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/err.h>
  23 #include <linux/slab.h>
  24 #include <linux/preempt.h>
  25 #include <linux/sched/signal.h>
  26 #include <linux/sched/stat.h>
  27 #include <linux/delay.h>
  28 #include <linux/export.h>
  29 #include <linux/fs.h>
  30 #include <linux/anon_inodes.h>
  31 #include <linux/cpu.h>
  32 #include <linux/cpumask.h>
  33 #include <linux/spinlock.h>
  34 #include <linux/page-flags.h>
  35 #include <linux/srcu.h>
  36 #include <linux/miscdevice.h>
  37 #include <linux/debugfs.h>
  38 #include <linux/gfp.h>
  39 #include <linux/vmalloc.h>
  40 #include <linux/highmem.h>
  41 #include <linux/hugetlb.h>
  42 #include <linux/kvm_irqfd.h>
  43 #include <linux/irqbypass.h>
  44 #include <linux/module.h>
  45 #include <linux/compiler.h>
  46 #include <linux/of.h>
  47
  48 #include <asm/reg.h>
  49 #include <asm/ppc-opcode.h>
  50 #include <asm/disassemble.h>
  51 #include <asm/cputable.h>
  52 #include <asm/cacheflush.h>
  53 #include <asm/tlbflush.h>
  54 #include <linux/uaccess.h>
  55 #include <asm/io.h>
  56 #include <asm/kvm_ppc.h>
  57 #include <asm/kvm_book3s.h>
  58 #include <asm/mmu_context.h>
  59 #include <asm/lppaca.h>
  60 #include <asm/processor.h>
  61 #include <asm/cputhreads.h>
  62 #include <asm/page.h>
  63 #include <asm/hvcall.h>
  64 #include <asm/switch_to.h>
  65 #include <asm/smp.h>
  66 #include <asm/dbell.h>
  67 #include <asm/hmi.h>
  68 #include <asm/pnv-pci.h>
  69 #include <asm/mmu.h>
  70 #include <asm/opal.h>
  71 #include <asm/xics.h>
  72 #include <asm/xive.h>
  73
  74 #include "book3s.h"
  75
  76 #define CREATE_TRACE_POINTS
  77 #include "trace_hv.h"
  78
  79 /* #define EXIT_DEBUG */
  80 /* #define EXIT_DEBUG_SIMPLE */
  81 /* #define EXIT_DEBUG_INT */
  82
  83 /* Used to indicate that a guest page fault needs to be handled */
  84 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  85 /* Used to indicate that a guest passthrough interrupt needs to be handled */
  86 #define RESUME_PASSTHROUGH      (RESUME_GUEST | RESUME_FLAG_ARCH2)
  87
  88 /* Used as a "null" value for timebase values */
  89 #define TB_NIL  (~(u64)0)
  90
  91 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
  92
  93 static int dynamic_mt_modes = 6;
  94 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
  95 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
  96 static int target_smt_mode;
  97 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
  98 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
  99
 100 #ifdef CONFIG_KVM_XICS
 101 static struct kernel_param_ops module_param_ops = {
 102         .set = param_set_int,
 103         .get = param_get_int,
 104 };
 105
 106 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass,
 107                                                         S_IRUGO | S_IWUSR);
 108 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
 109
 110 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect,
 111                                                         S_IRUGO | S_IWUSR);
 112 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
 113 #endif
 114
 115 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
 116 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
 117
 118 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
 119                 int *ip)
 120 {
 121         int i = *ip;
 122         struct kvm_vcpu *vcpu;
 123
 124         while (++i < MAX_SMT_THREADS) {
 125                 vcpu = READ_ONCE(vc->runnable_threads[i]);
 126                 if (vcpu) {
 127                         *ip = i;
 128                         return vcpu;
 129                 }
 130         }
 131         return NULL;
 132 }
 133
 134 /* Used to traverse the list of runnable threads for a given vcore */
 135 #define for_each_runnable_thread(i, vcpu, vc) \
 136         for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
 137
 138 static bool kvmppc_ipi_thread(int cpu)
 139 {
 140         unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
 141
 142         /* On POWER9 we can use msgsnd to IPI any cpu */
 143         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
 144                 msg |= get_hard_smp_processor_id(cpu);
 145                 smp_mb();
 146                 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 147                 return true;
 148         }
 149
 150         /* On POWER8 for IPIs to threads in the same core, use msgsnd */
 151         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
 152                 preempt_disable();
 153                 if (cpu_first_thread_sibling(cpu) ==
 154                     cpu_first_thread_sibling(smp_processor_id())) {
 155                         msg |= cpu_thread_in_core(cpu);
 156                         smp_mb();
 157                         __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 158                         preempt_enable();
 159                         return true;
 160                 }
 161                 preempt_enable();
 162         }
 163
 164 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
 165         if (cpu >= 0 && cpu < nr_cpu_ids) {
 166                 if (paca[cpu].kvm_hstate.xics_phys) {
 167                         xics_wake_cpu(cpu);
 168                         return true;
 169                 }
 170                 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
 171                 return true;
 172         }
 173 #endif
 174
 175         return false;
 176 }
 177
 178 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
 179 {
 180         int cpu;
 181         struct swait_queue_head *wqp;
 182
 183         wqp = kvm_arch_vcpu_wq(vcpu);
 184         if (swait_active(wqp)) {
 185                 swake_up(wqp);
 186                 ++vcpu->stat.halt_wakeup;
 187         }
 188
 189         cpu = READ_ONCE(vcpu->arch.thread_cpu);
 190         if (cpu >= 0 && kvmppc_ipi_thread(cpu))
 191                 return;
 192
 193         /* CPU points to the first thread of the core */
 194         cpu = vcpu->cpu;
 195         if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
 196                 smp_send_reschedule(cpu);
 197 }
 198
 199 /*
 200  * We use the vcpu_load/put functions to measure stolen time.
 201  * Stolen time is counted as time when either the vcpu is able to
 202  * run as part of a virtual core, but the task running the vcore
 203  * is preempted or sleeping, or when the vcpu needs something done
 204  * in the kernel by the task running the vcpu, but that task is
 205  * preempted or sleeping.  Those two things have to be counted
 206  * separately, since one of the vcpu tasks will take on the job
 207  * of running the core, and the other vcpu tasks in the vcore will
 208  * sleep waiting for it to do that, but that sleep shouldn't count
 209  * as stolen time.
 210  *
 211  * Hence we accumulate stolen time when the vcpu can run as part of
 212  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 213  * needs its task to do other things in the kernel (for example,
 214  * service a page fault) in busy_stolen.  We don't accumulate
 215  * stolen time for a vcore when it is inactive, or for a vcpu
 216  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 217  * a misnomer; it means that the vcpu task is not executing in
 218  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 219  * the kernel.  We don't have any way of dividing up that time
 220  * between time that the vcpu is genuinely stopped, time that
 221  * the task is actively working on behalf of the vcpu, and time
 222  * that the task is preempted, so we don't count any of it as
 223  * stolen.
 224  *
 225  * Updates to busy_stolen are protected by arch.tbacct_lock;
 226  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
 227  * lock.  The stolen times are measured in units of timebase ticks.
 228  * (Note that the != TB_NIL checks below are purely defensive;
 229  * they should never fail.)
 230  */
 231
 232 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
 233 {
 234         unsigned long flags;
 235
 236         spin_lock_irqsave(&vc->stoltb_lock, flags);
 237         vc->preempt_tb = mftb();
 238         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 239 }
 240
 241 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
 242 {
 243         unsigned long flags;
 244
 245         spin_lock_irqsave(&vc->stoltb_lock, flags);
 246         if (vc->preempt_tb != TB_NIL) {
 247                 vc->stolen_tb += mftb() - vc->preempt_tb;
 248                 vc->preempt_tb = TB_NIL;
 249         }
 250         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 251 }
 252
 253 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 254 {
 255         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 256         unsigned long flags;
 257
 258         /*
 259          * We can test vc->runner without taking the vcore lock,
 260          * because only this task ever sets vc->runner to this
 261          * vcpu, and once it is set to this vcpu, only this task
 262          * ever sets it to NULL.
 263          */
 264         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 265                 kvmppc_core_end_stolen(vc);
 266
 267         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 268         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 269             vcpu->arch.busy_preempt != TB_NIL) {
 270                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 271                 vcpu->arch.busy_preempt = TB_NIL;
 272         }
 273         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 274 }
 275
 276 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 277 {
 278         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 279         unsigned long flags;
 280
 281         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 282                 kvmppc_core_start_stolen(vc);
 283
 284         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 285         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 286                 vcpu->arch.busy_preempt = mftb();
 287         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 288 }
 289
 290 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
 291 {
 292         /*
 293          * Check for illegal transactional state bit combination
 294          * and if we find it, force the TS field to a safe state.
 295          */
 296         if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
 297                 msr &= ~MSR_TS_MASK;
 298         vcpu->arch.shregs.msr = msr;
 299         kvmppc_end_cede(vcpu);
 300 }
 301
 302 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 303 {
 304         vcpu->arch.pvr = pvr;
 305 }
 306
 307 /* Dummy value used in computing PCR value below */
 308 #define PCR_ARCH_300    (PCR_ARCH_207 << 1)
 309
 310 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 311 {
 312         unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
 313         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 314
 315         /* We can (emulate) our own architecture version and anything older */
 316         if (cpu_has_feature(CPU_FTR_ARCH_300))
 317                 host_pcr_bit = PCR_ARCH_300;
 318         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
 319                 host_pcr_bit = PCR_ARCH_207;
 320         else if (cpu_has_feature(CPU_FTR_ARCH_206))
 321                 host_pcr_bit = PCR_ARCH_206;
 322         else
 323                 host_pcr_bit = PCR_ARCH_205;
 324
 325         /* Determine lowest PCR bit needed to run guest in given PVR level */
 326         guest_pcr_bit = host_pcr_bit;
 327         if (arch_compat) {
 328                 switch (arch_compat) {
 329                 case PVR_ARCH_205:
 330                         guest_pcr_bit = PCR_ARCH_205;
 331                         break;
 332                 case PVR_ARCH_206:
 333                 case PVR_ARCH_206p:
 334                         guest_pcr_bit = PCR_ARCH_206;
 335                         break;
 336                 case PVR_ARCH_207:
 337                         guest_pcr_bit = PCR_ARCH_207;
 338                         break;
 339                 case PVR_ARCH_300:
 340                         guest_pcr_bit = PCR_ARCH_300;
 341                         break;
 342                 default:
 343                         return -EINVAL;
 344                 }
 345         }
 346
 347         /* Check requested PCR bits don't exceed our capabilities */
 348         if (guest_pcr_bit > host_pcr_bit)
 349                 return -EINVAL;
 350
 351         spin_lock(&vc->lock);
 352         vc->arch_compat = arch_compat;
 353         /* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
 354         vc->pcr = host_pcr_bit - guest_pcr_bit;
 355         spin_unlock(&vc->lock);
 356
 357         return 0;
 358 }
 359
 360 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 361 {
 362         int r;
 363
 364         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 365         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 366                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
 367         for (r = 0; r < 16; ++r)
 368                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 369                        r, kvmppc_get_gpr(vcpu, r),
 370                        r+16, kvmppc_get_gpr(vcpu, r+16));
 371         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 372                vcpu->arch.ctr, vcpu->arch.lr);
 373         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 374                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 375         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 376                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 377         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 378                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 379         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 380                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 381         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 382         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 383                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 384         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 385         for (r = 0; r < vcpu->arch.slb_max; ++r)
 386                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 387                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 388         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 389                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 390                vcpu->arch.last_inst);
 391 }
 392
 393 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 394 {
 395         struct kvm_vcpu *ret;
 396
 397         mutex_lock(&kvm->lock);
 398         ret = kvm_get_vcpu_by_id(kvm, id);
 399         mutex_unlock(&kvm->lock);
 400         return ret;
 401 }
 402
 403 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 404 {
 405         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 406         vpa->yield_count = cpu_to_be32(1);
 407 }
 408
 409 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 410                    unsigned long addr, unsigned long len)
 411 {
 412         /* check address is cacheline aligned */
 413         if (addr & (L1_CACHE_BYTES - 1))
 414                 return -EINVAL;
 415         spin_lock(&vcpu->arch.vpa_update_lock);
 416         if (v->next_gpa != addr || v->len != len) {
 417                 v->next_gpa = addr;
 418                 v->len = addr ? len : 0;
 419                 v->update_pending = 1;
 420         }
 421         spin_unlock(&vcpu->arch.vpa_update_lock);
 422         return 0;
 423 }
 424
 425 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 426 struct reg_vpa {
 427         u32 dummy;
 428         union {
 429                 __be16 hword;
 430                 __be32 word;
 431         } length;
 432 };
 433
 434 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 435 {
 436         if (vpap->update_pending)
 437                 return vpap->next_gpa != 0;
 438         return vpap->pinned_addr != NULL;
 439 }
 440
 441 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 442                                        unsigned long flags,
 443                                        unsigned long vcpuid, unsigned long vpa)
 444 {
 445         struct kvm *kvm = vcpu->kvm;
 446         unsigned long len, nb;
 447         void *va;
 448         struct kvm_vcpu *tvcpu;
 449         int err;
 450         int subfunc;
 451         struct kvmppc_vpa *vpap;
 452
 453         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 454         if (!tvcpu)
 455                 return H_PARAMETER;
 456
 457         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 458         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 459             subfunc == H_VPA_REG_SLB) {
 460                 /* Registering new area - address must be cache-line aligned */
 461                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 462                         return H_PARAMETER;
 463
 464                 /* convert logical addr to kernel addr and read length */
 465                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 466                 if (va == NULL)
 467                         return H_PARAMETER;
 468                 if (subfunc == H_VPA_REG_VPA)
 469                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 470                 else
 471                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 472                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 473
 474                 /* Check length */
 475                 if (len > nb || len < sizeof(struct reg_vpa))
 476                         return H_PARAMETER;
 477         } else {
 478                 vpa = 0;
 479                 len = 0;
 480         }
 481
 482         err = H_PARAMETER;
 483         vpap = NULL;
 484         spin_lock(&tvcpu->arch.vpa_update_lock);
 485
 486         switch (subfunc) {
 487         case H_VPA_REG_VPA:             /* register VPA */
 488                 if (len < sizeof(struct lppaca))
 489                         break;
 490                 vpap = &tvcpu->arch.vpa;
 491                 err = 0;
 492                 break;
 493
 494         case H_VPA_REG_DTL:             /* register DTL */
 495                 if (len < sizeof(struct dtl_entry))
 496                         break;
 497                 len -= len % sizeof(struct dtl_entry);
 498
 499                 /* Check that they have previously registered a VPA */
 500                 err = H_RESOURCE;
 501                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 502                         break;
 503
 504                 vpap = &tvcpu->arch.dtl;
 505                 err = 0;
 506                 break;
 507
 508         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 509                 /* Check that they have previously registered a VPA */
 510                 err = H_RESOURCE;
 511                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 512                         break;
 513
 514                 vpap = &tvcpu->arch.slb_shadow;
 515                 err = 0;
 516                 break;
 517
 518         case H_VPA_DEREG_VPA:           /* deregister VPA */
 519                 /* Check they don't still have a DTL or SLB buf registered */
 520                 err = H_RESOURCE;
 521                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 522                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 523                         break;
 524
 525                 vpap = &tvcpu->arch.vpa;
 526                 err = 0;
 527                 break;
 528
 529         case H_VPA_DEREG_DTL:           /* deregister DTL */
 530                 vpap = &tvcpu->arch.dtl;
 531                 err = 0;
 532                 break;
 533
 534         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 535                 vpap = &tvcpu->arch.slb_shadow;
 536                 err = 0;
 537                 break;
 538         }
 539
 540         if (vpap) {
 541                 vpap->next_gpa = vpa;
 542                 vpap->len = len;
 543                 vpap->update_pending = 1;
 544         }
 545
 546         spin_unlock(&tvcpu->arch.vpa_update_lock);
 547
 548         return err;
 549 }
 550
 551 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 552 {
 553         struct kvm *kvm = vcpu->kvm;
 554         void *va;
 555         unsigned long nb;
 556         unsigned long gpa;
 557
 558         /*
 559          * We need to pin the page pointed to by vpap->next_gpa,
 560          * but we can't call kvmppc_pin_guest_page under the lock
 561          * as it does get_user_pages() and down_read().  So we
 562          * have to drop the lock, pin the page, then get the lock
 563          * again and check that a new area didn't get registered
 564          * in the meantime.
 565          */
 566         for (;;) {
 567                 gpa = vpap->next_gpa;
 568                 spin_unlock(&vcpu->arch.vpa_update_lock);
 569                 va = NULL;
 570                 nb = 0;
 571                 if (gpa)
 572                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 573                 spin_lock(&vcpu->arch.vpa_update_lock);
 574                 if (gpa == vpap->next_gpa)
 575                         break;
 576                 /* sigh... unpin that one and try again */
 577                 if (va)
 578                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 579         }
 580
 581         vpap->update_pending = 0;
 582         if (va && nb < vpap->len) {
 583                 /*
 584                  * If it's now too short, it must be that userspace
 585                  * has changed the mappings underlying guest memory,
 586                  * so unregister the region.
