch.secure_guest)
4977                         uv_svm_terminate(kvm->arch.lpid);
4978                 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
4979         }
4980
4981         kvmppc_free_lpid(kvm->arch.lpid);
4982
4983         kvmppc_free_pimap(kvm);
4984 }
4985
4986 /* We don't need to emulate any privileged instructions or dcbz */
4987 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4988                                      unsigned int inst, int *advance)
4989 {
4990         return EMULATE_FAIL;
4991 }
4992
4993 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4994                                         ulong spr_val)
4995 {
4996         return EMULATE_FAIL;
4997 }
4998
4999 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5000                                         ulong *spr_val)
5001 {
5002         return EMULATE_FAIL;
5003 }
5004
5005 static int kvmppc_core_check_processor_compat_hv(void)
5006 {
5007         if (cpu_has_feature(CPU_FTR_HVMODE) &&
5008             cpu_has_feature(CPU_FTR_ARCH_206))
5009                 return 0;
5010
5011         /* POWER9 in radix mode is capable of being a nested hypervisor. */
5012         if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5013                 return 0;
5014
5015         return -EIO;
5016 }
5017
5018 #ifdef CONFIG_KVM_XICS
5019
5020 void kvmppc_free_pimap(struct kvm *kvm)
5021 {
5022         kfree(kvm->arch.pimap);
5023 }
5024
5025 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5026 {
5027         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5028 }
5029
5030 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5031 {
5032         struct irq_desc *desc;
5033         struct kvmppc_irq_map *irq_map;
5034         struct kvmppc_passthru_irqmap *pimap;
5035         struct irq_chip *chip;
5036         int i, rc = 0;
5037
5038         if (!kvm_irq_bypass)
5039                 return 1;
5040
5041         desc = irq_to_desc(host_irq);
5042         if (!desc)
5043                 return -EIO;
5044
5045         mutex_lock(&kvm->lock);
5046
5047         pimap = kvm->arch.pimap;
5048         if (pimap == NULL) {
5049                 /* First call, allocate structure to hold IRQ map */
5050                 pimap = kvmppc_alloc_pimap();
5051                 if (pimap == NULL) {
5052                         mutex_unlock(&kvm->lock);
5053                         return -ENOMEM;
5054                 }
5055                 kvm->arch.pimap = pimap;
5056         }
5057
5058         /*
5059          * For now, we only support interrupts for which the EOI operation
5060          * is an OPAL call followed by a write to XIRR, since that's
5061          * what our real-mode EOI code does, or a XIVE interrupt
5062          */
5063         chip = irq_data_get_irq_chip(&desc->irq_data);
5064         if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5065                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5066                         host_irq, guest_gsi);
5067                 mutex_unlock(&kvm->lock);
5068                 return -ENOENT;
5069         }
5070
5071         /*
5072          * See if we already have an entry for this guest IRQ number.
5073          * If it's mapped to a hardware IRQ number, that's an error,
5074          * otherwise re-use this entry.
5075          */
5076         for (i = 0; i < pimap->n_mapped; i++) {
5077                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5078                         if (pimap->mapped[i].r_hwirq) {
5079                                 mutex_unlock(&kvm->lock);
5080                                 return -EINVAL;
5081                         }
5082                         break;
5083                 }
5084         }
5085
5086         if (i == KVMPPC_PIRQ_MAPPED) {
5087                 mutex_unlock(&kvm->lock);
5088                 return -EAGAIN;         /* table is full */
5089         }
5090
5091         irq_map = &pimap->mapped[i];
5092
5093         irq_map->v_hwirq = guest_gsi;
5094         irq_map->desc = desc;
5095
5096         /*
5097          * Order the above two stores before the next to serialize with
5098          * the KVM real mode handler.
