arch/powerpc/kvm/book3s_hv.c

   1 /*
   2  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
   3  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
   4  *
   5  * Authors:
   6  *    Paul Mackerras <paulus@au1.ibm.com>
   7  *    Alexander Graf <agraf@suse.de>
   8  *    Kevin Wolf <mail@kevin-wolf.de>
   9  *
  10  * Description: KVM functions specific to running on Book 3S
  11  * processors in hypervisor mode (specifically POWER7 and later).
  12  *
  13  * This file is derived from arch/powerpc/kvm/book3s.c,
  14  * by Alexander Graf <agraf@suse.de>.
  15  *
  16  * This program is free software; you can redistribute it and/or modify
  17  * it under the terms of the GNU General Public License, version 2, as
  18  * published by the Free Software Foundation.
  19  */
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kernel.h>
  23 #include <linux/err.h>
  24 #include <linux/slab.h>
  25 #include <linux/preempt.h>
  26 #include <linux/sched/signal.h>
  27 #include <linux/sched/stat.h>
  28 #include <linux/delay.h>
  29 #include <linux/export.h>
  30 #include <linux/fs.h>
  31 #include <linux/anon_inodes.h>
  32 #include <linux/cpu.h>
  33 #include <linux/cpumask.h>
  34 #include <linux/spinlock.h>
  35 #include <linux/page-flags.h>
  36 #include <linux/srcu.h>
  37 #include <linux/miscdevice.h>
  38 #include <linux/debugfs.h>
  39 #include <linux/gfp.h>
  40 #include <linux/vmalloc.h>
  41 #include <linux/highmem.h>
  42 #include <linux/hugetlb.h>
  43 #include <linux/kvm_irqfd.h>
  44 #include <linux/irqbypass.h>
  45 #include <linux/module.h>
  46 #include <linux/compiler.h>
  47 #include <linux/of.h>
  48
  49 #include <asm/reg.h>
  50 #include <asm/ppc-opcode.h>
  51 #include <asm/asm-prototypes.h>
  52 #include <asm/debug.h>
  53 #include <asm/disassemble.h>
  54 #include <asm/cputable.h>
  55 #include <asm/cacheflush.h>
  56 #include <asm/tlbflush.h>
  57 #include <linux/uaccess.h>
  58 #include <asm/io.h>
  59 #include <asm/kvm_ppc.h>
  60 #include <asm/kvm_book3s.h>
  61 #include <asm/mmu_context.h>
  62 #include <asm/lppaca.h>
  63 #include <asm/processor.h>
  64 #include <asm/cputhreads.h>
  65 #include <asm/page.h>
  66 #include <asm/hvcall.h>
  67 #include <asm/switch_to.h>
  68 #include <asm/smp.h>
  69 #include <asm/dbell.h>
  70 #include <asm/hmi.h>
  71 #include <asm/pnv-pci.h>
  72 #include <asm/mmu.h>
  73 #include <asm/opal.h>
  74 #include <asm/xics.h>
  75 #include <asm/xive.h>
  76
  77 #include "book3s.h"
  78
  79 #define CREATE_TRACE_POINTS
  80 #include "trace_hv.h"
  81
  82 /* #define EXIT_DEBUG */
  83 /* #define EXIT_DEBUG_SIMPLE */
  84 /* #define EXIT_DEBUG_INT */
  85
  86 /* Used to indicate that a guest page fault needs to be handled */
  87 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  88 /* Used to indicate that a guest passthrough interrupt needs to be handled */
  89 #define RESUME_PASSTHROUGH      (RESUME_GUEST | RESUME_FLAG_ARCH2)
  90
  91 /* Used as a "null" value for timebase values */
  92 #define TB_NIL  (~(u64)0)
  93
  94 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
  95
  96 static int dynamic_mt_modes = 6;
  97 module_param(dynamic_mt_modes, int, 0644);
  98 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
  99 static int target_smt_mode;
 100 module_param(target_smt_mode, int, 0644);
 101 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
 102
 103 static bool indep_threads_mode = true;
 104 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
 105 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
 106
 107 #ifdef CONFIG_KVM_XICS
 108 static struct kernel_param_ops module_param_ops = {
 109         .set = param_set_int,
 110         .get = param_get_int,
 111 };
 112
 113 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
 114 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
 115
 116 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
 117 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
 118 #endif
 119
 120 /* If set, the threads on each CPU core have to be in the same MMU mode */
 121 static bool no_mixing_hpt_and_radix;
 122
 123 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
 124 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
 125
 126 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
 127                 int *ip)
 128 {
 129         int i = *ip;
 130         struct kvm_vcpu *vcpu;
 131
 132         while (++i < MAX_SMT_THREADS) {
 133                 vcpu = READ_ONCE(vc->runnable_threads[i]);
 134                 if (vcpu) {
 135                         *ip = i;
 136                         return vcpu;
 137                 }
 138         }
 139         return NULL;
 140 }
 141
 142 /* Used to traverse the list of runnable threads for a given vcore */
 143 #define for_each_runnable_thread(i, vcpu, vc) \
 144         for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
 145
 146 static bool kvmppc_ipi_thread(int cpu)
 147 {
 148         unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
 149
 150         /* On POWER9 we can use msgsnd to IPI any cpu */
 151         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
 152                 msg |= get_hard_smp_processor_id(cpu);
 153                 smp_mb();
 154                 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 155                 return true;
 156         }
 157
 158         /* On POWER8 for IPIs to threads in the same core, use msgsnd */
 159         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
 160                 preempt_disable();
 161                 if (cpu_first_thread_sibling(cpu) ==
 162                     cpu_first_thread_sibling(smp_processor_id())) {
 163                         msg |= cpu_thread_in_core(cpu);
 164                         smp_mb();
 165                         __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 166                         preempt_enable();
 167                         return true;
 168                 }
 169                 preempt_enable();
 170         }
 171
 172 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
 173         if (cpu >= 0 && cpu < nr_cpu_ids) {
 174                 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
 175                         xics_wake_cpu(cpu);
 176                         return true;
 177                 }
 178                 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
 179                 return true;
 180         }
 181 #endif
 182
 183         return false;
 184 }
 185
 186 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
 187 {
 188         int cpu;
 189         struct swait_queue_head *wqp;
 190
 191         wqp = kvm_arch_vcpu_wq(vcpu);
 192         if (swq_has_sleeper(wqp)) {
 193                 swake_up(wqp);
 194                 ++vcpu->stat.halt_wakeup;
 195         }
 196
 197         cpu = READ_ONCE(vcpu->arch.thread_cpu);
 198         if (cpu >= 0 && kvmppc_ipi_thread(cpu))
 199                 return;
 200
 201         /* CPU points to the first thread of the core */
 202         cpu = vcpu->cpu;
 203         if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
 204                 smp_send_reschedule(cpu);
 205 }
 206
 207 /*
 208  * We use the vcpu_load/put functions to measure stolen time.
 209  * Stolen time is counted as time when either the vcpu is able to
 210  * run as part of a virtual core, but the task running the vcore
 211  * is preempted or sleeping, or when the vcpu needs something done
 212  * in the kernel by the task running the vcpu, but that task is
 213  * preempted or sleeping.  Those two things have to be counted
 214  * separately, since one of the vcpu tasks will take on the job
 215  * of running the core, and the other vcpu tasks in the vcore will
 216  * sleep waiting for it to do that, but that sleep shouldn't count
 217  * as stolen time.
 218  *
 219  * Hence we accumulate stolen time when the vcpu can run as part of
 220  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 221  * needs its task to do other things in the kernel (for example,
 222  * service a page fault) in busy_stolen.  We don't accumulate
 223  * stolen time for a vcore when it is inactive, or for a vcpu
 224  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 225  * a misnomer; it means that the vcpu task is not executing in
 226  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 227  * the kernel.  We don't have any way of dividing up that time
 228  * between time that the vcpu is genuinely stopped, time that
 229  * the task is actively working on behalf of the vcpu, and time
 230  * that the task is preempted, so we don't count any of it as
 231  * stolen.
 232  *
 233  * Updates to busy_stolen are protected by arch.tbacct_lock;
 234  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
 235  * lock.  The stolen times are measured in units of timebase ticks.
 236  * (Note that the != TB_NIL checks below are purely defensive;
 237  * they should never fail.)
 238  */
 239
 240 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
 241 {
 242         unsigned long flags;
 243
 244         spin_lock_irqsave(&vc->stoltb_lock, flags);
 245         vc->preempt_tb = mftb();
 246         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 247 }
 248
 249 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
 250 {
 251         unsigned long flags;
 252
 253         spin_lock_irqsave(&vc->stoltb_lock, flags);
 254         if (vc->preempt_tb != TB_NIL) {
 255                 vc->stolen_tb += mftb() - vc->preempt_tb;
 256                 vc->preempt_tb = TB_NIL;
 257         }
 258         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 259 }
 260
 261 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 262 {
 263         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 264         unsigned long flags;
 265
 266         /*
 267          * We can test vc->runner without taking the vcore lock,
 268          * because only this task ever sets vc->runner to this
 269          * vcpu, and once it is set to this vcpu, only this task
 270          * ever sets it to NULL.
 271          */
 272         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 273                 kvmppc_core_end_stolen(vc);
 274
 275         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 276         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 277             vcpu->arch.busy_preempt != TB_NIL) {
 278                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 279                 vcpu->arch.busy_preempt = TB_NIL;
 280         }
 281         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 282 }
 283
 284 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 285 {
 286         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 287         unsigned long flags;
 288
 289         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 290                 kvmppc_core_start_stolen(vc);
 291
 292         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 293         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 294                 vcpu->arch.busy_preempt = mftb();
 295         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 296 }
 297
 298 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
 299 {
 300         /*
 301          * Check for illegal transactional state bit combination
 302          * and if we find it, force the TS field to a safe state.
 303          */
 304         if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
 305                 msr &= ~MSR_TS_MASK;
 306         vcpu->arch.shregs.msr = msr;
 307         kvmppc_end_cede(vcpu);
 308 }
 309
 310 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 311 {
 312         vcpu->arch.pvr = pvr;
 313 }
 314
 315 /* Dummy value used in computing PCR value below */
 316 #define PCR_ARCH_300    (PCR_ARCH_207 << 1)
 317
 318 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 319 {
 320         unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
 321         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 322
 323         /* We can (emulate) our own architecture version and anything older */
 324         if (cpu_has_feature(CPU_FTR_ARCH_300))
 325                 host_pcr_bit = PCR_ARCH_300;
 326         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
 327                 host_pcr_bit = PCR_ARCH_207;
 328         else if (cpu_has_feature(CPU_FTR_ARCH_206))
 329                 host_pcr_bit = PCR_ARCH_206;
 330         else
 331                 host_pcr_bit = PCR_ARCH_205;
 332
 333         /* Determine lowest PCR bit needed to run guest in given PVR level */
 334         guest_pcr_bit = host_pcr_bit;
 335         if (arch_compat) {
 336                 switch (arch_compat) {
 337                 case PVR_ARCH_205:
 338                         guest_pcr_bit = PCR_ARCH_205;
 339                         break;
 340                 case PVR_ARCH_206:
 341                 case PVR_ARCH_206p:
 342                         guest_pcr_bit = PCR_ARCH_206;
 343                         break;
 344                 case PVR_ARCH_207:
 345                         guest_pcr_bit = PCR_ARCH_207;
 346                         break;
 347                 case PVR_ARCH_300:
 348                         guest_pcr_bit = PCR_ARCH_300;
 349                         break;
 350                 default:
 351                         return -EINVAL;
 352                 }
 353         }
 354
 355         /* Check requested PCR bits don't exceed our capabilities */
 356         if (guest_pcr_bit > host_pcr_bit)
 357                 return -EINVAL;
 358
 359         spin_lock(&vc->lock);
 360         vc->arch_compat = arch_compat;
 361         /* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
 362         vc->pcr = host_pcr_bit - guest_pcr_bit;
 363         spin_unlock(&vc->lock);
 364
 365         return 0;
 366 }
 367
 368 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 369 {
 370         int r;
 371
 372         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 373         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 374                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
 375         for (r = 0; r < 16; ++r)
 376                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 377                        r, kvmppc_get_gpr(vcpu, r),
 378                        r+16, kvmppc_get_gpr(vcpu, r+16));
 379         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 380                vcpu->arch.ctr, vcpu->arch.lr);
 381         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 382                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 383         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 384                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 385         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 386                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 387         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 388                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 389         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 390         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 391                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 392         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 393         for (r = 0; r < vcpu->arch.slb_max; ++r)
 394                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 395                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 396         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 397                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 398                vcpu->arch.last_inst);
 399 }
 400
 401 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 402 {
 403         struct kvm_vcpu *ret;
 404
 405         mutex_lock(&kvm->lock);
 406         ret = kvm_get_vcpu_by_id(kvm, id);
 407         mutex_unlock(&kvm->lock);
 408         return ret;
 409 }
 410
 411 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 412 {
 413         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 414         vpa->yield_count = cpu_to_be32(1);
 415 }
 416
 417 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 418                    unsigned long addr, unsigned long len)
 419 {
 420         /* check address is cacheline aligned */
 421         if (addr & (L1_CACHE_BYTES - 1))
 422                 return -EINVAL;
 423         spin_lock(&vcpu->arch.vpa_update_lock);
 424         if (v->next_gpa != addr || v->len != len) {
 425                 v->next_gpa = addr;
 426                 v->len = addr ? len : 0;
 427                 v->update_pending = 1;
 428         }
 429         spin_unlock(&vcpu->arch.vpa_update_lock);
 430         return 0;
 431 }
 432
 433 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 434 struct reg_vpa {
 435         u32 dummy;
 436         union {
 437                 __be16 hword;
 438                 __be32 word;
 439         } length;
 440 };
 441
 442 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 443 {
 444         if (vpap->update_pending)
 445                 return vpap->next_gpa != 0;
 446         return vpap->pinned_addr != NULL;
 447 }
 448
 449 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 450                                        unsigned long flags,
 451                                        unsigned long vcpuid, unsigned long vpa)
 452 {
 453         struct kvm *kvm = vcpu->kvm;
 454         unsigned long len, nb;
 455         void *va;
 456         struct kvm_vcpu *tvcpu;
 457         int err;
 458         int subfunc;
 459         struct kvmppc_vpa *vpap;
 460
 461         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 462         if (!tvcpu)
 463                 return H_PARAMETER;
 464
 465         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 466         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 467             subfunc == H_VPA_REG_SLB) {
 468                 /* Registering new area - address must be cache-line aligned */
 469                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 470                         return H_PARAMETER;
 471
 472                 /* convert logical addr to kernel addr and read length */
 473                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 474                 if (va == NULL)
 475                         return H_PARAMETER;
 476                 if (subfunc == H_VPA_REG_VPA)
 477                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 478                 else
 479                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 480                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 481
 482                 /* Check length */
 483                 if (len > nb || len < sizeof(struct reg_vpa))
 484                         return H_PARAMETER;
 485         } else {
 486                 vpa = 0;
 487                 len = 0;
 488         }
 489
 490         err = H_PARAMETER;
 491         vpap = NULL;
 492         spin_lock(&tvcpu->arch.vpa_update_lock);
 493
 494         switch (subfunc) {
 495         case H_VPA_REG_VPA:             /* register VPA */
 496                 /*
 497                  * The size of our lppaca is 1kB because of the way we align
 498                  * it for the guest to avoid crossing a 4kB boundary. We only
 499                  * use 640 bytes of the structure though, so we should accept
 500                  * clients that set a size of 640.
 501                  */
 502                 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
 503                 if (len < sizeof(struct lppaca))
 504                         break;
 505                 vpap = &tvcpu->arch.vpa;
 506                 err = 0;
 507                 break;
 508
 509         case H_VPA_REG_DTL:             /* register DTL */
 510                 if (len < sizeof(struct dtl_entry))
 511                         break;
 512                 len -= len % sizeof(struct dtl_entry);
 513
 514                 /* Check that they have previously registered a VPA */
 515                 err = H_RESOURCE;
 516                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 517                         break;
 518
 519                 vpap = &tvcpu->arch.dtl;
 520                 err = 0;
 521                 break;
 522
 523         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 524                 /* Check that they have previously registered a VPA */
 525                 err = H_RESOURCE;
 526                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 527                         break;
 528
 529                 vpap = &tvcpu->arch.slb_shadow;
 530                 err = 0;
 531                 break;
 532
 533         case H_VPA_DEREG_VPA:           /* deregister VPA */
 534                 /* Check they don't still have a DTL or SLB buf registered */
 535                 err = H_RESOURCE;
 536                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 537                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 538                         break;
 539
 540                 vpap = &tvcpu->arch.vpa;
 541                 err = 0;
 542                 break;
 543
 544         case H_VPA_DEREG_DTL:           /* deregister DTL */
 545                 vpap = &tvcpu->arch.dtl;
 546                 err = 0;
 547                 break;
 548
 549         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 550                 vpap = &tvcpu->arch.slb_shadow;
 551                 err = 0;
 552                 break;
 553         }
 554
 555         if (vpap) {
 556                 vpap->next_gpa = vpa;
 557                 vpap->len = len;
 558                 vpap->update_pending = 1;
 559         }
 560
 561         spin_unlock(&tvcpu->arch.vpa_update_lock);
 562
 563         return err;
 564 }
 565
 566 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 567 {
 568         struct kvm *kvm = vcpu->kvm;
 569         void *va;
 570         unsigned long nb;
 571         unsigned long gpa;
 572
 573         /*
 574          * We need to pin the page pointed to by vpap->next_gpa,
 575          * but we can't call kvmppc_pin_guest_page under the lock
 576          * as it does get_user_pages() and down_read().  So we
 577          * have to drop the lock, pin the page, then get the lock
 578          * again and check that a new area didn't get registered
 579          * in the meantime.
 580          */
 581         for (;;) {
 582                 gpa = vpap->next_gpa;
 583                 spin_unlock(&vcpu->arch.vpa_update_lock);
 584                 va = NULL;
 585                 nb = 0;
 586                 if (gpa)
 587                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 588                 spin_lock(&vcpu->arch.vpa_update_lock);
 589                 if (gpa == vpap->next_gpa)
 590                         break;
 591                 /* sigh... unpin that one and try again */
 592                 if (va)
 593                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 594         }
 595
 596         vpap->update_pending = 0;
 597         if (va && nb < vpap->len) {
 598                 /*
 599                  * If it's now too short, it must be that userspace
 600                  * has changed the mappings underlying guest memory,
 601                  * so unregister the region.