 587                  */
 588                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 589                 va = NULL;
 590         }
 591         if (vpap->pinned_addr)
 592                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 593                                         vpap->dirty);
 594         vpap->gpa = gpa;
 595         vpap->pinned_addr = va;
 596         vpap->dirty = false;
 597         if (va)
 598                 vpap->pinned_end = va + vpap->len;
 599 }
 600
 601 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 602 {
 603         if (!(vcpu->arch.vpa.update_pending ||
 604               vcpu->arch.slb_shadow.update_pending ||
 605               vcpu->arch.dtl.update_pending))
 606                 return;
 607
 608         spin_lock(&vcpu->arch.vpa_update_lock);
 609         if (vcpu->arch.vpa.update_pending) {
 610                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 611                 if (vcpu->arch.vpa.pinned_addr)
 612                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 613         }
 614         if (vcpu->arch.dtl.update_pending) {
 615                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 616                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 617                 vcpu->arch.dtl_index = 0;
 618         }
 619         if (vcpu->arch.slb_shadow.update_pending)
 620                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 621         spin_unlock(&vcpu->arch.vpa_update_lock);
 622 }
 623
 624 /*
 625  * Return the accumulated stolen time for the vcore up until `now'.
 626  * The caller should hold the vcore lock.
 627  */
 628 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 629 {
 630         u64 p;
 631         unsigned long flags;
 632
 633         spin_lock_irqsave(&vc->stoltb_lock, flags);
 634         p = vc->stolen_tb;
 635         if (vc->vcore_state != VCORE_INACTIVE &&
 636             vc->preempt_tb != TB_NIL)
 637                 p += now - vc->preempt_tb;
 638         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 639         return p;
 640 }
 641
 642 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 643                                     struct kvmppc_vcore *vc)
 644 {
 645         struct dtl_entry *dt;
 646         struct lppaca *vpa;
 647         unsigned long stolen;
 648         unsigned long core_stolen;
 649         u64 now;
 650         unsigned long flags;
 651
 652         dt = vcpu->arch.dtl_ptr;
 653         vpa = vcpu->arch.vpa.pinned_addr;
 654         now = mftb();
 655         core_stolen = vcore_stolen_time(vc, now);
 656         stolen = core_stolen - vcpu->arch.stolen_logged;
 657         vcpu->arch.stolen_logged = core_stolen;
 658         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 659         stolen += vcpu->arch.busy_stolen;
 660         vcpu->arch.busy_stolen = 0;
 661         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 662         if (!dt || !vpa)
 663                 return;
 664         memset(dt, 0, sizeof(struct dtl_entry));
 665         dt->dispatch_reason = 7;
 666         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 667         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 668         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 669         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 670         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 671         ++dt;
 672         if (dt == vcpu->arch.dtl.pinned_end)
 673                 dt = vcpu->arch.dtl.pinned_addr;
 674         vcpu->arch.dtl_ptr = dt;
 675         /* order writing *dt vs. writing vpa->dtl_idx */
 676         smp_wmb();
 677         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 678         vcpu->arch.dtl.dirty = true;
 679 }
 680
 681 /* See if there is a doorbell interrupt pending for a vcpu */
 682 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
 683 {
 684         int thr;
 685         struct kvmppc_vcore *vc;
 686
 687         if (vcpu->arch.doorbell_request)
 688                 return true;
 689         /*
 690          * Ensure that the read of vcore->dpdes comes after the read
 691          * of vcpu->doorbell_request.  This barrier matches the
 692          * lwsync in book3s_hv_rmhandlers.S just before the
 693          * fast_guest_return label.
 694          */
 695         smp_rmb();
 696         vc = vcpu->arch.vcore;
 697         thr = vcpu->vcpu_id - vc->first_vcpuid;
 698         return !!(vc->dpdes & (1 << thr));
 699 }
 700
 701 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 702 {
 703         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 704                 return true;
 705         if ((!vcpu->arch.vcore->arch_compat) &&
 706             cpu_has_feature(CPU_FTR_ARCH_207S))
 707                 return true;
 708         return false;
 709 }
 710
 711 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 712                              unsigned long resource, unsigned long value1,
 713                              unsigned long value2)
 714 {
 715         switch (resource) {
 716         case H_SET_MODE_RESOURCE_SET_CIABR:
 717                 if (!kvmppc_power8_compatible(vcpu))
 718                         return H_P2;
 719                 if (value2)
 720                         return H_P4;
 721                 if (mflags)
 722                         return H_UNSUPPORTED_FLAG_START;
 723                 /* Guests can't breakpoint the hypervisor */
 724                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 725                         return H_P3;
 726                 vcpu->arch.ciabr  = value1;
 727                 return H_SUCCESS;
 728         case H_SET_MODE_RESOURCE_SET_DAWR:
 729                 if (!kvmppc_power8_compatible(vcpu))
 730                         return H_P2;
 731                 if (mflags)
 732                         return H_UNSUPPORTED_FLAG_START;
 733                 if (value2 & DABRX_HYP)
 734                         return H_P4;
 735                 vcpu->arch.dawr  = value1;
 736                 vcpu->arch.dawrx = value2;
 737                 return H_SUCCESS;
 738         default:
 739                 return H_TOO_HARD;
 740         }
 741 }
 742
 743 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 744 {
 745         struct kvmppc_vcore *vcore = target->arch.vcore;
 746
 747         /*
 748          * We expect to have been called by the real mode handler
 749          * (kvmppc_rm_h_confer()) which would have directly returned
 750          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 751          * have useful work to do and should not confer) so we don't
 752          * recheck that here.
 753          */
 754
 755         spin_lock(&vcore->lock);
 756         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 757             vcore->vcore_state != VCORE_INACTIVE &&
 758             vcore->runner)
 759                 target = vcore->runner;
 760         spin_unlock(&vcore->lock);
 761
 762         return kvm_vcpu_yield_to(target);
 763 }
 764
 765 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 766 {
 767         int yield_count = 0;
 768         struct lppaca *lppaca;
 769
 770         spin_lock(&vcpu->arch.vpa_update_lock);
 771         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 772         if (lppaca)
 773                 yield_count = be32_to_cpu(lppaca->yield_count);
 774         spin_unlock(&vcpu->arch.vpa_update_lock);
 775         return yield_count;
 776 }
 777
 778 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 779 {
 780         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 781         unsigned long target, ret = H_SUCCESS;
 782         int yield_count;
 783         struct kvm_vcpu *tvcpu;
 784         int idx, rc;
 785
 786         if (req <= MAX_HCALL_OPCODE &&
 787             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 788                 return RESUME_HOST;
 789
 790         switch (req) {
 791         case H_CEDE:
 792                 break;
 793         case H_PROD:
 794                 target = kvmppc_get_gpr(vcpu, 4);
 795                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 796                 if (!tvcpu) {
 797                         ret = H_PARAMETER;
 798                         break;
 799                 }
 800                 tvcpu->arch.prodded = 1;
 801                 smp_mb();
 802                 if (tvcpu->arch.ceded)
 803                         kvmppc_fast_vcpu_kick_hv(tvcpu);
 804                 break;
 805         case H_CONFER:
 806                 target = kvmppc_get_gpr(vcpu, 4);
 807                 if (target == -1)
 808                         break;
 809                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 810                 if (!tvcpu) {
 811                         ret = H_PARAMETER;
 812                         break;
 813                 }
 814                 yield_count = kvmppc_get_gpr(vcpu, 5);
 815                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 816                         break;
 817                 kvm_arch_vcpu_yield_to(tvcpu);
 818                 break;
 819         case H_REGISTER_VPA:
 820                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 821                                         kvmppc_get_gpr(vcpu, 5),
 822                                         kvmppc_get_gpr(vcpu, 6));
 823                 break;
 824         case H_RTAS:
 825                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 826                         return RESUME_HOST;
 827
 828                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 829                 rc = kvmppc_rtas_hcall(vcpu);
 830                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 831
 832                 if (rc == -ENOENT)
 833                         return RESUME_HOST;
 834                 else if (rc == 0)
 835                         break;
 836
 837                 /* Send the error out to userspace via KVM_RUN */
 838                 return rc;
 839         case H_LOGICAL_CI_LOAD:
 840                 ret = kvmppc_h_logical_ci_load(vcpu);
 841                 if (ret == H_TOO_HARD)
 842                         return RESUME_HOST;
 843                 break;
 844         case H_LOGICAL_CI_STORE:
 845                 ret = kvmppc_h_logical_ci_store(vcpu);
 846                 if (ret == H_TOO_HARD)
 847                         return RESUME_HOST;
 848                 break;
 849         case H_SET_MODE:
 850                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 851                                         kvmppc_get_gpr(vcpu, 5),
 852                                         kvmppc_get_gpr(vcpu, 6),
 853                                         kvmppc_get_gpr(vcpu, 7));
 854                 if (ret == H_TOO_HARD)
 855                         return RESUME_HOST;
 856                 break;
 857         case H_XIRR:
 858         case H_CPPR:
 859         case H_EOI:
 860         case H_IPI:
 861         case H_IPOLL:
 862         case H_XIRR_X:
 863                 if (kvmppc_xics_enabled(vcpu)) {
 864                         if (xive_enabled()) {
 865                                 ret = H_NOT_AVAILABLE;
 866                                 return RESUME_GUEST;
 867                         }
 868                         ret = kvmppc_xics_hcall(vcpu, req);
 869                         break;
 870                 }
 871                 return RESUME_HOST;
 872         case H_PUT_TCE:
 873                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 874                                                 kvmppc_get_gpr(vcpu, 5),
 875                                                 kvmppc_get_gpr(vcpu, 6));
 876                 if (ret == H_TOO_HARD)
 877                         return RESUME_HOST;
 878                 break;
 879         case H_PUT_TCE_INDIRECT:
 880                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
 881                                                 kvmppc_get_gpr(vcpu, 5),
 882                                                 kvmppc_get_gpr(vcpu, 6),
 883                                                 kvmppc_get_gpr(vcpu, 7));
 884                 if (ret == H_TOO_HARD)
 885                         return RESUME_HOST;
 886                 break;
 887         case H_STUFF_TCE:
 888                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 889                                                 kvmppc_get_gpr(vcpu, 5),
 890                                                 kvmppc_get_gpr(vcpu, 6),
 891                                                 kvmppc_get_gpr(vcpu, 7));
 892                 if (ret == H_TOO_HARD)
 893                         return RESUME_HOST;
 894                 break;
 895         default:
 896                 return RESUME_HOST;
 897         }
 898         kvmppc_set_gpr(vcpu, 3, ret);
 899         vcpu->arch.hcall_needed = 0;
 900         return RESUME_GUEST;
 901 }
 902
 903 static int kvmppc_hcall_impl_hv(unsigned long cmd)
 904 {
 905         switch (cmd) {
 906         case H_CEDE:
 907         case H_PROD:
 908         case H_CONFER:
 909         case H_REGISTER_VPA:
 910         case H_SET_MODE:
 911         case H_LOGICAL_CI_LOAD:
 912         case H_LOGICAL_CI_STORE:
 913 #ifdef CONFIG_KVM_XICS
 914         case H_XIRR:
 915         case H_CPPR:
 916         case H_EOI:
 917         case H_IPI:
 918         case H_IPOLL:
 919         case H_XIRR_X:
 920 #endif
 921                 return 1;
 922         }
 923
 924         /* See if it's in the real-mode table */
 925         return kvmppc_hcall_impl_hv_realmode(cmd);
 926 }
 927
 928 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
 929                                         struct kvm_vcpu *vcpu)
 930 {
 931         u32 last_inst;
 932
 933         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
 934                                         EMULATE_DONE) {
 935                 /*
 936                  * Fetch failed, so return to guest and
 937                  * try executing it again.
 938                  */
 939                 return RESUME_GUEST;
 940         }
 941
 942         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
 943                 run->exit_reason = KVM_EXIT_DEBUG;
 944                 run->debug.arch.address = kvmppc_get_pc(vcpu);
 945                 return RESUME_HOST;
 946         } else {
 947                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 948                 return RESUME_GUEST;
 949         }
 950 }
 951
 952 static void do_nothing(void *x)
 953 {
 954 }
 955
 956 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
 957 {
 958         int thr, cpu, pcpu, nthreads;
 959         struct kvm_vcpu *v;
 960         unsigned long dpdes;
 961
 962         nthreads = vcpu->kvm->arch.emul_smt_mode;
 963         dpdes = 0;
 964         cpu = vcpu->vcpu_id & ~(nthreads - 1);
 965         for (thr = 0; thr < nthreads; ++thr, ++cpu) {
 966                 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
 967                 if (!v)
 968                         continue;
 969                 /*
 970                  * If the vcpu is currently running on a physical cpu thread,
 971                  * interrupt it in order to pull it out of the guest briefly,
 972                  * which will update its vcore->dpdes value.
 973                  */
 974                 pcpu = READ_ONCE(v->cpu);
 975                 if (pcpu >= 0)
 976                         smp_call_function_single(pcpu, do_nothing, NULL, 1);
 977                 if (kvmppc_doorbell_pending(v))
 978                         dpdes |= 1 << thr;
 979         }
 980         return dpdes;
 981 }
 982
 983 /*
 984  * On POWER9, emulate doorbell-related instructions in order to
 985  * give the guest the illusion of running on a multi-threaded core.
 986  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
 987  * and mfspr DPDES.
 988  */
 989 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
 990 {
 991         u32 inst, rb, thr;
 992         unsigned long arg;
 993         struct kvm *kvm = vcpu->kvm;
 994         struct kvm_vcpu *tvcpu;
 995
 996         if (!cpu_has_feature(CPU_FTR_ARCH_300))
 997                 return EMULATE_FAIL;
 998         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
 999                 return RESUME_GUEST;
1000         if (get_op(inst) != 31)
1001                 return EMULATE_FAIL;
1002         rb = get_rb(inst);
1003         thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1004         switch (get_xop(inst)) {
1005         case OP_31_XOP_MSGSNDP:
1006                 arg = kvmppc_get_gpr(vcpu, rb);
1007                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1008                         break;
1009                 arg &= 0x3f;
1010                 if (arg >= kvm->arch.emul_smt_mode)
1011                         break;
1012                 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1013                 if (!tvcpu)
1014                         break;
1015                 if (!tvcpu->arch.doorbell_request) {
1016                         tvcpu->arch.doorbell_request = 1;
1017                         kvmppc_fast_vcpu_kick_hv(tvcpu);
1018                 }
1019                 break;
1020         case OP_31_XOP_MSGCLRP:
1021                 arg = kvmppc_get_gpr(vcpu, rb);
1022                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1023                         break;
1024                 vcpu->arch.vcore->dpdes = 0;
1025                 vcpu->arch.doorbell_request = 0;
1026                 break;
1027         case OP_31_XOP_MFSPR:
1028                 switch (get_sprn(inst)) {
1029                 case SPRN_TIR:
1030                         arg = thr;
1031                         break;
1032                 case SPRN_DPDES:
1033                         arg = kvmppc_read_dpdes(vcpu);
1034                         break;
1035                 default:
1036                         return EMULATE_FAIL;
1037                 }
1038                 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1039                 break;
1040         default:
1041                 return EMULATE_FAIL;
1042         }
1043         kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1044         return RESUME_GUEST;
1045 }
1046
1047 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1048                                  struct task_struct *tsk)
1049 {
1050         int r = RESUME_HOST;
1051
1052         vcpu->stat.sum_exits++;
1053
1054         /*
1055          * This can happen if an interrupt occurs in the last stages
1056          * of guest entry or the first stages of guest exit (i.e. after
1057          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1058          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1059          * That can happen due to a bug, or due to a machine check
1060          * occurring at just the wrong time.
1061          */
1062         if (vcpu->arch.shregs.msr & MSR_HV) {
1063                 printk(KERN_EMERG "KVM trap in HV mode!\n");
1064                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1065                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1066                         vcpu->arch.shregs.msr);
1067                 kvmppc_dump_regs(vcpu);
1068                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1069                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1070                 return RESUME_HOST;
1071         }
1072         run->exit_reason = KVM_EXIT_UNKNOWN;
1073         run->ready_for_interrupt_injection = 1;
1074         switch (vcpu->arch.trap) {
1075         /* We're good on these - the host merely wanted to get our attention */
1076         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1077                 vcpu->stat.dec_exits++;
1078                 r = RESUME_GUEST;
1079                 break;
1080         case BOOK3S_INTERRUPT_EXTERNAL:
1081         case BOOK3S_INTERRUPT_H_DOORBELL:
1082         case BOOK3S_INTERRUPT_H_VIRT:
1083                 vcpu->stat.ext_intr_exits++;
1084                 r = RESUME_GUEST;
1085                 break;
1086         /* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
1087         case BOOK3S_INTERRUPT_HMI:
1088         case BOOK3S_INTERRUPT_PERFMON:
1089                 r = RESUME_GUEST;
1090                 break;
1091         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1092                 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1093                 run->exit_reason = KVM_EXIT_NMI;
1094                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1095                 /* Clear out the old NMI status from run->flags */
1096                 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1097                 /* Now set the NMI status */
1098                 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1099                         run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1100                 else
1101                         run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1102
1103                 r = RESUME_HOST;
1104                 /* Print the MCE event to host console. */
1105                 machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1106                 break;
1107         case BOOK3S_INTERRUPT_PROGRAM:
1108         {
1109                 ulong flags;
1110                 /*
1111                  * Normally program interrupts are delivered directly
1112                  * to the guest by the hardware, but we can get here
1113                  * as a result of a hypervisor emulation interrupt
1114                  * (e40) getting turned into a 700 by BML RTAS.
1115                  */
1116                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1117                 kvmppc_core_queue_program(vcpu, flags);
1118                 r = RESUME_GUEST;
1119                 break;
1120         }
1121         case BOOK3S_INTERRUPT_SYSCALL:
1122         {
1123                 /* hcall - punt to userspace */
1124                 int i;
1125
1126                 /* hypercall with MSR_PR has already been handled in rmode,
1127                  * and never reaches here.