5099          */
5100         smp_wmb();
5101         irq_map->r_hwirq = desc->irq_data.hwirq;
5102
5103         if (i == pimap->n_mapped)
5104                 pimap->n_mapped++;
5105
5106         if (xics_on_xive())
5107                 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5108         else
5109                 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5110         if (rc)
5111                 irq_map->r_hwirq = 0;
5112
5113         mutex_unlock(&kvm->lock);
5114
5115         return 0;
5116 }
5117
5118 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5119 {
5120         struct irq_desc *desc;
5121         struct kvmppc_passthru_irqmap *pimap;
5122         int i, rc = 0;
5123
5124         if (!kvm_irq_bypass)
5125                 return 0;
5126
5127         desc = irq_to_desc(host_irq);
5128         if (!desc)
5129                 return -EIO;
5130
5131         mutex_lock(&kvm->lock);
5132         if (!kvm->arch.pimap)
5133                 goto unlock;
5134
5135         pimap = kvm->arch.pimap;
5136
5137         for (i = 0; i < pimap->n_mapped; i++) {
5138                 if (guest_gsi == pimap->mapped[i].v_hwirq)
5139                         break;
5140         }
5141
5142         if (i == pimap->n_mapped) {
5143                 mutex_unlock(&kvm->lock);
5144                 return -ENODEV;
5145         }
5146
5147         if (xics_on_xive())
5148                 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5149         else
5150                 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5151
5152         /* invalidate the entry (what do do on error from the above ?) */
5153         pimap->mapped[i].r_hwirq = 0;
5154
5155         /*
5156          * We don't free this structure even when the count goes to
5157          * zero. The structure is freed when we destroy the VM.
5158          */
5159  unlock:
5160         mutex_unlock(&kvm->lock);
5161         return rc;
5162 }
5163
5164 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5165                                              struct irq_bypass_producer *prod)
5166 {
5167         int ret = 0;
5168         struct kvm_kernel_irqfd *irqfd =
5169                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5170
5171         irqfd->producer = prod;
5172
5173         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5174         if (ret)
5175                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5176                         prod->irq, irqfd->gsi, ret);
5177
5178         return ret;
5179 }
5180
5181 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5182                                               struct irq_bypass_producer *prod)
5183 {
5184         int ret;
5185         struct kvm_kernel_irqfd *irqfd =
5186                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5187
5188         irqfd->producer = NULL;
5189
5190         /*
5191          * When producer of consumer is unregistered, we change back to
5192          * default external interrupt handling mode - KVM real mode
5193          * will switch back to host.
5194          */
5195         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5196         if (ret)
5197                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5198                         prod->irq, irqfd->gsi, ret);
5199 }
5200 #endif
5201
5202 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5203                                  unsigned int ioctl, unsigned long arg)
5204 {
5205         struct kvm *kvm __maybe_unused = filp->private_data;
5206         void __user *argp = (void __user *)arg;
5207         long r;
5208
5209         switch (ioctl) {
5210
5211         case KVM_PPC_ALLOCATE_HTAB: {
5212                 u32 htab_order;
5213
5214                 r = -EFAULT;
5215                 if (get_user(htab_order, (u32 __user *)argp))
5216                         break;
5217                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5218                 if (r)
5219                         break;
5220                 r = 0;
5221                 break;
5222         }
5223
5224         case KVM_PPC_GET_HTAB_FD: {
5225                 struct kvm_get_htab_fd ghf;
5226
5227                 r = -EFAULT;
5228                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5229                         break;
5230                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5231                 break;
5232         }
5233
5234         case KVM_PPC_RESIZE_HPT_PREPARE: {
5235                 struct kvm_ppc_resize_hpt rhpt;
5236
5237                 r = -EFAULT;
5238                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5239                         break;
5240
5241                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5242                 break;
5243         }
5244
5245         case KVM_PPC_RESIZE_HPT_COMMIT: {
5246                 struct kvm_ppc_resize_hpt rhpt;
5247
5248                 r = -EFAULT;
5249                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5250                         break;
5251
5252                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5253                 break;
5254         }
5255
5256         default:
5257                 r = -ENOTTY;
5258         }
5259
5260         return r;
5261 }
5262
5263 /*
5264  * List of hcall numbers to enable by default.
5265  * For compatibility with old userspace, we enable by default
5266  * all hcalls that were implemented before the hcall-enabling
5267  * facility was added.  Note this list should not include H_RTAS.