 602                  */
 603                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 604                 va = NULL;
 605         }
 606         if (vpap->pinned_addr)
 607                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 608                                         vpap->dirty);
 609         vpap->gpa = gpa;
 610         vpap->pinned_addr = va;
 611         vpap->dirty = false;
 612         if (va)
 613                 vpap->pinned_end = va + vpap->len;
 614 }
 615
 616 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 617 {
 618         if (!(vcpu->arch.vpa.update_pending ||
 619               vcpu->arch.slb_shadow.update_pending ||
 620               vcpu->arch.dtl.update_pending))
 621                 return;
 622
 623         spin_lock(&vcpu->arch.vpa_update_lock);
 624         if (vcpu->arch.vpa.update_pending) {
 625                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 626                 if (vcpu->arch.vpa.pinned_addr)
 627                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 628         }
 629         if (vcpu->arch.dtl.update_pending) {
 630                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 631                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 632                 vcpu->arch.dtl_index = 0;
 633         }
 634         if (vcpu->arch.slb_shadow.update_pending)
 635                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 636         spin_unlock(&vcpu->arch.vpa_update_lock);
 637 }
 638
 639 /*
 640  * Return the accumulated stolen time for the vcore up until `now'.
 641  * The caller should hold the vcore lock.
 642  */
 643 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 644 {
 645         u64 p;
 646         unsigned long flags;
 647
 648         spin_lock_irqsave(&vc->stoltb_lock, flags);
 649         p = vc->stolen_tb;
 650         if (vc->vcore_state != VCORE_INACTIVE &&
 651             vc->preempt_tb != TB_NIL)
 652                 p += now - vc->preempt_tb;
 653         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 654         return p;
 655 }
 656
 657 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 658                                     struct kvmppc_vcore *vc)
 659 {
 660         struct dtl_entry *dt;
 661         struct lppaca *vpa;
 662         unsigned long stolen;
 663         unsigned long core_stolen;
 664         u64 now;
 665         unsigned long flags;
 666
 667         dt = vcpu->arch.dtl_ptr;
 668         vpa = vcpu->arch.vpa.pinned_addr;
 669         now = mftb();
 670         core_stolen = vcore_stolen_time(vc, now);
 671         stolen = core_stolen - vcpu->arch.stolen_logged;
 672         vcpu->arch.stolen_logged = core_stolen;
 673         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 674         stolen += vcpu->arch.busy_stolen;
 675         vcpu->arch.busy_stolen = 0;
 676         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 677         if (!dt || !vpa)
 678                 return;
 679         memset(dt, 0, sizeof(struct dtl_entry));
 680         dt->dispatch_reason = 7;
 681         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 682         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 683         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 684         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 685         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 686         ++dt;
 687         if (dt == vcpu->arch.dtl.pinned_end)
 688                 dt = vcpu->arch.dtl.pinned_addr;
 689         vcpu->arch.dtl_ptr = dt;
 690         /* order writing *dt vs. writing vpa->dtl_idx */
 691         smp_wmb();
 692         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 693         vcpu->arch.dtl.dirty = true;
 694 }
 695
 696 /* See if there is a doorbell interrupt pending for a vcpu */
 697 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
 698 {
 699         int thr;
 700         struct kvmppc_vcore *vc;
 701
 702         if (vcpu->arch.doorbell_request)
 703                 return true;
 704         /*
 705          * Ensure that the read of vcore->dpdes comes after the read
 706          * of vcpu->doorbell_request.  This barrier matches the
 707          * lwsync in book3s_hv_rmhandlers.S just before the
 708          * fast_guest_return label.
 709          */
 710         smp_rmb();
 711         vc = vcpu->arch.vcore;
 712         thr = vcpu->vcpu_id - vc->first_vcpuid;
 713         return !!(vc->dpdes & (1 << thr));
 714 }
 715
 716 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 717 {
 718         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 719                 return true;
 720         if ((!vcpu->arch.vcore->arch_compat) &&
 721             cpu_has_feature(CPU_FTR_ARCH_207S))
 722                 return true;
 723         return false;
 724 }
 725
 726 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 727                              unsigned long resource, unsigned long value1,
 728                              unsigned long value2)
 729 {
 730         switch (resource) {
 731         case H_SET_MODE_RESOURCE_SET_CIABR:
 732                 if (!kvmppc_power8_compatible(vcpu))
 733                         return H_P2;
 734                 if (value2)
 735                         return H_P4;
 736                 if (mflags)
 737                         return H_UNSUPPORTED_FLAG_START;
 738                 /* Guests can't breakpoint the hypervisor */
 739                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 740                         return H_P3;
 741                 vcpu->arch.ciabr  = value1;
 742                 return H_SUCCESS;
 743         case H_SET_MODE_RESOURCE_SET_DAWR:
 744                 if (!kvmppc_power8_compatible(vcpu))
 745                         return H_P2;
 746                 if (!ppc_breakpoint_available())
 747                         return H_P2;
 748                 if (mflags)
 749                         return H_UNSUPPORTED_FLAG_START;
 750                 if (value2 & DABRX_HYP)
 751                         return H_P4;
 752                 vcpu->arch.dawr  = value1;
 753                 vcpu->arch.dawrx = value2;
 754                 return H_SUCCESS;
 755         default:
 756                 return H_TOO_HARD;
 757         }
 758 }
 759
 760 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 761 {
 762         struct kvmppc_vcore *vcore = target->arch.vcore;
 763
 764         /*
 765          * We expect to have been called by the real mode handler
 766          * (kvmppc_rm_h_confer()) which would have directly returned
 767          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 768          * have useful work to do and should not confer) so we don't
 769          * recheck that here.
 770          */
 771
 772         spin_lock(&vcore->lock);
 773         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 774             vcore->vcore_state != VCORE_INACTIVE &&
 775             vcore->runner)
 776                 target = vcore->runner;
 777         spin_unlock(&vcore->lock);
 778
 779         return kvm_vcpu_yield_to(target);
 780 }
 781
 782 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 783 {
 784         int yield_count = 0;
 785         struct lppaca *lppaca;
 786
 787         spin_lock(&vcpu->arch.vpa_update_lock);
 788         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 789         if (lppaca)
 790                 yield_count = be32_to_cpu(lppaca->yield_count);
 791         spin_unlock(&vcpu->arch.vpa_update_lock);
 792         return yield_count;
 793 }
 794
 795 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 796 {
 797         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 798         unsigned long target, ret = H_SUCCESS;
 799         int yield_count;
 800         struct kvm_vcpu *tvcpu;
 801         int idx, rc;
 802
 803         if (req <= MAX_HCALL_OPCODE &&
 804             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 805                 return RESUME_HOST;
 806
 807         switch (req) {
 808         case H_CEDE:
 809                 break;
 810         case H_PROD:
 811                 target = kvmppc_get_gpr(vcpu, 4);
 812                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 813                 if (!tvcpu) {
 814                         ret = H_PARAMETER;
 815                         break;
 816                 }
 817                 tvcpu->arch.prodded = 1;
 818                 smp_mb();
 819                 if (tvcpu->arch.ceded)
 820                         kvmppc_fast_vcpu_kick_hv(tvcpu);
 821                 break;
 822         case H_CONFER:
 823                 target = kvmppc_get_gpr(vcpu, 4);
 824                 if (target == -1)
 825                         break;
 826                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 827                 if (!tvcpu) {
 828                         ret = H_PARAMETER;
 829                         break;
 830                 }
 831                 yield_count = kvmppc_get_gpr(vcpu, 5);
 832                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 833                         break;
 834                 kvm_arch_vcpu_yield_to(tvcpu);
 835                 break;
 836         case H_REGISTER_VPA:
 837                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 838                                         kvmppc_get_gpr(vcpu, 5),
 839                                         kvmppc_get_gpr(vcpu, 6));
 840                 break;
 841         case H_RTAS:
 842                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 843                         return RESUME_HOST;
 844
 845                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 846                 rc = kvmppc_rtas_hcall(vcpu);
 847                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 848
 849                 if (rc == -ENOENT)
 850                         return RESUME_HOST;
 851                 else if (rc == 0)
 852                         break;
 853
 854                 /* Send the error out to userspace via KVM_RUN */
 855                 return rc;
 856         case H_LOGICAL_CI_LOAD:
 857                 ret = kvmppc_h_logical_ci_load(vcpu);
 858                 if (ret == H_TOO_HARD)
 859                         return RESUME_HOST;
 860                 break;
 861         case H_LOGICAL_CI_STORE:
 862                 ret = kvmppc_h_logical_ci_store(vcpu);
 863                 if (ret == H_TOO_HARD)
 864                         return RESUME_HOST;
 865                 break;
 866         case H_SET_MODE:
 867                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 868                                         kvmppc_get_gpr(vcpu, 5),
 869                                         kvmppc_get_gpr(vcpu, 6),
 870                                         kvmppc_get_gpr(vcpu, 7));
 871                 if (ret == H_TOO_HARD)
 872                         return RESUME_HOST;
 873                 break;
 874         case H_XIRR:
 875         case H_CPPR:
 876         case H_EOI:
 877         case H_IPI:
 878         case H_IPOLL:
 879         case H_XIRR_X:
 880                 if (kvmppc_xics_enabled(vcpu)) {
 881                         if (xive_enabled()) {
 882                                 ret = H_NOT_AVAILABLE;
 883                                 return RESUME_GUEST;
 884                         }
 885                         ret = kvmppc_xics_hcall(vcpu, req);
 886                         break;
 887                 }
 888                 return RESUME_HOST;
 889         case H_PUT_TCE:
 890                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 891                                                 kvmppc_get_gpr(vcpu, 5),
 892                                                 kvmppc_get_gpr(vcpu, 6));
 893                 if (ret == H_TOO_HARD)
 894                         return RESUME_HOST;
 895                 break;
 896         case H_PUT_TCE_INDIRECT:
 897                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
 898                                                 kvmppc_get_gpr(vcpu, 5),
 899                                                 kvmppc_get_gpr(vcpu, 6),
 900                                                 kvmppc_get_gpr(vcpu, 7));
 901                 if (ret == H_TOO_HARD)
 902                         return RESUME_HOST;
 903                 break;
 904         case H_STUFF_TCE:
 905                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 906                                                 kvmppc_get_gpr(vcpu, 5),
 907                                                 kvmppc_get_gpr(vcpu, 6),
 908                                                 kvmppc_get_gpr(vcpu, 7));
 909                 if (ret == H_TOO_HARD)
 910                         return RESUME_HOST;
 911                 break;
 912         default:
 913                 return RESUME_HOST;
 914         }
 915         kvmppc_set_gpr(vcpu, 3, ret);
 916         vcpu->arch.hcall_needed = 0;
 917         return RESUME_GUEST;
 918 }
 919
 920 static int kvmppc_hcall_impl_hv(unsigned long cmd)
 921 {
 922         switch (cmd) {
 923         case H_CEDE:
 924         case H_PROD:
 925         case H_CONFER:
 926         case H_REGISTER_VPA:
 927         case H_SET_MODE:
 928         case H_LOGICAL_CI_LOAD:
 929         case H_LOGICAL_CI_STORE:
 930 #ifdef CONFIG_KVM_XICS
 931         case H_XIRR:
 932         case H_CPPR:
 933         case H_EOI:
 934         case H_IPI:
 935         case H_IPOLL:
 936         case H_XIRR_X:
 937 #endif
 938                 return 1;
 939         }
 940
 941         /* See if it's in the real-mode table */
 942         return kvmppc_hcall_impl_hv_realmode(cmd);
 943 }
 944
 945 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
 946                                         struct kvm_vcpu *vcpu)
 947 {
 948         u32 last_inst;
 949
 950         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
 951                                         EMULATE_DONE) {
 952                 /*
 953                  * Fetch failed, so return to guest and
 954                  * try executing it again.
 955                  */
 956                 return RESUME_GUEST;
 957         }
 958
 959         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
 960                 run->exit_reason = KVM_EXIT_DEBUG;
 961                 run->debug.arch.address = kvmppc_get_pc(vcpu);
 962                 return RESUME_HOST;
 963         } else {
 964                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 965                 return RESUME_GUEST;
 966         }
 967 }
 968
 969 static void do_nothing(void *x)
 970 {
 971 }
 972
 973 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
 974 {
 975         int thr, cpu, pcpu, nthreads;
 976         struct kvm_vcpu *v;
 977         unsigned long dpdes;
 978
 979         nthreads = vcpu->kvm->arch.emul_smt_mode;
 980         dpdes = 0;
 981         cpu = vcpu->vcpu_id & ~(nthreads - 1);
 982         for (thr = 0; thr < nthreads; ++thr, ++cpu) {
 983                 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
 984                 if (!v)
 985                         continue;
 986                 /*
 987                  * If the vcpu is currently running on a physical cpu thread,
 988                  * interrupt it in order to pull it out of the guest briefly,
 989                  * which will update its vcore->dpdes value.
 990                  */
 991                 pcpu = READ_ONCE(v->cpu);
 992                 if (pcpu >= 0)
 993                         smp_call_function_single(pcpu, do_nothing, NULL, 1);
 994                 if (kvmppc_doorbell_pending(v))
 995                         dpdes |= 1 << thr;
 996         }
 997         return dpdes;
 998 }
 999
1000 /*
1001  * On POWER9, emulate doorbell-related instructions in order to
1002  * give the guest the illusion of running on a multi-threaded core.
1003  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1004  * and mfspr DPDES.
1005  */
1006 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1007 {
1008         u32 inst, rb, thr;
1009         unsigned long arg;
1010         struct kvm *kvm = vcpu->kvm;
1011         struct kvm_vcpu *tvcpu;
1012
1013         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1014                 return RESUME_GUEST;
1015         if (get_op(inst) != 31)
1016                 return EMULATE_FAIL;
1017         rb = get_rb(inst);
1018         thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1019         switch (get_xop(inst)) {
1020         case OP_31_XOP_MSGSNDP:
1021                 arg = kvmppc_get_gpr(vcpu, rb);
1022                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1023                         break;
1024                 arg &= 0x3f;
1025                 if (arg >= kvm->arch.emul_smt_mode)
1026                         break;
1027                 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1028                 if (!tvcpu)
1029                         break;
1030                 if (!tvcpu->arch.doorbell_request) {
1031                         tvcpu->arch.doorbell_request = 1;
1032                         kvmppc_fast_vcpu_kick_hv(tvcpu);
1033                 }
1034                 break;
1035         case OP_31_XOP_MSGCLRP:
1036                 arg = kvmppc_get_gpr(vcpu, rb);
1037                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1038                         break;
1039                 vcpu->arch.vcore->dpdes = 0;
1040                 vcpu->arch.doorbell_request = 0;
1041                 break;
1042         case OP_31_XOP_MFSPR:
1043                 switch (get_sprn(inst)) {
1044                 case SPRN_TIR:
1045                         arg = thr;
1046                         break;
1047                 case SPRN_DPDES:
1048                         arg = kvmppc_read_dpdes(vcpu);
1049                         break;
1050                 default:
1051                         return EMULATE_FAIL;
1052                 }
1053                 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1054                 break;
1055         default:
1056                 return EMULATE_FAIL;
1057         }
1058         kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1059         return RESUME_GUEST;
1060 }
1061
1062 /* Called with vcpu->arch.vcore->lock held */
1063 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1064                                  struct task_struct *tsk)
1065 {
1066         int r = RESUME_HOST;
1067
1068         vcpu->stat.sum_exits++;
1069
1070         /*
1071          * This can happen if an interrupt occurs in the last stages
1072          * of guest entry or the first stages of guest exit (i.e. after
1073          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1074          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1075          * That can happen due to a bug, or due to a machine check
1076          * occurring at just the wrong time.
1077          */
1078         if (vcpu->arch.shregs.msr & MSR_HV) {
1079                 printk(KERN_EMERG "KVM trap in HV mode!\n");
1080                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1081                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1082                         vcpu->arch.shregs.msr);
1083                 kvmppc_dump_regs(vcpu);
1084                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1085                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1086                 return RESUME_HOST;
1087         }
1088         run->exit_reason = KVM_EXIT_UNKNOWN;
1089         run->ready_for_interrupt_injection = 1;
1090         switch (vcpu->arch.trap) {
1091         /* We're good on these - the host merely wanted to get our attention */
1092         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1093                 vcpu->stat.dec_exits++;
1094                 r = RESUME_GUEST;
1095                 break;
1096         case BOOK3S_INTERRUPT_EXTERNAL:
1097         case BOOK3S_INTERRUPT_H_DOORBELL:
1098         case BOOK3S_INTERRUPT_H_VIRT:
1099                 vcpu->stat.ext_intr_exits++;
1100                 r = RESUME_GUEST;
1101                 break;
1102         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1103         case BOOK3S_INTERRUPT_HMI:
1104         case BOOK3S_INTERRUPT_PERFMON:
1105         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1106                 r = RESUME_GUEST;
1107                 break;
1108         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1109                 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1110                 run->exit_reason = KVM_EXIT_NMI;
1111                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1112                 /* Clear out the old NMI status from run->flags */
1113                 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1114                 /* Now set the NMI status */
1115                 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1116                         run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1117                 else
1118                         run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1119
1120                 r = RESUME_HOST;
1121                 /* Print the MCE event to host console. */
1122                 machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1123                 break;
1124         case BOOK3S_INTERRUPT_PROGRAM:
1125         {
1126                 ulong flags;
1127                 /*
1128                  * Normally program interrupts are delivered directly
1129                  * to the guest by the hardware, but we can get here
1130                  * as a result of a hypervisor emulation interrupt
1131                  * (e40) getting turned into a 700 by BML RTAS.
1132                  */
1133                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1134                 kvmppc_core_queue_program(vcpu, flags);
1135                 r = RESUME_GUEST;
1136                 break;
1137         }
1138         case BOOK3S_INTERRUPT_SYSCALL:
1139         {
1140                 /* hcall - punt to userspace */
1141                 int i;
1142
1143                 /* hypercall with MSR_PR has already been handled in rmode,
1144                  * and never reaches here.
1145                  */
1146
1147                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1148                 for (i = 0; i < 9; ++i)
1149                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1150                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1151                 vcpu->arch.hcall_needed = 1;
1152                 r = RESUME_HOST;
1153                 break;
1154         }
1155         /*
1156          * We get these next two if the guest accesses a page which it thinks
1157          * it has mapped but which is not actually present, either because
1158          * it is for an emulated I/O device or because the corresonding
1159          * host page has been paged out.  Any other HDSI/HISI interrupts
1160          * have been handled already.