1128                  */
1129
1130                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1131                 for (i = 0; i < 9; ++i)
1132                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1133                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1134                 vcpu->arch.hcall_needed = 1;
1135                 r = RESUME_HOST;
1136                 break;
1137         }
1138         /*
1139          * We get these next two if the guest accesses a page which it thinks
1140          * it has mapped but which is not actually present, either because
1141          * it is for an emulated I/O device or because the corresonding
1142          * host page has been paged out.  Any other HDSI/HISI interrupts
1143          * have been handled already.
1144          */
1145         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1146                 r = RESUME_PAGE_FAULT;
1147                 break;
1148         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1149                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1150                 vcpu->arch.fault_dsisr = 0;
1151                 r = RESUME_PAGE_FAULT;
1152                 break;
1153         /*
1154          * This occurs if the guest executes an illegal instruction.
1155          * If the guest debug is disabled, generate a program interrupt
1156          * to the guest. If guest debug is enabled, we need to check
1157          * whether the instruction is a software breakpoint instruction.
1158          * Accordingly return to Guest or Host.
1159          */
1160         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1161                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1162                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1163                                 swab32(vcpu->arch.emul_inst) :
1164                                 vcpu->arch.emul_inst;
1165                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1166                         r = kvmppc_emulate_debug_inst(run, vcpu);
1167                 } else {
1168                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1169                         r = RESUME_GUEST;
1170                 }
1171                 break;
1172         /*
1173          * This occurs if the guest (kernel or userspace), does something that
1174          * is prohibited by HFSCR.
1175          * On POWER9, this could be a doorbell instruction that we need
1176          * to emulate.
1177          * Otherwise, we just generate a program interrupt to the guest.
1178          */
1179         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1180                 r = EMULATE_FAIL;
1181                 if ((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG)
1182                         r = kvmppc_emulate_doorbell_instr(vcpu);
1183                 if (r == EMULATE_FAIL) {
1184                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1185                         r = RESUME_GUEST;
1186                 }
1187                 break;
1188         case BOOK3S_INTERRUPT_HV_RM_HARD:
1189                 r = RESUME_PASSTHROUGH;
1190                 break;
1191         default:
1192                 kvmppc_dump_regs(vcpu);
1193                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1194                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1195                         vcpu->arch.shregs.msr);
1196                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1197                 r = RESUME_HOST;
1198                 break;
1199         }
1200
1201         return r;
1202 }
1203
1204 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1205                                             struct kvm_sregs *sregs)
1206 {
1207         int i;
1208
1209         memset(sregs, 0, sizeof(struct kvm_sregs));
1210         sregs->pvr = vcpu->arch.pvr;
1211         for (i = 0; i < vcpu->arch.slb_max; i++) {
1212                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1213                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1214         }
1215
1216         return 0;
1217 }
1218
1219 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1220                                             struct kvm_sregs *sregs)
1221 {
1222         int i, j;
1223
1224         /* Only accept the same PVR as the host's, since we can't spoof it */
1225         if (sregs->pvr != vcpu->arch.pvr)
1226                 return -EINVAL;
1227
1228         j = 0;
1229         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1230                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1231                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1232                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1233                         ++j;
1234                 }
1235         }
1236         vcpu->arch.slb_max = j;
1237
1238         return 0;
1239 }
1240
1241 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1242                 bool preserve_top32)
1243 {
1244         struct kvm *kvm = vcpu->kvm;
1245         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1246         u64 mask;
1247
1248         mutex_lock(&kvm->lock);
1249         spin_lock(&vc->lock);
1250         /*
1251          * If ILE (interrupt little-endian) has changed, update the
1252          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1253          */
1254         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1255                 struct kvm_vcpu *vcpu;
1256                 int i;
1257
1258                 kvm_for_each_vcpu(i, vcpu, kvm) {
1259                         if (vcpu->arch.vcore != vc)
1260                                 continue;
1261                         if (new_lpcr & LPCR_ILE)
1262                                 vcpu->arch.intr_msr |= MSR_LE;
1263                         else
1264                                 vcpu->arch.intr_msr &= ~MSR_LE;
1265                 }
1266         }
1267
1268         /*
1269          * Userspace can only modify DPFD (default prefetch depth),
1270          * ILE (interrupt little-endian) and TC (translation control).
1271          * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1272          */
1273         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1274         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1275                 mask |= LPCR_AIL;
1276         /*
1277          * On POWER9, allow userspace to enable large decrementer for the
1278          * guest, whether or not the host has it enabled.
1279          */
1280         if (cpu_has_feature(CPU_FTR_ARCH_300))
1281                 mask |= LPCR_LD;
1282
1283         /* Broken 32-bit version of LPCR must not clear top bits */
1284         if (preserve_top32)
1285                 mask &= 0xFFFFFFFF;
1286         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1287         spin_unlock(&vc->lock);
1288         mutex_unlock(&kvm->lock);
1289 }
1290
1291 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1292                                  union kvmppc_one_reg *val)
1293 {
1294         int r = 0;
1295         long int i;
1296
1297         switch (id) {
1298         case KVM_REG_PPC_DEBUG_INST:
1299                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1300                 break;
1301         case KVM_REG_PPC_HIOR:
1302                 *val = get_reg_val(id, 0);
1303                 break;
1304         case KVM_REG_PPC_DABR:
1305                 *val = get_reg_val(id, vcpu->arch.dabr);
1306                 break;
1307         case KVM_REG_PPC_DABRX:
1308                 *val = get_reg_val(id, vcpu->arch.dabrx);
1309                 break;
1310         case KVM_REG_PPC_DSCR:
1311                 *val = get_reg_val(id, vcpu->arch.dscr);
1312                 break;
1313         case KVM_REG_PPC_PURR:
1314                 *val = get_reg_val(id, vcpu->arch.purr);
1315                 break;
1316         case KVM_REG_PPC_SPURR:
1317                 *val = get_reg_val(id, vcpu->arch.spurr);
1318                 break;
1319         case KVM_REG_PPC_AMR:
1320                 *val = get_reg_val(id, vcpu->arch.amr);
1321                 break;
1322         case KVM_REG_PPC_UAMOR:
1323                 *val = get_reg_val(id, vcpu->arch.uamor);
1324                 break;
1325         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1326                 i = id - KVM_REG_PPC_MMCR0;
1327                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1328                 break;
1329         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1330                 i = id - KVM_REG_PPC_PMC1;
1331                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1332                 break;
1333         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1334                 i = id - KVM_REG_PPC_SPMC1;
1335                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1336                 break;
1337         case KVM_REG_PPC_SIAR:
1338                 *val = get_reg_val(id, vcpu->arch.siar);
1339                 break;
1340         case KVM_REG_PPC_SDAR:
1341                 *val = get_reg_val(id, vcpu->arch.sdar);
1342                 break;
1343         case KVM_REG_PPC_SIER:
1344                 *val = get_reg_val(id, vcpu->arch.sier);
1345                 break;
1346         case KVM_REG_PPC_IAMR:
1347                 *val = get_reg_val(id, vcpu->arch.iamr);
1348                 break;
1349         case KVM_REG_PPC_PSPB:
1350                 *val = get_reg_val(id, vcpu->arch.pspb);
1351                 break;
1352         case KVM_REG_PPC_DPDES:
1353                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1354                 break;
1355         case KVM_REG_PPC_VTB:
1356                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1357                 break;
1358         case KVM_REG_PPC_DAWR:
1359                 *val = get_reg_val(id, vcpu->arch.dawr);
1360                 break;
1361         case KVM_REG_PPC_DAWRX:
1362                 *val = get_reg_val(id, vcpu->arch.dawrx);
1363                 break;
1364         case KVM_REG_PPC_CIABR:
1365                 *val = get_reg_val(id, vcpu->arch.ciabr);
1366                 break;
1367         case KVM_REG_PPC_CSIGR:
1368                 *val = get_reg_val(id, vcpu->arch.csigr);
1369                 break;
1370         case KVM_REG_PPC_TACR:
1371                 *val = get_reg_val(id, vcpu->arch.tacr);
1372                 break;
1373         case KVM_REG_PPC_TCSCR:
1374                 *val = get_reg_val(id, vcpu->arch.tcscr);
1375                 break;
1376         case KVM_REG_PPC_PID:
1377                 *val = get_reg_val(id, vcpu->arch.pid);
1378                 break;
1379         case KVM_REG_PPC_ACOP:
1380                 *val = get_reg_val(id, vcpu->arch.acop);
1381                 break;
1382         case KVM_REG_PPC_WORT:
1383                 *val = get_reg_val(id, vcpu->arch.wort);
1384                 break;
1385         case KVM_REG_PPC_TIDR:
1386                 *val = get_reg_val(id, vcpu->arch.tid);
1387                 break;
1388         case KVM_REG_PPC_PSSCR:
1389                 *val = get_reg_val(id, vcpu->arch.psscr);
1390                 break;
1391         case KVM_REG_PPC_VPA_ADDR:
1392                 spin_lock(&vcpu->arch.vpa_update_lock);
1393                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1394                 spin_unlock(&vcpu->arch.vpa_update_lock);
1395                 break;
1396         case KVM_REG_PPC_VPA_SLB:
1397                 spin_lock(&vcpu->arch.vpa_update_lock);
1398                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1399                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1400                 spin_unlock(&vcpu->arch.vpa_update_lock);
1401                 break;
1402         case KVM_REG_PPC_VPA_DTL:
1403                 spin_lock(&vcpu->arch.vpa_update_lock);
1404                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1405                 val->vpaval.length = vcpu->arch.dtl.len;
1406                 spin_unlock(&vcpu->arch.vpa_update_lock);
1407                 break;
1408         case KVM_REG_PPC_TB_OFFSET:
1409                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1410                 break;
1411         case KVM_REG_PPC_LPCR:
1412         case KVM_REG_PPC_LPCR_64:
1413                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1414                 break;
1415         case KVM_REG_PPC_PPR:
1416                 *val = get_reg_val(id, vcpu->arch.ppr);
1417                 break;
1418 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1419         case KVM_REG_PPC_TFHAR:
1420                 *val = get_reg_val(id, vcpu->arch.tfhar);
1421                 break;
1422         case KVM_REG_PPC_TFIAR:
1423                 *val = get_reg_val(id, vcpu->arch.tfiar);
1424                 break;
1425         case KVM_REG_PPC_TEXASR:
1426                 *val = get_reg_val(id, vcpu->arch.texasr);
1427                 break;
1428         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1429                 i = id - KVM_REG_PPC_TM_GPR0;
1430                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1431                 break;
1432         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1433         {
1434                 int j;
1435                 i = id - KVM_REG_PPC_TM_VSR0;
1436                 if (i < 32)
1437                         for (j = 0; j < TS_FPRWIDTH; j++)
1438                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1439                 else {
1440                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1441                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1442                         else
1443                                 r = -ENXIO;
1444                 }
1445                 break;
1446         }
1447         case KVM_REG_PPC_TM_CR:
1448                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1449                 break;
1450         case KVM_REG_PPC_TM_XER:
1451                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1452                 break;
1453         case KVM_REG_PPC_TM_LR:
1454                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1455                 break;
1456         case KVM_REG_PPC_TM_CTR:
1457                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1458                 break;
1459         case KVM_REG_PPC_TM_FPSCR:
1460                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1461                 break;
1462         case KVM_REG_PPC_TM_AMR:
1463                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1464                 break;
1465         case KVM_REG_PPC_TM_PPR:
1466                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1467                 break;
1468         case KVM_REG_PPC_TM_VRSAVE:
1469                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1470                 break;
1471         case KVM_REG_PPC_TM_VSCR:
1472                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1473                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1474                 else
1475                         r = -ENXIO;
1476                 break;
1477         case KVM_REG_PPC_TM_DSCR:
1478                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1479                 break;
1480         case KVM_REG_PPC_TM_TAR:
1481                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1482                 break;
1483 #endif
1484         case KVM_REG_PPC_ARCH_COMPAT:
1485                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1486                 break;
1487         default:
1488                 r = -EINVAL;
1489                 break;
1490         }
1491
1492         return r;
1493 }
1494
1495 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1496                                  union kvmppc_one_reg *val)
1497 {
1498         int r = 0;
1499         long int i;
1500         unsigned long addr, len;
1501
1502         switch (id) {
1503         case KVM_REG_PPC_HIOR:
1504                 /* Only allow this to be set to zero */
1505                 if (set_reg_val(id, *val))
1506                         r = -EINVAL;
1507                 break;
1508         case KVM_REG_PPC_DABR:
1509                 vcpu->arch.dabr = set_reg_val(id, *val);
1510                 break;
1511         case KVM_REG_PPC_DABRX:
1512                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1513                 break;
1514         case KVM_REG_PPC_DSCR:
1515                 vcpu->arch.dscr = set_reg_val(id, *val);
1516                 break;
1517         case KVM_REG_PPC_PURR:
1518                 vcpu->arch.purr = set_reg_val(id, *val);
1519                 break;
1520         case KVM_REG_PPC_SPURR:
1521                 vcpu->arch.spurr = set_reg_val(id, *val);
1522                 break;
1523         case KVM_REG_PPC_AMR:
1524                 vcpu->arch.amr = set_reg_val(id, *val);
1525                 break;
1526         case KVM_REG_PPC_UAMOR:
1527                 vcpu->arch.uamor = set_reg_val(id, *val);
1528                 break;
1529         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1530                 i = id - KVM_REG_PPC_MMCR0;
1531                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1532                 break;
1533         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1534                 i = id - KVM_REG_PPC_PMC1;
1535                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1536                 break;
1537         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1538                 i = id - KVM_REG_PPC_SPMC1;
1539                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1540                 break;
1541         case KVM_REG_PPC_SIAR:
1542                 vcpu->arch.siar = set_reg_val(id, *val);
1543                 break;
1544         case KVM_REG_PPC_SDAR:
1545                 vcpu->arch.sdar = set_reg_val(id, *val);
1546                 break;
1547         case KVM_REG_PPC_SIER:
1548                 vcpu->arch.sier = set_reg_val(id, *val);
1549                 break;
1550         case KVM_REG_PPC_IAMR:
1551                 vcpu->arch.iamr = set_reg_val(id, *val);
1552                 break;
1553         case KVM_REG_PPC_PSPB:
1554                 vcpu->arch.pspb = set_reg_val(id, *val);
1555                 break;
1556         case KVM_REG_PPC_DPDES:
1557                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1558                 break;
1559         case KVM_REG_PPC_VTB:
1560                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1561                 break;
1562         case KVM_REG_PPC_DAWR:
1563                 vcpu->arch.dawr = set_reg_val(id, *val);
1564                 break;
1565         case KVM_REG_PPC_DAWRX:
1566                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1567                 break;
1568         case KVM_REG_PPC_CIABR:
1569                 vcpu->arch.ciabr = set_reg_val(id, *val);
1570                 /* Don't allow setting breakpoints in hypervisor code */
1571                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1572                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1573                 break;
1574         case KVM_REG_PPC_CSIGR:
1575                 vcpu->arch.csigr = set_reg_val(id, *val);
1576                 break;
1577         case KVM_REG_PPC_TACR:
1578                 vcpu->arch.tacr = set_reg_val(id, *val);
1579                 break;
1580         case KVM_REG_PPC_TCSCR:
1581                 vcpu->arch.tcscr = set_reg_val(id, *val);
1582                 break;
1583         case KVM_REG_PPC_PID:
1584                 vcpu->arch.pid = set_reg_val(id, *val);
1585                 break;
1586         case KVM_REG_PPC_ACOP:
1587                 vcpu->arch.acop = set_reg_val(id, *val);
1588                 break;
1589         case KVM_REG_PPC_WORT:
1590                 vcpu->arch.wort = set_reg_val(id, *val);
1591                 break;
1592         case KVM_REG_PPC_TIDR:
1593                 vcpu->arch.tid = set_reg_val(id, *val);
1594                 break;
1595         case KVM_REG_PPC_PSSCR:
1596                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1597                 break;
1598         case KVM_REG_PPC_VPA_ADDR:
1599                 addr = set_reg_val(id, *val);
1600                 r = -EINVAL;
1601                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1602                               vcpu->arch.dtl.next_gpa))
1603                         break;
1604                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1605                 break;
1606         case KVM_REG_PPC_VPA_SLB:
1607                 addr = val->vpaval.addr;
1608                 len = val->vpaval.length;
1609                 r = -EINVAL;
1610                 if (addr && !vcpu->arch.vpa.next_gpa)
1611                         break;
1612                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1613                 break;
1614         case KVM_REG_PPC_VPA_DTL:
1615                 addr = val->vpaval.addr;
1616                 len = val->vpaval.length;
1617                 r = -EINVAL;
1618                 if (addr && (len < sizeof(struct dtl_entry) ||
1619                              !vcpu->arch.vpa.next_gpa))
1620                         break;
1621                 len -= len % sizeof(struct dtl_entry);
1622                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1623                 break;
1624         case KVM_REG_PPC_TB_OFFSET:
1625                 /*
1626                  * POWER9 DD1 has an erratum where writing TBU40 causes
1627                  * the timebase to lose ticks.  So we don't let the
1628                  * timebase offset be changed on P9 DD1.  (It is
1629                  * initialized to zero.)