5268  */
5269 static unsigned int default_hcall_list[] = {
5270         H_REMOVE,
5271         H_ENTER,
5272         H_READ,
5273         H_PROTECT,
5274         H_BULK_REMOVE,
5275         H_GET_TCE,
5276         H_PUT_TCE,
5277         H_SET_DABR,
5278         H_SET_XDABR,
5279         H_CEDE,
5280         H_PROD,
5281         H_CONFER,
5282         H_REGISTER_VPA,
5283 #ifdef CONFIG_KVM_XICS
5284         H_EOI,
5285         H_CPPR,
5286         H_IPI,
5287         H_IPOLL,
5288         H_XIRR,
5289         H_XIRR_X,
5290 #endif
5291         0
5292 };
5293
5294 static void init_default_hcalls(void)
5295 {
5296         int i;
5297         unsigned int hcall;
5298
5299         for (i = 0; default_hcall_list[i]; ++i) {
5300                 hcall = default_hcall_list[i];
5301                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5302                 __set_bit(hcall / 4, default_enabled_hcalls);
5303         }
5304 }
5305
5306 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5307 {
5308         unsigned long lpcr;
5309         int radix;
5310         int err;
5311
5312         /* If not on a POWER9, reject it */
5313         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5314                 return -ENODEV;
5315
5316         /* If any unknown flags set, reject it */
5317         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5318                 return -EINVAL;
5319
5320         /* GR (guest radix) bit in process_table field must match */
5321         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5322         if (!!(cfg->process_table & PATB_GR) != radix)
5323                 return -EINVAL;
5324
5325         /* Process table size field must be reasonable, i.e. <= 24 */
5326         if ((cfg->process_table & PRTS_MASK) > 24)
5327                 return -EINVAL;
5328
5329         /* We can change a guest to/from radix now, if the host is radix */
5330         if (radix && !radix_enabled())
5331                 return -EINVAL;
5332
5333         /* If we're a nested hypervisor, we currently only support radix */
5334         if (kvmhv_on_pseries() && !radix)
5335                 return -EINVAL;
5336
5337         mutex_lock(&kvm->arch.mmu_setup_lock);
5338         if (radix != kvm_is_radix(kvm)) {
5339                 if (kvm->arch.mmu_ready) {
5340                         kvm->arch.mmu_ready = 0;
5341                         /* order mmu_ready vs. vcpus_running */
5342                         smp_mb();
5343                         if (atomic_read(&kvm->arch.vcpus_running)) {
5344                                 kvm->arch.mmu_ready = 1;
5345                                 err = -EBUSY;
5346                                 goto out_unlock;
5347                         }
5348                 }
5349                 if (radix)
5350                         err = kvmppc_switch_mmu_to_radix(kvm);
5351                 else
5352                         err = kvmppc_switch_mmu_to_hpt(kvm);
5353                 if (err)
5354                         goto out_unlock;
5355         }
5356
5357         kvm->arch.process_table = cfg->process_table;
5358         kvmppc_setup_partition_table(kvm);
5359
5360         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5361         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5362         err = 0;
5363
5364  out_unlock:
5365         mutex_unlock(&kvm->arch.mmu_setup_lock);
5366         return err;
5367 }
5368
5369 static int kvmhv_enable_nested(struct kvm *kvm)
5370 {
5371         if (!nested)
5372                 return -EPERM;
5373         if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5374                 return -ENODEV;
5375
5376         /* kvm == NULL means the caller is testing if the capability exists */
5377         if (kvm)
5378                 kvm->arch.nested_enable = true;
5379         return 0;
5380 }
5381
5382 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5383                                  int size)
5384 {
5385         int rc = -EINVAL;
5386
5387         if (kvmhv_vcpu_is_radix(vcpu)) {
5388                 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5389
5390                 if (rc > 0)
5391                         rc = -EINVAL;
5392         }
5393
5394         /* For now quadrants are the only way to access nested guest memory */
5395         if (rc && vcpu->arch.nested)
5396                 rc = -EAGAIN;
5397
5398         return rc;
5399 }
5400
5401 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5402                                 int size)
5403 {
5404         int rc = -EINVAL;
5405
5406         if (kvmhv_vcpu_is_radix(vcpu)) {
5407                 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5408
5409                 if (rc > 0)
5410                         rc = -EINVAL;
5411         }
5412
5413         /* For now quadrants are the only way to access nested guest memory */
5414         if (rc && vcpu->arch.nested)
5415                 rc = -EAGAIN;
5416
5417         return rc;
5418 }
5419
5420 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5421 {
5422         unpin_vpa(kvm, vpa);
5423         vpa->gpa = 0;
5424         vpa->pinned_addr = NULL;
5425         vpa->dirty = false;
5426         vpa->update_pending = 0;
5427 }
5428
5429 /*
5430  *  IOCTL handler to turn off secure mode of guest
5431  *
5432  * - Release all device pages
5433  * - Issue ucall to terminate the guest on the UV side
5434  * - Unpin the VPA pages.