1161          */
1162         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1163                 r = RESUME_PAGE_FAULT;
1164                 break;
1165         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1166                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1167                 vcpu->arch.fault_dsisr = 0;
1168                 r = RESUME_PAGE_FAULT;
1169                 break;
1170         /*
1171          * This occurs if the guest executes an illegal instruction.
1172          * If the guest debug is disabled, generate a program interrupt
1173          * to the guest. If guest debug is enabled, we need to check
1174          * whether the instruction is a software breakpoint instruction.
1175          * Accordingly return to Guest or Host.
1176          */
1177         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1178                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1179                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1180                                 swab32(vcpu->arch.emul_inst) :
1181                                 vcpu->arch.emul_inst;
1182                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1183                         /* Need vcore unlocked to call kvmppc_get_last_inst */
1184                         spin_unlock(&vcpu->arch.vcore->lock);
1185                         r = kvmppc_emulate_debug_inst(run, vcpu);
1186                         spin_lock(&vcpu->arch.vcore->lock);
1187                 } else {
1188                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1189                         r = RESUME_GUEST;
1190                 }
1191                 break;
1192         /*
1193          * This occurs if the guest (kernel or userspace), does something that
1194          * is prohibited by HFSCR.
1195          * On POWER9, this could be a doorbell instruction that we need
1196          * to emulate.
1197          * Otherwise, we just generate a program interrupt to the guest.
1198          */
1199         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1200                 r = EMULATE_FAIL;
1201                 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1202                     cpu_has_feature(CPU_FTR_ARCH_300)) {
1203                         /* Need vcore unlocked to call kvmppc_get_last_inst */
1204                         spin_unlock(&vcpu->arch.vcore->lock);
1205                         r = kvmppc_emulate_doorbell_instr(vcpu);
1206                         spin_lock(&vcpu->arch.vcore->lock);
1207                 }
1208                 if (r == EMULATE_FAIL) {
1209                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1210                         r = RESUME_GUEST;
1211                 }
1212                 break;
1213
1214 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1215         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1216                 /*
1217                  * This occurs for various TM-related instructions that
1218                  * we need to emulate on POWER9 DD2.2.  We have already
1219                  * handled the cases where the guest was in real-suspend
1220                  * mode and was transitioning to transactional state.
1221                  */
1222                 r = kvmhv_p9_tm_emulation(vcpu);
1223                 break;
1224 #endif
1225
1226         case BOOK3S_INTERRUPT_HV_RM_HARD:
1227                 r = RESUME_PASSTHROUGH;
1228                 break;
1229         default:
1230                 kvmppc_dump_regs(vcpu);
1231                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1232                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1233                         vcpu->arch.shregs.msr);
1234                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1235                 r = RESUME_HOST;
1236                 break;
1237         }
1238
1239         return r;
1240 }
1241
1242 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1243                                             struct kvm_sregs *sregs)
1244 {
1245         int i;
1246
1247         memset(sregs, 0, sizeof(struct kvm_sregs));
1248         sregs->pvr = vcpu->arch.pvr;
1249         for (i = 0; i < vcpu->arch.slb_max; i++) {
1250                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1251                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1252         }
1253
1254         return 0;
1255 }
1256
1257 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1258                                             struct kvm_sregs *sregs)
1259 {
1260         int i, j;
1261
1262         /* Only accept the same PVR as the host's, since we can't spoof it */
1263         if (sregs->pvr != vcpu->arch.pvr)
1264                 return -EINVAL;
1265
1266         j = 0;
1267         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1268                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1269                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1270                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1271                         ++j;
1272                 }
1273         }
1274         vcpu->arch.slb_max = j;
1275
1276         return 0;
1277 }
1278
1279 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1280                 bool preserve_top32)
1281 {
1282         struct kvm *kvm = vcpu->kvm;
1283         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1284         u64 mask;
1285
1286         mutex_lock(&kvm->lock);
1287         spin_lock(&vc->lock);
1288         /*
1289          * If ILE (interrupt little-endian) has changed, update the
1290          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1291          */
1292         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1293                 struct kvm_vcpu *vcpu;
1294                 int i;
1295
1296                 kvm_for_each_vcpu(i, vcpu, kvm) {
1297                         if (vcpu->arch.vcore != vc)
1298                                 continue;
1299                         if (new_lpcr & LPCR_ILE)
1300                                 vcpu->arch.intr_msr |= MSR_LE;
1301                         else
1302                                 vcpu->arch.intr_msr &= ~MSR_LE;
1303                 }
1304         }
1305
1306         /*
1307          * Userspace can only modify DPFD (default prefetch depth),
1308          * ILE (interrupt little-endian) and TC (translation control).
1309          * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1310          */
1311         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1312         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1313                 mask |= LPCR_AIL;
1314         /*
1315          * On POWER9, allow userspace to enable large decrementer for the
1316          * guest, whether or not the host has it enabled.
1317          */
1318         if (cpu_has_feature(CPU_FTR_ARCH_300))
1319                 mask |= LPCR_LD;
1320
1321         /* Broken 32-bit version of LPCR must not clear top bits */
1322         if (preserve_top32)
1323                 mask &= 0xFFFFFFFF;
1324         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1325         spin_unlock(&vc->lock);
1326         mutex_unlock(&kvm->lock);
1327 }
1328
1329 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1330                                  union kvmppc_one_reg *val)
1331 {
1332         int r = 0;
1333         long int i;
1334
1335         switch (id) {
1336         case KVM_REG_PPC_DEBUG_INST:
1337                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1338                 break;
1339         case KVM_REG_PPC_HIOR:
1340                 *val = get_reg_val(id, 0);
1341                 break;
1342         case KVM_REG_PPC_DABR:
1343                 *val = get_reg_val(id, vcpu->arch.dabr);
1344                 break;
1345         case KVM_REG_PPC_DABRX:
1346                 *val = get_reg_val(id, vcpu->arch.dabrx);
1347                 break;
1348         case KVM_REG_PPC_DSCR:
1349                 *val = get_reg_val(id, vcpu->arch.dscr);
1350                 break;
1351         case KVM_REG_PPC_PURR:
1352                 *val = get_reg_val(id, vcpu->arch.purr);
1353                 break;
1354         case KVM_REG_PPC_SPURR:
1355                 *val = get_reg_val(id, vcpu->arch.spurr);
1356                 break;
1357         case KVM_REG_PPC_AMR:
1358                 *val = get_reg_val(id, vcpu->arch.amr);
1359                 break;
1360         case KVM_REG_PPC_UAMOR:
1361                 *val = get_reg_val(id, vcpu->arch.uamor);
1362                 break;
1363         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1364                 i = id - KVM_REG_PPC_MMCR0;
1365                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1366                 break;
1367         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1368                 i = id - KVM_REG_PPC_PMC1;
1369                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1370                 break;
1371         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1372                 i = id - KVM_REG_PPC_SPMC1;
1373                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1374                 break;
1375         case KVM_REG_PPC_SIAR:
1376                 *val = get_reg_val(id, vcpu->arch.siar);
1377                 break;
1378         case KVM_REG_PPC_SDAR:
1379                 *val = get_reg_val(id, vcpu->arch.sdar);
1380                 break;
1381         case KVM_REG_PPC_SIER:
1382                 *val = get_reg_val(id, vcpu->arch.sier);
1383                 break;
1384         case KVM_REG_PPC_IAMR:
1385                 *val = get_reg_val(id, vcpu->arch.iamr);
1386                 break;
1387         case KVM_REG_PPC_PSPB:
1388                 *val = get_reg_val(id, vcpu->arch.pspb);
1389                 break;
1390         case KVM_REG_PPC_DPDES:
1391                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1392                 break;
1393         case KVM_REG_PPC_VTB:
1394                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1395                 break;
1396         case KVM_REG_PPC_DAWR:
1397                 *val = get_reg_val(id, vcpu->arch.dawr);
1398                 break;
1399         case KVM_REG_PPC_DAWRX:
1400                 *val = get_reg_val(id, vcpu->arch.dawrx);
1401                 break;
1402         case KVM_REG_PPC_CIABR:
1403                 *val = get_reg_val(id, vcpu->arch.ciabr);
1404                 break;
1405         case KVM_REG_PPC_CSIGR:
1406                 *val = get_reg_val(id, vcpu->arch.csigr);
1407                 break;
1408         case KVM_REG_PPC_TACR:
1409                 *val = get_reg_val(id, vcpu->arch.tacr);
1410                 break;
1411         case KVM_REG_PPC_TCSCR:
1412                 *val = get_reg_val(id, vcpu->arch.tcscr);
1413                 break;
1414         case KVM_REG_PPC_PID:
1415                 *val = get_reg_val(id, vcpu->arch.pid);
1416                 break;
1417         case KVM_REG_PPC_ACOP:
1418                 *val = get_reg_val(id, vcpu->arch.acop);
1419                 break;
1420         case KVM_REG_PPC_WORT:
1421                 *val = get_reg_val(id, vcpu->arch.wort);
1422                 break;
1423         case KVM_REG_PPC_TIDR:
1424                 *val = get_reg_val(id, vcpu->arch.tid);
1425                 break;
1426         case KVM_REG_PPC_PSSCR:
1427                 *val = get_reg_val(id, vcpu->arch.psscr);
1428                 break;
1429         case KVM_REG_PPC_VPA_ADDR:
1430                 spin_lock(&vcpu->arch.vpa_update_lock);
1431                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1432                 spin_unlock(&vcpu->arch.vpa_update_lock);
1433                 break;
1434         case KVM_REG_PPC_VPA_SLB:
1435                 spin_lock(&vcpu->arch.vpa_update_lock);
1436                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1437                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1438                 spin_unlock(&vcpu->arch.vpa_update_lock);
1439                 break;
1440         case KVM_REG_PPC_VPA_DTL:
1441                 spin_lock(&vcpu->arch.vpa_update_lock);
1442                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1443                 val->vpaval.length = vcpu->arch.dtl.len;
1444                 spin_unlock(&vcpu->arch.vpa_update_lock);
1445                 break;
1446         case KVM_REG_PPC_TB_OFFSET:
1447                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1448                 break;
1449         case KVM_REG_PPC_LPCR:
1450         case KVM_REG_PPC_LPCR_64:
1451                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1452                 break;
1453         case KVM_REG_PPC_PPR:
1454                 *val = get_reg_val(id, vcpu->arch.ppr);
1455                 break;
1456 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1457         case KVM_REG_PPC_TFHAR:
1458                 *val = get_reg_val(id, vcpu->arch.tfhar);
1459                 break;
1460         case KVM_REG_PPC_TFIAR:
1461                 *val = get_reg_val(id, vcpu->arch.tfiar);
1462                 break;
1463         case KVM_REG_PPC_TEXASR:
1464                 *val = get_reg_val(id, vcpu->arch.texasr);
1465                 break;
1466         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1467                 i = id - KVM_REG_PPC_TM_GPR0;
1468                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1469                 break;
1470         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1471         {
1472                 int j;
1473                 i = id - KVM_REG_PPC_TM_VSR0;
1474                 if (i < 32)
1475                         for (j = 0; j < TS_FPRWIDTH; j++)
1476                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1477                 else {
1478                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1479                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1480                         else
1481                                 r = -ENXIO;
1482                 }
1483                 break;
1484         }
1485         case KVM_REG_PPC_TM_CR:
1486                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1487                 break;
1488         case KVM_REG_PPC_TM_XER:
1489                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1490                 break;
1491         case KVM_REG_PPC_TM_LR:
1492                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1493                 break;
1494         case KVM_REG_PPC_TM_CTR:
1495                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1496                 break;
1497         case KVM_REG_PPC_TM_FPSCR:
1498                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1499                 break;
1500         case KVM_REG_PPC_TM_AMR:
1501                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1502                 break;
1503         case KVM_REG_PPC_TM_PPR:
1504                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1505                 break;
1506         case KVM_REG_PPC_TM_VRSAVE:
1507                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1508                 break;
1509         case KVM_REG_PPC_TM_VSCR:
1510                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1511                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1512                 else
1513                         r = -ENXIO;
1514                 break;
1515         case KVM_REG_PPC_TM_DSCR:
1516                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1517                 break;
1518         case KVM_REG_PPC_TM_TAR:
1519                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1520                 break;
1521 #endif
1522         case KVM_REG_PPC_ARCH_COMPAT:
1523                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1524                 break;
1525         case KVM_REG_PPC_DEC_EXPIRY:
1526                 *val = get_reg_val(id, vcpu->arch.dec_expires +
1527                                    vcpu->arch.vcore->tb_offset);
1528                 break;
1529         default:
1530                 r = -EINVAL;
1531                 break;
1532         }
1533
1534         return r;
1535 }
1536
1537 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1538                                  union kvmppc_one_reg *val)
1539 {
1540         int r = 0;
1541         long int i;
1542         unsigned long addr, len;
1543
1544         switch (id) {
1545         case KVM_REG_PPC_HIOR:
1546                 /* Only allow this to be set to zero */
1547                 if (set_reg_val(id, *val))
1548                         r = -EINVAL;
1549                 break;
1550         case KVM_REG_PPC_DABR:
1551                 vcpu->arch.dabr = set_reg_val(id, *val);
1552                 break;
1553         case KVM_REG_PPC_DABRX:
1554                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1555                 break;
1556         case KVM_REG_PPC_DSCR:
1557                 vcpu->arch.dscr = set_reg_val(id, *val);
1558                 break;
1559         case KVM_REG_PPC_PURR:
1560                 vcpu->arch.purr = set_reg_val(id, *val);
1561                 break;
1562         case KVM_REG_PPC_SPURR:
1563                 vcpu->arch.spurr = set_reg_val(id, *val);
1564                 break;
1565         case KVM_REG_PPC_AMR:
1566                 vcpu->arch.amr = set_reg_val(id, *val);
1567                 break;
1568         case KVM_REG_PPC_UAMOR:
1569                 vcpu->arch.uamor = set_reg_val(id, *val);
1570                 break;
1571         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1572                 i = id - KVM_REG_PPC_MMCR0;
1573                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1574                 break;
1575         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1576                 i = id - KVM_REG_PPC_PMC1;
1577                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1578                 break;
1579         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1580                 i = id - KVM_REG_PPC_SPMC1;
1581                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1582                 break;
1583         case KVM_REG_PPC_SIAR:
1584                 vcpu->arch.siar = set_reg_val(id, *val);
1585                 break;
1586         case KVM_REG_PPC_SDAR:
1587                 vcpu->arch.sdar = set_reg_val(id, *val);
1588                 break;
1589         case KVM_REG_PPC_SIER:
1590                 vcpu->arch.sier = set_reg_val(id, *val);
1591                 break;
1592         case KVM_REG_PPC_IAMR:
1593                 vcpu->arch.iamr = set_reg_val(id, *val);
1594                 break;
1595         case KVM_REG_PPC_PSPB:
1596                 vcpu->arch.pspb = set_reg_val(id, *val);
1597                 break;
1598         case KVM_REG_PPC_DPDES:
1599                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1600                 break;
1601         case KVM_REG_PPC_VTB:
1602                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1603                 break;
1604         case KVM_REG_PPC_DAWR:
1605                 vcpu->arch.dawr = set_reg_val(id, *val);
1606                 break;
1607         case KVM_REG_PPC_DAWRX:
1608                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1609                 break;
1610         case KVM_REG_PPC_CIABR:
1611                 vcpu->arch.ciabr = set_reg_val(id, *val);
1612                 /* Don't allow setting breakpoints in hypervisor code */
1613                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1614                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1615                 break;
1616         case KVM_REG_PPC_CSIGR:
1617                 vcpu->arch.csigr = set_reg_val(id, *val);
1618                 break;
1619         case KVM_REG_PPC_TACR:
1620                 vcpu->arch.tacr = set_reg_val(id, *val);
1621                 break;
1622         case KVM_REG_PPC_TCSCR:
1623                 vcpu->arch.tcscr = set_reg_val(id, *val);
1624                 break;
1625         case KVM_REG_PPC_PID:
1626                 vcpu->arch.pid = set_reg_val(id, *val);
1627                 break;
1628         case KVM_REG_PPC_ACOP:
1629                 vcpu->arch.acop = set_reg_val(id, *val);
1630                 break;
1631         case KVM_REG_PPC_WORT:
1632                 vcpu->arch.wort = set_reg_val(id, *val);
1633                 break;
1634         case KVM_REG_PPC_TIDR:
1635                 vcpu->arch.tid = set_reg_val(id, *val);
1636                 break;
1637         case KVM_REG_PPC_PSSCR:
1638                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1639                 break;
1640         case KVM_REG_PPC_VPA_ADDR:
1641                 addr = set_reg_val(id, *val);
1642                 r = -EINVAL;
1643                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1644                               vcpu->arch.dtl.next_gpa))
1645                         break;
1646                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1647                 break;
1648         case KVM_REG_PPC_VPA_SLB:
1649                 addr = val->vpaval.addr;
1650                 len = val->vpaval.length;
1651                 r = -EINVAL;
1652                 if (addr && !vcpu->arch.vpa.next_gpa)
1653                         break;
1654                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1655                 break;
1656         case KVM_REG_PPC_VPA_DTL:
1657                 addr = val->vpaval.addr;
1658                 len = val->vpaval.length;
1659                 r = -EINVAL;
1660                 if (addr && (len < sizeof(struct dtl_entry) ||
1661                              !vcpu->arch.vpa.next_gpa))
1662                         break;
1663                 len -= len % sizeof(struct dtl_entry);
1664                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1665                 break;
1666         case KVM_REG_PPC_TB_OFFSET:
1667                 /*
1668                  * POWER9 DD1 has an erratum where writing TBU40 causes
1669                  * the timebase to lose ticks.  So we don't let the
1670                  * timebase offset be changed on P9 DD1.  (It is
1671                  * initialized to zero.)