1630                  */
1631                 if (cpu_has_feature(CPU_FTR_POWER9_DD1))
1632                         break;
1633                 /* round up to multiple of 2^24 */
1634                 vcpu->arch.vcore->tb_offset =
1635                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1636                 break;
1637         case KVM_REG_PPC_LPCR:
1638                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1639                 break;
1640         case KVM_REG_PPC_LPCR_64:
1641                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1642                 break;
1643         case KVM_REG_PPC_PPR:
1644                 vcpu->arch.ppr = set_reg_val(id, *val);
1645                 break;
1646 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1647         case KVM_REG_PPC_TFHAR:
1648                 vcpu->arch.tfhar = set_reg_val(id, *val);
1649                 break;
1650         case KVM_REG_PPC_TFIAR:
1651                 vcpu->arch.tfiar = set_reg_val(id, *val);
1652                 break;
1653         case KVM_REG_PPC_TEXASR:
1654                 vcpu->arch.texasr = set_reg_val(id, *val);
1655                 break;
1656         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1657                 i = id - KVM_REG_PPC_TM_GPR0;
1658                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1659                 break;
1660         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1661         {
1662                 int j;
1663                 i = id - KVM_REG_PPC_TM_VSR0;
1664                 if (i < 32)
1665                         for (j = 0; j < TS_FPRWIDTH; j++)
1666                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1667                 else
1668                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1669                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1670                         else
1671                                 r = -ENXIO;
1672                 break;
1673         }
1674         case KVM_REG_PPC_TM_CR:
1675                 vcpu->arch.cr_tm = set_reg_val(id, *val);
1676                 break;
1677         case KVM_REG_PPC_TM_XER:
1678                 vcpu->arch.xer_tm = set_reg_val(id, *val);
1679                 break;
1680         case KVM_REG_PPC_TM_LR:
1681                 vcpu->arch.lr_tm = set_reg_val(id, *val);
1682                 break;
1683         case KVM_REG_PPC_TM_CTR:
1684                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1685                 break;
1686         case KVM_REG_PPC_TM_FPSCR:
1687                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1688                 break;
1689         case KVM_REG_PPC_TM_AMR:
1690                 vcpu->arch.amr_tm = set_reg_val(id, *val);
1691                 break;
1692         case KVM_REG_PPC_TM_PPR:
1693                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1694                 break;
1695         case KVM_REG_PPC_TM_VRSAVE:
1696                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1697                 break;
1698         case KVM_REG_PPC_TM_VSCR:
1699                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1700                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1701                 else
1702                         r = - ENXIO;
1703                 break;
1704         case KVM_REG_PPC_TM_DSCR:
1705                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1706                 break;
1707         case KVM_REG_PPC_TM_TAR:
1708                 vcpu->arch.tar_tm = set_reg_val(id, *val);
1709                 break;
1710 #endif
1711         case KVM_REG_PPC_ARCH_COMPAT:
1712                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1713                 break;
1714         default:
1715                 r = -EINVAL;
1716                 break;
1717         }
1718
1719         return r;
1720 }
1721
1722 /*
1723  * On POWER9, threads are independent and can be in different partitions.
1724  * Therefore we consider each thread to be a subcore.
1725  * There is a restriction that all threads have to be in the same
1726  * MMU mode (radix or HPT), unfortunately, but since we only support
1727  * HPT guests on a HPT host so far, that isn't an impediment yet.
1728  */
1729 static int threads_per_vcore(void)
1730 {
1731         if (cpu_has_feature(CPU_FTR_ARCH_300))
1732                 return 1;
1733         return threads_per_subcore;
1734 }
1735
1736 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1737 {
1738         struct kvmppc_vcore *vcore;
1739
1740         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1741
1742         if (vcore == NULL)
1743                 return NULL;
1744
1745         spin_lock_init(&vcore->lock);
1746         spin_lock_init(&vcore->stoltb_lock);
1747         init_swait_queue_head(&vcore->wq);
1748         vcore->preempt_tb = TB_NIL;
1749         vcore->lpcr = kvm->arch.lpcr;
1750         vcore->first_vcpuid = core * kvm->arch.smt_mode;
1751         vcore->kvm = kvm;
1752         INIT_LIST_HEAD(&vcore->preempt_list);
1753
1754         return vcore;
1755 }
1756
1757 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1758 static struct debugfs_timings_element {
1759         const char *name;
1760         size_t offset;
1761 } timings[] = {
1762         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
1763         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
1764         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
1765         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
1766         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
1767 };
1768
1769 #define N_TIMINGS       (sizeof(timings) / sizeof(timings[0]))
1770
1771 struct debugfs_timings_state {
1772         struct kvm_vcpu *vcpu;
1773         unsigned int    buflen;
1774         char            buf[N_TIMINGS * 100];
1775 };
1776
1777 static int debugfs_timings_open(struct inode *inode, struct file *file)
1778 {
1779         struct kvm_vcpu *vcpu = inode->i_private;
1780         struct debugfs_timings_state *p;
1781
1782         p = kzalloc(sizeof(*p), GFP_KERNEL);
1783         if (!p)
1784                 return -ENOMEM;
1785
1786         kvm_get_kvm(vcpu->kvm);
1787         p->vcpu = vcpu;
1788         file->private_data = p;
1789
1790         return nonseekable_open(inode, file);
1791 }
1792
1793 static int debugfs_timings_release(struct inode *inode, struct file *file)
1794 {
1795         struct debugfs_timings_state *p = file->private_data;
1796
1797         kvm_put_kvm(p->vcpu->kvm);
1798         kfree(p);
1799         return 0;
1800 }
1801
1802 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1803                                     size_t len, loff_t *ppos)
1804 {
1805         struct debugfs_timings_state *p = file->private_data;
1806         struct kvm_vcpu *vcpu = p->vcpu;
1807         char *s, *buf_end;
1808         struct kvmhv_tb_accumulator tb;
1809         u64 count;
1810         loff_t pos;
1811         ssize_t n;
1812         int i, loops;
1813         bool ok;
1814
1815         if (!p->buflen) {
1816                 s = p->buf;
1817                 buf_end = s + sizeof(p->buf);
1818                 for (i = 0; i < N_TIMINGS; ++i) {
1819                         struct kvmhv_tb_accumulator *acc;
1820
1821                         acc = (struct kvmhv_tb_accumulator *)
1822                                 ((unsigned long)vcpu + timings[i].offset);
1823                         ok = false;
1824                         for (loops = 0; loops < 1000; ++loops) {
1825                                 count = acc->seqcount;
1826                                 if (!(count & 1)) {
1827                                         smp_rmb();
1828                                         tb = *acc;
1829                                         smp_rmb();
1830                                         if (count == acc->seqcount) {
1831                                                 ok = true;
1832                                                 break;
1833                                         }
1834                                 }
1835                                 udelay(1);
1836                         }
1837                         if (!ok)
1838                                 snprintf(s, buf_end - s, "%s: stuck\n",
1839                                         timings[i].name);
1840                         else
1841                                 snprintf(s, buf_end - s,
1842                                         "%s: %llu %llu %llu %llu\n",
1843                                         timings[i].name, count / 2,
1844                                         tb_to_ns(tb.tb_total),
1845                                         tb_to_ns(tb.tb_min),
1846                                         tb_to_ns(tb.tb_max));
1847                         s += strlen(s);
1848                 }
1849                 p->buflen = s - p->buf;
1850         }
1851
1852         pos = *ppos;
1853         if (pos >= p->buflen)
1854                 return 0;
1855         if (len > p->buflen - pos)
1856                 len = p->buflen - pos;
1857         n = copy_to_user(buf, p->buf + pos, len);
1858         if (n) {
1859                 if (n == len)
1860                         return -EFAULT;
1861                 len -= n;
1862         }
1863         *ppos = pos + len;
1864         return len;
1865 }
1866
1867 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1868                                      size_t len, loff_t *ppos)
1869 {
1870         return -EACCES;
1871 }
1872
1873 static const struct file_operations debugfs_timings_ops = {
1874         .owner   = THIS_MODULE,
1875         .open    = debugfs_timings_open,
1876         .release = debugfs_timings_release,
1877         .read    = debugfs_timings_read,
1878         .write   = debugfs_timings_write,
1879         .llseek  = generic_file_llseek,
1880 };
1881
1882 /* Create a debugfs directory for the vcpu */
1883 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1884 {
1885         char buf[16];
1886         struct kvm *kvm = vcpu->kvm;
1887
1888         snprintf(buf, sizeof(buf), "vcpu%u", id);
1889         if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1890                 return;
1891         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1892         if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1893                 return;
1894         vcpu->arch.debugfs_timings =
1895                 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1896                                     vcpu, &debugfs_timings_ops);
1897 }
1898
1899 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1900 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1901 {
1902 }
1903 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1904
1905 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1906                                                    unsigned int id)
1907 {
1908         struct kvm_vcpu *vcpu;
1909         int err;
1910         int core;
1911         struct kvmppc_vcore *vcore;
1912
1913         err = -ENOMEM;
1914         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1915         if (!vcpu)
1916                 goto out;
1917
1918         err = kvm_vcpu_init(vcpu, kvm, id);
1919         if (err)
1920                 goto free_vcpu;
1921
1922         vcpu->arch.shared = &vcpu->arch.shregs;
1923 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1924         /*
1925          * The shared struct is never shared on HV,
1926          * so we can always use host endianness
1927          */
1928 #ifdef __BIG_ENDIAN__
1929         vcpu->arch.shared_big_endian = true;
1930 #else
1931         vcpu->arch.shared_big_endian = false;
1932 #endif
1933 #endif
1934         vcpu->arch.mmcr[0] = MMCR0_FC;
1935         vcpu->arch.ctrl = CTRL_RUNLATCH;
1936         /* default to host PVR, since we can't spoof it */
1937         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1938         spin_lock_init(&vcpu->arch.vpa_update_lock);
1939         spin_lock_init(&vcpu->arch.tbacct_lock);
1940         vcpu->arch.busy_preempt = TB_NIL;
1941         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1942
1943         /*
1944          * Set the default HFSCR for the guest from the host value.
1945          * This value is only used on POWER9.
1946          * On POWER9 DD1, TM doesn't work, so we make sure to
1947          * prevent the guest from using it.
1948          * On POWER9, we want to virtualize the doorbell facility, so we
1949          * turn off the HFSCR bit, which causes those instructions to trap.
1950          */
1951         vcpu->arch.hfscr = mfspr(SPRN_HFSCR);
1952         if (!cpu_has_feature(CPU_FTR_TM))
1953                 vcpu->arch.hfscr &= ~HFSCR_TM;
1954         if (cpu_has_feature(CPU_FTR_ARCH_300))
1955                 vcpu->arch.hfscr &= ~HFSCR_MSGP;
1956
1957         kvmppc_mmu_book3s_hv_init(vcpu);
1958
1959         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1960
1961         init_waitqueue_head(&vcpu->arch.cpu_run);
1962
1963         mutex_lock(&kvm->lock);
1964         vcore = NULL;
1965         err = -EINVAL;
1966         core = id / kvm->arch.smt_mode;
1967         if (core < KVM_MAX_VCORES) {
1968                 vcore = kvm->arch.vcores[core];
1969                 if (!vcore) {
1970                         err = -ENOMEM;
1971                         vcore = kvmppc_vcore_create(kvm, core);
1972                         kvm->arch.vcores[core] = vcore;
1973                         kvm->arch.online_vcores++;
1974                 }
1975         }
1976         mutex_unlock(&kvm->lock);
1977
1978         if (!vcore)
1979                 goto free_vcpu;
1980
1981         spin_lock(&vcore->lock);
1982         ++vcore->num_threads;
1983         spin_unlock(&vcore->lock);
1984         vcpu->arch.vcore = vcore;
1985         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1986         vcpu->arch.thread_cpu = -1;
1987         vcpu->arch.prev_cpu = -1;
1988
1989         vcpu->arch.cpu_type = KVM_CPU_3S_64;
1990         kvmppc_sanity_check(vcpu);
1991
1992         debugfs_vcpu_init(vcpu, id);
1993
1994         return vcpu;
1995
1996 free_vcpu:
1997         kmem_cache_free(kvm_vcpu_cache, vcpu);
1998 out:
1999         return ERR_PTR(err);
2000 }
2001
2002 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2003                               unsigned long flags)
2004 {
2005         int err;
2006         int esmt = 0;
2007
2008         if (flags)
2009                 return -EINVAL;
2010         if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2011                 return -EINVAL;
2012         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2013                 /*
2014                  * On POWER8 (or POWER7), the threading mode is "strict",
2015                  * so we pack smt_mode vcpus per vcore.
2016                  */
2017                 if (smt_mode > threads_per_subcore)
2018                         return -EINVAL;
2019         } else {
2020                 /*
2021                  * On POWER9, the threading mode is "loose",
2022                  * so each vcpu gets its own vcore.
2023                  */
2024                 esmt = smt_mode;
2025                 smt_mode = 1;
2026         }
2027         mutex_lock(&kvm->lock);
2028         err = -EBUSY;
2029         if (!kvm->arch.online_vcores) {
2030                 kvm->arch.smt_mode = smt_mode;
2031                 kvm->arch.emul_smt_mode = esmt;
2032                 err = 0;
2033         }
2034         mutex_unlock(&kvm->lock);
2035
2036         return err;
2037 }
2038
2039 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2040 {
2041         if (vpa->pinned_addr)
2042                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2043                                         vpa->dirty);
2044 }
2045
2046 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2047 {
2048         spin_lock(&vcpu->arch.vpa_update_lock);
2049         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2050         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2051         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2052         spin_unlock(&vcpu->arch.vpa_update_lock);
2053         kvm_vcpu_uninit(vcpu);
2054         kmem_cache_free(kvm_vcpu_cache, vcpu);
2055 }
2056
2057 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2058 {
2059         /* Indicate we want to get back into the guest */
2060         return 1;
2061 }
2062
2063 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2064 {
2065         unsigned long dec_nsec, now;
2066
2067         now = get_tb();
2068         if (now > vcpu->arch.dec_expires) {
2069                 /* decrementer has already gone negative */
2070                 kvmppc_core_queue_dec(vcpu);
2071                 kvmppc_core_prepare_to_enter(vcpu);
2072                 return;
2073         }
2074         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
2075                    / tb_ticks_per_sec;
2076         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2077         vcpu->arch.timer_running = 1;
2078 }
2079
2080 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2081 {
2082         vcpu->arch.ceded = 0;
2083         if (vcpu->arch.timer_running) {
2084                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2085                 vcpu->arch.timer_running = 0;
2086         }
2087 }
2088
2089 extern int __kvmppc_vcore_entry(void);
2090
2091 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2092                                    struct kvm_vcpu *vcpu)
2093 {
2094         u64 now;
2095
2096         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2097                 return;
2098         spin_lock_irq(&vcpu->arch.tbacct_lock);
2099         now = mftb();
2100         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2101                 vcpu->arch.stolen_logged;
2102         vcpu->arch.busy_preempt = now;
2103         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2104         spin_unlock_irq(&vcpu->arch.tbacct_lock);
2105         --vc->n_runnable;
2106         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2107 }
2108
2109 static int kvmppc_grab_hwthread(int cpu)
2110 {
2111         struct paca_struct *tpaca;
2112         long timeout = 10000;
2113
2114         /*
2115          * ISA v3.0 idle routines do not set hwthread_state or test
2116          * hwthread_req, so they can not grab idle threads.
2117          */
2118         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2119                 WARN(1, "KVM: can not control sibling threads\n");
2120                 return -EBUSY;
2121         }
2122
2123         tpaca = &paca[cpu];
2124
2125         /* Ensure the thread won't go into the kernel if it wakes */
2126         tpaca->kvm_hstate.kvm_vcpu = NULL;
2127         tpaca->kvm_hstate.kvm_vcore = NULL;
2128         tpaca->kvm_hstate.napping = 0;
2129         smp_wmb();
2130         tpaca->kvm_hstate.hwthread_req = 1;
2131
2132         /*
2133          * If the thread is already executing in the kernel (e.g. handling
2134          * a stray interrupt), wait for it to get back to nap mode.
2135          * The smp_mb() is to ensure that our setting of hwthread_req
2136          * is visible before we look at hwthread_state, so if this
2137          * races with the code at system_reset_pSeries and the thread
2138          * misses our setting of hwthread_req, we are sure to see its
2139          * setting of hwthread_state, and vice versa.
2140          */
2141         smp_mb();
2142         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2143                 if (--timeout <= 0) {
2144                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
2145                         return -EBUSY;
2146                 }
2147                 udelay(1);
2148         }
2149         return 0;
2150 }
2151
2152 static void kvmppc_release_hwthread(int cpu)
2153 {
2154         struct paca_struct *tpaca;
2155
2156         tpaca = &paca[cpu];
2157         tpaca->kvm_hstate.kvm_vcpu = NULL;
2158         tpaca->kvm_hstate.kvm_vcore = NULL;
2159         tpaca->kvm_hstate.kvm_split_mode = NULL;
2160         if (!cpu_has_feature(CPU_FTR_ARCH_300))
2161                 tpaca->kvm_hstate.hwthread_req = 0;
2162
2163 }
2164
2165 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2166 {
2167         int i;
2168
2169         cpu = cpu_first_thread_sibling(cpu);
2170         cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2171         /*
2172          * Make sure setting of bit in need_tlb_flush precedes
2173          * testing of cpu_in_guest bits.  The matching barrier on
2174          * the other side is the first smp_mb() in kvmppc_run_core().