5435  * - Reinit the partition scoped page tables
5436  */
5437 static int kvmhv_svm_off(struct kvm *kvm)
5438 {
5439         struct kvm_vcpu *vcpu;
5440         int mmu_was_ready;
5441         int srcu_idx;
5442         int ret = 0;
5443         int i;
5444
5445         if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5446                 return ret;
5447
5448         mutex_lock(&kvm->arch.mmu_setup_lock);
5449         mmu_was_ready = kvm->arch.mmu_ready;
5450         if (kvm->arch.mmu_ready) {
5451                 kvm->arch.mmu_ready = 0;
5452                 /* order mmu_ready vs. vcpus_running */
5453                 smp_mb();
5454                 if (atomic_read(&kvm->arch.vcpus_running)) {
5455                         kvm->arch.mmu_ready = 1;
5456                         ret = -EBUSY;
5457                         goto out;
5458                 }
5459         }
5460
5461         srcu_idx = srcu_read_lock(&kvm->srcu);
5462         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5463                 struct kvm_memory_slot *memslot;
5464                 struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5465
5466                 if (!slots)
5467                         continue;
5468
5469                 kvm_for_each_memslot(memslot, slots) {
5470                         kvmppc_uvmem_drop_pages(memslot, kvm, true);
5471                         uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5472                 }
5473         }
5474         srcu_read_unlock(&kvm->srcu, srcu_idx);
5475
5476         ret = uv_svm_terminate(kvm->arch.lpid);
5477         if (ret != U_SUCCESS) {
5478                 ret = -EINVAL;
5479                 goto out;
5480         }
5481
5482         /*
5483          * When secure guest is reset, all the guest pages are sent
5484          * to UV via UV_PAGE_IN before the non-boot vcpus get a
5485          * chance to run and unpin their VPA pages. Unpinning of all
5486          * VPA pages is done here explicitly so that VPA pages
5487          * can be migrated to the secure side.
5488          *
5489          * This is required to for the secure SMP guest to reboot
5490          * correctly.
5491          */
5492         kvm_for_each_vcpu(i, vcpu, kvm) {
5493                 spin_lock(&vcpu->arch.vpa_update_lock);
5494                 unpin_vpa_reset(kvm, &vcpu->arch.dtl);
5495                 unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
5496                 unpin_vpa_reset(kvm, &vcpu->arch.vpa);
5497                 spin_unlock(&vcpu->arch.vpa_update_lock);
5498         }
5499
5500         kvmppc_setup_partition_table(kvm);
5501         kvm->arch.secure_guest = 0;
5502         kvm->arch.mmu_ready = mmu_was_ready;
5503 out:
5504         mutex_unlock(&kvm->arch.mmu_setup_lock);
5505         return ret;
5506 }
5507
5508 static struct kvmppc_ops kvm_ops_hv = {
5509         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5510         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5511         .get_one_reg = kvmppc_get_one_reg_hv,
5512         .set_one_reg = kvmppc_set_one_reg_hv,
5513         .vcpu_load   = kvmppc_core_vcpu_load_hv,
5514         .vcpu_put    = kvmppc_core_vcpu_put_hv,
5515         .inject_interrupt = kvmppc_inject_interrupt_hv,
5516         .set_msr     = kvmppc_set_msr_hv,
5517         .vcpu_run    = kvmppc_vcpu_run_hv,
5518         .vcpu_create = kvmppc_core_vcpu_create_hv,
5519         .vcpu_free   = kvmppc_core_vcpu_free_hv,
5520         .check_requests = kvmppc_core_check_requests_hv,
5521         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5522         .flush_memslot  = kvmppc_core_flush_memslot_hv,
5523         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5524         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5525         .unmap_hva_range = kvm_unmap_hva_range_hv,
5526         .age_hva  = kvm_age_hva_hv,
5527         .test_age_hva = kvm_test_age_hva_hv,
5528         .set_spte_hva = kvm_set_spte_hva_hv,
5529         .mmu_destroy  = kvmppc_mmu_destroy_hv,
5530         .free_memslot = kvmppc_core_free_memslot_hv,
5531         .create_memslot = kvmppc_core_create_memslot_hv,
5532         .init_vm =  kvmppc_core_init_vm_hv,
5533         .destroy_vm = kvmppc_core_destroy_vm_hv,
5534         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5535         .emulate_op = kvmppc_core_emulate_op_hv,
5536         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5537         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5538         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5539         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5540         .hcall_implemented = kvmppc_hcall_impl_hv,
5541 #ifdef CONFIG_KVM_XICS