1672                  */
1673                 if (cpu_has_feature(CPU_FTR_POWER9_DD1))
1674                         break;
1675                 /* round up to multiple of 2^24 */
1676                 vcpu->arch.vcore->tb_offset =
1677                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1678                 break;
1679         case KVM_REG_PPC_LPCR:
1680                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1681                 break;
1682         case KVM_REG_PPC_LPCR_64:
1683                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1684                 break;
1685         case KVM_REG_PPC_PPR:
1686                 vcpu->arch.ppr = set_reg_val(id, *val);
1687                 break;
1688 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1689         case KVM_REG_PPC_TFHAR:
1690                 vcpu->arch.tfhar = set_reg_val(id, *val);
1691                 break;
1692         case KVM_REG_PPC_TFIAR:
1693                 vcpu->arch.tfiar = set_reg_val(id, *val);
1694                 break;
1695         case KVM_REG_PPC_TEXASR:
1696                 vcpu->arch.texasr = set_reg_val(id, *val);
1697                 break;
1698         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1699                 i = id - KVM_REG_PPC_TM_GPR0;
1700                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1701                 break;
1702         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1703         {
1704                 int j;
1705                 i = id - KVM_REG_PPC_TM_VSR0;
1706                 if (i < 32)
1707                         for (j = 0; j < TS_FPRWIDTH; j++)
1708                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1709                 else
1710                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1711                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1712                         else
1713                                 r = -ENXIO;
1714                 break;
1715         }
1716         case KVM_REG_PPC_TM_CR:
1717                 vcpu->arch.cr_tm = set_reg_val(id, *val);
1718                 break;
1719         case KVM_REG_PPC_TM_XER:
1720                 vcpu->arch.xer_tm = set_reg_val(id, *val);
1721                 break;
1722         case KVM_REG_PPC_TM_LR:
1723                 vcpu->arch.lr_tm = set_reg_val(id, *val);
1724                 break;
1725         case KVM_REG_PPC_TM_CTR:
1726                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1727                 break;
1728         case KVM_REG_PPC_TM_FPSCR:
1729                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1730                 break;
1731         case KVM_REG_PPC_TM_AMR:
1732                 vcpu->arch.amr_tm = set_reg_val(id, *val);
1733                 break;
1734         case KVM_REG_PPC_TM_PPR:
1735                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1736                 break;
1737         case KVM_REG_PPC_TM_VRSAVE:
1738                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1739                 break;
1740         case KVM_REG_PPC_TM_VSCR:
1741                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1742                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1743                 else
1744                         r = - ENXIO;
1745                 break;
1746         case KVM_REG_PPC_TM_DSCR:
1747                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1748                 break;
1749         case KVM_REG_PPC_TM_TAR:
1750                 vcpu->arch.tar_tm = set_reg_val(id, *val);
1751                 break;
1752 #endif
1753         case KVM_REG_PPC_ARCH_COMPAT:
1754                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1755                 break;
1756         case KVM_REG_PPC_DEC_EXPIRY:
1757                 vcpu->arch.dec_expires = set_reg_val(id, *val) -
1758                         vcpu->arch.vcore->tb_offset;
1759                 break;
1760         default:
1761                 r = -EINVAL;
1762                 break;
1763         }
1764
1765         return r;
1766 }
1767
1768 /*
1769  * On POWER9, threads are independent and can be in different partitions.
1770  * Therefore we consider each thread to be a subcore.
1771  * There is a restriction that all threads have to be in the same
1772  * MMU mode (radix or HPT), unfortunately, but since we only support
1773  * HPT guests on a HPT host so far, that isn't an impediment yet.
1774  */
1775 static int threads_per_vcore(struct kvm *kvm)
1776 {
1777         if (kvm->arch.threads_indep)
1778                 return 1;
1779         return threads_per_subcore;
1780 }
1781
1782 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1783 {
1784         struct kvmppc_vcore *vcore;
1785
1786         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1787
1788         if (vcore == NULL)
1789                 return NULL;
1790
1791         spin_lock_init(&vcore->lock);
1792         spin_lock_init(&vcore->stoltb_lock);
1793         init_swait_queue_head(&vcore->wq);
1794         vcore->preempt_tb = TB_NIL;
1795         vcore->lpcr = kvm->arch.lpcr;
1796         vcore->first_vcpuid = core * kvm->arch.smt_mode;
1797         vcore->kvm = kvm;
1798         INIT_LIST_HEAD(&vcore->preempt_list);
1799
1800         return vcore;
1801 }
1802
1803 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1804 static struct debugfs_timings_element {
1805         const char *name;
1806         size_t offset;
1807 } timings[] = {
1808         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
1809         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
1810         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
1811         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
1812         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
1813 };
1814
1815 #define N_TIMINGS       (ARRAY_SIZE(timings))
1816
1817 struct debugfs_timings_state {
1818         struct kvm_vcpu *vcpu;
1819         unsigned int    buflen;
1820         char            buf[N_TIMINGS * 100];
1821 };
1822
1823 static int debugfs_timings_open(struct inode *inode, struct file *file)
1824 {
1825         struct kvm_vcpu *vcpu = inode->i_private;
1826         struct debugfs_timings_state *p;
1827
1828         p = kzalloc(sizeof(*p), GFP_KERNEL);
1829         if (!p)
1830                 return -ENOMEM;
1831
1832         kvm_get_kvm(vcpu->kvm);
1833         p->vcpu = vcpu;
1834         file->private_data = p;
1835
1836         return nonseekable_open(inode, file);
1837 }
1838
1839 static int debugfs_timings_release(struct inode *inode, struct file *file)
1840 {
1841         struct debugfs_timings_state *p = file->private_data;
1842
1843         kvm_put_kvm(p->vcpu->kvm);
1844         kfree(p);
1845         return 0;
1846 }
1847
1848 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1849                                     size_t len, loff_t *ppos)
1850 {
1851         struct debugfs_timings_state *p = file->private_data;
1852         struct kvm_vcpu *vcpu = p->vcpu;
1853         char *s, *buf_end;
1854         struct kvmhv_tb_accumulator tb;
1855         u64 count;
1856         loff_t pos;
1857         ssize_t n;
1858         int i, loops;
1859         bool ok;
1860
1861         if (!p->buflen) {
1862                 s = p->buf;
1863                 buf_end = s + sizeof(p->buf);
1864                 for (i = 0; i < N_TIMINGS; ++i) {
1865                         struct kvmhv_tb_accumulator *acc;
1866
1867                         acc = (struct kvmhv_tb_accumulator *)
1868                                 ((unsigned long)vcpu + timings[i].offset);
1869                         ok = false;
1870                         for (loops = 0; loops < 1000; ++loops) {
1871                                 count = acc->seqcount;
1872                                 if (!(count & 1)) {
1873                                         smp_rmb();
1874                                         tb = *acc;
1875                                         smp_rmb();
1876                                         if (count == acc->seqcount) {
1877                                                 ok = true;
1878                                                 break;
1879                                         }
1880                                 }
1881                                 udelay(1);
1882                         }
1883                         if (!ok)
1884                                 snprintf(s, buf_end - s, "%s: stuck\n",
1885                                         timings[i].name);
1886                         else
1887                                 snprintf(s, buf_end - s,
1888                                         "%s: %llu %llu %llu %llu\n",
1889                                         timings[i].name, count / 2,
1890                                         tb_to_ns(tb.tb_total),
1891                                         tb_to_ns(tb.tb_min),
1892                                         tb_to_ns(tb.tb_max));
1893                         s += strlen(s);
1894                 }
1895                 p->buflen = s - p->buf;
1896         }
1897
1898         pos = *ppos;
1899         if (pos >= p->buflen)
1900                 return 0;
1901         if (len > p->buflen - pos)
1902                 len = p->buflen - pos;
1903         n = copy_to_user(buf, p->buf + pos, len);
1904         if (n) {
1905                 if (n == len)
1906                         return -EFAULT;
1907                 len -= n;
1908         }
1909         *ppos = pos + len;
1910         return len;
1911 }
1912
1913 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1914                                      size_t len, loff_t *ppos)
1915 {
1916         return -EACCES;
1917 }
1918
1919 static const struct file_operations debugfs_timings_ops = {
1920         .owner   = THIS_MODULE,
1921         .open    = debugfs_timings_open,
1922         .release = debugfs_timings_release,
1923         .read    = debugfs_timings_read,
1924         .write   = debugfs_timings_write,
1925         .llseek  = generic_file_llseek,
1926 };
1927
1928 /* Create a debugfs directory for the vcpu */
1929 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1930 {
1931         char buf[16];
1932         struct kvm *kvm = vcpu->kvm;
1933
1934         snprintf(buf, sizeof(buf), "vcpu%u", id);
1935         if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1936                 return;
1937         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1938         if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1939                 return;
1940         vcpu->arch.debugfs_timings =
1941                 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1942                                     vcpu, &debugfs_timings_ops);
1943 }
1944
1945 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1946 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1947 {
1948 }
1949 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1950
1951 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1952                                                    unsigned int id)
1953 {
1954         struct kvm_vcpu *vcpu;
1955         int err;
1956         int core;
1957         struct kvmppc_vcore *vcore;
1958
1959         err = -ENOMEM;
1960         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1961         if (!vcpu)
1962                 goto out;
1963
1964         err = kvm_vcpu_init(vcpu, kvm, id);
1965         if (err)
1966                 goto free_vcpu;
1967
1968         vcpu->arch.shared = &vcpu->arch.shregs;
1969 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1970         /*
1971          * The shared struct is never shared on HV,
1972          * so we can always use host endianness
1973          */
1974 #ifdef __BIG_ENDIAN__
1975         vcpu->arch.shared_big_endian = true;
1976 #else
1977         vcpu->arch.shared_big_endian = false;
1978 #endif
1979 #endif
1980         vcpu->arch.mmcr[0] = MMCR0_FC;
1981         vcpu->arch.ctrl = CTRL_RUNLATCH;
1982         /* default to host PVR, since we can't spoof it */
1983         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1984         spin_lock_init(&vcpu->arch.vpa_update_lock);
1985         spin_lock_init(&vcpu->arch.tbacct_lock);
1986         vcpu->arch.busy_preempt = TB_NIL;
1987         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1988
1989         /*
1990          * Set the default HFSCR for the guest from the host value.
1991          * This value is only used on POWER9.
1992          * On POWER9 DD1, TM doesn't work, so we make sure to
1993          * prevent the guest from using it.
1994          * On POWER9, we want to virtualize the doorbell facility, so we
1995          * turn off the HFSCR bit, which causes those instructions to trap.
1996          */
1997         vcpu->arch.hfscr = mfspr(SPRN_HFSCR);
1998         if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
1999                 vcpu->arch.hfscr |= HFSCR_TM;
2000         else if (!cpu_has_feature(CPU_FTR_TM_COMP))
2001                 vcpu->arch.hfscr &= ~HFSCR_TM;
2002         if (cpu_has_feature(CPU_FTR_ARCH_300))
2003                 vcpu->arch.hfscr &= ~HFSCR_MSGP;
2004
2005         kvmppc_mmu_book3s_hv_init(vcpu);
2006
2007         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2008
2009         init_waitqueue_head(&vcpu->arch.cpu_run);
2010
2011         mutex_lock(&kvm->lock);
2012         vcore = NULL;
2013         err = -EINVAL;
2014         core = id / kvm->arch.smt_mode;
2015         if (core < KVM_MAX_VCORES) {
2016                 vcore = kvm->arch.vcores[core];
2017                 if (!vcore) {
2018                         err = -ENOMEM;
2019                         vcore = kvmppc_vcore_create(kvm, core);
2020                         kvm->arch.vcores[core] = vcore;
2021                         kvm->arch.online_vcores++;
2022                 }
2023         }
2024         mutex_unlock(&kvm->lock);
2025
2026         if (!vcore)
2027                 goto free_vcpu;
2028
2029         spin_lock(&vcore->lock);
2030         ++vcore->num_threads;
2031         spin_unlock(&vcore->lock);
2032         vcpu->arch.vcore = vcore;
2033         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2034         vcpu->arch.thread_cpu = -1;
2035         vcpu->arch.prev_cpu = -1;
2036
2037         vcpu->arch.cpu_type = KVM_CPU_3S_64;
2038         kvmppc_sanity_check(vcpu);
2039
2040         debugfs_vcpu_init(vcpu, id);
2041
2042         return vcpu;
2043
2044 free_vcpu:
2045         kmem_cache_free(kvm_vcpu_cache, vcpu);
2046 out:
2047         return ERR_PTR(err);
2048 }
2049
2050 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2051                               unsigned long flags)
2052 {
2053         int err;
2054         int esmt = 0;
2055
2056         if (flags)
2057                 return -EINVAL;
2058         if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2059                 return -EINVAL;
2060         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2061                 /*
2062                  * On POWER8 (or POWER7), the threading mode is "strict",
2063                  * so we pack smt_mode vcpus per vcore.
2064                  */
2065                 if (smt_mode > threads_per_subcore)
2066                         return -EINVAL;
2067         } else {
2068                 /*
2069                  * On POWER9, the threading mode is "loose",
2070                  * so each vcpu gets its own vcore.
2071                  */
2072                 esmt = smt_mode;
2073                 smt_mode = 1;
2074         }
2075         mutex_lock(&kvm->lock);
2076         err = -EBUSY;
2077         if (!kvm->arch.online_vcores) {
2078                 kvm->arch.smt_mode = smt_mode;
2079                 kvm->arch.emul_smt_mode = esmt;
2080                 err = 0;
2081         }
2082         mutex_unlock(&kvm->lock);
2083
2084         return err;
2085 }
2086
2087 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2088 {
2089         if (vpa->pinned_addr)
2090                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2091                                         vpa->dirty);
2092 }
2093
2094 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2095 {
2096         spin_lock(&vcpu->arch.vpa_update_lock);
2097         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2098         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2099         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2100         spin_unlock(&vcpu->arch.vpa_update_lock);
2101         kvm_vcpu_uninit(vcpu);
2102         kmem_cache_free(kvm_vcpu_cache, vcpu);
2103 }
2104
2105 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2106 {
2107         /* Indicate we want to get back into the guest */
2108         return 1;
2109 }
2110
2111 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2112 {
2113         unsigned long dec_nsec, now;
2114
2115         now = get_tb();
2116         if (now > vcpu->arch.dec_expires) {
2117                 /* decrementer has already gone negative */
2118                 kvmppc_core_queue_dec(vcpu);
2119                 kvmppc_core_prepare_to_enter(vcpu);
2120                 return;
2121         }
2122         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
2123                    / tb_ticks_per_sec;
2124         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2125         vcpu->arch.timer_running = 1;
2126 }
2127
2128 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2129 {
2130         vcpu->arch.ceded = 0;
2131         if (vcpu->arch.timer_running) {
2132                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2133                 vcpu->arch.timer_running = 0;
2134         }
2135 }
2136
2137 extern int __kvmppc_vcore_entry(void);
2138
2139 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2140                                    struct kvm_vcpu *vcpu)
2141 {
2142         u64 now;
2143
2144         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2145                 return;
2146         spin_lock_irq(&vcpu->arch.tbacct_lock);
2147         now = mftb();
2148         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2149                 vcpu->arch.stolen_logged;
2150         vcpu->arch.busy_preempt = now;
2151         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2152         spin_unlock_irq(&vcpu->arch.tbacct_lock);
2153         --vc->n_runnable;
2154         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2155 }
2156
2157 static int kvmppc_grab_hwthread(int cpu)
2158 {
2159         struct paca_struct *tpaca;
2160         long timeout = 10000;
2161
2162         tpaca = paca_ptrs[cpu];
2163
2164         /* Ensure the thread won't go into the kernel if it wakes */
2165         tpaca->kvm_hstate.kvm_vcpu = NULL;
2166         tpaca->kvm_hstate.kvm_vcore = NULL;
2167         tpaca->kvm_hstate.napping = 0;
2168         smp_wmb();
2169         tpaca->kvm_hstate.hwthread_req = 1;
2170
2171         /*
2172          * If the thread is already executing in the kernel (e.g. handling
2173          * a stray interrupt), wait for it to get back to nap mode.
2174          * The smp_mb() is to ensure that our setting of hwthread_req
2175          * is visible before we look at hwthread_state, so if this
2176          * races with the code at system_reset_pSeries and the thread
2177          * misses our setting of hwthread_req, we are sure to see its
2178          * setting of hwthread_state, and vice versa.
2179          */
2180         smp_mb();
2181         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2182                 if (--timeout <= 0) {
2183                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
2184                         return -EBUSY;
2185                 }
2186                 udelay(1);
2187         }
2188         return 0;
2189 }
2190
2191 static void kvmppc_release_hwthread(int cpu)
2192 {
2193         struct paca_struct *tpaca;
2194
2195         tpaca = paca_ptrs[cpu];
2196         tpaca->kvm_hstate.hwthread_req = 0;
2197         tpaca->kvm_hstate.kvm_vcpu = NULL;
2198         tpaca->kvm_hstate.kvm_vcore = NULL;
2199         tpaca->kvm_hstate.kvm_split_mode = NULL;
2200 }
2201
2202 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2203 {
2204         int i;
2205
2206         cpu = cpu_first_thread_sibling(cpu);
2207         cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2208         /*
2209          * Make sure setting of bit in need_tlb_flush precedes
2210          * testing of cpu_in_guest bits.  The matching barrier on
2211          * the other side is the first smp_mb() in kvmppc_run_core().
2212          */
2213         smp_mb();
2214         for (i = 0; i < threads_per_core; ++i)
2215                 if (cpumask_test_cpu(cpu + i, &kvm->arch.cpu_in_guest))
2216                         smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2217 }
2218
2219 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2220 {
2221         struct kvm *kvm = vcpu->kvm;
2222
2223         /*
2224          * With radix, the guest can do TLB invalidations itself,
2225          * and it could choose to use the local form (tlbiel) if
2226          * it is invalidating a translation that has only ever been
2227          * used on one vcpu.  However, that doesn't mean it has
2228          * only ever been used on one physical cpu, since vcpus
2229          * can move around between pcpus.  To cope with this, when
2230          * a vcpu moves from one pcpu to another, we need to tell
2231          * any vcpus running on the same core as this vcpu previously
2232          * ran to flush the TLB.  The TLB is shared between threads,
2233          * so we use a single bit in .need_tlb_flush for all 4 threads.