2175          */
2176         smp_mb();
2177         for (i = 0; i < threads_per_core; ++i)
2178                 if (cpumask_test_cpu(cpu + i, &kvm->arch.cpu_in_guest))
2179                         smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2180 }
2181
2182 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2183 {
2184         struct kvm *kvm = vcpu->kvm;
2185
2186         /*
2187          * With radix, the guest can do TLB invalidations itself,
2188          * and it could choose to use the local form (tlbiel) if
2189          * it is invalidating a translation that has only ever been
2190          * used on one vcpu.  However, that doesn't mean it has
2191          * only ever been used on one physical cpu, since vcpus
2192          * can move around between pcpus.  To cope with this, when
2193          * a vcpu moves from one pcpu to another, we need to tell
2194          * any vcpus running on the same core as this vcpu previously
2195          * ran to flush the TLB.  The TLB is shared between threads,
2196          * so we use a single bit in .need_tlb_flush for all 4 threads.
2197          */
2198         if (vcpu->arch.prev_cpu != pcpu) {
2199                 if (vcpu->arch.prev_cpu >= 0 &&
2200                     cpu_first_thread_sibling(vcpu->arch.prev_cpu) !=
2201                     cpu_first_thread_sibling(pcpu))
2202                         radix_flush_cpu(kvm, vcpu->arch.prev_cpu, vcpu);
2203                 vcpu->arch.prev_cpu = pcpu;
2204         }
2205 }
2206
2207 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2208 {
2209         int cpu;
2210         struct paca_struct *tpaca;
2211         struct kvm *kvm = vc->kvm;
2212
2213         cpu = vc->pcpu;
2214         if (vcpu) {
2215                 if (vcpu->arch.timer_running) {
2216                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2217                         vcpu->arch.timer_running = 0;
2218                 }
2219                 cpu += vcpu->arch.ptid;
2220                 vcpu->cpu = vc->pcpu;
2221                 vcpu->arch.thread_cpu = cpu;
2222                 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2223         }
2224         tpaca = &paca[cpu];
2225         tpaca->kvm_hstate.kvm_vcpu = vcpu;
2226         tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2227         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2228         smp_wmb();
2229         tpaca->kvm_hstate.kvm_vcore = vc;
2230         if (cpu != smp_processor_id())
2231                 kvmppc_ipi_thread(cpu);
2232 }
2233
2234 static void kvmppc_wait_for_nap(void)
2235 {
2236         int cpu = smp_processor_id();
2237         int i, loops;
2238         int n_threads = threads_per_vcore();
2239
2240         if (n_threads <= 1)
2241                 return;
2242         for (loops = 0; loops < 1000000; ++loops) {
2243                 /*
2244                  * Check if all threads are finished.
2245                  * We set the vcore pointer when starting a thread
2246                  * and the thread clears it when finished, so we look
2247                  * for any threads that still have a non-NULL vcore ptr.
2248                  */
2249                 for (i = 1; i < n_threads; ++i)
2250                         if (paca[cpu + i].kvm_hstate.kvm_vcore)
2251                                 break;
2252                 if (i == n_threads) {
2253                         HMT_medium();
2254                         return;
2255                 }
2256                 HMT_low();
2257         }
2258         HMT_medium();
2259         for (i = 1; i < n_threads; ++i)
2260                 if (paca[cpu + i].kvm_hstate.kvm_vcore)
2261                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2262 }
2263
2264 /*
2265  * Check that we are on thread 0 and that any other threads in
2266  * this core are off-line.  Then grab the threads so they can't
2267  * enter the kernel.
2268  */
2269 static int on_primary_thread(void)
2270 {
2271         int cpu = smp_processor_id();
2272         int thr;
2273
2274         /* Are we on a primary subcore? */
2275         if (cpu_thread_in_subcore(cpu))
2276                 return 0;
2277
2278         thr = 0;
2279         while (++thr < threads_per_subcore)
2280                 if (cpu_online(cpu + thr))
2281                         return 0;
2282
2283         /* Grab all hw threads so they can't go into the kernel */
2284         for (thr = 1; thr < threads_per_subcore; ++thr) {
2285                 if (kvmppc_grab_hwthread(cpu + thr)) {
2286                         /* Couldn't grab one; let the others go */
2287                         do {
2288                                 kvmppc_release_hwthread(cpu + thr);
2289                         } while (--thr > 0);
2290                         return 0;
2291                 }
2292         }
2293         return 1;
2294 }
2295
2296 /*
2297  * A list of virtual cores for each physical CPU.
2298  * These are vcores that could run but their runner VCPU tasks are
2299  * (or may be) preempted.
2300  */
2301 struct preempted_vcore_list {
2302         struct list_head        list;
2303         spinlock_t              lock;
2304 };
2305
2306 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2307
2308 static void init_vcore_lists(void)
2309 {
2310         int cpu;
2311
2312         for_each_possible_cpu(cpu) {
2313                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2314                 spin_lock_init(&lp->lock);
2315                 INIT_LIST_HEAD(&lp->list);
2316         }
2317 }
2318
2319 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2320 {
2321         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2322
2323         vc->vcore_state = VCORE_PREEMPT;
2324         vc->pcpu = smp_processor_id();
2325         if (vc->num_threads < threads_per_vcore()) {
2326                 spin_lock(&lp->lock);
2327                 list_add_tail(&vc->preempt_list, &lp->list);
2328                 spin_unlock(&lp->lock);
2329         }
2330
2331         /* Start accumulating stolen time */
2332         kvmppc_core_start_stolen(vc);
2333 }
2334
2335 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2336 {
2337         struct preempted_vcore_list *lp;
2338
2339         kvmppc_core_end_stolen(vc);
2340         if (!list_empty(&vc->preempt_list)) {
2341                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2342                 spin_lock(&lp->lock);
2343                 list_del_init(&vc->preempt_list);
2344                 spin_unlock(&lp->lock);
2345         }
2346         vc->vcore_state = VCORE_INACTIVE;
2347 }
2348
2349 /*
2350  * This stores information about the virtual cores currently
2351  * assigned to a physical core.
2352  */
2353 struct core_info {
2354         int             n_subcores;
2355         int             max_subcore_threads;
2356         int             total_threads;
2357         int             subcore_threads[MAX_SUBCORES];
2358         struct kvmppc_vcore *vc[MAX_SUBCORES];
2359 };
2360
2361 /*
2362  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2363  * respectively in 2-way micro-threading (split-core) mode.
2364  */
2365 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2366
2367 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2368 {
2369         memset(cip, 0, sizeof(*cip));
2370         cip->n_subcores = 1;
2371         cip->max_subcore_threads = vc->num_threads;
2372         cip->total_threads = vc->num_threads;
2373         cip->subcore_threads[0] = vc->num_threads;
2374         cip->vc[0] = vc;
2375 }
2376
2377 static bool subcore_config_ok(int n_subcores, int n_threads)
2378 {
2379         /* Can only dynamically split if unsplit to begin with */
2380         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2381                 return false;
2382         if (n_subcores > MAX_SUBCORES)
2383                 return false;
2384         if (n_subcores > 1) {
2385                 if (!(dynamic_mt_modes & 2))
2386                         n_subcores = 4;
2387                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2388                         return false;
2389         }
2390
2391         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2392 }
2393
2394 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2395 {
2396         vc->entry_exit_map = 0;
2397         vc->in_guest = 0;
2398         vc->napping_threads = 0;
2399         vc->conferring_threads = 0;
2400 }
2401
2402 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2403 {
2404         int n_threads = vc->num_threads;
2405         int sub;
2406
2407         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2408                 return false;
2409
2410         if (n_threads < cip->max_subcore_threads)
2411                 n_threads = cip->max_subcore_threads;
2412         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2413                 return false;
2414         cip->max_subcore_threads = n_threads;
2415
2416         sub = cip->n_subcores;
2417         ++cip->n_subcores;
2418         cip->total_threads += vc->num_threads;
2419         cip->subcore_threads[sub] = vc->num_threads;
2420         cip->vc[sub] = vc;
2421         init_vcore_to_run(vc);
2422         list_del_init(&vc->preempt_list);
2423
2424         return true;
2425 }
2426
2427 /*
2428  * Work out whether it is possible to piggyback the execution of
2429  * vcore *pvc onto the execution of the other vcores described in *cip.
2430  */
2431 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2432                           int target_threads)
2433 {
2434         if (cip->total_threads + pvc->num_threads > target_threads)
2435                 return false;
2436
2437         return can_dynamic_split(pvc, cip);
2438 }
2439
2440 static void prepare_threads(struct kvmppc_vcore *vc)
2441 {
2442         int i;
2443         struct kvm_vcpu *vcpu;
2444
2445         for_each_runnable_thread(i, vcpu, vc) {
2446                 if (signal_pending(vcpu->arch.run_task))
2447                         vcpu->arch.ret = -EINTR;
2448                 else if (vcpu->arch.vpa.update_pending ||
2449                          vcpu->arch.slb_shadow.update_pending ||
2450                          vcpu->arch.dtl.update_pending)
2451                         vcpu->arch.ret = RESUME_GUEST;
2452                 else
2453                         continue;
2454                 kvmppc_remove_runnable(vc, vcpu);
2455                 wake_up(&vcpu->arch.cpu_run);
2456         }
2457 }
2458
2459 static void collect_piggybacks(struct core_info *cip, int target_threads)
2460 {
2461         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2462         struct kvmppc_vcore *pvc, *vcnext;
2463
2464         spin_lock(&lp->lock);
2465         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2466                 if (!spin_trylock(&pvc->lock))
2467                         continue;
2468                 prepare_threads(pvc);
2469                 if (!pvc->n_runnable) {
2470                         list_del_init(&pvc->preempt_list);
2471                         if (pvc->runner == NULL) {
2472                                 pvc->vcore_state = VCORE_INACTIVE;
2473                                 kvmppc_core_end_stolen(pvc);
2474                         }
2475                         spin_unlock(&pvc->lock);
2476                         continue;
2477                 }
2478                 if (!can_piggyback(pvc, cip, target_threads)) {
2479                         spin_unlock(&pvc->lock);
2480                         continue;
2481                 }
2482                 kvmppc_core_end_stolen(pvc);
2483                 pvc->vcore_state = VCORE_PIGGYBACK;
2484                 if (cip->total_threads >= target_threads)
2485                         break;
2486         }
2487         spin_unlock(&lp->lock);
2488 }
2489
2490 static bool recheck_signals(struct core_info *cip)
2491 {
2492         int sub, i;
2493         struct kvm_vcpu *vcpu;
2494
2495         for (sub = 0; sub < cip->n_subcores; ++sub)
2496                 for_each_runnable_thread(i, vcpu, cip->vc[sub])
2497                         if (signal_pending(vcpu->arch.run_task))
2498                                 return true;
2499         return false;
2500 }
2501
2502 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2503 {
2504         int still_running = 0, i;
2505         u64 now;
2506         long ret;
2507         struct kvm_vcpu *vcpu;
2508
2509         spin_lock(&vc->lock);
2510         now = get_tb();
2511         for_each_runnable_thread(i, vcpu, vc) {
2512                 /* cancel pending dec exception if dec is positive */
2513                 if (now < vcpu->arch.dec_expires &&
2514                     kvmppc_core_pending_dec(vcpu))
2515                         kvmppc_core_dequeue_dec(vcpu);
2516
2517                 trace_kvm_guest_exit(vcpu);
2518
2519                 ret = RESUME_GUEST;
2520                 if (vcpu->arch.trap)
2521                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2522                                                     vcpu->arch.run_task);
2523
2524                 vcpu->arch.ret = ret;
2525                 vcpu->arch.trap = 0;
2526
2527                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2528                         if (vcpu->arch.pending_exceptions)
2529                                 kvmppc_core_prepare_to_enter(vcpu);
2530                         if (vcpu->arch.ceded)
2531                                 kvmppc_set_timer(vcpu);
2532                         else
2533                                 ++still_running;
2534                 } else {
2535                         kvmppc_remove_runnable(vc, vcpu);
2536                         wake_up(&vcpu->arch.cpu_run);
2537                 }
2538         }
2539         if (!is_master) {
2540                 if (still_running > 0) {
2541                         kvmppc_vcore_preempt(vc);
2542                 } else if (vc->runner) {
2543                         vc->vcore_state = VCORE_PREEMPT;
2544                         kvmppc_core_start_stolen(vc);
2545                 } else {
2546                         vc->vcore_state = VCORE_INACTIVE;
2547                 }
2548                 if (vc->n_runnable > 0 && vc->runner == NULL) {
2549                         /* make sure there's a candidate runner awake */
2550                         i = -1;
2551                         vcpu = next_runnable_thread(vc, &i);
2552                         wake_up(&vcpu->arch.cpu_run);
2553                 }
2554         }
2555         spin_unlock(&vc->lock);
2556 }
2557
2558 /*
2559  * Clear core from the list of active host cores as we are about to
2560  * enter the guest. Only do this if it is the primary thread of the
2561  * core (not if a subcore) that is entering the guest.
2562  */
2563 static inline int kvmppc_clear_host_core(unsigned int cpu)
2564 {
2565         int core;
2566
2567         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2568                 return 0;
2569         /*
2570          * Memory barrier can be omitted here as we will do a smp_wmb()
2571          * later in kvmppc_start_thread and we need ensure that state is
2572          * visible to other CPUs only after we enter guest.
2573          */
2574         core = cpu >> threads_shift;
2575         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2576         return 0;
2577 }
2578
2579 /*
2580  * Advertise this core as an active host core since we exited the guest
2581  * Only need to do this if it is the primary thread of the core that is
2582  * exiting.
2583  */
2584 static inline int kvmppc_set_host_core(unsigned int cpu)
2585 {
2586         int core;
2587
2588         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2589                 return 0;
2590
2591         /*
2592          * Memory barrier can be omitted here because we do a spin_unlock
2593          * immediately after this which provides the memory barrier.
2594          */
2595         core = cpu >> threads_shift;
2596         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2597         return 0;
2598 }
2599
2600 static void set_irq_happened(int trap)
2601 {
2602         switch (trap) {
2603         case BOOK3S_INTERRUPT_EXTERNAL:
2604                 local_paca->irq_happened |= PACA_IRQ_EE;
2605                 break;
2606         case BOOK3S_INTERRUPT_H_DOORBELL:
2607                 local_paca->irq_happened |= PACA_IRQ_DBELL;
2608                 break;
2609         case BOOK3S_INTERRUPT_HMI:
2610                 local_paca->irq_happened |= PACA_IRQ_HMI;
2611                 break;
2612         }
2613 }
2614
2615 /*
2616  * Run a set of guest threads on a physical core.
2617  * Called with vc->lock held.
2618  */
2619 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2620 {
2621         struct kvm_vcpu *vcpu;
2622         int i;
2623         int srcu_idx;
2624         struct core_info core_info;
2625         struct kvmppc_vcore *pvc;
2626         struct kvm_split_mode split_info, *sip;
2627         int split, subcore_size, active;
2628         int sub;
2629         bool thr0_done;
2630         unsigned long cmd_bit, stat_bit;
2631         int pcpu, thr;
2632         int target_threads;
2633         int controlled_threads;
2634         int trap;
2635
2636         /*
2637          * Remove from the list any threads that have a signal pending
2638          * or need a VPA update done
2639          */
2640         prepare_threads(vc);
2641
2642         /* if the runner is no longer runnable, let the caller pick a new one */
2643         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2644                 return;
2645
2646         /*
2647          * Initialize *vc.
2648          */
2649         init_vcore_to_run(vc);
2650         vc->preempt_tb = TB_NIL;
2651
2652         /*
2653          * Number of threads that we will be controlling: the same as
2654          * the number of threads per subcore, except on POWER9,
2655          * where it's 1 because the threads are (mostly) independent.
2656          */
2657         controlled_threads = threads_per_vcore();
2658
2659         /*
2660          * Make sure we are running on primary threads, and that secondary
2661          * threads are offline.  Also check if the number of threads in this
2662          * guest are greater than the current system threads per guest.
2663          */
2664         if ((controlled_threads > 1) &&
2665             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2666                 for_each_runnable_thread(i, vcpu, vc) {
2667                         vcpu->arch.ret = -EBUSY;
2668                         kvmppc_remove_runnable(vc, vcpu);
2669                         wake_up(&vcpu->arch.cpu_run);
2670                 }
2671                 goto out;
2672         }
2673
2674         /*
2675          * See if we could run any other vcores on the physical core
2676          * along with this one.
2677          */
2678         init_core_info(&core_info, vc);
2679         pcpu = smp_processor_id();
2680         target_threads = controlled_threads;
2681         if (target_smt_mode && target_smt_mode < target_threads)
2682                 target_threads = target_smt_mode;
2683         if (vc->num_threads < target_threads)
2684                 collect_piggybacks(&core_info, target_threads);
2685
2686         /*
2687          * On radix, arrange for TLB flushing if necessary.
2688          * This has to be done before disabling interrupts since
2689          * it uses smp_call_function().
2690          */
2691         pcpu = smp_processor_id();
2692         if (kvm_is_radix(vc->kvm)) {
2693                 for (sub = 0; sub < core_info.n_subcores; ++sub)
2694                         for_each_runnable_thread(i, vcpu, core_info.vc[sub])
2695                                 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
2696         }
2697
2698         /*
2699          * Hard-disable interrupts, and check resched flag and signals.
2700          * If we need to reschedule or deliver a signal, clean up
2701          * and return without going into the guest(s).