5542         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5543         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5544 #endif
5545         .configure_mmu = kvmhv_configure_mmu,
5546         .get_rmmu_info = kvmhv_get_rmmu_info,
5547         .set_smt_mode = kvmhv_set_smt_mode,
5548         .enable_nested = kvmhv_enable_nested,
5549         .load_from_eaddr = kvmhv_load_from_eaddr,
5550         .store_to_eaddr = kvmhv_store_to_eaddr,
5551         .svm_off = kvmhv_svm_off,
5552 };
5553
5554 static int kvm_init_subcore_bitmap(void)
5555 {
5556         int i, j;
5557         int nr_cores = cpu_nr_cores();
5558         struct sibling_subcore_state *sibling_subcore_state;
5559
5560         for (i = 0; i < nr_cores; i++) {
5561                 int first_cpu = i * threads_per_core;
5562                 int node = cpu_to_node(first_cpu);
5563
5564                 /* Ignore if it is already allocated. */
5565                 if (paca_ptrs[first_cpu]->sibling_subcore_state)
5566                         continue;
5567
5568                 sibling_subcore_state =
5569                         kzalloc_node(sizeof(struct sibling_subcore_state),
5570                                                         GFP_KERNEL, node);
5571                 if (!sibling_subcore_state)
5572                         return -ENOMEM;
5573
5574
5575                 for (j = 0; j < threads_per_core; j++) {
5576                         int cpu = first_cpu + j;
5577
5578                         paca_ptrs[cpu]->sibling_subcore_state =
5579                                                 sibling_subcore_state;
5580                 }
5581         }
5582         return 0;
5583 }
5584
5585 static int kvmppc_radix_possible(void)
5586 {
5587         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5588 }
5589
5590 static int kvmppc_book3s_init_hv(void)
5591 {
5592         int r;
5593
5594         if (!tlbie_capable) {
5595                 pr_err("KVM-HV: Host does not support TLBIE\n");
5596                 return -ENODEV;
5597         }
5598
5599         /*
5600          * FIXME!! Do we need to check on all cpus ?
5601          */
5602         r = kvmppc_core_check_processor_compat_hv();
5603         if (r < 0)
5604                 return -ENODEV;
5605
5606         r = kvmhv_nested_init();
5607         if (r)
5608                 return r;
5609
5610         r = kvm_init_subcore_bitmap();
5611         if (r)
5612                 return r;
5613
5614         /*
5615          * We need a way of accessing the XICS interrupt controller,
5616          * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5617          * indirectly, via OPAL.
5618          */
5619 #ifdef CONFIG_SMP
5620         if (!xics_on_xive() && !kvmhv_on_pseries() &&
5621             !local_paca->kvm_hstate.xics_phys) {
5622                 struct device_node *np;
5623
5624                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5625                 if (!np) {
5626                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5627                         return -ENODEV;
5628                 }
5629                 /* presence of intc confirmed - node can be dropped again */
5630                 of_node_put(np);
5631         }
5632 #endif
5633
5634         kvm_ops_hv.owner = THIS_MODULE;
5635         kvmppc_hv_ops = &kvm_ops_hv;
5636
5637         init_default_hcalls();
5638
5639         init_vcore_lists();
5640
5641         r = kvmppc_mmu_hv_init();
5642         if (r)
5643                 return r;
5644
5645         if (kvmppc_radix_possible())
5646                 r = kvmppc_radix_init();
5647
5648         /*
5649          * POWER9 chips before version 2.02 can't have some threads in
5650          * HPT mode and some in radix mode on the same core.
5651          */
5652         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5653                 unsigned int pvr = mfspr(SPRN_PVR);
5654                 if ((pvr >> 16) == PVR_POWER9 &&
5655                     (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5656                      ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5657                         no_mixing_hpt_and_radix = true;
5658         }
5659
5660         r = kvmppc_uvmem_init();
5661         if (r < 0)
5662                 pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
5663
5664         return r;
5665 }
5666
5667 static void kvmppc_book3s_exit_hv(void)
5668 {
5669         kvmppc_uvmem_free();
5670         kvmppc_free_host_rm_ops();
5671         if (kvmppc_radix_possible())
5672                 kvmppc_radix_exit();
5673         kvmppc_hv_ops = NULL;
5674         kvmhv_nested_exit();
5675 }
5676
5677 module_init(kvmppc_book3s_init_hv);
5678 module_exit(kvmppc_book3s_exit_hv);
5679 MODULE_LICENSE("GPL");
5680 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5681 MODULE_ALIAS("devname:kvm");