2234          */
2235         if (vcpu->arch.prev_cpu != pcpu) {
2236                 if (vcpu->arch.prev_cpu >= 0 &&
2237                     cpu_first_thread_sibling(vcpu->arch.prev_cpu) !=
2238                     cpu_first_thread_sibling(pcpu))
2239                         radix_flush_cpu(kvm, vcpu->arch.prev_cpu, vcpu);
2240                 vcpu->arch.prev_cpu = pcpu;
2241         }
2242 }
2243
2244 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2245 {
2246         int cpu;
2247         struct paca_struct *tpaca;
2248         struct kvm *kvm = vc->kvm;
2249
2250         cpu = vc->pcpu;
2251         if (vcpu) {
2252                 if (vcpu->arch.timer_running) {
2253                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2254                         vcpu->arch.timer_running = 0;
2255                 }
2256                 cpu += vcpu->arch.ptid;
2257                 vcpu->cpu = vc->pcpu;
2258                 vcpu->arch.thread_cpu = cpu;
2259                 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2260         }
2261         tpaca = paca_ptrs[cpu];
2262         tpaca->kvm_hstate.kvm_vcpu = vcpu;
2263         tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2264         tpaca->kvm_hstate.fake_suspend = 0;
2265         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2266         smp_wmb();
2267         tpaca->kvm_hstate.kvm_vcore = vc;
2268         if (cpu != smp_processor_id())
2269                 kvmppc_ipi_thread(cpu);
2270 }
2271
2272 static void kvmppc_wait_for_nap(int n_threads)
2273 {
2274         int cpu = smp_processor_id();
2275         int i, loops;
2276
2277         if (n_threads <= 1)
2278                 return;
2279         for (loops = 0; loops < 1000000; ++loops) {
2280                 /*
2281                  * Check if all threads are finished.
2282                  * We set the vcore pointer when starting a thread
2283                  * and the thread clears it when finished, so we look
2284                  * for any threads that still have a non-NULL vcore ptr.
2285                  */
2286                 for (i = 1; i < n_threads; ++i)
2287                         if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2288                                 break;
2289                 if (i == n_threads) {
2290                         HMT_medium();
2291                         return;
2292                 }
2293                 HMT_low();
2294         }
2295         HMT_medium();
2296         for (i = 1; i < n_threads; ++i)
2297                 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2298                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2299 }
2300
2301 /*
2302  * Check that we are on thread 0 and that any other threads in
2303  * this core are off-line.  Then grab the threads so they can't
2304  * enter the kernel.
2305  */
2306 static int on_primary_thread(void)
2307 {
2308         int cpu = smp_processor_id();
2309         int thr;
2310
2311         /* Are we on a primary subcore? */
2312         if (cpu_thread_in_subcore(cpu))
2313                 return 0;
2314
2315         thr = 0;
2316         while (++thr < threads_per_subcore)
2317                 if (cpu_online(cpu + thr))
2318                         return 0;
2319
2320         /* Grab all hw threads so they can't go into the kernel */
2321         for (thr = 1; thr < threads_per_subcore; ++thr) {
2322                 if (kvmppc_grab_hwthread(cpu + thr)) {
2323                         /* Couldn't grab one; let the others go */
2324                         do {
2325                                 kvmppc_release_hwthread(cpu + thr);
2326                         } while (--thr > 0);
2327                         return 0;
2328                 }
2329         }
2330         return 1;
2331 }
2332
2333 /*
2334  * A list of virtual cores for each physical CPU.
2335  * These are vcores that could run but their runner VCPU tasks are
2336  * (or may be) preempted.
2337  */
2338 struct preempted_vcore_list {
2339         struct list_head        list;
2340         spinlock_t              lock;
2341 };
2342
2343 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2344
2345 static void init_vcore_lists(void)
2346 {
2347         int cpu;
2348
2349         for_each_possible_cpu(cpu) {
2350                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2351                 spin_lock_init(&lp->lock);
2352                 INIT_LIST_HEAD(&lp->list);
2353         }
2354 }
2355
2356 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2357 {
2358         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2359
2360         vc->vcore_state = VCORE_PREEMPT;
2361         vc->pcpu = smp_processor_id();
2362         if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2363                 spin_lock(&lp->lock);
2364                 list_add_tail(&vc->preempt_list, &lp->list);
2365                 spin_unlock(&lp->lock);
2366         }
2367
2368         /* Start accumulating stolen time */
2369         kvmppc_core_start_stolen(vc);
2370 }
2371
2372 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2373 {
2374         struct preempted_vcore_list *lp;
2375
2376         kvmppc_core_end_stolen(vc);
2377         if (!list_empty(&vc->preempt_list)) {
2378                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2379                 spin_lock(&lp->lock);
2380                 list_del_init(&vc->preempt_list);
2381                 spin_unlock(&lp->lock);
2382         }
2383         vc->vcore_state = VCORE_INACTIVE;
2384 }
2385
2386 /*
2387  * This stores information about the virtual cores currently
2388  * assigned to a physical core.
2389  */
2390 struct core_info {
2391         int             n_subcores;
2392         int             max_subcore_threads;
2393         int             total_threads;
2394         int             subcore_threads[MAX_SUBCORES];
2395         struct kvmppc_vcore *vc[MAX_SUBCORES];
2396 };
2397
2398 /*
2399  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2400  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2401  */
2402 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2403
2404 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2405 {
2406         memset(cip, 0, sizeof(*cip));
2407         cip->n_subcores = 1;
2408         cip->max_subcore_threads = vc->num_threads;
2409         cip->total_threads = vc->num_threads;
2410         cip->subcore_threads[0] = vc->num_threads;
2411         cip->vc[0] = vc;
2412 }
2413
2414 static bool subcore_config_ok(int n_subcores, int n_threads)
2415 {
2416         /*
2417          * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2418          * split-core mode, with one thread per subcore.
2419          */
2420         if (cpu_has_feature(CPU_FTR_ARCH_300))
2421                 return n_subcores <= 4 && n_threads == 1;
2422
2423         /* On POWER8, can only dynamically split if unsplit to begin with */
2424         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2425                 return false;
2426         if (n_subcores > MAX_SUBCORES)
2427                 return false;
2428         if (n_subcores > 1) {
2429                 if (!(dynamic_mt_modes & 2))
2430                         n_subcores = 4;
2431                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2432                         return false;
2433         }
2434
2435         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2436 }
2437
2438 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2439 {
2440         vc->entry_exit_map = 0;
2441         vc->in_guest = 0;
2442         vc->napping_threads = 0;
2443         vc->conferring_threads = 0;
2444 }
2445
2446 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2447 {
2448         int n_threads = vc->num_threads;
2449         int sub;
2450
2451         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2452                 return false;
2453
2454         /* Some POWER9 chips require all threads to be in the same MMU mode */
2455         if (no_mixing_hpt_and_radix &&
2456             kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2457                 return false;
2458
2459         if (n_threads < cip->max_subcore_threads)
2460                 n_threads = cip->max_subcore_threads;
2461         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2462                 return false;
2463         cip->max_subcore_threads = n_threads;
2464
2465         sub = cip->n_subcores;
2466         ++cip->n_subcores;
2467         cip->total_threads += vc->num_threads;
2468         cip->subcore_threads[sub] = vc->num_threads;
2469         cip->vc[sub] = vc;
2470         init_vcore_to_run(vc);
2471         list_del_init(&vc->preempt_list);
2472
2473         return true;
2474 }
2475
2476 /*
2477  * Work out whether it is possible to piggyback the execution of
2478  * vcore *pvc onto the execution of the other vcores described in *cip.
2479  */
2480 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2481                           int target_threads)
2482 {
2483         if (cip->total_threads + pvc->num_threads > target_threads)
2484                 return false;
2485
2486         return can_dynamic_split(pvc, cip);
2487 }
2488
2489 static void prepare_threads(struct kvmppc_vcore *vc)
2490 {
2491         int i;
2492         struct kvm_vcpu *vcpu;
2493
2494         for_each_runnable_thread(i, vcpu, vc) {
2495                 if (signal_pending(vcpu->arch.run_task))
2496                         vcpu->arch.ret = -EINTR;
2497                 else if (vcpu->arch.vpa.update_pending ||
2498                          vcpu->arch.slb_shadow.update_pending ||
2499                          vcpu->arch.dtl.update_pending)
2500                         vcpu->arch.ret = RESUME_GUEST;
2501                 else
2502                         continue;
2503                 kvmppc_remove_runnable(vc, vcpu);
2504                 wake_up(&vcpu->arch.cpu_run);
2505         }
2506 }
2507
2508 static void collect_piggybacks(struct core_info *cip, int target_threads)
2509 {
2510         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2511         struct kvmppc_vcore *pvc, *vcnext;
2512
2513         spin_lock(&lp->lock);
2514         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2515                 if (!spin_trylock(&pvc->lock))
2516                         continue;
2517                 prepare_threads(pvc);
2518                 if (!pvc->n_runnable) {
2519                         list_del_init(&pvc->preempt_list);
2520                         if (pvc->runner == NULL) {
2521                                 pvc->vcore_state = VCORE_INACTIVE;
2522                                 kvmppc_core_end_stolen(pvc);
2523                         }
2524                         spin_unlock(&pvc->lock);
2525                         continue;
2526                 }
2527                 if (!can_piggyback(pvc, cip, target_threads)) {
2528                         spin_unlock(&pvc->lock);
2529                         continue;
2530                 }
2531                 kvmppc_core_end_stolen(pvc);
2532                 pvc->vcore_state = VCORE_PIGGYBACK;
2533                 if (cip->total_threads >= target_threads)
2534                         break;
2535         }
2536         spin_unlock(&lp->lock);
2537 }
2538
2539 static bool recheck_signals(struct core_info *cip)
2540 {
2541         int sub, i;
2542         struct kvm_vcpu *vcpu;
2543
2544         for (sub = 0; sub < cip->n_subcores; ++sub)
2545                 for_each_runnable_thread(i, vcpu, cip->vc[sub])
2546                         if (signal_pending(vcpu->arch.run_task))
2547                                 return true;
2548         return false;
2549 }
2550
2551 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2552 {
2553         int still_running = 0, i;
2554         u64 now;
2555         long ret;
2556         struct kvm_vcpu *vcpu;
2557
2558         spin_lock(&vc->lock);
2559         now = get_tb();
2560         for_each_runnable_thread(i, vcpu, vc) {
2561                 /* cancel pending dec exception if dec is positive */
2562                 if (now < vcpu->arch.dec_expires &&
2563                     kvmppc_core_pending_dec(vcpu))
2564                         kvmppc_core_dequeue_dec(vcpu);
2565
2566                 trace_kvm_guest_exit(vcpu);
2567
2568                 ret = RESUME_GUEST;
2569                 if (vcpu->arch.trap)
2570                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2571                                                     vcpu->arch.run_task);
2572
2573                 vcpu->arch.ret = ret;
2574                 vcpu->arch.trap = 0;
2575
2576                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2577                         if (vcpu->arch.pending_exceptions)
2578                                 kvmppc_core_prepare_to_enter(vcpu);
2579                         if (vcpu->arch.ceded)
2580                                 kvmppc_set_timer(vcpu);
2581                         else
2582                                 ++still_running;
2583                 } else {
2584                         kvmppc_remove_runnable(vc, vcpu);
2585                         wake_up(&vcpu->arch.cpu_run);
2586                 }
2587         }
2588         if (!is_master) {
2589                 if (still_running > 0) {
2590                         kvmppc_vcore_preempt(vc);
2591                 } else if (vc->runner) {
2592                         vc->vcore_state = VCORE_PREEMPT;
2593                         kvmppc_core_start_stolen(vc);
2594                 } else {
2595                         vc->vcore_state = VCORE_INACTIVE;
2596                 }
2597                 if (vc->n_runnable > 0 && vc->runner == NULL) {
2598                         /* make sure there's a candidate runner awake */
2599                         i = -1;
2600                         vcpu = next_runnable_thread(vc, &i);
2601                         wake_up(&vcpu->arch.cpu_run);
2602                 }
2603         }
2604         spin_unlock(&vc->lock);
2605 }
2606
2607 /*
2608  * Clear core from the list of active host cores as we are about to
2609  * enter the guest. Only do this if it is the primary thread of the
2610  * core (not if a subcore) that is entering the guest.
2611  */
2612 static inline int kvmppc_clear_host_core(unsigned int cpu)
2613 {
2614         int core;
2615
2616         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2617                 return 0;
2618         /*
2619          * Memory barrier can be omitted here as we will do a smp_wmb()
2620          * later in kvmppc_start_thread and we need ensure that state is
2621          * visible to other CPUs only after we enter guest.
2622          */
2623         core = cpu >> threads_shift;
2624         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2625         return 0;
2626 }
2627
2628 /*
2629  * Advertise this core as an active host core since we exited the guest
2630  * Only need to do this if it is the primary thread of the core that is
2631  * exiting.
2632  */
2633 static inline int kvmppc_set_host_core(unsigned int cpu)
2634 {
2635         int core;
2636
2637         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2638                 return 0;
2639
2640         /*
2641          * Memory barrier can be omitted here because we do a spin_unlock
2642          * immediately after this which provides the memory barrier.
2643          */
2644         core = cpu >> threads_shift;
2645         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2646         return 0;
2647 }
2648
2649 static void set_irq_happened(int trap)
2650 {
2651         switch (trap) {
2652         case BOOK3S_INTERRUPT_EXTERNAL:
2653                 local_paca->irq_happened |= PACA_IRQ_EE;
2654                 break;
2655         case BOOK3S_INTERRUPT_H_DOORBELL:
2656                 local_paca->irq_happened |= PACA_IRQ_DBELL;
2657                 break;
2658         case BOOK3S_INTERRUPT_HMI:
2659                 local_paca->irq_happened |= PACA_IRQ_HMI;
2660                 break;
2661         case BOOK3S_INTERRUPT_SYSTEM_RESET:
2662                 replay_system_reset();
2663                 break;
2664         }
2665 }
2666
2667 /*
2668  * Run a set of guest threads on a physical core.
2669  * Called with vc->lock held.
2670  */
2671 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2672 {
2673         struct kvm_vcpu *vcpu;
2674         int i;
2675         int srcu_idx;
2676         struct core_info core_info;
2677         struct kvmppc_vcore *pvc;
2678         struct kvm_split_mode split_info, *sip;
2679         int split, subcore_size, active;
2680         int sub;
2681         bool thr0_done;
2682         unsigned long cmd_bit, stat_bit;
2683         int pcpu, thr;
2684         int target_threads;
2685         int controlled_threads;
2686         int trap;
2687         bool is_power8;
2688         bool hpt_on_radix;
2689
2690         /*
2691          * Remove from the list any threads that have a signal pending
2692          * or need a VPA update done
2693          */
2694         prepare_threads(vc);
2695
2696         /* if the runner is no longer runnable, let the caller pick a new one */
2697         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2698                 return;
2699
2700         /*
2701          * Initialize *vc.
2702          */
2703         init_vcore_to_run(vc);
2704         vc->preempt_tb = TB_NIL;
2705
2706         /*
2707          * Number of threads that we will be controlling: the same as
2708          * the number of threads per subcore, except on POWER9,
2709          * where it's 1 because the threads are (mostly) independent.
2710          */
2711         controlled_threads = threads_per_vcore(vc->kvm);
2712
2713         /*
2714          * Make sure we are running on primary threads, and that secondary
2715          * threads are offline.  Also check if the number of threads in this
2716          * guest are greater than the current system threads per guest.
2717          * On POWER9, we need to be not in independent-threads mode if
2718          * this is a HPT guest on a radix host machine where the
2719          * CPU threads may not be in different MMU modes.
2720          */
2721         hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
2722                 !kvm_is_radix(vc->kvm);
2723         if (((controlled_threads > 1) &&
2724              ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
2725             (hpt_on_radix && vc->kvm->arch.threads_indep)) {
2726                 for_each_runnable_thread(i, vcpu, vc) {
2727                         vcpu->arch.ret = -EBUSY;
2728                         kvmppc_remove_runnable(vc, vcpu);
2729                         wake_up(&vcpu->arch.cpu_run);
2730                 }
2731                 goto out;
2732         }
2733
2734         /*
2735          * See if we could run any other vcores on the physical core
2736          * along with this one.
2737          */
2738         init_core_info(&core_info, vc);
2739         pcpu = smp_processor_id();
2740         target_threads = controlled_threads;
2741         if (target_smt_mode && target_smt_mode < target_threads)
2742                 target_threads = target_smt_mode;
2743         if (vc->num_threads < target_threads)
2744                 collect_piggybacks(&core_info, target_threads);
2745
2746         /*
2747          * On radix, arrange for TLB flushing if necessary.
2748          * This has to be done before disabling interrupts since
2749          * it uses smp_call_function().
2750          */
2751         pcpu = smp_processor_id();
2752         if (kvm_is_radix(vc->kvm)) {
2753                 for (sub = 0; sub < core_info.n_subcores; ++sub)
2754                         for_each_runnable_thread(i, vcpu, core_info.vc[sub])
2755                                 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
2756         }
2757
2758         /*
2759          * Hard-disable interrupts, and check resched flag and signals.
2760          * If we need to reschedule or deliver a signal, clean up
2761          * and return without going into the guest(s).
2762          * If the mmu_ready flag has been cleared, don't go into the
2763          * guest because that means a HPT resize operation is in progress.