2702          */
2703         local_irq_disable();
2704         hard_irq_disable();
2705         if (lazy_irq_pending() || need_resched() ||
2706             recheck_signals(&core_info)) {
2707                 local_irq_enable();
2708                 vc->vcore_state = VCORE_INACTIVE;
2709                 /* Unlock all except the primary vcore */
2710                 for (sub = 1; sub < core_info.n_subcores; ++sub) {
2711                         pvc = core_info.vc[sub];
2712                         /* Put back on to the preempted vcores list */
2713                         kvmppc_vcore_preempt(pvc);
2714                         spin_unlock(&pvc->lock);
2715                 }
2716                 for (i = 0; i < controlled_threads; ++i)
2717                         kvmppc_release_hwthread(pcpu + i);
2718                 return;
2719         }
2720
2721         kvmppc_clear_host_core(pcpu);
2722
2723         /* Decide on micro-threading (split-core) mode */
2724         subcore_size = threads_per_subcore;
2725         cmd_bit = stat_bit = 0;
2726         split = core_info.n_subcores;
2727         sip = NULL;
2728         if (split > 1) {
2729                 /* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2730                 if (split == 2 && (dynamic_mt_modes & 2)) {
2731                         cmd_bit = HID0_POWER8_1TO2LPAR;
2732                         stat_bit = HID0_POWER8_2LPARMODE;
2733                 } else {
2734                         split = 4;
2735                         cmd_bit = HID0_POWER8_1TO4LPAR;
2736                         stat_bit = HID0_POWER8_4LPARMODE;
2737                 }
2738                 subcore_size = MAX_SMT_THREADS / split;
2739                 sip = &split_info;
2740                 memset(&split_info, 0, sizeof(split_info));
2741                 split_info.rpr = mfspr(SPRN_RPR);
2742                 split_info.pmmar = mfspr(SPRN_PMMAR);
2743                 split_info.ldbar = mfspr(SPRN_LDBAR);
2744                 split_info.subcore_size = subcore_size;
2745                 for (sub = 0; sub < core_info.n_subcores; ++sub)
2746                         split_info.vc[sub] = core_info.vc[sub];
2747                 /* order writes to split_info before kvm_split_mode pointer */
2748                 smp_wmb();
2749         }
2750         for (thr = 0; thr < controlled_threads; ++thr)
2751                 paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2752
2753         /* Initiate micro-threading (split-core) if required */
2754         if (cmd_bit) {
2755                 unsigned long hid0 = mfspr(SPRN_HID0);
2756
2757                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2758                 mb();
2759                 mtspr(SPRN_HID0, hid0);
2760                 isync();
2761                 for (;;) {
2762                         hid0 = mfspr(SPRN_HID0);
2763                         if (hid0 & stat_bit)
2764                                 break;
2765                         cpu_relax();
2766                 }
2767         }
2768
2769         /* Start all the threads */
2770         active = 0;
2771         for (sub = 0; sub < core_info.n_subcores; ++sub) {
2772                 thr = subcore_thread_map[sub];
2773                 thr0_done = false;
2774                 active |= 1 << thr;
2775                 pvc = core_info.vc[sub];
2776                 pvc->pcpu = pcpu + thr;
2777                 for_each_runnable_thread(i, vcpu, pvc) {
2778                         kvmppc_start_thread(vcpu, pvc);
2779                         kvmppc_create_dtl_entry(vcpu, pvc);
2780                         trace_kvm_guest_enter(vcpu);
2781                         if (!vcpu->arch.ptid)
2782                                 thr0_done = true;
2783                         active |= 1 << (thr + vcpu->arch.ptid);
2784                 }
2785                 /*
2786                  * We need to start the first thread of each subcore
2787                  * even if it doesn't have a vcpu.
2788                  */
2789                 if (!thr0_done)
2790                         kvmppc_start_thread(NULL, pvc);
2791                 thr += pvc->num_threads;
2792         }
2793
2794         /*
2795          * Ensure that split_info.do_nap is set after setting
2796          * the vcore pointer in the PACA of the secondaries.
2797          */
2798         smp_mb();
2799         if (cmd_bit)
2800                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
2801
2802         /*
2803          * When doing micro-threading, poke the inactive threads as well.
2804          * This gets them to the nap instruction after kvm_do_nap,
2805          * which reduces the time taken to unsplit later.
2806          */
2807         if (split > 1)
2808                 for (thr = 1; thr < threads_per_subcore; ++thr)
2809                         if (!(active & (1 << thr)))
2810                                 kvmppc_ipi_thread(pcpu + thr);
2811
2812         vc->vcore_state = VCORE_RUNNING;
2813         preempt_disable();
2814
2815         trace_kvmppc_run_core(vc, 0);
2816
2817         for (sub = 0; sub < core_info.n_subcores; ++sub)
2818                 spin_unlock(&core_info.vc[sub]->lock);
2819
2820         /*
2821          * Interrupts will be enabled once we get into the guest,
2822          * so tell lockdep that we're about to enable interrupts.
2823          */
2824         trace_hardirqs_on();
2825
2826         guest_enter();
2827
2828         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2829
2830         trap = __kvmppc_vcore_entry();
2831
2832         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2833
2834         guest_exit();
2835
2836         trace_hardirqs_off();
2837         set_irq_happened(trap);
2838
2839         spin_lock(&vc->lock);
2840         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2841         vc->vcore_state = VCORE_EXITING;
2842
2843         /* wait for secondary threads to finish writing their state to memory */
2844         kvmppc_wait_for_nap();
2845
2846         /* Return to whole-core mode if we split the core earlier */
2847         if (split > 1) {
2848                 unsigned long hid0 = mfspr(SPRN_HID0);
2849                 unsigned long loops = 0;
2850
2851                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2852                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2853                 mb();
2854                 mtspr(SPRN_HID0, hid0);
2855                 isync();
2856                 for (;;) {
2857                         hid0 = mfspr(SPRN_HID0);
2858                         if (!(hid0 & stat_bit))
2859                                 break;
2860                         cpu_relax();
2861                         ++loops;
2862                 }
2863                 split_info.do_nap = 0;
2864         }
2865
2866         kvmppc_set_host_core(pcpu);
2867
2868         local_irq_enable();
2869
2870         /* Let secondaries go back to the offline loop */
2871         for (i = 0; i < controlled_threads; ++i) {
2872                 kvmppc_release_hwthread(pcpu + i);
2873                 if (sip && sip->napped[i])
2874                         kvmppc_ipi_thread(pcpu + i);
2875                 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
2876         }
2877
2878         spin_unlock(&vc->lock);
2879
2880         /* make sure updates to secondary vcpu structs are visible now */
2881         smp_mb();
2882
2883         for (sub = 0; sub < core_info.n_subcores; ++sub) {
2884                 pvc = core_info.vc[sub];
2885                 post_guest_process(pvc, pvc == vc);
2886         }
2887
2888         spin_lock(&vc->lock);
2889         preempt_enable();
2890
2891  out:
2892         vc->vcore_state = VCORE_INACTIVE;
2893         trace_kvmppc_run_core(vc, 1);
2894 }
2895
2896 /*
2897  * Wait for some other vcpu thread to execute us, and
2898  * wake us up when we need to handle something in the host.
2899  */
2900 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2901                                  struct kvm_vcpu *vcpu, int wait_state)
2902 {
2903         DEFINE_WAIT(wait);
2904
2905         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2906         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2907                 spin_unlock(&vc->lock);
2908                 schedule();
2909                 spin_lock(&vc->lock);
2910         }
2911         finish_wait(&vcpu->arch.cpu_run, &wait);
2912 }
2913
2914 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
2915 {
2916         /* 10us base */
2917         if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
2918                 vc->halt_poll_ns = 10000;
2919         else
2920                 vc->halt_poll_ns *= halt_poll_ns_grow;
2921 }
2922
2923 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
2924 {
2925         if (halt_poll_ns_shrink == 0)
2926                 vc->halt_poll_ns = 0;
2927         else
2928                 vc->halt_poll_ns /= halt_poll_ns_shrink;
2929 }
2930
2931 #ifdef CONFIG_KVM_XICS
2932 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
2933 {
2934         if (!xive_enabled())
2935                 return false;
2936         return vcpu->arch.xive_saved_state.pipr <
2937                 vcpu->arch.xive_saved_state.cppr;
2938 }
2939 #else
2940 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
2941 {
2942         return false;
2943 }
2944 #endif /* CONFIG_KVM_XICS */
2945
2946 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
2947 {
2948         if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
2949             kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
2950                 return true;
2951
2952         return false;
2953 }
2954
2955 /*
2956  * Check to see if any of the runnable vcpus on the vcore have pending
2957  * exceptions or are no longer ceded
2958  */
2959 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
2960 {
2961         struct kvm_vcpu *vcpu;
2962         int i;
2963
2964         for_each_runnable_thread(i, vcpu, vc) {
2965                 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
2966                         return 1;
2967         }
2968
2969         return 0;
2970 }
2971
2972 /*
2973  * All the vcpus in this vcore are idle, so wait for a decrementer
2974  * or external interrupt to one of the vcpus.  vc->lock is held.
2975  */
2976 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2977 {
2978         ktime_t cur, start_poll, start_wait;
2979         int do_sleep = 1;
2980         u64 block_ns;
2981         DECLARE_SWAITQUEUE(wait);
2982
2983         /* Poll for pending exceptions and ceded state */
2984         cur = start_poll = ktime_get();
2985         if (vc->halt_poll_ns) {
2986                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
2987                 ++vc->runner->stat.halt_attempted_poll;
2988
2989                 vc->vcore_state = VCORE_POLLING;
2990                 spin_unlock(&vc->lock);
2991
2992                 do {
2993                         if (kvmppc_vcore_check_block(vc)) {
2994                                 do_sleep = 0;
2995                                 break;
2996                         }
2997                         cur = ktime_get();
2998                 } while (single_task_running() && ktime_before(cur, stop));
2999
3000                 spin_lock(&vc->lock);
3001                 vc->vcore_state = VCORE_INACTIVE;
3002
3003                 if (!do_sleep) {
3004                         ++vc->runner->stat.halt_successful_poll;
3005                         goto out;
3006                 }
3007         }
3008
3009         prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3010
3011         if (kvmppc_vcore_check_block(vc)) {
3012                 finish_swait(&vc->wq, &wait);
3013                 do_sleep = 0;
3014                 /* If we polled, count this as a successful poll */
3015                 if (vc->halt_poll_ns)
3016                         ++vc->runner->stat.halt_successful_poll;
3017                 goto out;
3018         }
3019
3020         start_wait = ktime_get();
3021
3022         vc->vcore_state = VCORE_SLEEPING;
3023         trace_kvmppc_vcore_blocked(vc, 0);
3024         spin_unlock(&vc->lock);
3025         schedule();
3026         finish_swait(&vc->wq, &wait);
3027         spin_lock(&vc->lock);
3028         vc->vcore_state = VCORE_INACTIVE;
3029         trace_kvmppc_vcore_blocked(vc, 1);
3030         ++vc->runner->stat.halt_successful_wait;
3031
3032         cur = ktime_get();
3033
3034 out:
3035         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3036
3037         /* Attribute wait time */
3038         if (do_sleep) {
3039                 vc->runner->stat.halt_wait_ns +=
3040                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
3041                 /* Attribute failed poll time */
3042                 if (vc->halt_poll_ns)
3043                         vc->runner->stat.halt_poll_fail_ns +=
3044                                 ktime_to_ns(start_wait) -
3045                                 ktime_to_ns(start_poll);
3046         } else {
3047                 /* Attribute successful poll time */
3048                 if (vc->halt_poll_ns)
3049                         vc->runner->stat.halt_poll_success_ns +=
3050                                 ktime_to_ns(cur) -
3051                                 ktime_to_ns(start_poll);
3052         }
3053
3054         /* Adjust poll time */
3055         if (halt_poll_ns) {
3056                 if (block_ns <= vc->halt_poll_ns)
3057                         ;
3058                 /* We slept and blocked for longer than the max halt time */
3059                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3060                         shrink_halt_poll_ns(vc);
3061                 /* We slept and our poll time is too small */
3062                 else if (vc->halt_poll_ns < halt_poll_ns &&
3063                                 block_ns < halt_poll_ns)
3064                         grow_halt_poll_ns(vc);
3065                 if (vc->halt_poll_ns > halt_poll_ns)
3066                         vc->halt_poll_ns = halt_poll_ns;
3067         } else
3068                 vc->halt_poll_ns = 0;
3069
3070         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3071 }
3072
3073 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3074 {
3075         int n_ceded, i;
3076         struct kvmppc_vcore *vc;
3077         struct kvm_vcpu *v;
3078
3079         trace_kvmppc_run_vcpu_enter(vcpu);
3080
3081         kvm_run->exit_reason = 0;
3082         vcpu->arch.ret = RESUME_GUEST;
3083         vcpu->arch.trap = 0;
3084         kvmppc_update_vpas(vcpu);
3085
3086         /*
3087          * Synchronize with other threads in this virtual core
3088          */
3089         vc = vcpu->arch.vcore;
3090         spin_lock(&vc->lock);
3091         vcpu->arch.ceded = 0;
3092         vcpu->arch.run_task = current;
3093         vcpu->arch.kvm_run = kvm_run;
3094         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3095         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3096         vcpu->arch.busy_preempt = TB_NIL;
3097         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3098         ++vc->n_runnable;
3099
3100         /*
3101          * This happens the first time this is called for a vcpu.
3102          * If the vcore is already running, we may be able to start
3103          * this thread straight away and have it join in.
3104          */
3105         if (!signal_pending(current)) {
3106                 if (vc->vcore_state == VCORE_PIGGYBACK) {
3107                         if (spin_trylock(&vc->lock)) {
3108                                 if (vc->vcore_state == VCORE_RUNNING &&
3109                                     !VCORE_IS_EXITING(vc)) {
3110                                         kvmppc_create_dtl_entry(vcpu, vc);
3111                                         kvmppc_start_thread(vcpu, vc);
3112                                         trace_kvm_guest_enter(vcpu);
3113                                 }
3114                                 spin_unlock(&vc->lock);
3115                         }
3116                 } else if (vc->vcore_state == VCORE_RUNNING &&
3117                            !VCORE_IS_EXITING(vc)) {
3118                         kvmppc_create_dtl_entry(vcpu, vc);
3119                         kvmppc_start_thread(vcpu, vc);
3120                         trace_kvm_guest_enter(vcpu);
3121                 } else if (vc->vcore_state == VCORE_SLEEPING) {
3122                         swake_up(&vc->wq);
3123                 }
3124
3125         }
3126
3127         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3128                !signal_pending(current)) {
3129                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3130                         kvmppc_vcore_end_preempt(vc);
3131
3132                 if (vc->vcore_state != VCORE_INACTIVE) {
3133                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3134                         continue;
3135                 }
3136                 for_each_runnable_thread(i, v, vc) {
3137                         kvmppc_core_prepare_to_enter(v);
3138                         if (signal_pending(v->arch.run_task)) {
3139                                 kvmppc_remove_runnable(vc, v);
3140                                 v->stat.signal_exits++;
3141                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3142                                 v->arch.ret = -EINTR;
3143                                 wake_up(&v->arch.cpu_run);
3144                         }
3145                 }
3146                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3147                         break;
3148                 n_ceded = 0;
3149                 for_each_runnable_thread(i, v, vc) {
3150                         if (!kvmppc_vcpu_woken(v))
3151                                 n_ceded += v->arch.ceded;
3152                         else
3153                                 v->arch.ceded = 0;
3154                 }
3155                 vc->runner = vcpu;
3156                 if (n_ceded == vc->n_runnable) {
3157                         kvmppc_vcore_blocked(vc);
3158                 } else if (need_resched()) {
3159                         kvmppc_vcore_preempt(vc);
3160                         /* Let something else run */
3161                         cond_resched_lock(&vc->lock);
3162                         if (vc->vcore_state == VCORE_PREEMPT)
3163                                 kvmppc_vcore_end_preempt(vc);
3164                 } else {
3165                         kvmppc_run_core(vc);
3166                 }
3167                 vc->runner = NULL;
3168         }
3169
3170         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3171                (vc->vcore_state == VCORE_RUNNING ||
3172                 vc->vcore_state == VCORE_EXITING ||
3173                 vc->vcore_state == VCORE_PIGGYBACK))
3174                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3175
3176         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3177                 kvmppc_vcore_end_preempt(vc);
3178
3179         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3180                 kvmppc_remove_runnable(vc, vcpu);
3181                 vcpu->stat.signal_exits++;
3182                 kvm_run->exit_reason = KVM_EXIT_INTR;
3183                 vcpu->arch.ret = -EINTR;
3184         }
3185
3186         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3187                 /* Wake up some vcpu to run the core */
3188                 i = -1;
3189                 v = next_runnable_thread(vc, &i);
3190                 wake_up(&v->arch.cpu_run);
3191         }
3192
3193         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3194         spin_unlock(&vc->lock);
3195         return vcpu->arch.ret;
3196 }
3197
3198 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
3199 {
3200         int r;
3201         int srcu_idx;
3202         unsigned long ebb_regs[3] = {}; /* shut up GCC */
3203         unsigned long user_tar = 0;
3204         unsigned int user_vrsave;
3205
3206         if (!vcpu->arch.sane) {
3207                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3208                 return -EINVAL;
3209         }
3210
3211         /*
3212          * Don't allow entry with a suspended transaction, because
3213          * the guest entry/exit code will lose it.
3214          * If the guest has TM enabled, save away their TM-related SPRs
3215          * (they will get restored by the TM unavailable interrupt).