2764          */
2765         local_irq_disable();
2766         hard_irq_disable();
2767         if (lazy_irq_pending() || need_resched() ||
2768             recheck_signals(&core_info) || !vc->kvm->arch.mmu_ready) {
2769                 local_irq_enable();
2770                 vc->vcore_state = VCORE_INACTIVE;
2771                 /* Unlock all except the primary vcore */
2772                 for (sub = 1; sub < core_info.n_subcores; ++sub) {
2773                         pvc = core_info.vc[sub];
2774                         /* Put back on to the preempted vcores list */
2775                         kvmppc_vcore_preempt(pvc);
2776                         spin_unlock(&pvc->lock);
2777                 }
2778                 for (i = 0; i < controlled_threads; ++i)
2779                         kvmppc_release_hwthread(pcpu + i);
2780                 return;
2781         }
2782
2783         kvmppc_clear_host_core(pcpu);
2784
2785         /* Decide on micro-threading (split-core) mode */
2786         subcore_size = threads_per_subcore;
2787         cmd_bit = stat_bit = 0;
2788         split = core_info.n_subcores;
2789         sip = NULL;
2790         is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
2791                 && !cpu_has_feature(CPU_FTR_ARCH_300);
2792
2793         if (split > 1 || hpt_on_radix) {
2794                 sip = &split_info;
2795                 memset(&split_info, 0, sizeof(split_info));
2796                 for (sub = 0; sub < core_info.n_subcores; ++sub)
2797                         split_info.vc[sub] = core_info.vc[sub];
2798
2799                 if (is_power8) {
2800                         if (split == 2 && (dynamic_mt_modes & 2)) {
2801                                 cmd_bit = HID0_POWER8_1TO2LPAR;
2802                                 stat_bit = HID0_POWER8_2LPARMODE;
2803                         } else {
2804                                 split = 4;
2805                                 cmd_bit = HID0_POWER8_1TO4LPAR;
2806                                 stat_bit = HID0_POWER8_4LPARMODE;
2807                         }
2808                         subcore_size = MAX_SMT_THREADS / split;
2809                         split_info.rpr = mfspr(SPRN_RPR);
2810                         split_info.pmmar = mfspr(SPRN_PMMAR);
2811                         split_info.ldbar = mfspr(SPRN_LDBAR);
2812                         split_info.subcore_size = subcore_size;
2813                 } else {
2814                         split_info.subcore_size = 1;
2815                         if (hpt_on_radix) {
2816                                 /* Use the split_info for LPCR/LPIDR changes */
2817                                 split_info.lpcr_req = vc->lpcr;
2818                                 split_info.lpidr_req = vc->kvm->arch.lpid;
2819                                 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
2820                                 split_info.do_set = 1;
2821                         }
2822                 }
2823
2824                 /* order writes to split_info before kvm_split_mode pointer */
2825                 smp_wmb();
2826         }
2827
2828         for (thr = 0; thr < controlled_threads; ++thr) {
2829                 struct paca_struct *paca = paca_ptrs[pcpu + thr];
2830
2831                 paca->kvm_hstate.tid = thr;
2832                 paca->kvm_hstate.napping = 0;
2833                 paca->kvm_hstate.kvm_split_mode = sip;
2834         }
2835
2836         /* Initiate micro-threading (split-core) on POWER8 if required */
2837         if (cmd_bit) {
2838                 unsigned long hid0 = mfspr(SPRN_HID0);
2839
2840                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2841                 mb();
2842                 mtspr(SPRN_HID0, hid0);
2843                 isync();
2844                 for (;;) {
2845                         hid0 = mfspr(SPRN_HID0);
2846                         if (hid0 & stat_bit)
2847                                 break;
2848                         cpu_relax();
2849                 }
2850         }
2851
2852         /* Start all the threads */
2853         active = 0;
2854         for (sub = 0; sub < core_info.n_subcores; ++sub) {
2855                 thr = is_power8 ? subcore_thread_map[sub] : sub;
2856                 thr0_done = false;
2857                 active |= 1 << thr;
2858                 pvc = core_info.vc[sub];
2859                 pvc->pcpu = pcpu + thr;
2860                 for_each_runnable_thread(i, vcpu, pvc) {
2861                         kvmppc_start_thread(vcpu, pvc);
2862                         kvmppc_create_dtl_entry(vcpu, pvc);
2863                         trace_kvm_guest_enter(vcpu);
2864                         if (!vcpu->arch.ptid)
2865                                 thr0_done = true;
2866                         active |= 1 << (thr + vcpu->arch.ptid);
2867                 }
2868                 /*
2869                  * We need to start the first thread of each subcore
2870                  * even if it doesn't have a vcpu.
2871                  */
2872                 if (!thr0_done)
2873                         kvmppc_start_thread(NULL, pvc);
2874         }
2875
2876         /*
2877          * Ensure that split_info.do_nap is set after setting
2878          * the vcore pointer in the PACA of the secondaries.
2879          */
2880         smp_mb();
2881
2882         /*
2883          * When doing micro-threading, poke the inactive threads as well.
2884          * This gets them to the nap instruction after kvm_do_nap,
2885          * which reduces the time taken to unsplit later.
2886          * For POWER9 HPT guest on radix host, we need all the secondary
2887          * threads woken up so they can do the LPCR/LPIDR change.
2888          */
2889         if (cmd_bit || hpt_on_radix) {
2890                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
2891                 for (thr = 1; thr < threads_per_subcore; ++thr)
2892                         if (!(active & (1 << thr)))
2893                                 kvmppc_ipi_thread(pcpu + thr);
2894         }
2895
2896         vc->vcore_state = VCORE_RUNNING;
2897         preempt_disable();
2898
2899         trace_kvmppc_run_core(vc, 0);
2900
2901         for (sub = 0; sub < core_info.n_subcores; ++sub)
2902                 spin_unlock(&core_info.vc[sub]->lock);
2903
2904         /*
2905          * Interrupts will be enabled once we get into the guest,
2906          * so tell lockdep that we're about to enable interrupts.
2907          */
2908         trace_hardirqs_on();
2909
2910         guest_enter_irqoff();
2911
2912         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2913
2914         trap = __kvmppc_vcore_entry();
2915
2916         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2917
2918         trace_hardirqs_off();
2919         set_irq_happened(trap);
2920
2921         spin_lock(&vc->lock);
2922         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2923         vc->vcore_state = VCORE_EXITING;
2924
2925         /* wait for secondary threads to finish writing their state to memory */
2926         kvmppc_wait_for_nap(controlled_threads);
2927
2928         /* Return to whole-core mode if we split the core earlier */
2929         if (cmd_bit) {
2930                 unsigned long hid0 = mfspr(SPRN_HID0);
2931                 unsigned long loops = 0;
2932
2933                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2934                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2935                 mb();
2936                 mtspr(SPRN_HID0, hid0);
2937                 isync();
2938                 for (;;) {
2939                         hid0 = mfspr(SPRN_HID0);
2940                         if (!(hid0 & stat_bit))
2941                                 break;
2942                         cpu_relax();
2943                         ++loops;
2944                 }
2945         } else if (hpt_on_radix) {
2946                 /* Wait for all threads to have seen final sync */
2947                 for (thr = 1; thr < controlled_threads; ++thr) {
2948                         struct paca_struct *paca = paca_ptrs[pcpu + thr];
2949
2950                         while (paca->kvm_hstate.kvm_split_mode) {
2951                                 HMT_low();
2952                                 barrier();
2953                         }
2954                         HMT_medium();
2955                 }
2956         }
2957         split_info.do_nap = 0;
2958
2959         kvmppc_set_host_core(pcpu);
2960
2961         local_irq_enable();
2962         guest_exit();
2963
2964         /* Let secondaries go back to the offline loop */
2965         for (i = 0; i < controlled_threads; ++i) {
2966                 kvmppc_release_hwthread(pcpu + i);
2967                 if (sip && sip->napped[i])
2968                         kvmppc_ipi_thread(pcpu + i);
2969                 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
2970         }
2971
2972         spin_unlock(&vc->lock);
2973
2974         /* make sure updates to secondary vcpu structs are visible now */
2975         smp_mb();
2976
2977         preempt_enable();
2978
2979         for (sub = 0; sub < core_info.n_subcores; ++sub) {
2980                 pvc = core_info.vc[sub];
2981                 post_guest_process(pvc, pvc == vc);
2982         }
2983
2984         spin_lock(&vc->lock);
2985
2986  out:
2987         vc->vcore_state = VCORE_INACTIVE;
2988         trace_kvmppc_run_core(vc, 1);
2989 }
2990
2991 /*
2992  * Wait for some other vcpu thread to execute us, and
2993  * wake us up when we need to handle something in the host.
2994  */
2995 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2996                                  struct kvm_vcpu *vcpu, int wait_state)
2997 {
2998         DEFINE_WAIT(wait);
2999
3000         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3001         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3002                 spin_unlock(&vc->lock);
3003                 schedule();
3004                 spin_lock(&vc->lock);
3005         }
3006         finish_wait(&vcpu->arch.cpu_run, &wait);
3007 }
3008
3009 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3010 {
3011         /* 10us base */
3012         if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
3013                 vc->halt_poll_ns = 10000;
3014         else
3015                 vc->halt_poll_ns *= halt_poll_ns_grow;
3016 }
3017
3018 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3019 {
3020         if (halt_poll_ns_shrink == 0)
3021                 vc->halt_poll_ns = 0;
3022         else
3023                 vc->halt_poll_ns /= halt_poll_ns_shrink;
3024 }
3025
3026 #ifdef CONFIG_KVM_XICS
3027 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3028 {
3029         if (!xive_enabled())
3030                 return false;
3031         return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3032                 vcpu->arch.xive_saved_state.cppr;
3033 }
3034 #else
3035 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3036 {
3037         return false;
3038 }
3039 #endif /* CONFIG_KVM_XICS */
3040
3041 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3042 {
3043         if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3044             kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3045                 return true;
3046
3047         return false;
3048 }
3049
3050 /*
3051  * Check to see if any of the runnable vcpus on the vcore have pending
3052  * exceptions or are no longer ceded
3053  */
3054 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3055 {
3056         struct kvm_vcpu *vcpu;
3057         int i;
3058
3059         for_each_runnable_thread(i, vcpu, vc) {
3060                 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3061                         return 1;
3062         }
3063
3064         return 0;
3065 }
3066
3067 /*
3068  * All the vcpus in this vcore are idle, so wait for a decrementer
3069  * or external interrupt to one of the vcpus.  vc->lock is held.
3070  */
3071 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3072 {
3073         ktime_t cur, start_poll, start_wait;
3074         int do_sleep = 1;
3075         u64 block_ns;
3076         DECLARE_SWAITQUEUE(wait);
3077
3078         /* Poll for pending exceptions and ceded state */
3079         cur = start_poll = ktime_get();
3080         if (vc->halt_poll_ns) {
3081                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3082                 ++vc->runner->stat.halt_attempted_poll;
3083
3084                 vc->vcore_state = VCORE_POLLING;
3085                 spin_unlock(&vc->lock);
3086
3087                 do {
3088                         if (kvmppc_vcore_check_block(vc)) {
3089                                 do_sleep = 0;
3090                                 break;
3091                         }
3092                         cur = ktime_get();
3093                 } while (single_task_running() && ktime_before(cur, stop));
3094
3095                 spin_lock(&vc->lock);
3096                 vc->vcore_state = VCORE_INACTIVE;
3097
3098                 if (!do_sleep) {
3099                         ++vc->runner->stat.halt_successful_poll;
3100                         goto out;
3101                 }
3102         }
3103
3104         prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3105
3106         if (kvmppc_vcore_check_block(vc)) {
3107                 finish_swait(&vc->wq, &wait);
3108                 do_sleep = 0;
3109                 /* If we polled, count this as a successful poll */
3110                 if (vc->halt_poll_ns)
3111                         ++vc->runner->stat.halt_successful_poll;
3112                 goto out;
3113         }
3114
3115         start_wait = ktime_get();
3116
3117         vc->vcore_state = VCORE_SLEEPING;
3118         trace_kvmppc_vcore_blocked(vc, 0);
3119         spin_unlock(&vc->lock);
3120         schedule();
3121         finish_swait(&vc->wq, &wait);
3122         spin_lock(&vc->lock);
3123         vc->vcore_state = VCORE_INACTIVE;
3124         trace_kvmppc_vcore_blocked(vc, 1);
3125         ++vc->runner->stat.halt_successful_wait;
3126
3127         cur = ktime_get();
3128
3129 out:
3130         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3131
3132         /* Attribute wait time */
3133         if (do_sleep) {
3134                 vc->runner->stat.halt_wait_ns +=
3135                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
3136                 /* Attribute failed poll time */
3137                 if (vc->halt_poll_ns)
3138                         vc->runner->stat.halt_poll_fail_ns +=
3139                                 ktime_to_ns(start_wait) -
3140                                 ktime_to_ns(start_poll);
3141         } else {
3142                 /* Attribute successful poll time */
3143                 if (vc->halt_poll_ns)
3144                         vc->runner->stat.halt_poll_success_ns +=
3145                                 ktime_to_ns(cur) -
3146                                 ktime_to_ns(start_poll);
3147         }
3148
3149         /* Adjust poll time */
3150         if (halt_poll_ns) {
3151                 if (block_ns <= vc->halt_poll_ns)
3152                         ;
3153                 /* We slept and blocked for longer than the max halt time */
3154                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3155                         shrink_halt_poll_ns(vc);
3156                 /* We slept and our poll time is too small */
3157                 else if (vc->halt_poll_ns < halt_poll_ns &&
3158                                 block_ns < halt_poll_ns)
3159                         grow_halt_poll_ns(vc);
3160                 if (vc->halt_poll_ns > halt_poll_ns)
3161                         vc->halt_poll_ns = halt_poll_ns;
3162         } else
3163                 vc->halt_poll_ns = 0;
3164
3165         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3166 }
3167
3168 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3169 {
3170         int r = 0;
3171         struct kvm *kvm = vcpu->kvm;
3172
3173         mutex_lock(&kvm->lock);
3174         if (!kvm->arch.mmu_ready) {
3175                 if (!kvm_is_radix(kvm))
3176                         r = kvmppc_hv_setup_htab_rma(vcpu);
3177                 if (!r) {
3178                         if (cpu_has_feature(CPU_FTR_ARCH_300))
3179                                 kvmppc_setup_partition_table(kvm);
3180                         kvm->arch.mmu_ready = 1;
3181                 }
3182         }
3183         mutex_unlock(&kvm->lock);
3184         return r;
3185 }
3186
3187 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3188 {
3189         int n_ceded, i, r;
3190         struct kvmppc_vcore *vc;
3191         struct kvm_vcpu *v;
3192
3193         trace_kvmppc_run_vcpu_enter(vcpu);
3194
3195         kvm_run->exit_reason = 0;
3196         vcpu->arch.ret = RESUME_GUEST;
3197         vcpu->arch.trap = 0;
3198         kvmppc_update_vpas(vcpu);
3199
3200         /*
3201          * Synchronize with other threads in this virtual core
3202          */
3203         vc = vcpu->arch.vcore;
3204         spin_lock(&vc->lock);
3205         vcpu->arch.ceded = 0;
3206         vcpu->arch.run_task = current;
3207         vcpu->arch.kvm_run = kvm_run;
3208         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3209         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3210         vcpu->arch.busy_preempt = TB_NIL;
3211         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3212         ++vc->n_runnable;
3213
3214         /*
3215          * This happens the first time this is called for a vcpu.
3216          * If the vcore is already running, we may be able to start
3217          * this thread straight away and have it join in.
3218          */
3219         if (!signal_pending(current)) {
3220                 if ((vc->vcore_state == VCORE_PIGGYBACK ||
3221                      vc->vcore_state == VCORE_RUNNING) &&
3222                            !VCORE_IS_EXITING(vc)) {
3223                         kvmppc_create_dtl_entry(vcpu, vc);
3224                         kvmppc_start_thread(vcpu, vc);
3225                         trace_kvm_guest_enter(vcpu);
3226                 } else if (vc->vcore_state == VCORE_SLEEPING) {
3227                         swake_up(&vc->wq);
3228                 }
3229
3230         }
3231
3232         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3233                !signal_pending(current)) {
3234                 /* See if the MMU is ready to go */
3235                 if (!vcpu->kvm->arch.mmu_ready) {
3236                         spin_unlock(&vc->lock);
3237                         r = kvmhv_setup_mmu(vcpu);
3238                         spin_lock(&vc->lock);
3239                         if (r) {
3240                                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3241                                 kvm_run->fail_entry.
3242                                         hardware_entry_failure_reason = 0;
3243                                 vcpu->arch.ret = r;
3244                                 break;
3245                         }
3246                 }
3247
3248                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3249                         kvmppc_vcore_end_preempt(vc);
3250
3251                 if (vc->vcore_state != VCORE_INACTIVE) {
3252                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3253                         continue;
3254                 }
3255                 for_each_runnable_thread(i, v, vc) {
3256                         kvmppc_core_prepare_to_enter(v);
3257                         if (signal_pending(v->arch.run_task)) {
3258                                 kvmppc_remove_runnable(vc, v);
3259                                 v->stat.signal_exits++;
3260                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3261                                 v->arch.ret = -EINTR;
3262                                 wake_up(&v->arch.cpu_run);
3263                         }
3264                 }
3265                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3266                         break;
3267                 n_ceded = 0;
3268                 for_each_runnable_thread(i, v, vc) {
3269                         if (!kvmppc_vcpu_woken(v))
3270                                 n_ceded += v->arch.ceded;
3271                         else
3272                                 v->arch.ceded = 0;
3273                 }
3274                 vc->runner = vcpu;
3275                 if (n_ceded == vc->n_runnable) {
3276                         kvmppc_vcore_blocked(vc);
3277                 } else if (need_resched()) {
3278                         kvmppc_vcore_preempt(vc);
3279                         /* Let something else run */
3280                         cond_resched_lock(&vc->lock);
3281                         if (vc->vcore_state == VCORE_PREEMPT)
3282                                 kvmppc_vcore_end_preempt(vc);
3283                 } else {
3284                         kvmppc_run_core(vc);
3285                 }
3286                 vc->runner = NULL;
3287         }
3288
3289         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3290                (vc->vcore_state == VCORE_RUNNING ||
3291                 vc->vcore_state == VCORE_EXITING ||
3292                 vc->vcore_state == VCORE_PIGGYBACK))
3293                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3294
3295         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3296                 kvmppc_vcore_end_preempt(vc);
3297
3298         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3299                 kvmppc_remove_runnable(vc, vcpu);
3300                 vcpu->stat.signal_exits++;
3301                 kvm_run->exit_reason = KVM_EXIT_INTR;
3302                 vcpu->arch.ret = -EINTR;
3303         }
3304
3305         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3306                 /* Wake up some vcpu to run the core */
3307                 i = -1;
3308                 v = next_runnable_thread(vc, &i);
3309                 wake_up(&v->arch.cpu_run);
3310         }
3311
3312         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3313         spin_unlock(&vc->lock);
3314         return vcpu->arch.ret;
3315 }
3316
3317 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
3318 {
3319         int r;
3320         int srcu_idx;
3321         unsigned long ebb_regs[3] = {}; /* shut up GCC */
3322         unsigned long user_tar = 0;
3323         unsigned int user_vrsave;
3324         struct kvm *kvm;
3325
3326         if (!vcpu->arch.sane) {
3327                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3328                 return -EINVAL;
3329         }
3330
3331         /*
3332          * Don't allow entry with a suspended transaction, because
3333          * the guest entry/exit code will lose it.
3334          * If the guest has TM enabled, save away their TM-related SPRs
3335          * (they will get restored by the TM unavailable interrupt).
3336          */
3337 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3338         if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
3339             (current->thread.regs->msr & MSR_TM)) {
3340                 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
3341                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3342                         run->fail_entry.hardware_entry_failure_reason = 0;
3343                         return -EINVAL;
3344                 }
3345                 /* Enable TM so we can read the TM SPRs */
3346                 mtmsr(mfmsr() | MSR_TM);
3347                 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
3348                 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
3349                 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
3350                 current->thread.regs->msr &= ~MSR_TM;
3351         }
3352 #endif
3353
3354         kvmppc_core_prepare_to_enter(vcpu);
3355
3356         /* No need to go into the guest when all we'll do is come back out */
3357         if (signal_pending(current)) {
3358                 run->exit_reason = KVM_EXIT_INTR;
3359                 return -EINTR;
3360         }
3361
3362         kvm = vcpu->kvm;
3363         atomic_inc(&kvm->arch.vcpus_running);
3364         /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
3365         smp_mb();
3366
3367         flush_all_to_thread(current);
3368
3369         /* Save userspace EBB and other register values */
3370         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3371                 ebb_regs[0] = mfspr(SPRN_EBBHR);
3372                 ebb_regs[1] = mfspr(SPRN_EBBRR);
3373                 ebb_regs[2] = mfspr(SPRN_BESCR);
3374                 user_tar = mfspr(SPRN_TAR);
3375         }
3376         user_vrsave = mfspr(SPRN_VRSAVE);
3377
3378         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
3379         vcpu->arch.pgdir = current->mm->pgd;
3380         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3381
3382         do {
3383                 r = kvmppc_run_vcpu(run, vcpu);
3384
3385                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
3386                     !(vcpu->arch.shregs.msr & MSR_PR)) {
3387                         trace_kvm_hcall_enter(vcpu);
3388                         r = kvmppc_pseries_do_hcall(vcpu);
3389                         trace_kvm_hcall_exit(vcpu, r);
3390                         kvmppc_core_prepare_to_enter(vcpu);
3391                 } else if (r == RESUME_PAGE_FAULT) {
3392                         srcu_idx = srcu_read_lock(&kvm->srcu);
3393                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
3394                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
3395                         srcu_read_unlock(&kvm->srcu, srcu_idx);
3396                 } else if (r == RESUME_PASSTHROUGH) {
3397                         if (WARN_ON(xive_enabled()))
3398                                 r = H_SUCCESS;
3399                         else
3400                                 r = kvmppc_xics_rm_complete(vcpu, 0);
3401                 }
3402         } while (is_kvmppc_resume_guest(r));
3403
3404         /* Restore userspace EBB and other register values */
3405         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3406                 mtspr(SPRN_EBBHR, ebb_regs[0]);
3407                 mtspr(SPRN_EBBRR, ebb_regs[1]);
3408                 mtspr(SPRN_BESCR, ebb_regs[2]);
3409                 mtspr(SPRN_TAR, user_tar);
3410                 mtspr(SPRN_FSCR, current->thread.fscr);
3411         }
3412         mtspr(SPRN_VRSAVE, user_vrsave);
3413
3414         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
3415         atomic_dec(&kvm->arch.vcpus_running);
3416         return r;
3417 }
3418
3419 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
3420                                      int shift, int sllp)
3421 {
3422         (*sps)->page_shift = shift;
3423         (*sps)->slb_enc = sllp;
3424         (*sps)->enc[0].page_shift = shift;
3425         (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
3426         /*
3427          * Add 16MB MPSS support (may get filtered out by userspace)
3428          */
3429         if (shift != 24) {
3430                 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
3431                 if (penc != -1) {
3432                         (*sps)->enc[1].page_shift = 24;
3433                         (*sps)->enc[1].pte_enc = penc;
3434                 }
3435         }
3436         (*sps)++;
3437 }
3438
3439 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
3440                                          struct kvm_ppc_smmu_info *info)
3441 {
3442         struct kvm_ppc_one_seg_page_size *sps;
3443
3444         /*
3445          * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
3446          * POWER7 doesn't support keys for instruction accesses,
3447          * POWER8 and POWER9 do.
3448          */
3449         info->data_keys = 32;
3450         info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
3451
3452         /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
3453         info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
3454         info->slb_size = 32;
3455
3456         /* We only support these sizes for now, and no muti-size segments */
3457         sps = &info->sps[0];
3458         kvmppc_add_seg_page_size(&sps, 12, 0);
3459         kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
3460         kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
3461
3462         return 0;
3463 }
3464
3465 /*
3466  * Get (and clear) the dirty memory log for a memory slot.
3467  */
3468 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
3469                                          struct kvm_dirty_log *log)
3470 {
3471         struct kvm_memslots *slots;
3472         struct kvm_memory_slot *memslot;
3473         int i, r;
3474         unsigned long n;
3475         unsigned long *buf, *p;
3476         struct kvm_vcpu *vcpu;
3477
3478         mutex_lock(&kvm->slots_lock);
3479
3480         r = -EINVAL;
3481         if (log->slot >= KVM_USER_MEM_SLOTS)
3482                 goto out;
3483
3484         slots = kvm_memslots(kvm);
3485         memslot = id_to_memslot(slots, log->slot);
3486         r = -ENOENT;
3487         if (!memslot->dirty_bitmap)
3488                 goto out;
3489
3490         /*
3491          * Use second half of bitmap area because both HPT and radix
3492          * accumulate bits in the first half.
3493          */
3494         n = kvm_dirty_bitmap_bytes(memslot);
3495         buf = memslot->dirty_bitmap + n / sizeof(long);
3496         memset(buf, 0, n);
3497
3498         if (kvm_is_radix(kvm))
3499                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
3500         else
3501                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
3502         if (r)
3503                 goto out;
3504
3505         /*
3506          * We accumulate dirty bits in the first half of the
3507          * memslot's dirty_bitmap area, for when pages are paged
3508          * out or modified by the host directly.  Pick up these
3509          * bits and add them to the map.
3510          */
3511         p = memslot->dirty_bitmap;
3512         for (i = 0; i < n / sizeof(long); ++i)
3513                 buf[i] |= xchg(&p[i], 0);
3514
3515         /* Harvest dirty bits from VPA and DTL updates */
3516         /* Note: we never modify the SLB shadow buffer areas */
3517         kvm_for_each_vcpu(i, vcpu, kvm) {
3518                 spin_lock(&vcpu->arch.vpa_update_lock);
3519                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
3520                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
3521                 spin_unlock(&vcpu->arch.vpa_update_lock);
3522         }
3523
3524         r = -EFAULT;
3525         if (copy_to_user(log->dirty_bitmap, buf, n))
3526                 goto out;
3527
3528         r = 0;
3529 out:
3530         mutex_unlock(&kvm->slots_lock);
3531         return r;
3532 }
3533
3534 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
3535                                         struct kvm_memory_slot *dont)
3536 {
3537         if (!dont || free->arch.rmap != dont->arch.rmap) {
3538                 vfree(free->arch.rmap);
3539                 free->arch.rmap = NULL;
3540         }
3541 }
3542
3543 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3544                                          unsigned long npages)
3545 {
3546         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
3547         if (!slot->arch.rmap)
3548                 return -ENOMEM;
3549
3550         return 0;
3551 }
3552
3553 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3554                                         struct kvm_memory_slot *memslot,
3555                                         const struct kvm_userspace_memory_region *mem)
3556 {
3557         return 0;
3558 }
3559
3560 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3561                                 const struct kvm_userspace_memory_region *mem,
3562                                 const struct kvm_memory_slot *old,
3563                                 const struct kvm_memory_slot *new)
3564 {
3565         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3566
3567         /*
3568          * If we are making a new memslot, it might make
3569          * some address that was previously cached as emulated
3570          * MMIO be no longer emulated MMIO, so invalidate
3571          * all the caches of emulated MMIO translations.
3572          */
3573         if (npages)
3574                 atomic64_inc(&kvm->arch.mmio_update);
3575 }
3576
3577 /*
3578  * Update LPCR values in kvm->arch and in vcores.
3579  * Caller must hold kvm->lock.
3580  */
3581 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3582 {
3583         long int i;
3584         u32 cores_done = 0;
3585
3586         if ((kvm->arch.lpcr & mask) == lpcr)
3587                 return;
3588
3589         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3590
3591         for (i = 0; i < KVM_MAX_VCORES; ++i) {
3592                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3593                 if (!vc)
3594                         continue;
3595                 spin_lock(&vc->lock);
3596                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3597                 spin_unlock(&vc->lock);
3598                 if (++cores_done >= kvm->arch.online_vcores)
3599                         break;
3600         }
3601 }
3602
3603 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3604 {
3605         return;
3606 }
3607
3608 void kvmppc_setup_partition_table(struct kvm *kvm)
3609 {
3610         unsigned long dw0, dw1;
3611
3612         if (!kvm_is_radix(kvm)) {
3613                 /* PS field - page size for VRMA */
3614                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
3615                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
3616                 /* HTABSIZE and HTABORG fields */
3617                 dw0 |= kvm->arch.sdr1;
3618
3619                 /* Second dword as set by userspace */
3620                 dw1 = kvm->arch.process_table;
3621         } else {
3622                 dw0 = PATB_HR | radix__get_tree_size() |
3623                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
3624                 dw1 = PATB_GR | kvm->arch.process_table;
3625         }
3626
3627         mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
3628 }
3629
3630 /*
3631  * Set up HPT (hashed page table) and RMA (real-mode area).
3632  * Must be called with kvm->lock held.
3633  */
3634 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3635 {
3636         int err = 0;
3637         struct kvm *kvm = vcpu->kvm;
3638         unsigned long hva;
3639         struct kvm_memory_slot *memslot;
3640         struct vm_area_struct *vma;
3641         unsigned long lpcr = 0, senc;
3642         unsigned long psize, porder;
3643         int srcu_idx;
3644
3645         /* Allocate hashed page table (if not done already) and reset it */
3646         if (!kvm->arch.hpt.virt) {
3647                 int order = KVM_DEFAULT_HPT_ORDER;
3648                 struct kvm_hpt_info info;
3649
3650                 err = kvmppc_allocate_hpt(&info, order);
3651                 /* If we get here, it means userspace didn't specify a
3652                  * size explicitly.  So, try successively smaller
3653                  * sizes if the default failed. */
3654                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
3655                         err  = kvmppc_allocate_hpt(&info, order);
3656
3657                 if (err < 0) {
3658                         pr_err("KVM: Couldn't alloc HPT\n");
3659                         goto out;
3660                 }
3661
3662                 kvmppc_set_hpt(kvm, &info);
3663         }
3664
3665         /* Look up the memslot for guest physical address 0 */
3666         srcu_idx = srcu_read_lock(&kvm->srcu);
3667         memslot = gfn_to_memslot(kvm, 0);
3668
3669         /* We must have some memory at 0 by now */
3670         err = -EINVAL;
3671         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3672                 goto out_srcu;
3673
3674         /* Look up the VMA for the start of this memory slot */
3675         hva = memslot->userspace_addr;
3676         down_read(&current->mm->mmap_sem);
3677         vma = find_vma(current->mm, hva);
3678         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3679                 goto up_out;
3680
3681         psize = vma_kernel_pagesize(vma);
3682
3683         up_read(&current->mm->mmap_sem);
3684
3685         /* We can handle 4k, 64k or 16M pages in the VRMA */
3686         if (psize >= 0x1000000)
3687                 psize = 0x1000000;
3688         else if (psize >= 0x10000)
3689                 psize = 0x10000;
3690         else
3691                 psize = 0x1000;
3692         porder = __ilog2(psize);
3693
3694         senc = slb_pgsize_encoding(psize);
3695         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3696                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3697         /* Create HPTEs in the hash page table for the VRMA */
3698         kvmppc_map_vrma(vcpu, memslot, porder);
3699
3700         /* Update VRMASD field in the LPCR */
3701         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3702                 /* the -4 is to account for senc values starting at 0x10 */
3703                 lpcr = senc << (LPCR_VRMASD_SH - 4);
3704                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3705         }
3706
3707         /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
3708         smp_wmb();
3709         err = 0;
3710  out_srcu:
3711         srcu_read_unlock(&kvm->srcu, srcu_idx);
3712  out:
3713         return err;
3714
3715  up_out:
3716         up_read(&current->mm->mmap_sem);
3717         goto out_srcu;
3718 }
3719
3720 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
3721 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
3722 {
3723         kvmppc_free_radix(kvm);
3724         kvmppc_update_lpcr(kvm, LPCR_VPM1,
3725                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
3726         kvmppc_rmap_reset(kvm);
3727         kvm->arch.radix = 0;
3728         kvm->arch.process_table = 0;
3729         return 0;
3730 }
3731
3732 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
3733 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
3734 {
3735         int err;
3736
3737         err = kvmppc_init_vm_radix(kvm);
3738         if (err)
3739                 return err;
3740
3741         kvmppc_free_hpt(&kvm->arch.hpt);
3742         kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
3743                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
3744         kvm->arch.radix = 1;
3745         return 0;
3746 }
3747
3748 #ifdef CONFIG_KVM_XICS
3749 /*
3750  * Allocate a per-core structure for managing state about which cores are
3751  * running in the host versus the guest and for exchanging data between
3752  * real mode KVM and CPU running in the host.
3753  * This is only done for the first VM.
3754  * The allocated structure stays even if all VMs have stopped.
3755  * It is only freed when the kvm-hv module is unloaded.
3756  * It's OK for this routine to fail, we just don't support host
3757  * core operations like redirecting H_IPI wakeups.
3758  */
3759 void kvmppc_alloc_host_rm_ops(void)
3760 {
3761         struct kvmppc_host_rm_ops *ops;
3762         unsigned long l_ops;
3763         int cpu, core;
3764         int size;
3765
3766         /* Not the first time here ? */
3767         if (kvmppc_host_rm_ops_hv != NULL)
3768                 return;
3769
3770         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3771         if (!ops)
3772                 return;
3773
3774         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3775         ops->rm_core = kzalloc(size, GFP_KERNEL);
3776
3777         if (!ops->rm_core) {
3778                 kfree(ops);
3779                 return;
3780         }
3781
3782         cpus_read_lock();
3783
3784         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3785                 if (!cpu_online(cpu))
3786                         continue;
3787
3788                 core = cpu >> threads_shift;
3789                 ops->rm_core[core].rm_state.in_host = 1;
3790         }
3791
3792         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3793
3794         /*
3795          * Make the contents of the kvmppc_host_rm_ops structure visible
3796          * to other CPUs before we assign it to the global variable.
3797          * Do an atomic assignment (no locks used here), but if someone
3798          * beats us to it, just free our copy and return.
3799          */
3800         smp_wmb();
3801         l_ops = (unsigned long) ops;
3802
3803         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3804                 cpus_read_unlock();
3805                 kfree(ops->rm_core);
3806                 kfree(ops);
3807                 return;
3808         }
3809
3810         cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
3811                                              "ppc/kvm_book3s:prepare",
3812                                              kvmppc_set_host_core,
3813                                              kvmppc_clear_host_core);
3814         cpus_read_unlock();
3815 }
3816
3817 void kvmppc_free_host_rm_ops(void)
3818 {
3819         if (kvmppc_host_rm_ops_hv) {
3820                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
3821                 kfree(kvmppc_host_rm_ops_hv->rm_core);
3822                 kfree(kvmppc_host_rm_ops_hv);
3823                 kvmppc_host_rm_ops_hv = NULL;
3824         }
3825 }
3826 #endif
3827
3828 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3829 {
3830         unsigned long lpcr, lpid;
3831         char buf[32];
3832         int ret;
3833
3834         /* Allocate the guest's logical partition ID */
3835
3836         lpid = kvmppc_alloc_lpid();
3837         if ((long)lpid < 0)
3838                 return -ENOMEM;
3839         kvm->arch.lpid = lpid;
3840
3841         kvmppc_alloc_host_rm_ops();
3842
3843         /*
3844          * Since we don't flush the TLB when tearing down a VM,
3845          * and this lpid might have previously been used,
3846          * make sure we flush on each core before running the new VM.
3847          * On POWER9, the tlbie in mmu_partition_table_set_entry()
3848          * does this flush for us.
3849          */
3850         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3851                 cpumask_setall(&kvm->arch.need_tlb_flush);
3852
3853         /* Start out with the default set of hcalls enabled */
3854         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3855                sizeof(kvm->arch.enabled_hcalls));
3856
3857         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3858                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3859
3860         /* Init LPCR for virtual RMA mode */
3861         kvm->arch.host_lpid = mfspr(SPRN_LPID);
3862         kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3863         lpcr &= LPCR_PECE | LPCR_LPES;
3864         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3865                 LPCR_VPM0 | LPCR_VPM1;
3866         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3867                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3868         /* On POWER8 turn on online bit to enable PURR/SPURR */
3869         if (cpu_has_feature(CPU_FTR_ARCH_207S))
3870                 lpcr |= LPCR_ONL;
3871         /*
3872          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
3873          * Set HVICE bit to enable hypervisor virtualization interrupts.
3874          * Set HEIC to prevent OS interrupts to go to hypervisor (should
3875          * be unnecessary but better safe than sorry in case we re-enable
3876          * EE in HV mode with this LPCR still set)
3877          */
3878         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3879                 lpcr &= ~LPCR_VPM0;
3880                 lpcr |= LPCR_HVICE | LPCR_HEIC;
3881
3882                 /*
3883                  * If xive is enabled, we route 0x500 interrupts directly
3884                  * to the guest.
3885                  */
3886                 if (xive_enabled())
3887                         lpcr |= LPCR_LPES;
3888         }
3889
3890         /*
3891          * If the host uses radix, the guest starts out as radix.
3892          */
3893         if (radix_enabled()) {
3894                 kvm->arch.radix = 1;
3895                 kvm->arch.mmu_ready = 1;
3896                 lpcr &= ~LPCR_VPM1;
3897                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
3898                 ret = kvmppc_init_vm_radix(kvm);
3899                 if (ret) {
3900                         kvmppc_free_lpid(kvm->arch.lpid);
3901                         return ret;
3902                 }
3903                 kvmppc_setup_partition_table(kvm);
3904         }
3905
3906         kvm->arch.lpcr = lpcr;
3907
3908         /* Initialization for future HPT resizes */
3909         kvm->arch.resize_hpt = NULL;
3910
3911         /*
3912          * Work out how many sets the TLB has, for the use of
3913          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
3914          */
3915         if (radix_enabled())
3916                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
3917         else if (cpu_has_feature(CPU_FTR_ARCH_300))
3918                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
3919         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
3920                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
3921         else
3922                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
3923
3924         /*
3925          * Track that we now have a HV mode VM active. This blocks secondary
3926          * CPU threads from coming online.