3216          */
3217 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3218         if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
3219             (current->thread.regs->msr & MSR_TM)) {
3220                 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
3221                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3222                         run->fail_entry.hardware_entry_failure_reason = 0;
3223                         return -EINVAL;
3224                 }
3225                 /* Enable TM so we can read the TM SPRs */
3226                 mtmsr(mfmsr() | MSR_TM);
3227                 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
3228                 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
3229                 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
3230                 current->thread.regs->msr &= ~MSR_TM;
3231         }
3232 #endif
3233
3234         kvmppc_core_prepare_to_enter(vcpu);
3235
3236         /* No need to go into the guest when all we'll do is come back out */
3237         if (signal_pending(current)) {
3238                 run->exit_reason = KVM_EXIT_INTR;
3239                 return -EINTR;
3240         }
3241
3242         atomic_inc(&vcpu->kvm->arch.vcpus_running);
3243         /* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
3244         smp_mb();
3245
3246         /* On the first time here, set up HTAB and VRMA */
3247         if (!kvm_is_radix(vcpu->kvm) && !vcpu->kvm->arch.hpte_setup_done) {
3248                 r = kvmppc_hv_setup_htab_rma(vcpu);
3249                 if (r)
3250                         goto out;
3251         }
3252
3253         flush_all_to_thread(current);
3254
3255         /* Save userspace EBB and other register values */
3256         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3257                 ebb_regs[0] = mfspr(SPRN_EBBHR);
3258                 ebb_regs[1] = mfspr(SPRN_EBBRR);
3259                 ebb_regs[2] = mfspr(SPRN_BESCR);
3260                 user_tar = mfspr(SPRN_TAR);
3261         }
3262         user_vrsave = mfspr(SPRN_VRSAVE);
3263
3264         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
3265         vcpu->arch.pgdir = current->mm->pgd;
3266         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3267
3268         do {
3269                 r = kvmppc_run_vcpu(run, vcpu);
3270
3271                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
3272                     !(vcpu->arch.shregs.msr & MSR_PR)) {
3273                         trace_kvm_hcall_enter(vcpu);
3274                         r = kvmppc_pseries_do_hcall(vcpu);
3275                         trace_kvm_hcall_exit(vcpu, r);
3276                         kvmppc_core_prepare_to_enter(vcpu);
3277                 } else if (r == RESUME_PAGE_FAULT) {
3278                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3279                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
3280                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
3281                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
3282                 } else if (r == RESUME_PASSTHROUGH) {
3283                         if (WARN_ON(xive_enabled()))
3284                                 r = H_SUCCESS;
3285                         else
3286                                 r = kvmppc_xics_rm_complete(vcpu, 0);
3287                 }
3288         } while (is_kvmppc_resume_guest(r));
3289
3290         /* Restore userspace EBB and other register values */
3291         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3292                 mtspr(SPRN_EBBHR, ebb_regs[0]);
3293                 mtspr(SPRN_EBBRR, ebb_regs[1]);
3294                 mtspr(SPRN_BESCR, ebb_regs[2]);
3295                 mtspr(SPRN_TAR, user_tar);
3296                 mtspr(SPRN_FSCR, current->thread.fscr);
3297         }
3298         mtspr(SPRN_VRSAVE, user_vrsave);
3299
3300  out:
3301         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
3302         atomic_dec(&vcpu->kvm->arch.vcpus_running);
3303         return r;
3304 }
3305
3306 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
3307                                      int linux_psize)
3308 {
3309         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
3310
3311         if (!def->shift)
3312                 return;
3313         (*sps)->page_shift = def->shift;
3314         (*sps)->slb_enc = def->sllp;
3315         (*sps)->enc[0].page_shift = def->shift;
3316         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
3317         /*
3318          * Add 16MB MPSS support if host supports it
3319          */
3320         if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
3321                 (*sps)->enc[1].page_shift = 24;
3322                 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
3323         }
3324         (*sps)++;
3325 }
3326
3327 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
3328                                          struct kvm_ppc_smmu_info *info)
3329 {
3330         struct kvm_ppc_one_seg_page_size *sps;
3331
3332         /*
3333          * Since we don't yet support HPT guests on a radix host,
3334          * return an error if the host uses radix.
3335          */
3336         if (radix_enabled())
3337                 return -EINVAL;
3338
3339         info->flags = KVM_PPC_PAGE_SIZES_REAL;
3340         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
3341                 info->flags |= KVM_PPC_1T_SEGMENTS;
3342         info->slb_size = mmu_slb_size;
3343
3344         /* We only support these sizes for now, and no muti-size segments */
3345         sps = &info->sps[0];
3346         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
3347         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
3348         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
3349
3350         return 0;
3351 }
3352
3353 /*
3354  * Get (and clear) the dirty memory log for a memory slot.
3355  */
3356 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
3357                                          struct kvm_dirty_log *log)
3358 {
3359         struct kvm_memslots *slots;
3360         struct kvm_memory_slot *memslot;
3361         int i, r;
3362         unsigned long n;
3363         unsigned long *buf;
3364         struct kvm_vcpu *vcpu;
3365
3366         mutex_lock(&kvm->slots_lock);
3367
3368         r = -EINVAL;
3369         if (log->slot >= KVM_USER_MEM_SLOTS)
3370                 goto out;
3371
3372         slots = kvm_memslots(kvm);
3373         memslot = id_to_memslot(slots, log->slot);
3374         r = -ENOENT;
3375         if (!memslot->dirty_bitmap)
3376                 goto out;
3377
3378         /*
3379          * Use second half of bitmap area because radix accumulates
3380          * bits in the first half.
3381          */
3382         n = kvm_dirty_bitmap_bytes(memslot);
3383         buf = memslot->dirty_bitmap + n / sizeof(long);
3384         memset(buf, 0, n);
3385
3386         if (kvm_is_radix(kvm))
3387                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
3388         else
3389                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
3390         if (r)
3391                 goto out;
3392
3393         /* Harvest dirty bits from VPA and DTL updates */
3394         /* Note: we never modify the SLB shadow buffer areas */
3395         kvm_for_each_vcpu(i, vcpu, kvm) {
3396                 spin_lock(&vcpu->arch.vpa_update_lock);
3397                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
3398                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
3399                 spin_unlock(&vcpu->arch.vpa_update_lock);
3400         }
3401
3402         r = -EFAULT;
3403         if (copy_to_user(log->dirty_bitmap, buf, n))
3404                 goto out;
3405
3406         r = 0;
3407 out:
3408         mutex_unlock(&kvm->slots_lock);
3409         return r;
3410 }
3411
3412 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
3413                                         struct kvm_memory_slot *dont)
3414 {
3415         if (!dont || free->arch.rmap != dont->arch.rmap) {
3416                 vfree(free->arch.rmap);
3417                 free->arch.rmap = NULL;
3418         }
3419 }
3420
3421 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3422                                          unsigned long npages)
3423 {
3424         /*
3425          * For now, if radix_enabled() then we only support radix guests,
3426          * and in that case we don't need the rmap array.
3427          */
3428         if (radix_enabled()) {
3429                 slot->arch.rmap = NULL;
3430                 return 0;
3431         }
3432
3433         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
3434         if (!slot->arch.rmap)
3435                 return -ENOMEM;
3436
3437         return 0;
3438 }
3439
3440 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3441                                         struct kvm_memory_slot *memslot,
3442                                         const struct kvm_userspace_memory_region *mem)
3443 {
3444         return 0;
3445 }
3446
3447 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3448                                 const struct kvm_userspace_memory_region *mem,
3449                                 const struct kvm_memory_slot *old,
3450                                 const struct kvm_memory_slot *new)
3451 {
3452         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3453         struct kvm_memslots *slots;
3454         struct kvm_memory_slot *memslot;
3455
3456         /*
3457          * If we are making a new memslot, it might make
3458          * some address that was previously cached as emulated
3459          * MMIO be no longer emulated MMIO, so invalidate
3460          * all the caches of emulated MMIO translations.
3461          */
3462         if (npages)
3463                 atomic64_inc(&kvm->arch.mmio_update);
3464
3465         if (npages && old->npages && !kvm_is_radix(kvm)) {
3466                 /*
3467                  * If modifying a memslot, reset all the rmap dirty bits.
3468                  * If this is a new memslot, we don't need to do anything
3469                  * since the rmap array starts out as all zeroes,
3470                  * i.e. no pages are dirty.
3471                  */
3472                 slots = kvm_memslots(kvm);
3473                 memslot = id_to_memslot(slots, mem->slot);
3474                 kvmppc_hv_get_dirty_log_hpt(kvm, memslot, NULL);
3475         }
3476 }
3477
3478 /*
3479  * Update LPCR values in kvm->arch and in vcores.
3480  * Caller must hold kvm->lock.
3481  */
3482 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3483 {
3484         long int i;
3485         u32 cores_done = 0;
3486
3487         if ((kvm->arch.lpcr & mask) == lpcr)
3488                 return;
3489
3490         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3491
3492         for (i = 0; i < KVM_MAX_VCORES; ++i) {
3493                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3494                 if (!vc)
3495                         continue;
3496                 spin_lock(&vc->lock);
3497                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3498                 spin_unlock(&vc->lock);
3499                 if (++cores_done >= kvm->arch.online_vcores)
3500                         break;
3501         }
3502 }
3503
3504 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3505 {
3506         return;
3507 }
3508
3509 static void kvmppc_setup_partition_table(struct kvm *kvm)
3510 {
3511         unsigned long dw0, dw1;
3512
3513         if (!kvm_is_radix(kvm)) {
3514                 /* PS field - page size for VRMA */
3515                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
3516                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
3517                 /* HTABSIZE and HTABORG fields */
3518                 dw0 |= kvm->arch.sdr1;
3519
3520                 /* Second dword as set by userspace */
3521                 dw1 = kvm->arch.process_table;
3522         } else {
3523                 dw0 = PATB_HR | radix__get_tree_size() |
3524                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
3525                 dw1 = PATB_GR | kvm->arch.process_table;
3526         }
3527
3528         mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
3529 }
3530
3531 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3532 {
3533         int err = 0;
3534         struct kvm *kvm = vcpu->kvm;
3535         unsigned long hva;
3536         struct kvm_memory_slot *memslot;
3537         struct vm_area_struct *vma;
3538         unsigned long lpcr = 0, senc;
3539         unsigned long psize, porder;
3540         int srcu_idx;
3541
3542         mutex_lock(&kvm->lock);
3543         if (kvm->arch.hpte_setup_done)
3544                 goto out;       /* another vcpu beat us to it */
3545
3546         /* Allocate hashed page table (if not done already) and reset it */
3547         if (!kvm->arch.hpt.virt) {
3548                 int order = KVM_DEFAULT_HPT_ORDER;
3549                 struct kvm_hpt_info info;
3550
3551                 err = kvmppc_allocate_hpt(&info, order);
3552                 /* If we get here, it means userspace didn't specify a
3553                  * size explicitly.  So, try successively smaller
3554                  * sizes if the default failed. */
3555                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
3556                         err  = kvmppc_allocate_hpt(&info, order);
3557
3558                 if (err < 0) {
3559                         pr_err("KVM: Couldn't alloc HPT\n");
3560                         goto out;
3561                 }
3562
3563                 kvmppc_set_hpt(kvm, &info);
3564         }
3565
3566         /* Look up the memslot for guest physical address 0 */
3567         srcu_idx = srcu_read_lock(&kvm->srcu);
3568         memslot = gfn_to_memslot(kvm, 0);
3569
3570         /* We must have some memory at 0 by now */
3571         err = -EINVAL;
3572         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3573                 goto out_srcu;
3574
3575         /* Look up the VMA for the start of this memory slot */
3576         hva = memslot->userspace_addr;
3577         down_read(&current->mm->mmap_sem);
3578         vma = find_vma(current->mm, hva);
3579         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3580                 goto up_out;
3581
3582         psize = vma_kernel_pagesize(vma);
3583         porder = __ilog2(psize);
3584
3585         up_read(&current->mm->mmap_sem);
3586
3587         /* We can handle 4k, 64k or 16M pages in the VRMA */
3588         err = -EINVAL;
3589         if (!(psize == 0x1000 || psize == 0x10000 ||
3590               psize == 0x1000000))
3591                 goto out_srcu;
3592
3593         senc = slb_pgsize_encoding(psize);
3594         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3595                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3596         /* Create HPTEs in the hash page table for the VRMA */
3597         kvmppc_map_vrma(vcpu, memslot, porder);
3598
3599         /* Update VRMASD field in the LPCR */
3600         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3601                 /* the -4 is to account for senc values starting at 0x10 */
3602                 lpcr = senc << (LPCR_VRMASD_SH - 4);
3603                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3604         } else {
3605                 kvmppc_setup_partition_table(kvm);
3606         }
3607
3608         /* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
3609         smp_wmb();
3610         kvm->arch.hpte_setup_done = 1;
3611         err = 0;
3612  out_srcu:
3613         srcu_read_unlock(&kvm->srcu, srcu_idx);
3614  out:
3615         mutex_unlock(&kvm->lock);
3616         return err;
3617
3618  up_out:
3619         up_read(&current->mm->mmap_sem);
3620         goto out_srcu;
3621 }
3622
3623 #ifdef CONFIG_KVM_XICS
3624 /*
3625  * Allocate a per-core structure for managing state about which cores are
3626  * running in the host versus the guest and for exchanging data between
3627  * real mode KVM and CPU running in the host.
3628  * This is only done for the first VM.
3629  * The allocated structure stays even if all VMs have stopped.
3630  * It is only freed when the kvm-hv module is unloaded.
3631  * It's OK for this routine to fail, we just don't support host
3632  * core operations like redirecting H_IPI wakeups.
3633  */
3634 void kvmppc_alloc_host_rm_ops(void)
3635 {
3636         struct kvmppc_host_rm_ops *ops;
3637         unsigned long l_ops;
3638         int cpu, core;
3639         int size;
3640
3641         /* Not the first time here ? */
3642         if (kvmppc_host_rm_ops_hv != NULL)
3643                 return;
3644
3645         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3646         if (!ops)
3647                 return;
3648
3649         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3650         ops->rm_core = kzalloc(size, GFP_KERNEL);
3651
3652         if (!ops->rm_core) {
3653                 kfree(ops);
3654                 return;
3655         }
3656
3657         cpus_read_lock();
3658
3659         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3660                 if (!cpu_online(cpu))
3661                         continue;
3662
3663                 core = cpu >> threads_shift;
3664                 ops->rm_core[core].rm_state.in_host = 1;
3665         }
3666
3667         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3668
3669         /*
3670          * Make the contents of the kvmppc_host_rm_ops structure visible
3671          * to other CPUs before we assign it to the global variable.
3672          * Do an atomic assignment (no locks used here), but if someone
3673          * beats us to it, just free our copy and return.
3674          */
3675         smp_wmb();
3676         l_ops = (unsigned long) ops;
3677
3678         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3679                 cpus_read_unlock();
3680                 kfree(ops->rm_core);
3681                 kfree(ops);
3682                 return;
3683         }
3684
3685         cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
3686                                              "ppc/kvm_book3s:prepare",
3687                                              kvmppc_set_host_core,
3688                                              kvmppc_clear_host_core);
3689         cpus_read_unlock();
3690 }
3691
3692 void kvmppc_free_host_rm_ops(void)
3693 {
3694         if (kvmppc_host_rm_ops_hv) {
3695                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
3696                 kfree(kvmppc_host_rm_ops_hv->rm_core);
3697                 kfree(kvmppc_host_rm_ops_hv);
3698                 kvmppc_host_rm_ops_hv = NULL;
3699         }
3700 }
3701 #endif
3702
3703 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3704 {
3705         unsigned long lpcr, lpid;
3706         char buf[32];
3707         int ret;
3708
3709         /* Allocate the guest's logical partition ID */
3710
3711         lpid = kvmppc_alloc_lpid();
3712         if ((long)lpid < 0)
3713                 return -ENOMEM;
3714         kvm->arch.lpid = lpid;
3715
3716         kvmppc_alloc_host_rm_ops();
3717
3718         /*
3719          * Since we don't flush the TLB when tearing down a VM,
3720          * and this lpid might have previously been used,
3721          * make sure we flush on each core before running the new VM.
3722          * On POWER9, the tlbie in mmu_partition_table_set_entry()
3723          * does this flush for us.
3724          */
3725         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3726                 cpumask_setall(&kvm->arch.need_tlb_flush);
3727
3728         /* Start out with the default set of hcalls enabled */
3729         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3730                sizeof(kvm->arch.enabled_hcalls));
3731
3732         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3733                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3734
3735         /* Init LPCR for virtual RMA mode */
3736         kvm->arch.host_lpid = mfspr(SPRN_LPID);
3737         kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3738         lpcr &= LPCR_PECE | LPCR_LPES;
3739         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3740                 LPCR_VPM0 | LPCR_VPM1;
3741         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3742                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3743         /* On POWER8 turn on online bit to enable PURR/SPURR */
3744         if (cpu_has_feature(CPU_FTR_ARCH_207S))
3745                 lpcr |= LPCR_ONL;
3746         /*
3747          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
3748          * Set HVICE bit to enable hypervisor virtualization interrupts.
3749          * Set HEIC to prevent OS interrupts to go to hypervisor (should
3750          * be unnecessary but better safe than sorry in case we re-enable
3751          * EE in HV mode with this LPCR still set)
3752          */
3753         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3754                 lpcr &= ~LPCR_VPM0;
3755                 lpcr |= LPCR_HVICE | LPCR_HEIC;
3756
3757                 /*
3758                  * If xive is enabled, we route 0x500 interrupts directly
3759                  * to the guest.
3760                  */
3761                 if (xive_enabled())
3762                         lpcr |= LPCR_LPES;
3763         }
3764
3765         /*
3766          * For now, if the host uses radix, the guest must be radix.