3927          * On POWER9, we only need to do this if the "indep_threads_mode"
3928          * module parameter has been set to N.
3929          */
3930         if (cpu_has_feature(CPU_FTR_ARCH_300))
3931                 kvm->arch.threads_indep = indep_threads_mode;
3932         if (!kvm->arch.threads_indep)
3933                 kvm_hv_vm_activated();
3934
3935         /*
3936          * Initialize smt_mode depending on processor.
3937          * POWER8 and earlier have to use "strict" threading, where
3938          * all vCPUs in a vcore have to run on the same (sub)core,
3939          * whereas on POWER9 the threads can each run a different
3940          * guest.
3941          */
3942         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3943                 kvm->arch.smt_mode = threads_per_subcore;
3944         else
3945                 kvm->arch.smt_mode = 1;
3946         kvm->arch.emul_smt_mode = 1;
3947
3948         /*
3949          * Create a debugfs directory for the VM
3950          */
3951         snprintf(buf, sizeof(buf), "vm%d", current->pid);
3952         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3953         if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3954                 kvmppc_mmu_debugfs_init(kvm);
3955
3956         return 0;
3957 }
3958
3959 static void kvmppc_free_vcores(struct kvm *kvm)
3960 {
3961         long int i;
3962
3963         for (i = 0; i < KVM_MAX_VCORES; ++i)
3964                 kfree(kvm->arch.vcores[i]);
3965         kvm->arch.online_vcores = 0;
3966 }
3967
3968 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3969 {
3970         debugfs_remove_recursive(kvm->arch.debugfs_dir);
3971
3972         if (!kvm->arch.threads_indep)
3973                 kvm_hv_vm_deactivated();
3974
3975         kvmppc_free_vcores(kvm);
3976
3977         kvmppc_free_lpid(kvm->arch.lpid);
3978
3979         if (kvm_is_radix(kvm))
3980                 kvmppc_free_radix(kvm);
3981         else
3982                 kvmppc_free_hpt(&kvm->arch.hpt);
3983
3984         kvmppc_free_pimap(kvm);
3985 }
3986
3987 /* We don't need to emulate any privileged instructions or dcbz */
3988 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3989                                      unsigned int inst, int *advance)
3990 {
3991         return EMULATE_FAIL;
3992 }
3993
3994 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3995                                         ulong spr_val)
3996 {
3997         return EMULATE_FAIL;
3998 }
3999
4000 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4001                                         ulong *spr_val)
4002 {
4003         return EMULATE_FAIL;
4004 }
4005
4006 static int kvmppc_core_check_processor_compat_hv(void)
4007 {
4008         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
4009             !cpu_has_feature(CPU_FTR_ARCH_206))
4010                 return -EIO;
4011
4012         return 0;
4013 }
4014
4015 #ifdef CONFIG_KVM_XICS
4016
4017 void kvmppc_free_pimap(struct kvm *kvm)
4018 {
4019         kfree(kvm->arch.pimap);
4020 }
4021
4022 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
4023 {
4024         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
4025 }
4026
4027 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4028 {
4029         struct irq_desc *desc;
4030         struct kvmppc_irq_map *irq_map;
4031         struct kvmppc_passthru_irqmap *pimap;
4032         struct irq_chip *chip;
4033         int i, rc = 0;
4034
4035         if (!kvm_irq_bypass)
4036                 return 1;
4037
4038         desc = irq_to_desc(host_irq);
4039         if (!desc)
4040                 return -EIO;
4041
4042         mutex_lock(&kvm->lock);
4043
4044         pimap = kvm->arch.pimap;
4045         if (pimap == NULL) {
4046                 /* First call, allocate structure to hold IRQ map */
4047                 pimap = kvmppc_alloc_pimap();
4048                 if (pimap == NULL) {
4049                         mutex_unlock(&kvm->lock);
4050                         return -ENOMEM;
4051                 }
4052                 kvm->arch.pimap = pimap;
4053         }
4054
4055         /*
4056          * For now, we only support interrupts for which the EOI operation
4057          * is an OPAL call followed by a write to XIRR, since that's
4058          * what our real-mode EOI code does, or a XIVE interrupt
4059          */
4060         chip = irq_data_get_irq_chip(&desc->irq_data);
4061         if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
4062                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
4063                         host_irq, guest_gsi);
4064                 mutex_unlock(&kvm->lock);
4065                 return -ENOENT;
4066         }
4067
4068         /*
4069          * See if we already have an entry for this guest IRQ number.
4070          * If it's mapped to a hardware IRQ number, that's an error,
4071          * otherwise re-use this entry.
4072          */
4073         for (i = 0; i < pimap->n_mapped; i++) {
4074                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
4075                         if (pimap->mapped[i].r_hwirq) {
4076                                 mutex_unlock(&kvm->lock);
4077                                 return -EINVAL;
4078                         }
4079                         break;
4080                 }
4081         }
4082
4083         if (i == KVMPPC_PIRQ_MAPPED) {
4084                 mutex_unlock(&kvm->lock);
4085                 return -EAGAIN;         /* table is full */
4086         }
4087
4088         irq_map = &pimap->mapped[i];
4089
4090         irq_map->v_hwirq = guest_gsi;
4091         irq_map->desc = desc;
4092
4093         /*
4094          * Order the above two stores before the next to serialize with
4095          * the KVM real mode handler.
4096          */
4097         smp_wmb();
4098         irq_map->r_hwirq = desc->irq_data.hwirq;
4099
4100         if (i == pimap->n_mapped)
4101                 pimap->n_mapped++;
4102
4103         if (xive_enabled())
4104                 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
4105         else
4106                 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
4107         if (rc)
4108                 irq_map->r_hwirq = 0;
4109
4110         mutex_unlock(&kvm->lock);
4111
4112         return 0;
4113 }
4114
4115 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4116 {
4117         struct irq_desc *desc;
4118         struct kvmppc_passthru_irqmap *pimap;
4119         int i, rc = 0;
4120
4121         if (!kvm_irq_bypass)
4122                 return 0;
4123
4124         desc = irq_to_desc(host_irq);
4125         if (!desc)
4126                 return -EIO;
4127
4128         mutex_lock(&kvm->lock);
4129         if (!kvm->arch.pimap)
4130                 goto unlock;
4131
4132         pimap = kvm->arch.pimap;
4133
4134         for (i = 0; i < pimap->n_mapped; i++) {
4135                 if (guest_gsi == pimap->mapped[i].v_hwirq)
4136                         break;
4137         }
4138
4139         if (i == pimap->n_mapped) {
4140                 mutex_unlock(&kvm->lock);
4141                 return -ENODEV;
4142         }
4143
4144         if (xive_enabled())
4145                 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
4146         else
4147                 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
4148
4149         /* invalidate the entry (what do do on error from the above ?) */
4150         pimap->mapped[i].r_hwirq = 0;
4151
4152         /*
4153          * We don't free this structure even when the count goes to
4154          * zero. The structure is freed when we destroy the VM.
4155          */
4156  unlock:
4157         mutex_unlock(&kvm->lock);
4158         return rc;
4159 }
4160
4161 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
4162                                              struct irq_bypass_producer *prod)
4163 {
4164         int ret = 0;
4165         struct kvm_kernel_irqfd *irqfd =
4166                 container_of(cons, struct kvm_kernel_irqfd, consumer);
4167
4168         irqfd->producer = prod;
4169
4170         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4171         if (ret)
4172                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
4173                         prod->irq, irqfd->gsi, ret);
4174
4175         return ret;
4176 }
4177
4178 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
4179                                               struct irq_bypass_producer *prod)
4180 {
4181         int ret;
4182         struct kvm_kernel_irqfd *irqfd =
4183                 container_of(cons, struct kvm_kernel_irqfd, consumer);
4184
4185         irqfd->producer = NULL;
4186
4187         /*
4188          * When producer of consumer is unregistered, we change back to
4189          * default external interrupt handling mode - KVM real mode
4190          * will switch back to host.
4191          */
4192         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4193         if (ret)
4194                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
4195                         prod->irq, irqfd->gsi, ret);
4196 }
4197 #endif
4198
4199 static long kvm_arch_vm_ioctl_hv(struct file *filp,
4200                                  unsigned int ioctl, unsigned long arg)
4201 {
4202         struct kvm *kvm __maybe_unused = filp->private_data;
4203         void __user *argp = (void __user *)arg;
4204         long r;
4205
4206         switch (ioctl) {
4207
4208         case KVM_PPC_ALLOCATE_HTAB: {
4209                 u32 htab_order;
4210
4211                 r = -EFAULT;
4212                 if (get_user(htab_order, (u32 __user *)argp))
4213                         break;
4214                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
4215                 if (r)
4216                         break;
4217                 r = 0;
4218                 break;
4219         }
4220
4221         case KVM_PPC_GET_HTAB_FD: {
4222                 struct kvm_get_htab_fd ghf;
4223
4224                 r = -EFAULT;
4225                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
4226                         break;
4227                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
4228                 break;
4229         }
4230
4231         case KVM_PPC_RESIZE_HPT_PREPARE: {
4232                 struct kvm_ppc_resize_hpt rhpt;
4233
4234                 r = -EFAULT;
4235                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4236                         break;
4237
4238                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
4239                 break;
4240         }
4241
4242         case KVM_PPC_RESIZE_HPT_COMMIT: {
4243                 struct kvm_ppc_resize_hpt rhpt;
4244
4245                 r = -EFAULT;
4246                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4247                         break;
4248
4249                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
4250                 break;
4251         }
4252
4253         default:
4254                 r = -ENOTTY;
4255         }
4256
4257         return r;
4258 }
4259
4260 /*
4261  * List of hcall numbers to enable by default.
4262  * For compatibility with old userspace, we enable by default
4263  * all hcalls that were implemented before the hcall-enabling
4264  * facility was added.  Note this list should not include H_RTAS.
4265  */
4266 static unsigned int default_hcall_list[] = {
4267         H_REMOVE,
4268         H_ENTER,
4269         H_READ,
4270         H_PROTECT,
4271         H_BULK_REMOVE,
4272         H_GET_TCE,
4273         H_PUT_TCE,
4274         H_SET_DABR,
4275         H_SET_XDABR,
4276         H_CEDE,
4277         H_PROD,
4278         H_CONFER,
4279         H_REGISTER_VPA,
4280 #ifdef CONFIG_KVM_XICS
4281         H_EOI,
4282         H_CPPR,
4283         H_IPI,
4284         H_IPOLL,
4285         H_XIRR,
4286         H_XIRR_X,
4287 #endif
4288         0
4289 };
4290
4291 static void init_default_hcalls(void)
4292 {
4293         int i;
4294         unsigned int hcall;
4295
4296         for (i = 0; default_hcall_list[i]; ++i) {
4297                 hcall = default_hcall_list[i];
4298                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
4299                 __set_bit(hcall / 4, default_enabled_hcalls);
4300         }
4301 }
4302
4303 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
4304 {
4305         unsigned long lpcr;
4306         int radix;
4307         int err;
4308
4309         /* If not on a POWER9, reject it */
4310         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4311                 return -ENODEV;
4312
4313         /* If any unknown flags set, reject it */
4314         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
4315                 return -EINVAL;
4316
4317         /* GR (guest radix) bit in process_table field must match */
4318         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
4319         if (!!(cfg->process_table & PATB_GR) != radix)
4320                 return -EINVAL;
4321
4322         /* Process table size field must be reasonable, i.e. <= 24 */
4323         if ((cfg->process_table & PRTS_MASK) > 24)
4324                 return -EINVAL;
4325
4326         /* We can change a guest to/from radix now, if the host is radix */
4327         if (radix && !radix_enabled())
4328                 return -EINVAL;
4329
4330         mutex_lock(&kvm->lock);
4331         if (radix != kvm_is_radix(kvm)) {
4332                 if (kvm->arch.mmu_ready) {
4333                         kvm->arch.mmu_ready = 0;
4334                         /* order mmu_ready vs. vcpus_running */
4335                         smp_mb();
4336                         if (atomic_read(&kvm->arch.vcpus_running)) {
4337                                 kvm->arch.mmu_ready = 1;
4338                                 err = -EBUSY;
4339                                 goto out_unlock;
4340                         }
4341                 }
4342                 if (radix)
4343                         err = kvmppc_switch_mmu_to_radix(kvm);
4344                 else
4345                         err = kvmppc_switch_mmu_to_hpt(kvm);
4346                 if (err)
4347                         goto out_unlock;
4348         }
4349
4350         kvm->arch.process_table = cfg->process_table;
4351         kvmppc_setup_partition_table(kvm);
4352
4353         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
4354         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
4355         err = 0;
4356
4357  out_unlock:
4358         mutex_unlock(&kvm->lock);
4359         return err;
4360 }
4361
4362 static struct kvmppc_ops kvm_ops_hv = {
4363         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
4364         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
4365         .get_one_reg = kvmppc_get_one_reg_hv,
4366         .set_one_reg = kvmppc_set_one_reg_hv,
4367         .vcpu_load   = kvmppc_core_vcpu_load_hv,
4368         .vcpu_put    = kvmppc_core_vcpu_put_hv,
4369         .set_msr     = kvmppc_set_msr_hv,
4370         .vcpu_run    = kvmppc_vcpu_run_hv,
4371         .vcpu_create = kvmppc_core_vcpu_create_hv,
4372         .vcpu_free   = kvmppc_core_vcpu_free_hv,
4373         .check_requests = kvmppc_core_check_requests_hv,
4374         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
4375         .flush_memslot  = kvmppc_core_flush_memslot_hv,
4376         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
4377         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
4378         .unmap_hva = kvm_unmap_hva_hv,
4379         .unmap_hva_range = kvm_unmap_hva_range_hv,
4380         .age_hva  = kvm_age_hva_hv,
4381         .test_age_hva = kvm_test_age_hva_hv,
4382         .set_spte_hva = kvm_set_spte_hva_hv,
4383         .mmu_destroy  = kvmppc_mmu_destroy_hv,
4384         .free_memslot = kvmppc_core_free_memslot_hv,
4385         .create_memslot = kvmppc_core_create_memslot_hv,
4386         .init_vm =  kvmppc_core_init_vm_hv,
4387         .destroy_vm = kvmppc_core_destroy_vm_hv,
4388         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
4389         .emulate_op = kvmppc_core_emulate_op_hv,
4390         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
4391         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
4392         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
4393         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
4394         .hcall_implemented = kvmppc_hcall_impl_hv,
4395 #ifdef CONFIG_KVM_XICS
4396         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
4397         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
4398 #endif
4399         .configure_mmu = kvmhv_configure_mmu,
4400         .get_rmmu_info = kvmhv_get_rmmu_info,
4401         .set_smt_mode = kvmhv_set_smt_mode,
4402 };
4403
4404 static int kvm_init_subcore_bitmap(void)
4405 {
4406         int i, j;
4407         int nr_cores = cpu_nr_cores();
4408         struct sibling_subcore_state *sibling_subcore_state;
4409
4410         for (i = 0; i < nr_cores; i++) {
4411                 int first_cpu = i * threads_per_core;
4412                 int node = cpu_to_node(first_cpu);
4413
4414                 /* Ignore if it is already allocated. */
4415                 if (paca_ptrs[first_cpu]->sibling_subcore_state)
4416                         continue;
4417
4418                 sibling_subcore_state =
4419                         kmalloc_node(sizeof(struct sibling_subcore_state),
4420                                                         GFP_KERNEL, node);
4421                 if (!sibling_subcore_state)
4422                         return -ENOMEM;
4423
4424                 memset(sibling_subcore_state, 0,
4425                                 sizeof(struct sibling_subcore_state));
4426
4427                 for (j = 0; j < threads_per_core; j++) {
4428                         int cpu = first_cpu + j;
4429
4430                         paca_ptrs[cpu]->sibling_subcore_state =
4431                                                 sibling_subcore_state;
4432                 }
4433         }
4434         return 0;
4435 }
4436
4437 static int kvmppc_radix_possible(void)
4438 {
4439         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
4440 }
4441
4442 static int kvmppc_book3s_init_hv(void)
4443 {
4444         int r;
4445         /*
4446          * FIXME!! Do we need to check on all cpus ?
4447          */
4448         r = kvmppc_core_check_processor_compat_hv();
4449         if (r < 0)
4450                 return -ENODEV;
4451
4452         r = kvm_init_subcore_bitmap();
4453         if (r)
4454                 return r;
4455
4456         /*
4457          * We need a way of accessing the XICS interrupt controller,
4458          * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
4459          * indirectly, via OPAL.
4460          */
4461 #ifdef CONFIG_SMP
4462         if (!xive_enabled() && !local_paca->kvm_hstate.xics_phys) {
4463                 struct device_node *np;
4464
4465                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
4466                 if (!np) {
4467                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
4468                         return -ENODEV;
4469                 }
4470         }
4471 #endif
4472
4473         kvm_ops_hv.owner = THIS_MODULE;
4474         kvmppc_hv_ops = &kvm_ops_hv;
4475
4476         init_default_hcalls();
4477
4478         init_vcore_lists();
4479
4480         r = kvmppc_mmu_hv_init();
4481         if (r)
4482                 return r;
4483
4484         if (kvmppc_radix_possible())
4485                 r = kvmppc_radix_init();
4486
4487         /*
4488          * POWER9 chips before version 2.02 can't have some threads in
4489          * HPT mode and some in radix mode on the same core.
4490          */
4491         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4492                 unsigned int pvr = mfspr(SPRN_PVR);
4493                 if ((pvr >> 16) == PVR_POWER9 &&
4494                     (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
4495                      ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
4496                         no_mixing_hpt_and_radix = true;
4497         }
4498
4499         return r;
4500 }
4501
4502 static void kvmppc_book3s_exit_hv(void)
4503 {
4504         kvmppc_free_host_rm_ops();
4505         if (kvmppc_radix_possible())
4506                 kvmppc_radix_exit();
4507         kvmppc_hv_ops = NULL;
4508 }
4509
4510 module_init(kvmppc_book3s_init_hv);
4511 module_exit(kvmppc_book3s_exit_hv);
4512 MODULE_LICENSE("GPL");
4513 MODULE_ALIAS_MISCDEV(KVM_MINOR);
4514 MODULE_ALIAS("devname:kvm");