3767          */
3768         if (radix_enabled()) {
3769                 kvm->arch.radix = 1;
3770                 lpcr &= ~LPCR_VPM1;
3771                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
3772                 ret = kvmppc_init_vm_radix(kvm);
3773                 if (ret) {
3774                         kvmppc_free_lpid(kvm->arch.lpid);
3775                         return ret;
3776                 }
3777                 kvmppc_setup_partition_table(kvm);
3778         }
3779
3780         kvm->arch.lpcr = lpcr;
3781
3782         /* Initialization for future HPT resizes */
3783         kvm->arch.resize_hpt = NULL;
3784
3785         /*
3786          * Work out how many sets the TLB has, for the use of
3787          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
3788          */
3789         if (kvm_is_radix(kvm))
3790                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
3791         else if (cpu_has_feature(CPU_FTR_ARCH_300))
3792                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
3793         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
3794                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
3795         else
3796                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
3797
3798         /*
3799          * Track that we now have a HV mode VM active. This blocks secondary
3800          * CPU threads from coming online.
3801          * On POWER9, we only need to do this for HPT guests on a radix
3802          * host, which is not yet supported.
3803          */
3804         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3805                 kvm_hv_vm_activated();
3806
3807         /*
3808          * Initialize smt_mode depending on processor.
3809          * POWER8 and earlier have to use "strict" threading, where
3810          * all vCPUs in a vcore have to run on the same (sub)core,
3811          * whereas on POWER9 the threads can each run a different
3812          * guest.
3813          */
3814         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3815                 kvm->arch.smt_mode = threads_per_subcore;
3816         else
3817                 kvm->arch.smt_mode = 1;
3818         kvm->arch.emul_smt_mode = 1;
3819
3820         /*
3821          * Create a debugfs directory for the VM
3822          */
3823         snprintf(buf, sizeof(buf), "vm%d", current->pid);
3824         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3825         if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3826                 kvmppc_mmu_debugfs_init(kvm);
3827
3828         return 0;
3829 }
3830
3831 static void kvmppc_free_vcores(struct kvm *kvm)
3832 {
3833         long int i;
3834
3835         for (i = 0; i < KVM_MAX_VCORES; ++i)
3836                 kfree(kvm->arch.vcores[i]);
3837         kvm->arch.online_vcores = 0;
3838 }
3839
3840 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3841 {
3842         debugfs_remove_recursive(kvm->arch.debugfs_dir);
3843
3844         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3845                 kvm_hv_vm_deactivated();
3846
3847         kvmppc_free_vcores(kvm);
3848
3849         kvmppc_free_lpid(kvm->arch.lpid);
3850
3851         if (kvm_is_radix(kvm))
3852                 kvmppc_free_radix(kvm);
3853         else
3854                 kvmppc_free_hpt(&kvm->arch.hpt);
3855
3856         kvmppc_free_pimap(kvm);
3857 }
3858
3859 /* We don't need to emulate any privileged instructions or dcbz */
3860 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3861                                      unsigned int inst, int *advance)
3862 {
3863         return EMULATE_FAIL;
3864 }
3865
3866 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3867                                         ulong spr_val)
3868 {
3869         return EMULATE_FAIL;
3870 }
3871
3872 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3873                                         ulong *spr_val)
3874 {
3875         return EMULATE_FAIL;
3876 }
3877
3878 static int kvmppc_core_check_processor_compat_hv(void)
3879 {
3880         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3881             !cpu_has_feature(CPU_FTR_ARCH_206))
3882                 return -EIO;
3883
3884         return 0;
3885 }
3886
3887 #ifdef CONFIG_KVM_XICS
3888
3889 void kvmppc_free_pimap(struct kvm *kvm)
3890 {
3891         kfree(kvm->arch.pimap);
3892 }
3893
3894 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
3895 {
3896         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
3897 }
3898
3899 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3900 {
3901         struct irq_desc *desc;
3902         struct kvmppc_irq_map *irq_map;
3903         struct kvmppc_passthru_irqmap *pimap;
3904         struct irq_chip *chip;
3905         int i, rc = 0;
3906
3907         if (!kvm_irq_bypass)
3908                 return 1;
3909
3910         desc = irq_to_desc(host_irq);
3911         if (!desc)
3912                 return -EIO;
3913
3914         mutex_lock(&kvm->lock);
3915
3916         pimap = kvm->arch.pimap;
3917         if (pimap == NULL) {
3918                 /* First call, allocate structure to hold IRQ map */
3919                 pimap = kvmppc_alloc_pimap();
3920                 if (pimap == NULL) {
3921                         mutex_unlock(&kvm->lock);
3922                         return -ENOMEM;
3923                 }
3924                 kvm->arch.pimap = pimap;
3925         }
3926
3927         /*
3928          * For now, we only support interrupts for which the EOI operation
3929          * is an OPAL call followed by a write to XIRR, since that's
3930          * what our real-mode EOI code does, or a XIVE interrupt
3931          */
3932         chip = irq_data_get_irq_chip(&desc->irq_data);
3933         if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
3934                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
3935                         host_irq, guest_gsi);
3936                 mutex_unlock(&kvm->lock);
3937                 return -ENOENT;
3938         }
3939
3940         /*
3941          * See if we already have an entry for this guest IRQ number.
3942          * If it's mapped to a hardware IRQ number, that's an error,
3943          * otherwise re-use this entry.
3944          */
3945         for (i = 0; i < pimap->n_mapped; i++) {
3946                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
3947                         if (pimap->mapped[i].r_hwirq) {
3948                                 mutex_unlock(&kvm->lock);
3949                                 return -EINVAL;
3950                         }
3951                         break;
3952                 }
3953         }
3954
3955         if (i == KVMPPC_PIRQ_MAPPED) {
3956                 mutex_unlock(&kvm->lock);
3957                 return -EAGAIN;         /* table is full */
3958         }
3959
3960         irq_map = &pimap->mapped[i];
3961
3962         irq_map->v_hwirq = guest_gsi;
3963         irq_map->desc = desc;
3964
3965         /*
3966          * Order the above two stores before the next to serialize with
3967          * the KVM real mode handler.
3968          */
3969         smp_wmb();
3970         irq_map->r_hwirq = desc->irq_data.hwirq;
3971
3972         if (i == pimap->n_mapped)
3973                 pimap->n_mapped++;
3974
3975         if (xive_enabled())
3976                 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
3977         else
3978                 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
3979         if (rc)
3980                 irq_map->r_hwirq = 0;
3981
3982         mutex_unlock(&kvm->lock);
3983
3984         return 0;
3985 }
3986
3987 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3988 {
3989         struct irq_desc *desc;
3990         struct kvmppc_passthru_irqmap *pimap;
3991         int i, rc = 0;
3992
3993         if (!kvm_irq_bypass)
3994                 return 0;
3995
3996         desc = irq_to_desc(host_irq);
3997         if (!desc)
3998                 return -EIO;
3999
4000         mutex_lock(&kvm->lock);
4001         if (!kvm->arch.pimap)
4002                 goto unlock;
4003
4004         pimap = kvm->arch.pimap;
4005
4006         for (i = 0; i < pimap->n_mapped; i++) {
4007                 if (guest_gsi == pimap->mapped[i].v_hwirq)
4008                         break;
4009         }
4010
4011         if (i == pimap->n_mapped) {
4012                 mutex_unlock(&kvm->lock);
4013                 return -ENODEV;
4014         }
4015
4016         if (xive_enabled())
4017                 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
4018         else
4019                 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
4020
4021         /* invalidate the entry (what do do on error from the above ?) */
4022         pimap->mapped[i].r_hwirq = 0;
4023
4024         /*
4025          * We don't free this structure even when the count goes to
4026          * zero. The structure is freed when we destroy the VM.
4027          */
4028  unlock:
4029         mutex_unlock(&kvm->lock);
4030         return rc;
4031 }
4032
4033 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
4034                                              struct irq_bypass_producer *prod)
4035 {
4036         int ret = 0;
4037         struct kvm_kernel_irqfd *irqfd =
4038                 container_of(cons, struct kvm_kernel_irqfd, consumer);
4039
4040         irqfd->producer = prod;
4041
4042         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4043         if (ret)
4044                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
4045                         prod->irq, irqfd->gsi, ret);
4046
4047         return ret;
4048 }
4049
4050 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
4051                                               struct irq_bypass_producer *prod)
4052 {
4053         int ret;
4054         struct kvm_kernel_irqfd *irqfd =
4055                 container_of(cons, struct kvm_kernel_irqfd, consumer);
4056
4057         irqfd->producer = NULL;
4058
4059         /*
4060          * When producer of consumer is unregistered, we change back to
4061          * default external interrupt handling mode - KVM real mode
4062          * will switch back to host.
4063          */
4064         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4065         if (ret)
4066                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
4067                         prod->irq, irqfd->gsi, ret);
4068 }
4069 #endif
4070
4071 static long kvm_arch_vm_ioctl_hv(struct file *filp,
4072                                  unsigned int ioctl, unsigned long arg)
4073 {
4074         struct kvm *kvm __maybe_unused = filp->private_data;
4075         void __user *argp = (void __user *)arg;
4076         long r;
4077
4078         switch (ioctl) {
4079
4080         case KVM_PPC_ALLOCATE_HTAB: {
4081                 u32 htab_order;
4082
4083                 r = -EFAULT;
4084                 if (get_user(htab_order, (u32 __user *)argp))
4085                         break;
4086                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
4087                 if (r)
4088                         break;
4089                 r = 0;
4090                 break;
4091         }
4092
4093         case KVM_PPC_GET_HTAB_FD: {
4094                 struct kvm_get_htab_fd ghf;
4095
4096                 r = -EFAULT;
4097                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
4098                         break;
4099                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
4100                 break;
4101         }
4102
4103         case KVM_PPC_RESIZE_HPT_PREPARE: {
4104                 struct kvm_ppc_resize_hpt rhpt;
4105
4106                 r = -EFAULT;
4107                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4108                         break;
4109
4110                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
4111                 break;
4112         }
4113
4114         case KVM_PPC_RESIZE_HPT_COMMIT: {
4115                 struct kvm_ppc_resize_hpt rhpt;
4116
4117                 r = -EFAULT;
4118                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4119                         break;
4120
4121                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
4122                 break;
4123         }
4124
4125         default:
4126                 r = -ENOTTY;
4127         }
4128
4129         return r;
4130 }
4131
4132 /*
4133  * List of hcall numbers to enable by default.
4134  * For compatibility with old userspace, we enable by default
4135  * all hcalls that were implemented before the hcall-enabling
4136  * facility was added.  Note this list should not include H_RTAS.
4137  */
4138 static unsigned int default_hcall_list[] = {
4139         H_REMOVE,
4140         H_ENTER,
4141         H_READ,
4142         H_PROTECT,
4143         H_BULK_REMOVE,
4144         H_GET_TCE,
4145         H_PUT_TCE,
4146         H_SET_DABR,
4147         H_SET_XDABR,
4148         H_CEDE,
4149         H_PROD,
4150         H_CONFER,
4151         H_REGISTER_VPA,
4152 #ifdef CONFIG_KVM_XICS
4153         H_EOI,
4154         H_CPPR,
4155         H_IPI,
4156         H_IPOLL,
4157         H_XIRR,
4158         H_XIRR_X,
4159 #endif
4160         0
4161 };
4162
4163 static void init_default_hcalls(void)
4164 {
4165         int i;
4166         unsigned int hcall;
4167
4168         for (i = 0; default_hcall_list[i]; ++i) {
4169                 hcall = default_hcall_list[i];
4170                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
4171                 __set_bit(hcall / 4, default_enabled_hcalls);
4172         }
4173 }
4174
4175 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
4176 {
4177         unsigned long lpcr;
4178         int radix;
4179
4180         /* If not on a POWER9, reject it */
4181         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4182                 return -ENODEV;
4183
4184         /* If any unknown flags set, reject it */
4185         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
4186                 return -EINVAL;
4187
4188         /* We can't change a guest to/from radix yet */
4189         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
4190         if (radix != kvm_is_radix(kvm))
4191                 return -EINVAL;
4192
4193         /* GR (guest radix) bit in process_table field must match */
4194         if (!!(cfg->process_table & PATB_GR) != radix)
4195                 return -EINVAL;
4196
4197         /* Process table size field must be reasonable, i.e. <= 24 */
4198         if ((cfg->process_table & PRTS_MASK) > 24)
4199                 return -EINVAL;
4200
4201         kvm->arch.process_table = cfg->process_table;
4202         kvmppc_setup_partition_table(kvm);
4203
4204         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
4205         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
4206
4207         return 0;
4208 }
4209
4210 static struct kvmppc_ops kvm_ops_hv = {
4211         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
4212         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
4213         .get_one_reg = kvmppc_get_one_reg_hv,
4214         .set_one_reg = kvmppc_set_one_reg_hv,
4215         .vcpu_load   = kvmppc_core_vcpu_load_hv,
4216         .vcpu_put    = kvmppc_core_vcpu_put_hv,
4217         .set_msr     = kvmppc_set_msr_hv,
4218         .vcpu_run    = kvmppc_vcpu_run_hv,
4219         .vcpu_create = kvmppc_core_vcpu_create_hv,
4220         .vcpu_free   = kvmppc_core_vcpu_free_hv,
4221         .check_requests = kvmppc_core_check_requests_hv,
4222         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
4223         .flush_memslot  = kvmppc_core_flush_memslot_hv,
4224         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
4225         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
4226         .unmap_hva = kvm_unmap_hva_hv,
4227         .unmap_hva_range = kvm_unmap_hva_range_hv,
4228         .age_hva  = kvm_age_hva_hv,
4229         .test_age_hva = kvm_test_age_hva_hv,
4230         .set_spte_hva = kvm_set_spte_hva_hv,
4231         .mmu_destroy  = kvmppc_mmu_destroy_hv,
4232         .free_memslot = kvmppc_core_free_memslot_hv,
4233         .create_memslot = kvmppc_core_create_memslot_hv,
4234         .init_vm =  kvmppc_core_init_vm_hv,
4235         .destroy_vm = kvmppc_core_destroy_vm_hv,
4236         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
4237         .emulate_op = kvmppc_core_emulate_op_hv,
4238         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
4239         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
4240         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
4241         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
4242         .hcall_implemented = kvmppc_hcall_impl_hv,
4243 #ifdef CONFIG_KVM_XICS
4244         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
4245         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
4246 #endif
4247         .configure_mmu = kvmhv_configure_mmu,
4248         .get_rmmu_info = kvmhv_get_rmmu_info,
4249         .set_smt_mode = kvmhv_set_smt_mode,
4250 };
4251
4252 static int kvm_init_subcore_bitmap(void)
4253 {
4254         int i, j;
4255         int nr_cores = cpu_nr_cores();
4256         struct sibling_subcore_state *sibling_subcore_state;
4257
4258         for (i = 0; i < nr_cores; i++) {
4259                 int first_cpu = i * threads_per_core;
4260                 int node = cpu_to_node(first_cpu);
4261
4262                 /* Ignore if it is already allocated. */
4263                 if (paca[first_cpu].sibling_subcore_state)
4264                         continue;
4265
4266                 sibling_subcore_state =
4267                         kmalloc_node(sizeof(struct sibling_subcore_state),
4268                                                         GFP_KERNEL, node);
4269                 if (!sibling_subcore_state)
4270                         return -ENOMEM;
4271
4272                 memset(sibling_subcore_state, 0,
4273                                 sizeof(struct sibling_subcore_state));
4274
4275                 for (j = 0; j < threads_per_core; j++) {
4276                         int cpu = first_cpu + j;
4277
4278                         paca[cpu].sibling_subcore_state = sibling_subcore_state;
4279                 }
4280         }
4281         return 0;
4282 }
4283
4284 static int kvmppc_radix_possible(void)
4285 {
4286         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
4287 }
4288
4289 static int kvmppc_book3s_init_hv(void)
4290 {
4291         int r;
4292         /*
4293          * FIXME!! Do we need to check on all cpus ?
4294          */
4295         r = kvmppc_core_check_processor_compat_hv();
4296         if (r < 0)
4297                 return -ENODEV;
4298
4299         r = kvm_init_subcore_bitmap();
4300         if (r)
4301                 return r;
4302
4303         /*
4304          * We need a way of accessing the XICS interrupt controller,
4305          * either directly, via paca[cpu].kvm_hstate.xics_phys, or
4306          * indirectly, via OPAL.
4307          */
4308 #ifdef CONFIG_SMP
4309         if (!xive_enabled() && !local_paca->kvm_hstate.xics_phys) {
4310                 struct device_node *np;
4311
4312                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
4313                 if (!np) {
4314                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
4315                         return -ENODEV;
4316                 }
4317         }
4318 #endif
4319
4320         kvm_ops_hv.owner = THIS_MODULE;
4321         kvmppc_hv_ops = &kvm_ops_hv;
4322
4323         init_default_hcalls();
4324
4325         init_vcore_lists();
4326
4327         r = kvmppc_mmu_hv_init();
4328         if (r)
4329                 return r;
4330
4331         if (kvmppc_radix_possible())
4332                 r = kvmppc_radix_init();
4333         return r;
4334 }
4335
4336 static void kvmppc_book3s_exit_hv(void)
4337 {
4338         kvmppc_free_host_rm_ops();
4339         if (kvmppc_radix_possible())
4340                 kvmppc_radix_exit();
4341         kvmppc_hv_ops = NULL;
4342 }
4343
4344 module_init(kvmppc_book3s_init_hv);
4345 module_exit(kvmppc_book3s_exit_hv);
4346 MODULE_LICENSE("GPL");
4347 MODULE_ALIAS_MISCDEV(KVM_MINOR);
4348 MODULE_ALIAS("devname:kvm");
4349